Interpreting Convolutional Neural Networks#
Learning Goals#
In this tutorial, you will practice analyzing and interpreting a convolutional neural network.
This tutorial assumes a basic knowledge of convolutional neural networks. We will utilize the model described in Classifying_JWST-HST_galaxy_mergers_with_CNNs
, so it is recommended to complete that notebook before reading this one.
Introduction#
Machine learning techniques can be powerful tools for categorizing data and performing data analysis questions. However, machine learning techniques often involve a lot of hidden computation that is not immediately meaningful. The black-box nature of intermediary processes, especially in layered neural networks, can make it difficult to interpret and understand. The goal of this notebook is to familiarize you with some of the various techniques used to make sense of machine learning and convolutional neural networks (CNNs) in particular. CNNs in particular can be very difficult to interpret due to their multi-layered structure and convolutional layers. In this notebook, we will examine two methods of visualizing CNN results (Backpropagation and Grad-CAM) and another method for testing model architecture.
Load the data
Split the data into training, validation, and testing sets
Build and train a model
Apply some interpretation technique to understand your results from a physical perspective.
Dependencies#
This notebook uses the following packages:
numpy
to handle array functionsastropy
for downloading and accessing FITS filesmatplotlib.pyplot
for plotting datakeras
andtensorflow
for building the CNNsklearn
for some utility functions
If you do not have these packages installed, you can install them using pip
or conda
.
Further information about the original model can be found at the Hello Universe codebase.
Author:
Oliver Lin, oliverlin2004@gmail.com
Additional Contributors:
Daisuke Nagai, daisuke.nagai@yale.edu.
Michelle Ntampaka, mntampaka@stsci.edu.
Published: 2024-05-08
# arrays
import numpy as np
# fits
from astropy.io import fits
from astropy.utils.data import download_file
from astropy.visualization import simple_norm
# plotting
from matplotlib import pyplot as plt
# keras
from keras.models import Model
from keras.layers import Input, Flatten, Dense, Dropout, BatchNormalization, Convolution2D, MaxPooling2D
# from keras.layers.convolutional import Convolution2D, MaxPooling2D
from keras.regularizers import l2
# sklearn
from sklearn.model_selection import train_test_split
# tensorflow for saliency
import tensorflow as tf
import cv2
2024-06-04 17:39:04.164296: I external/local_tsl/tsl/cuda/cudart_stub.cc:31] Could not find cuda drivers on your machine, GPU will not be used.
2024-06-04 17:39:04.189500: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:9261] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2024-06-04 17:39:04.189521: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:607] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2024-06-04 17:39:04.190263: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1515] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
2024-06-04 17:39:04.194692: I external/local_tsl/tsl/cuda/cudart_stub.cc:31] Could not find cuda drivers on your machine, GPU will not be used.
2024-06-04 17:39:04.195181: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-06-04 17:39:05.328964: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
Reloading our Model#
To start, we need to reload our model from the previous galaxy classification notebook from the Mikulski Archive for Space Telescopes (MAST). The following code is directly copied over from that notebook. For a quick refresher, the model intakes a FITS file from a high level science product hosted by MAST. There are 15,426 observations in total, each taken with three filters (F814W from the Advanced Camera for Surveys and F160W from the Wide Field Camera 3 on the Hubble Space Telescope (HST), and F160W and F356W from Near Infrared Camera on the James Webb Space Telescope (JWST)). The model then applies a Convolutional Neural Network to classify whether a galaxy has undergone a merger.
version = 'pristine'
file_url = 'https://archive.stsci.edu/hlsps/deepmerge/hlsp_deepmerge_hst-jwst_acs-wfc3-nircam_illustris-z2_f814w-f160w-f356w_v1_sim-'+version+'.fits'
hdu = fits.open(download_file(file_url, cache=True, show_progress=True))
Build and Compile the Convolutional Model#
For the sake of transparency, we will rebuild the model using the same architecture as the original notebook. The model can also be loaded directly by using save_model
and load_model
from the Keras package.
X = hdu[0].data
y = hdu[1].data
Following the authors, we will split the data into 70:10:20 ratio of train:validate:test. As above, set the random seed to randomly split the images in a repeatable way. Feel free to try different values!
random_state = 42
X = np.asarray(X).astype('float32')
y = np.asarray(y).astype('float32')
# First split off 30% of the data for validation+testing
X_train, X_split, y_train, y_split = train_test_split(X, y, test_size=0.3, random_state=random_state, shuffle=True)
# Then divide this subset into training and testing sets
X_valid, X_test, y_valid, y_test = train_test_split(X_split, y_split, test_size=0.666, random_state=random_state, shuffle=True)
imsize = np.shape(X_train)[2]
X_train = np.array([np.stack(x, axis=2) for x in X_train])
X_valid = np.array([np.stack(x, axis=2) for x in X_valid])
X_test = np.array([np.stack(x, axis=2) for x in X_test])
Generate the model architecture (written for Keras 2)#
# Define architecture for model
data_shape = np.shape(X)
input_shape = (imsize, imsize, 3)
x_in = Input(shape=input_shape)
c0 = Convolution2D(8, (5, 5), activation='relu', strides=(1, 1), padding='same')(x_in)
b0 = BatchNormalization()(c0)
d0 = MaxPooling2D(pool_size=(2, 2), strides=None, padding='valid')(b0)
e0 = Dropout(0.5)(d0)
c1 = Convolution2D(16, (3, 3), activation='relu', strides=(1, 1), padding='same')(e0)
b1 = BatchNormalization()(c1)
d1 = MaxPooling2D(pool_size=(2, 2), strides=None, padding='valid')(b1)
e1 = Dropout(0.5)(d1)
c2 = Convolution2D(32, (3, 3), activation='relu', strides=(1, 1), padding='same')(e1)
b2 = BatchNormalization()(c2)
d2 = MaxPooling2D(pool_size=(2, 2), strides=None, padding='valid')(b2)
e2 = Dropout(0.5)(d2)
f = Flatten()(e2)
z0 = Dense(64, activation='softmax', kernel_regularizer=l2(0.0001))(f)
z1 = Dense(32, activation='softmax', kernel_regularizer=l2(0.0001))(z0)
y_out = Dense(1, activation='sigmoid')(z1)
cnn = Model(inputs=x_in, outputs=y_out)
Compile Model#
optimizer = 'adam'
fit_metrics = ['accuracy']
loss = 'binary_crossentropy'
cnn.compile(loss=loss, optimizer=optimizer, metrics=fit_metrics)
Load pretrained weights#
file_url = 'https://archive.stsci.edu/hlsps/hellouniverse/hellouniverse_interpretability_best_weights.hdf5'
cnn.load_weights(download_file(file_url, cache=True, show_progress=True))
2024-06-04 17:39:57.384384: W tensorflow/core/util/tensor_slice_reader.cc:98] Could not open /home/runner/.astropy/cache/download/url/75479cba1a5e1313befcce7b9c2f3f36/contents: DATA_LOSS: not an sstable (bad magic number): perhaps your file is in a different file format and you need to use a different restore operator?
1. Saliency Maps#
We will start by introducing the most popular and well known method of interpreting CNNs. A saliency map can help us identify which pixels are significant to the models final prediction. There are many methods of calculating saliency maps, but the most popular method utilizes gradient backpropagation to determine the significance of pixels at each layer of the model. To calculate the saliency map, the error gradient at each layer is calculated and then fed into the previous layer, repeating until we reach the original image. Then the pixels with the highest gradient values will also have the most effect on the model’s activation. This methodology is described in detail by Simonyan et al. 2013.
Dependencies#
tensorflow.GradientTape()
is used to track the gradient of the function.
# Choose the image to analyze
img_idx = 1
# We can change the index to any number in range of the test set
orig_img = X_test[img_idx]
img = orig_img
img = img.reshape((1, *img.shape))
norm = simple_norm(orig_img, 'log', max_percent=99.75)
scaled_img = norm(orig_img)
images = tf.Variable(img, dtype=float)
# Make a prediction and track gradients
with tf.GradientTape() as tape:
pred = cnn(images, training=False)
class_idxs_sorted = np.argsort(pred.numpy().flatten())[::-1]
loss = pred[class_idxs_sorted[0]]
grads = tape.gradient(loss, images)
Plot the original image and the saliency map#
Saliency maps provide an intuitive understanding of how the model works. The hot pixels represent higher activation and more importance. In the below model, the saliency maps demonstrates that the model focuses on the area around the center of the galaxy for the majority of galaxies. Our results are in line with a corroborating result by Ntampaka et al. 2018, suggesting that the key features of a galaxy are found the ring around the galaxy rather than in the center of the galaxy.
y_pred = cnn.predict(img)
dgrad_abs = tf.math.abs(grads)
dgrad_max_ = np.max(dgrad_abs, axis=3)[0]
# normalize to range between 0 and 1
arr_min, arr_max = np.min(dgrad_max_), np.max(dgrad_max_)
grad_eval = (dgrad_max_ - arr_min) / (arr_max - arr_min + 1e-18)
# Plot the results next to the original image
fig, axes = plt.subplots(1, 3, figsize=(14, 5))
axes[0].imshow(orig_img)
axes[0].set_title("orig_img")
axes[1].imshow(scaled_img)
axes[1].set_title("scaled_img")
i = axes[2].imshow(grad_eval, cmap="turbo")
fig.colorbar(i)
axes[2].set_title("heat_map")
fig.suptitle("prediction_val=" + str(y_pred))
1/1 [==============================] - ETA: 0s
1/1 [==============================] - 0s 121ms/step
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Got range [1.822696e-08..8.880155].
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Got range [0.0..1.2109969226408994].
Text(0.5, 0.98, 'prediction_val=[[0.44489214]]')
Image caption: The above image contains three panels in a horizontal row. The first panel shows the original image of a merging galaxy candidate, the second panel shows a logarithmically-scaled version of the original image, and the third panel shows a heat map of the saliency map.
Create a stack of saliency images#
To understand the overall behavior of our algorithm, we can stack some or all of the saliency maps in the test set to generate an overarching estimate of important pixels. For the sake of simplicity, we will stack the saliency maps for the first 100 images in the test set. Our results once again indicate that the region around a galaxy is particularly important to the model.
sum_map = np.zeros((75, 75))
# Summing the first 100 saliencies. We can change
# the range to sum more or less saliencies or pick
# specific ones
for i in range(100):
img = X_test[img_idx]
img = img.reshape((1, *img.shape))
images = tf.Variable(img, dtype=float)
# Make a prediction and track gradients
with tf.GradientTape() as tape:
pred = cnn(images, training=False)
class_idxs_sorted = np.argsort(pred.numpy().flatten())[::-1]
loss = pred[class_idxs_sorted[0]]
grads = tape.gradient(loss, images)
y_pred = cnn.predict(images, verbose=0)
dgrad_abs = tf.math.abs(grads)
dgrad_max_ = np.max(dgrad_abs, axis=3)[0]
# normalize to range between 0 and 1
arr_min, arr_max = np.min(dgrad_max_), np.max(dgrad_max_)
grad_eval = (dgrad_max_ - arr_min) / (arr_max - arr_min)
sum_map += grad_eval
plt.imshow(sum_map, cmap='turbo')
<matplotlib.image.AxesImage at 0x7f72684aec90>
Image caption: The above image contains a single panel, and shows a stacked version of the saliency maps from 100 images.
2. Grad-CAM#
While gradient backpropagation has historically been the most popular type of saliency map, the highly connected nature of backtracking has been shown to produce high variance under small changes to inputs. As such, gradient backpropagation is extremely sensitive to data manipulation (preprocessing, sensitivity analysis, GANs), raising questions about its reliability and validity. Gradient Class Activation Mapping (Grad-CAM) is an alternative method for generating saliency models that only examines the gradient of the final convolutional layer when producing the map. As a consequence, Grad-CAM maps have lower (coarser) resolution than backpropagation but are far more resilient to small changes and therefore more reliable when tuning a model. A full description of the technique can be found in Selveraju et al. 2016.
The code for Grad-CAM comes from a useful tutorial on the subject by Daniel Reiff. For more information, please visit the full tutorial.
Dependencies#
Open_CV
and astropy.simple_norm
are used to do manipulate the image for display. Alternatively, we could allow Python to automatically clip the image when the heatmap is out of range.
# Choose the image to analyze
img_idx = 1
# We can change the index to any number in range of the test set
orig_img = X_test[img_idx]
img = orig_img
img = img.reshape((1, *img.shape))
norm = simple_norm(orig_img, 'log', max_percent=99.75)
scaled_img = norm(orig_img)
# Note: recompiling the model will change the layer
# name. In that case, you can either restart the
# kernel or change the layer_name.
# We can also change the layer selected here to pull out any layer of our model
gradModel = Model(inputs=[cnn.inputs], outputs=[cnn.get_layer("conv2d_2").output, cnn.output])
with tf.GradientTape() as tape:
# get the loss with associated with the prediction
inputs = tf.cast(X_test, tf.float32)
(convOutputs, predictions) = gradModel(inputs)
loss = predictions[:, 0]
# use automatic differentiation to compute the gradients
grads = tape.gradient(loss, convOutputs)
# compute the guided gradients by removing all nonpositive
# gradients
castConvOutputs = tf.cast(convOutputs > 0, "float32")
castGrads = tf.cast(grads > 0, "float32")
guidedGrads = castConvOutputs * castGrads * grads
# pick out the convolution and gradient of the chosen image
convOutputs = convOutputs[img_idx]
guidedGrads = guidedGrads[img_idx]
# compute the average of the gradient values, and using them
# as weights, compute the importance of the pieces
weights = tf.reduce_mean(guidedGrads, axis=(0, 1))
cam = tf.reduce_sum(tf.multiply(weights, convOutputs), axis=-1)
# grab the spatial dimensions of the input image and resize
# the output class activation map to match the input image
# dimensions
(w, h) = (X_test.shape[2], X_test.shape[1])
heatmap = cv2.resize(cam.numpy(), (w, h))
# normalize the heatmap such that all values lie in the range
# [0, 1], scale the resulting values to the range [0, 255],
# and then convert to an unsigned 8-bit integer
y_pred = cnn.predict(img)
# Plot the results next to the original image
fig, axes = plt.subplots(1, 3, figsize=(14, 5))
axes[0].imshow(orig_img)
axes[0].set_title("orig_img")
axes[1].imshow(scaled_img)
axes[1].set_title("scaled_img")
i = axes[2].imshow(heatmap, cmap="turbo")
fig.colorbar(i)
axes[2].set_title("heat_map")
fig.suptitle("prediction_val=" + str(y_pred))
1/1 [==============================] - ETA: 0s
1/1 [==============================] - 0s 24ms/step
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Got range [1.822696e-08..8.880155].
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Got range [0.0..1.2109969226408994].
Text(0.5, 0.98, 'prediction_val=[[0.44489214]]')
Image caption: The above image contains three panels in a horizontal row. The first panel shows the original image of a merging galaxy candidate, the second panel shows a logarithmically-scaled version of the original image, and the third panel shows a heat map of the grad-cam map.
The result is very similar to our saliency map from backpropagation, although the image is coarser and also shows both the top and bottom of the galaxy. We can also play around with the selected layer to calculate the output at different steps in the model and see how activation changes throughout the model.
3. RISE Algorithm#
The RISE (Randomized Input Sampling for Explanation) Algorithm is another interpretation technique for calculating saliency maps. Instead of calculating gradients from within the model, the RISE implementation works by covering up pieces of the input image, running it through the model, and calculating the average activation in order to determine what parts of the image are most important. As such, this method does not require any access to the inner workings of the model. The algorithm first generates a random sequence of binary grids (called masks), which are placed onto the image. Everything not covered by the mask is removed by multiplying the images together, and the resultant activations are averaged to get our final heatmap. A full description of the algorithm and its variations is provided by Petsiuk et al. 2018.
# Choose the image to analyze
img_idx = 6
# We can change the index to any number in range of the test set
image = X_test[img_idx]
N = 1000 # Number of masks
s = 8 # Size of the grid
p1 = 0.5 # Probability of the cell being set to 1
cell_size = np.ceil(np.array(input_shape[:2]) / s).astype(int)
up_size = (s * cell_size).astype(int)
grid = np.random.rand(N, s, s) < p1
masks = np.empty((N, *input_shape[:2]))
for i in range(N):
# Randomly place the grid on the image
x = np.random.randint(0, input_shape[0]-s)
y = np.random.randint(0, input_shape[1]-s)
mask = np.pad(grid[i], ((x, input_shape[0]-x-s), (y, input_shape[0]-y-s)), 'constant', constant_values=(0, 0))
mask = mask[:input_shape[0], :input_shape[1]]
masks[i] = mask
masks = masks.reshape(-1, *input_shape[:2], 1)
N = len(masks)
pred_masks = cnn.predict(image * masks)
pred_masks = np.expand_dims(pred_masks, axis=-1)
pred_masks = np.expand_dims(pred_masks, axis=-1) # Reshape pred_masks for broadcasting
heatmap = (pred_masks * masks).sum(axis=0)
heatmap = heatmap / N / p1
1/32 [..............................] - ETA: 0s
5/32 [===>..........................] - ETA: 0s
9/32 [=======>......................] - ETA: 0s
13/32 [===========>..................] - ETA: 0s
17/32 [==============>...............] - ETA: 0s
20/32 [=================>............] - ETA: 0s
23/32 [====================>.........] - ETA: 0s
27/32 [========================>.....] - ETA: 0s
31/32 [============================>.] - ETA: 0s
32/32 [==============================] - 1s 16ms/step
# Plot the results next to the original image
fig, axes = plt.subplots(1, 2, figsize=(14, 5))
axes[0].imshow(image)
axes[0].set_title("orig_img")
i = axes[1].imshow(heatmap, cmap="turbo")
fig.colorbar(i)
axes[1].set_title("heat_map")
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Got range [0.00022207294..674.50696].
Text(0.5, 1.0, 'heat_map')
Image caption: The above image contains two panels in a horizontal row. The first panel shows the original image of a merginig galaxy candidate, and the second panel shows a heat map of the RISE map.
When examining the selected image, we see a ring around the galactic center. Note that this is not the case for all astronomical images, or even all images in this dataset. Try playing around with the selected image to generate different saliency maps. For images where the central feature is relatively small, RISE’s occlusion-based methodology can be unreliable.
4. Ablation Analysis#
Saliency maps provide an intuitive visual understanding of our model’s focus and can be useful for understanding the physical relevance of our classification scheme. In order to understand the efficacy of our model’s architecture, we can instead use ablation analysis to determine the most important layers of our model. Ablation analysis works by rebuilding our model without a specified layer of interest and testing and comparing the performance of a partial model. Since we are focusing on our model’s internal architecture rather than the features of the dataset we are looking at, we want to use this technique when trying to improve the training metrics of our model by editing its layers. This method allows us to determine which layers of the model are most important, or if some layers are hindering the learning capabilities of our mode
In the exercise below, we will build and train four mini-models on the same data set as before. As this is an educational notebook, we will limit the training time of each model to five epochs. Results with these models may vary considerably due to these training constraints, but we highly encourage you to try modifying this section of the notebook for different results (see Exercises)
Note that performing an ablation analysis will require training multiple models with the same architecture. This can be quite compute intensive on personal computers, so if you are running this notebook locally it is recommended that your device be plugged in before running the analysis.
Dependencies#
Tensorflow
is used build our model and train it.
def create_model(ablate=None):
x_in = Input(shape=input_shape)
if ablate != 'c0':
c0 = Convolution2D(8, (5, 5), activation='relu', strides=(1, 1), padding='same')(x_in)
else:
c0 = x_in
b0 = BatchNormalization()(c0)
d0 = MaxPooling2D(pool_size=(2, 2), strides=None, padding='valid')(b0)
e0 = Dropout(0.5)(d0)
if ablate != 'c1':
c1 = Convolution2D(16, (3, 3), activation='relu', strides=(1, 1), padding='same')(e0)
else:
c1 = e0
b1 = BatchNormalization()(c1)
d1 = MaxPooling2D(pool_size=(2, 2), strides=None, padding='valid')(b1)
e1 = Dropout(0.5)(d1)
if ablate != 'c2':
c2 = Convolution2D(32, (3, 3), activation='relu', strides=(1, 1), padding='same')(e1)
else:
c2 = e1
b2 = BatchNormalization()(c2)
d2 = MaxPooling2D(pool_size=(2, 2), strides=None, padding='valid')(b2)
e2 = Dropout(0.5)(d2)
f = Flatten()(e2)
z0 = Dense(64, activation='softmax', kernel_regularizer=l2(0.0001))(f)
z1 = Dense(32, activation='softmax', kernel_regularizer=l2(0.0001))(z0)
y_out = Dense(1, activation='sigmoid')(z1)
cnn = Model(inputs=x_in, outputs=y_out)
return cnn
Since ablation analysis requires training multiple models, it can often be more resource intensive than other methods. However, it can also provide useful information on the way features are organized during training. The following cell can be edited to change how much we want to train our mini-models.
# You can change how much to train each model
# 5 epochs is chosen due to time and computation constraints
num_epochs = 5
# Train the original model
model = create_model()
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
model.fit(X_train, y_train, epochs=num_epochs, batch_size=32, validation_data=(X_test, y_test))
# Get baseline performance
baseline_score = model.evaluate(X_test, y_test)
# Ablate each layer and compare performance
layers_to_ablate = ['c0', 'c1', 'c2']
for layer in layers_to_ablate:
model_ablated = create_model(ablate=layer)
model_ablated.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
model_ablated.fit(X_train, y_train, epochs=num_epochs, batch_size=32, validation_data=(X_test, y_test))
ablated_score = model_ablated.evaluate(X_test, y_test)
print(f"Performance drop after ablating {layer}: {baseline_score[1] - ablated_score[1]}")
Epoch 1/5
2024-06-04 17:40:08.586796: W external/local_tsl/tsl/framework/cpu_allocator_impl.cc:83] Allocation of 728865000 exceeds 10% of free system memory.
1/338 [..............................] - ETA: 6:19 - loss: 0.7102 - accuracy: 0.2500
2/338 [..............................] - ETA: 20s - loss: 0.7097 - accuracy: 0.4062
3/338 [..............................] - ETA: 21s - loss: 0.7093 - accuracy: 0.4688
4/338 [..............................] - ETA: 20s - loss: 0.7091 - accuracy: 0.4688
5/338 [..............................] - ETA: 19s - loss: 0.7091 - accuracy: 0.4375
6/338 [..............................] - ETA: 19s - loss: 0.7085 - accuracy: 0.4844
7/338 [..............................] - ETA: 19s - loss: 0.7082 - accuracy: 0.4911
8/338 [..............................] - ETA: 18s - loss: 0.7079 - accuracy: 0.4961
9/338 [..............................] - ETA: 18s - loss: 0.7077 - accuracy: 0.4931
10/338 [..............................] - ETA: 18s - loss: 0.7075 - accuracy: 0.4906
11/338 [..............................] - ETA: 18s - loss: 0.7073 - accuracy: 0.4858
12/338 [>.............................] - ETA: 18s - loss: 0.7068 - accuracy: 0.5104
13/338 [>.............................] - ETA: 18s - loss: 0.7065 - accuracy: 0.5144
14/338 [>.............................] - ETA: 18s - loss: 0.7062 - accuracy: 0.5179
15/338 [>.............................] - ETA: 18s - loss: 0.7061 - accuracy: 0.5104
16/338 [>.............................] - ETA: 18s - loss: 0.7059 - accuracy: 0.5098
17/338 [>.............................] - ETA: 18s - loss: 0.7056 - accuracy: 0.5092
18/338 [>.............................] - ETA: 18s - loss: 0.7056 - accuracy: 0.5000
19/338 [>.............................] - ETA: 18s - loss: 0.7052 - accuracy: 0.5099
20/338 [>.............................] - ETA: 18s - loss: 0.7050 - accuracy: 0.5141
21/338 [>.............................] - ETA: 18s - loss: 0.7048 - accuracy: 0.5104
22/338 [>.............................] - ETA: 17s - loss: 0.7046 - accuracy: 0.5142
23/338 [=>............................] - ETA: 17s - loss: 0.7043 - accuracy: 0.5204
24/338 [=>............................] - ETA: 17s - loss: 0.7040 - accuracy: 0.5208
25/338 [=>............................] - ETA: 17s - loss: 0.7039 - accuracy: 0.5200
26/338 [=>............................] - ETA: 17s - loss: 0.7036 - accuracy: 0.5228
27/338 [=>............................] - ETA: 17s - loss: 0.7035 - accuracy: 0.5208
28/338 [=>............................] - ETA: 17s - loss: 0.7032 - accuracy: 0.5257
29/338 [=>............................] - ETA: 17s - loss: 0.7029 - accuracy: 0.5312
30/338 [=>............................] - ETA: 17s - loss: 0.7028 - accuracy: 0.5312
31/338 [=>............................] - ETA: 17s - loss: 0.7025 - accuracy: 0.5343
32/338 [=>............................] - ETA: 17s - loss: 0.7024 - accuracy: 0.5352
33/338 [=>............................] - ETA: 17s - loss: 0.7022 - accuracy: 0.5360
34/338 [==>...........................] - ETA: 17s - loss: 0.7021 - accuracy: 0.5349
35/338 [==>...........................] - ETA: 17s - loss: 0.7019 - accuracy: 0.5357
36/338 [==>...........................] - ETA: 16s - loss: 0.7017 - accuracy: 0.5373
37/338 [==>...........................] - ETA: 16s - loss: 0.7016 - accuracy: 0.5346
38/338 [==>...........................] - ETA: 16s - loss: 0.7015 - accuracy: 0.5354
39/338 [==>...........................] - ETA: 16s - loss: 0.7015 - accuracy: 0.5312
40/338 [==>...........................] - ETA: 16s - loss: 0.7016 - accuracy: 0.5250
41/338 [==>...........................] - ETA: 16s - loss: 0.7016 - accuracy: 0.5236
42/338 [==>...........................] - ETA: 16s - loss: 0.7014 - accuracy: 0.5238
43/338 [==>...........................] - ETA: 16s - loss: 0.7013 - accuracy: 0.5247
44/338 [==>...........................] - ETA: 16s - loss: 0.7012 - accuracy: 0.5249
45/338 [==>...........................] - ETA: 16s - loss: 0.7010 - accuracy: 0.5250
46/338 [===>..........................] - ETA: 16s - loss: 0.7009 - accuracy: 0.5265
47/338 [===>..........................] - ETA: 16s - loss: 0.7006 - accuracy: 0.5293
48/338 [===>..........................] - ETA: 16s - loss: 0.7004 - accuracy: 0.5319
49/338 [===>..........................] - ETA: 16s - loss: 0.7002 - accuracy: 0.5351
50/338 [===>..........................] - ETA: 16s - loss: 0.7001 - accuracy: 0.5350
51/338 [===>..........................] - ETA: 16s - loss: 0.7000 - accuracy: 0.5343
52/338 [===>..........................] - ETA: 15s - loss: 0.6999 - accuracy: 0.5349
53/338 [===>..........................] - ETA: 15s - loss: 0.6998 - accuracy: 0.5342
54/338 [===>..........................] - ETA: 15s - loss: 0.6996 - accuracy: 0.5370
55/338 [===>..........................] - ETA: 15s - loss: 0.6995 - accuracy: 0.5381
56/338 [===>..........................] - ETA: 15s - loss: 0.6994 - accuracy: 0.5385
57/338 [====>.........................] - ETA: 15s - loss: 0.6993 - accuracy: 0.5389
58/338 [====>.........................] - ETA: 15s - loss: 0.6992 - accuracy: 0.5383
59/338 [====>.........................] - ETA: 15s - loss: 0.6992 - accuracy: 0.5376
60/338 [====>.........................] - ETA: 15s - loss: 0.6991 - accuracy: 0.5375
61/338 [====>.........................] - ETA: 15s - loss: 0.6991 - accuracy: 0.5369
62/338 [====>.........................] - ETA: 15s - loss: 0.6990 - accuracy: 0.5363
63/338 [====>.........................] - ETA: 15s - loss: 0.6991 - accuracy: 0.5342
64/338 [====>.........................] - ETA: 15s - loss: 0.6991 - accuracy: 0.5327
65/338 [====>.........................] - ETA: 15s - loss: 0.6991 - accuracy: 0.5312
66/338 [====>.........................] - ETA: 15s - loss: 0.6990 - accuracy: 0.5317
67/338 [====>.........................] - ETA: 15s - loss: 0.6989 - accuracy: 0.5317
68/338 [=====>........................] - ETA: 15s - loss: 0.6989 - accuracy: 0.5312
69/338 [=====>........................] - ETA: 15s - loss: 0.6989 - accuracy: 0.5299
70/338 [=====>........................] - ETA: 14s - loss: 0.6986 - accuracy: 0.5326
71/338 [=====>........................] - ETA: 14s - loss: 0.6985 - accuracy: 0.5330
72/338 [=====>........................] - ETA: 14s - loss: 0.6985 - accuracy: 0.5326
73/338 [=====>........................] - ETA: 14s - loss: 0.6984 - accuracy: 0.5330
74/338 [=====>........................] - ETA: 14s - loss: 0.6984 - accuracy: 0.5317
75/338 [=====>........................] - ETA: 14s - loss: 0.6983 - accuracy: 0.5321
76/338 [=====>........................] - ETA: 14s - loss: 0.6982 - accuracy: 0.5333
77/338 [=====>........................] - ETA: 14s - loss: 0.6982 - accuracy: 0.5317
78/338 [=====>........................] - ETA: 14s - loss: 0.6981 - accuracy: 0.5317
79/338 [======>.......................] - ETA: 14s - loss: 0.6981 - accuracy: 0.5309
80/338 [======>.......................] - ETA: 14s - loss: 0.6981 - accuracy: 0.5309
81/338 [======>.......................] - ETA: 14s - loss: 0.6981 - accuracy: 0.5293
82/338 [======>.......................] - ETA: 14s - loss: 0.6981 - accuracy: 0.5282
83/338 [======>.......................] - ETA: 14s - loss: 0.6980 - accuracy: 0.5297
84/338 [======>.......................] - ETA: 14s - loss: 0.6979 - accuracy: 0.5301
85/338 [======>.......................] - ETA: 14s - loss: 0.6979 - accuracy: 0.5298
86/338 [======>.......................] - ETA: 14s - loss: 0.6979 - accuracy: 0.5291
87/338 [======>.......................] - ETA: 14s - loss: 0.6978 - accuracy: 0.5295
88/338 [======>.......................] - ETA: 13s - loss: 0.6978 - accuracy: 0.5288
89/338 [======>.......................] - ETA: 13s - loss: 0.6977 - accuracy: 0.5295
90/338 [======>.......................] - ETA: 13s - loss: 0.6977 - accuracy: 0.5295
91/338 [=======>......................] - ETA: 13s - loss: 0.6977 - accuracy: 0.5288
92/338 [=======>......................] - ETA: 13s - loss: 0.6976 - accuracy: 0.5289
93/338 [=======>......................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5292
94/338 [=======>......................] - ETA: 13s - loss: 0.6976 - accuracy: 0.5279
95/338 [=======>......................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5286
96/338 [=======>......................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5270
97/338 [=======>......................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5261
98/338 [=======>......................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5252
99/338 [=======>......................] - ETA: 13s - loss: 0.6976 - accuracy: 0.5240
100/338 [=======>......................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5234
101/338 [=======>......................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5232
102/338 [========>.....................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5230
103/338 [========>.....................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5221
104/338 [========>.....................] - ETA: 13s - loss: 0.6975 - accuracy: 0.5213
105/338 [========>.....................] - ETA: 12s - loss: 0.6975 - accuracy: 0.5208
106/338 [========>.....................] - ETA: 12s - loss: 0.6975 - accuracy: 0.5203
107/338 [========>.....................] - ETA: 12s - loss: 0.6975 - accuracy: 0.5190
108/338 [========>.....................] - ETA: 12s - loss: 0.6975 - accuracy: 0.5197
109/338 [========>.....................] - ETA: 12s - loss: 0.6974 - accuracy: 0.5201
110/338 [========>.....................] - ETA: 12s - loss: 0.6974 - accuracy: 0.5202
111/338 [========>.....................] - ETA: 12s - loss: 0.6973 - accuracy: 0.5206
112/338 [========>.....................] - ETA: 12s - loss: 0.6973 - accuracy: 0.5204
113/338 [=========>....................] - ETA: 12s - loss: 0.6973 - accuracy: 0.5194
114/338 [=========>....................] - ETA: 12s - loss: 0.6972 - accuracy: 0.5200
115/338 [=========>....................] - ETA: 12s - loss: 0.6973 - accuracy: 0.5188
116/338 [=========>....................] - ETA: 12s - loss: 0.6973 - accuracy: 0.5170
117/338 [=========>....................] - ETA: 12s - loss: 0.6972 - accuracy: 0.5176
118/338 [=========>....................] - ETA: 12s - loss: 0.6972 - accuracy: 0.5177
119/338 [=========>....................] - ETA: 12s - loss: 0.6972 - accuracy: 0.5171
120/338 [=========>....................] - ETA: 12s - loss: 0.6972 - accuracy: 0.5172
121/338 [=========>....................] - ETA: 12s - loss: 0.6971 - accuracy: 0.5186
122/338 [=========>....................] - ETA: 12s - loss: 0.6971 - accuracy: 0.5177
123/338 [=========>....................] - ETA: 11s - loss: 0.6971 - accuracy: 0.5178
124/338 [==========>...................] - ETA: 11s - loss: 0.6971 - accuracy: 0.5176
125/338 [==========>...................] - ETA: 11s - loss: 0.6970 - accuracy: 0.5185
126/338 [==========>...................] - ETA: 11s - loss: 0.6969 - accuracy: 0.5191
127/338 [==========>...................] - ETA: 11s - loss: 0.6969 - accuracy: 0.5187
128/338 [==========>...................] - ETA: 11s - loss: 0.6969 - accuracy: 0.5183
129/338 [==========>...................] - ETA: 11s - loss: 0.6969 - accuracy: 0.5177
130/338 [==========>...................] - ETA: 11s - loss: 0.6969 - accuracy: 0.5183
131/338 [==========>...................] - ETA: 11s - loss: 0.6968 - accuracy: 0.5186
132/338 [==========>...................] - ETA: 11s - loss: 0.6968 - accuracy: 0.5173
133/338 [==========>...................] - ETA: 11s - loss: 0.6968 - accuracy: 0.5186
134/338 [==========>...................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5184
135/338 [==========>...................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5181
136/338 [===========>..................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5182
137/338 [===========>..................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5185
138/338 [===========>..................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5186
139/338 [===========>..................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5180
140/338 [===========>..................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5179
141/338 [===========>..................] - ETA: 11s - loss: 0.6967 - accuracy: 0.5175
142/338 [===========>..................] - ETA: 11s - loss: 0.6966 - accuracy: 0.5172
143/338 [===========>..................] - ETA: 10s - loss: 0.6966 - accuracy: 0.5179
144/338 [===========>..................] - ETA: 10s - loss: 0.6966 - accuracy: 0.5178
145/338 [===========>..................] - ETA: 10s - loss: 0.6965 - accuracy: 0.5177
146/338 [===========>..................] - ETA: 10s - loss: 0.6965 - accuracy: 0.5167
147/338 [============>.................] - ETA: 10s - loss: 0.6965 - accuracy: 0.5166
148/338 [============>.................] - ETA: 10s - loss: 0.6965 - accuracy: 0.5165
149/338 [============>.................] - ETA: 10s - loss: 0.6965 - accuracy: 0.5170
150/338 [============>.................] - ETA: 10s - loss: 0.6965 - accuracy: 0.5167
151/338 [============>.................] - ETA: 10s - loss: 0.6965 - accuracy: 0.5155
152/338 [============>.................] - ETA: 10s - loss: 0.6964 - accuracy: 0.5160
153/338 [============>.................] - ETA: 10s - loss: 0.6964 - accuracy: 0.5161
154/338 [============>.................] - ETA: 10s - loss: 0.6964 - accuracy: 0.5156
155/338 [============>.................] - ETA: 10s - loss: 0.6964 - accuracy: 0.5151
156/338 [============>.................] - ETA: 10s - loss: 0.6964 - accuracy: 0.5156
157/338 [============>.................] - ETA: 10s - loss: 0.6963 - accuracy: 0.5157
158/338 [=============>................] - ETA: 10s - loss: 0.6963 - accuracy: 0.5162
159/338 [=============>................] - ETA: 10s - loss: 0.6962 - accuracy: 0.5165
160/338 [=============>................] - ETA: 10s - loss: 0.6962 - accuracy: 0.5172
161/338 [=============>................] - ETA: 9s - loss: 0.6962 - accuracy: 0.5171
162/338 [=============>................] - ETA: 9s - loss: 0.6962 - accuracy: 0.5168
163/338 [=============>................] - ETA: 9s - loss: 0.6962 - accuracy: 0.5165
164/338 [=============>................] - ETA: 9s - loss: 0.6961 - accuracy: 0.5158
165/338 [=============>................] - ETA: 9s - loss: 0.6961 - accuracy: 0.5167
166/338 [=============>................] - ETA: 9s - loss: 0.6960 - accuracy: 0.5168
167/338 [=============>................] - ETA: 9s - loss: 0.6960 - accuracy: 0.5165
168/338 [=============>................] - ETA: 9s - loss: 0.6960 - accuracy: 0.5166
169/338 [==============>...............] - ETA: 9s - loss: 0.6960 - accuracy: 0.5159
170/338 [==============>...............] - ETA: 9s - loss: 0.6960 - accuracy: 0.5154
171/338 [==============>...............] - ETA: 9s - loss: 0.6960 - accuracy: 0.5159
172/338 [==============>...............] - ETA: 9s - loss: 0.6959 - accuracy: 0.5154
173/338 [==============>...............] - ETA: 9s - loss: 0.6959 - accuracy: 0.5159
174/338 [==============>...............] - ETA: 9s - loss: 0.6959 - accuracy: 0.5158
175/338 [==============>...............] - ETA: 9s - loss: 0.6959 - accuracy: 0.5159
176/338 [==============>...............] - ETA: 9s - loss: 0.6959 - accuracy: 0.5156
177/338 [==============>...............] - ETA: 9s - loss: 0.6959 - accuracy: 0.5150
178/338 [==============>...............] - ETA: 9s - loss: 0.6958 - accuracy: 0.5162
179/338 [==============>...............] - ETA: 8s - loss: 0.6958 - accuracy: 0.5157
180/338 [==============>...............] - ETA: 8s - loss: 0.6958 - accuracy: 0.5153
181/338 [===============>..............] - ETA: 8s - loss: 0.6958 - accuracy: 0.5155
182/338 [===============>..............] - ETA: 8s - loss: 0.6958 - accuracy: 0.5155
183/338 [===============>..............] - ETA: 8s - loss: 0.6957 - accuracy: 0.5157
184/338 [===============>..............] - ETA: 8s - loss: 0.6957 - accuracy: 0.5155
185/338 [===============>..............] - ETA: 8s - loss: 0.6957 - accuracy: 0.5152
186/338 [===============>..............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5156
187/338 [===============>..............] - ETA: 8s - loss: 0.6957 - accuracy: 0.5147
188/338 [===============>..............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5148
189/338 [===============>..............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5146
190/338 [===============>..............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5150
191/338 [===============>..............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5146
192/338 [================>.............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5156
193/338 [================>.............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5154
194/338 [================>.............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5143
195/338 [================>.............] - ETA: 8s - loss: 0.6956 - accuracy: 0.5139
196/338 [================>.............] - ETA: 7s - loss: 0.6956 - accuracy: 0.5136
197/338 [================>.............] - ETA: 7s - loss: 0.6955 - accuracy: 0.5140
198/338 [================>.............] - ETA: 7s - loss: 0.6955 - accuracy: 0.5142
199/338 [================>.............] - ETA: 7s - loss: 0.6955 - accuracy: 0.5143
200/338 [================>.............] - ETA: 7s - loss: 0.6955 - accuracy: 0.5139
201/338 [================>.............] - ETA: 7s - loss: 0.6955 - accuracy: 0.5143
202/338 [================>.............] - ETA: 7s - loss: 0.6954 - accuracy: 0.5149
203/338 [=================>............] - ETA: 7s - loss: 0.6954 - accuracy: 0.5162
204/338 [=================>............] - ETA: 7s - loss: 0.6954 - accuracy: 0.5165
205/338 [=================>............] - ETA: 7s - loss: 0.6953 - accuracy: 0.5171
206/338 [=================>............] - ETA: 7s - loss: 0.6953 - accuracy: 0.5173
207/338 [=================>............] - ETA: 7s - loss: 0.6952 - accuracy: 0.5178
208/338 [=================>............] - ETA: 7s - loss: 0.6952 - accuracy: 0.5182
209/338 [=================>............] - ETA: 7s - loss: 0.6952 - accuracy: 0.5187
210/338 [=================>............] - ETA: 7s - loss: 0.6952 - accuracy: 0.5189
211/338 [=================>............] - ETA: 7s - loss: 0.6951 - accuracy: 0.5191
212/338 [=================>............] - ETA: 7s - loss: 0.6951 - accuracy: 0.5192
213/338 [=================>............] - ETA: 7s - loss: 0.6951 - accuracy: 0.5194
214/338 [=================>............] - ETA: 6s - loss: 0.6951 - accuracy: 0.5199
215/338 [==================>...........] - ETA: 6s - loss: 0.6951 - accuracy: 0.5208
216/338 [==================>...........] - ETA: 6s - loss: 0.6951 - accuracy: 0.5207
217/338 [==================>...........] - ETA: 6s - loss: 0.6950 - accuracy: 0.5212
218/338 [==================>...........] - ETA: 6s - loss: 0.6950 - accuracy: 0.5214
219/338 [==================>...........] - ETA: 6s - loss: 0.6950 - accuracy: 0.5217
220/338 [==================>...........] - ETA: 6s - loss: 0.6949 - accuracy: 0.5222
221/338 [==================>...........] - ETA: 6s - loss: 0.6949 - accuracy: 0.5221
222/338 [==================>...........] - ETA: 6s - loss: 0.6949 - accuracy: 0.5224
223/338 [==================>...........] - ETA: 6s - loss: 0.6948 - accuracy: 0.5228
224/338 [==================>...........] - ETA: 6s - loss: 0.6948 - accuracy: 0.5227
225/338 [==================>...........] - ETA: 6s - loss: 0.6948 - accuracy: 0.5228
226/338 [===================>..........] - ETA: 6s - loss: 0.6948 - accuracy: 0.5234
227/338 [===================>..........] - ETA: 6s - loss: 0.6948 - accuracy: 0.5234
228/338 [===================>..........] - ETA: 6s - loss: 0.6947 - accuracy: 0.5238
229/338 [===================>..........] - ETA: 6s - loss: 0.6947 - accuracy: 0.5236
230/338 [===================>..........] - ETA: 6s - loss: 0.6947 - accuracy: 0.5235
231/338 [===================>..........] - ETA: 6s - loss: 0.6947 - accuracy: 0.5239
232/338 [===================>..........] - ETA: 5s - loss: 0.6946 - accuracy: 0.5242
233/338 [===================>..........] - ETA: 5s - loss: 0.6946 - accuracy: 0.5244
234/338 [===================>..........] - ETA: 5s - loss: 0.6946 - accuracy: 0.5240
235/338 [===================>..........] - ETA: 5s - loss: 0.6946 - accuracy: 0.5246
236/338 [===================>..........] - ETA: 5s - loss: 0.6946 - accuracy: 0.5246
237/338 [====================>.........] - ETA: 5s - loss: 0.6946 - accuracy: 0.5244
238/338 [====================>.........] - ETA: 5s - loss: 0.6946 - accuracy: 0.5251
239/338 [====================>.........] - ETA: 5s - loss: 0.6945 - accuracy: 0.5254
240/338 [====================>.........] - ETA: 5s - loss: 0.6945 - accuracy: 0.5249
241/338 [====================>.........] - ETA: 5s - loss: 0.6945 - accuracy: 0.5252
242/338 [====================>.........] - ETA: 5s - loss: 0.6945 - accuracy: 0.5251
243/338 [====================>.........] - ETA: 5s - loss: 0.6945 - accuracy: 0.5251
244/338 [====================>.........] - ETA: 5s - loss: 0.6945 - accuracy: 0.5251
245/338 [====================>.........] - ETA: 5s - loss: 0.6944 - accuracy: 0.5255
246/338 [====================>.........] - ETA: 5s - loss: 0.6944 - accuracy: 0.5258
247/338 [====================>.........] - ETA: 5s - loss: 0.6944 - accuracy: 0.5254
248/338 [=====================>........] - ETA: 5s - loss: 0.6943 - accuracy: 0.5251
249/338 [=====================>........] - ETA: 5s - loss: 0.6943 - accuracy: 0.5245
250/338 [=====================>........] - ETA: 4s - loss: 0.6943 - accuracy: 0.5250
251/338 [=====================>........] - ETA: 4s - loss: 0.6943 - accuracy: 0.5250
252/338 [=====================>........] - ETA: 4s - loss: 0.6942 - accuracy: 0.5248
253/338 [=====================>........] - ETA: 4s - loss: 0.6942 - accuracy: 0.5245
254/338 [=====================>........] - ETA: 4s - loss: 0.6941 - accuracy: 0.5253
255/338 [=====================>........] - ETA: 4s - loss: 0.6942 - accuracy: 0.5249
256/338 [=====================>........] - ETA: 4s - loss: 0.6942 - accuracy: 0.5249
257/338 [=====================>........] - ETA: 4s - loss: 0.6941 - accuracy: 0.5250
258/338 [=====================>........] - ETA: 4s - loss: 0.6941 - accuracy: 0.5253
259/338 [=====================>........] - ETA: 4s - loss: 0.6940 - accuracy: 0.5256
260/338 [======================>.......] - ETA: 4s - loss: 0.6940 - accuracy: 0.5258
261/338 [======================>.......] - ETA: 4s - loss: 0.6940 - accuracy: 0.5259
262/338 [======================>.......] - ETA: 4s - loss: 0.6940 - accuracy: 0.5258
263/338 [======================>.......] - ETA: 4s - loss: 0.6940 - accuracy: 0.5260
264/338 [======================>.......] - ETA: 4s - loss: 0.6939 - accuracy: 0.5259
265/338 [======================>.......] - ETA: 4s - loss: 0.6938 - accuracy: 0.5270
266/338 [======================>.......] - ETA: 4s - loss: 0.6938 - accuracy: 0.5271
267/338 [======================>.......] - ETA: 3s - loss: 0.6938 - accuracy: 0.5270
268/338 [======================>.......] - ETA: 3s - loss: 0.6938 - accuracy: 0.5264
269/338 [======================>.......] - ETA: 3s - loss: 0.6938 - accuracy: 0.5267
270/338 [======================>.......] - ETA: 3s - loss: 0.6938 - accuracy: 0.5267
271/338 [=======================>......] - ETA: 3s - loss: 0.6937 - accuracy: 0.5269
272/338 [=======================>......] - ETA: 3s - loss: 0.6937 - accuracy: 0.5269
273/338 [=======================>......] - ETA: 3s - loss: 0.6937 - accuracy: 0.5270
274/338 [=======================>......] - ETA: 3s - loss: 0.6937 - accuracy: 0.5269
275/338 [=======================>......] - ETA: 3s - loss: 0.6936 - accuracy: 0.5272
276/338 [=======================>......] - ETA: 3s - loss: 0.6936 - accuracy: 0.5274
277/338 [=======================>......] - ETA: 3s - loss: 0.6936 - accuracy: 0.5273
278/338 [=======================>......] - ETA: 3s - loss: 0.6935 - accuracy: 0.5277
279/338 [=======================>......] - ETA: 3s - loss: 0.6934 - accuracy: 0.5282
280/338 [=======================>......] - ETA: 3s - loss: 0.6934 - accuracy: 0.5281
281/338 [=======================>......] - ETA: 3s - loss: 0.6933 - accuracy: 0.5286
282/338 [========================>.....] - ETA: 3s - loss: 0.6933 - accuracy: 0.5284
283/338 [========================>.....] - ETA: 3s - loss: 0.6933 - accuracy: 0.5285
284/338 [========================>.....] - ETA: 3s - loss: 0.6933 - accuracy: 0.5282
285/338 [========================>.....] - ETA: 2s - loss: 0.6932 - accuracy: 0.5286
286/338 [========================>.....] - ETA: 2s - loss: 0.6932 - accuracy: 0.5286
287/338 [========================>.....] - ETA: 2s - loss: 0.6931 - accuracy: 0.5287
288/338 [========================>.....] - ETA: 2s - loss: 0.6931 - accuracy: 0.5286
289/338 [========================>.....] - ETA: 2s - loss: 0.6931 - accuracy: 0.5289
290/338 [========================>.....] - ETA: 2s - loss: 0.6930 - accuracy: 0.5292
291/338 [========================>.....] - ETA: 2s - loss: 0.6930 - accuracy: 0.5293
292/338 [========================>.....] - ETA: 2s - loss: 0.6930 - accuracy: 0.5290
293/338 [=========================>....] - ETA: 2s - loss: 0.6929 - accuracy: 0.5292
294/338 [=========================>....] - ETA: 2s - loss: 0.6929 - accuracy: 0.5293
295/338 [=========================>....] - ETA: 2s - loss: 0.6929 - accuracy: 0.5297
296/338 [=========================>....] - ETA: 2s - loss: 0.6928 - accuracy: 0.5301
297/338 [=========================>....] - ETA: 2s - loss: 0.6927 - accuracy: 0.5302
298/338 [=========================>....] - ETA: 2s - loss: 0.6926 - accuracy: 0.5308
299/338 [=========================>....] - ETA: 2s - loss: 0.6926 - accuracy: 0.5308
300/338 [=========================>....] - ETA: 2s - loss: 0.6926 - accuracy: 0.5305
301/338 [=========================>....] - ETA: 2s - loss: 0.6926 - accuracy: 0.5303
302/338 [=========================>....] - ETA: 2s - loss: 0.6926 - accuracy: 0.5301
303/338 [=========================>....] - ETA: 1s - loss: 0.6926 - accuracy: 0.5296
304/338 [=========================>....] - ETA: 1s - loss: 0.6925 - accuracy: 0.5299
305/338 [==========================>...] - ETA: 1s - loss: 0.6926 - accuracy: 0.5296
306/338 [==========================>...] - ETA: 1s - loss: 0.6925 - accuracy: 0.5297
307/338 [==========================>...] - ETA: 1s - loss: 0.6925 - accuracy: 0.5297
308/338 [==========================>...] - ETA: 1s - loss: 0.6924 - accuracy: 0.5302
309/338 [==========================>...] - ETA: 1s - loss: 0.6924 - accuracy: 0.5303
310/338 [==========================>...] - ETA: 1s - loss: 0.6923 - accuracy: 0.5304
311/338 [==========================>...] - ETA: 1s - loss: 0.6923 - accuracy: 0.5306
312/338 [==========================>...] - ETA: 1s - loss: 0.6922 - accuracy: 0.5308
313/338 [==========================>...] - ETA: 1s - loss: 0.6922 - accuracy: 0.5306
314/338 [==========================>...] - ETA: 1s - loss: 0.6922 - accuracy: 0.5306
315/338 [==========================>...] - ETA: 1s - loss: 0.6922 - accuracy: 0.5307
316/338 [===========================>..] - ETA: 1s - loss: 0.6921 - accuracy: 0.5309
317/338 [===========================>..] - ETA: 1s - loss: 0.6922 - accuracy: 0.5303
318/338 [===========================>..] - ETA: 1s - loss: 0.6921 - accuracy: 0.5306
319/338 [===========================>..] - ETA: 1s - loss: 0.6922 - accuracy: 0.5301
320/338 [===========================>..] - ETA: 1s - loss: 0.6921 - accuracy: 0.5305
321/338 [===========================>..] - ETA: 0s - loss: 0.6921 - accuracy: 0.5307
322/338 [===========================>..] - ETA: 0s - loss: 0.6921 - accuracy: 0.5305
323/338 [===========================>..] - ETA: 0s - loss: 0.6921 - accuracy: 0.5302
324/338 [===========================>..] - ETA: 0s - loss: 0.6921 - accuracy: 0.5301
325/338 [===========================>..] - ETA: 0s - loss: 0.6921 - accuracy: 0.5300
326/338 [===========================>..] - ETA: 0s - loss: 0.6921 - accuracy: 0.5297
327/338 [============================>.] - ETA: 0s - loss: 0.6920 - accuracy: 0.5302
328/338 [============================>.] - ETA: 0s - loss: 0.6919 - accuracy: 0.5304
329/338 [============================>.] - ETA: 0s - loss: 0.6919 - accuracy: 0.5303
330/338 [============================>.] - ETA: 0s - loss: 0.6919 - accuracy: 0.5307
331/338 [============================>.] - ETA: 0s - loss: 0.6918 - accuracy: 0.5308
332/338 [============================>.] - ETA: 0s - loss: 0.6918 - accuracy: 0.5308
333/338 [============================>.] - ETA: 0s - loss: 0.6918 - accuracy: 0.5309
334/338 [============================>.] - ETA: 0s - loss: 0.6919 - accuracy: 0.5306
335/338 [============================>.] - ETA: 0s - loss: 0.6918 - accuracy: 0.5307
336/338 [============================>.] - ETA: 0s - loss: 0.6918 - accuracy: 0.5305
337/338 [============================>.] - ETA: 0s - loss: 0.6918 - accuracy: 0.5309
338/338 [==============================] - 22s 61ms/step - loss: 0.6917 - accuracy: 0.5311 - val_loss: 0.6935 - val_accuracy: 0.5235
Epoch 2/5
1/338 [..............................] - ETA: 19s - loss: 0.6757 - accuracy: 0.5312
2/338 [..............................] - ETA: 18s - loss: 0.6819 - accuracy: 0.5156
3/338 [..............................] - ETA: 18s - loss: 0.6772 - accuracy: 0.5312
4/338 [..............................] - ETA: 18s - loss: 0.6809 - accuracy: 0.5234
5/338 [..............................] - ETA: 18s - loss: 0.6827 - accuracy: 0.5188
6/338 [..............................] - ETA: 18s - loss: 0.6872 - accuracy: 0.4948
7/338 [..............................] - ETA: 18s - loss: 0.6870 - accuracy: 0.5000
8/338 [..............................] - ETA: 18s - loss: 0.6843 - accuracy: 0.5117
9/338 [..............................] - ETA: 18s - loss: 0.6851 - accuracy: 0.5069
10/338 [..............................] - ETA: 18s - loss: 0.6846 - accuracy: 0.5000
11/338 [..............................] - ETA: 18s - loss: 0.6846 - accuracy: 0.5057
12/338 [>.............................] - ETA: 18s - loss: 0.6857 - accuracy: 0.4948
13/338 [>.............................] - ETA: 18s - loss: 0.6869 - accuracy: 0.4904
14/338 [>.............................] - ETA: 18s - loss: 0.6874 - accuracy: 0.4888
15/338 [>.............................] - ETA: 17s - loss: 0.6861 - accuracy: 0.4958
16/338 [>.............................] - ETA: 17s - loss: 0.6867 - accuracy: 0.4961
17/338 [>.............................] - ETA: 17s - loss: 0.6878 - accuracy: 0.4871
18/338 [>.............................] - ETA: 17s - loss: 0.6874 - accuracy: 0.4896
19/338 [>.............................] - ETA: 17s - loss: 0.6861 - accuracy: 0.4951
20/338 [>.............................] - ETA: 17s - loss: 0.6863 - accuracy: 0.5031
21/338 [>.............................] - ETA: 17s - loss: 0.6864 - accuracy: 0.5000
22/338 [>.............................] - ETA: 17s - loss: 0.6860 - accuracy: 0.5057
23/338 [=>............................] - ETA: 17s - loss: 0.6860 - accuracy: 0.5054
24/338 [=>............................] - ETA: 17s - loss: 0.6846 - accuracy: 0.5156
25/338 [=>............................] - ETA: 17s - loss: 0.6846 - accuracy: 0.5188
26/338 [=>............................] - ETA: 17s - loss: 0.6840 - accuracy: 0.5240
27/338 [=>............................] - ETA: 17s - loss: 0.6847 - accuracy: 0.5208
28/338 [=>............................] - ETA: 17s - loss: 0.6854 - accuracy: 0.5145
29/338 [=>............................] - ETA: 17s - loss: 0.6856 - accuracy: 0.5162
30/338 [=>............................] - ETA: 17s - loss: 0.6858 - accuracy: 0.5167
31/338 [=>............................] - ETA: 17s - loss: 0.6858 - accuracy: 0.5171
32/338 [=>............................] - ETA: 16s - loss: 0.6856 - accuracy: 0.5186
33/338 [=>............................] - ETA: 16s - loss: 0.6856 - accuracy: 0.5170
34/338 [==>...........................] - ETA: 16s - loss: 0.6858 - accuracy: 0.5147
35/338 [==>...........................] - ETA: 16s - loss: 0.6854 - accuracy: 0.5170
36/338 [==>...........................] - ETA: 16s - loss: 0.6856 - accuracy: 0.5148
37/338 [==>...........................] - ETA: 16s - loss: 0.6853 - accuracy: 0.5169
38/338 [==>...........................] - ETA: 16s - loss: 0.6854 - accuracy: 0.5189
39/338 [==>...........................] - ETA: 16s - loss: 0.6855 - accuracy: 0.5160
40/338 [==>...........................] - ETA: 16s - loss: 0.6856 - accuracy: 0.5164
41/338 [==>...........................] - ETA: 16s - loss: 0.6851 - accuracy: 0.5183
42/338 [==>...........................] - ETA: 16s - loss: 0.6844 - accuracy: 0.5216
43/338 [==>...........................] - ETA: 16s - loss: 0.6847 - accuracy: 0.5203
44/338 [==>...........................] - ETA: 16s - loss: 0.6846 - accuracy: 0.5199
45/338 [==>...........................] - ETA: 16s - loss: 0.6846 - accuracy: 0.5201
46/338 [===>..........................] - ETA: 16s - loss: 0.6847 - accuracy: 0.5197
47/338 [===>..........................] - ETA: 16s - loss: 0.6858 - accuracy: 0.5140
48/338 [===>..........................] - ETA: 16s - loss: 0.6857 - accuracy: 0.5150
49/338 [===>..........................] - ETA: 16s - loss: 0.6860 - accuracy: 0.5140
50/338 [===>..........................] - ETA: 16s - loss: 0.6862 - accuracy: 0.5125
51/338 [===>..........................] - ETA: 15s - loss: 0.6862 - accuracy: 0.5110
52/338 [===>..........................] - ETA: 15s - loss: 0.6862 - accuracy: 0.5144
53/338 [===>..........................] - ETA: 15s - loss: 0.6860 - accuracy: 0.5171
54/338 [===>..........................] - ETA: 15s - loss: 0.6859 - accuracy: 0.5197
55/338 [===>..........................] - ETA: 15s - loss: 0.6855 - accuracy: 0.5222
56/338 [===>..........................] - ETA: 15s - loss: 0.6851 - accuracy: 0.5273
57/338 [====>.........................] - ETA: 15s - loss: 0.6850 - accuracy: 0.5302
58/338 [====>.........................] - ETA: 15s - loss: 0.6850 - accuracy: 0.5312
59/338 [====>.........................] - ETA: 15s - loss: 0.6851 - accuracy: 0.5318
60/338 [====>.........................] - ETA: 15s - loss: 0.6850 - accuracy: 0.5344
61/338 [====>.........................] - ETA: 15s - loss: 0.6849 - accuracy: 0.5364
62/338 [====>.........................] - ETA: 15s - loss: 0.6848 - accuracy: 0.5388
63/338 [====>.........................] - ETA: 15s - loss: 0.6848 - accuracy: 0.5397
64/338 [====>.........................] - ETA: 15s - loss: 0.6846 - accuracy: 0.5415
65/338 [====>.........................] - ETA: 15s - loss: 0.6845 - accuracy: 0.5437
66/338 [====>.........................] - ETA: 15s - loss: 0.6847 - accuracy: 0.5436
67/338 [====>.........................] - ETA: 15s - loss: 0.6845 - accuracy: 0.5434
68/338 [=====>........................] - ETA: 15s - loss: 0.6844 - accuracy: 0.5455
69/338 [=====>........................] - ETA: 14s - loss: 0.6841 - accuracy: 0.5485
70/338 [=====>........................] - ETA: 14s - loss: 0.6843 - accuracy: 0.5482
71/338 [=====>........................] - ETA: 14s - loss: 0.6843 - accuracy: 0.5489
72/338 [=====>........................] - ETA: 14s - loss: 0.6844 - accuracy: 0.5482
73/338 [=====>........................] - ETA: 14s - loss: 0.6843 - accuracy: 0.5497
74/338 [=====>........................] - ETA: 14s - loss: 0.6844 - accuracy: 0.5498
75/338 [=====>........................] - ETA: 14s - loss: 0.6843 - accuracy: 0.5504
76/338 [=====>........................] - ETA: 14s - loss: 0.6842 - accuracy: 0.5518
77/338 [=====>........................] - ETA: 14s - loss: 0.6841 - accuracy: 0.5536
78/338 [=====>........................] - ETA: 14s - loss: 0.6841 - accuracy: 0.5549
79/338 [======>.......................] - ETA: 14s - loss: 0.6841 - accuracy: 0.5562
80/338 [======>.......................] - ETA: 14s - loss: 0.6840 - accuracy: 0.5570
81/338 [======>.......................] - ETA: 14s - loss: 0.6838 - accuracy: 0.5590
82/338 [======>.......................] - ETA: 14s - loss: 0.6837 - accuracy: 0.5606
83/338 [======>.......................] - ETA: 14s - loss: 0.6837 - accuracy: 0.5614
84/338 [======>.......................] - ETA: 14s - loss: 0.6837 - accuracy: 0.5621
85/338 [======>.......................] - ETA: 14s - loss: 0.6836 - accuracy: 0.5643
86/338 [======>.......................] - ETA: 13s - loss: 0.6838 - accuracy: 0.5640
87/338 [======>.......................] - ETA: 13s - loss: 0.6837 - accuracy: 0.5650
88/338 [======>.......................] - ETA: 13s - loss: 0.6837 - accuracy: 0.5657
89/338 [======>.......................] - ETA: 13s - loss: 0.6837 - accuracy: 0.5660
90/338 [======>.......................] - ETA: 13s - loss: 0.6840 - accuracy: 0.5649
91/338 [=======>......................] - ETA: 13s - loss: 0.6840 - accuracy: 0.5649
92/338 [=======>......................] - ETA: 13s - loss: 0.6838 - accuracy: 0.5666
93/338 [=======>......................] - ETA: 13s - loss: 0.6834 - accuracy: 0.5689
94/338 [=======>......................] - ETA: 13s - loss: 0.6834 - accuracy: 0.5688
95/338 [=======>......................] - ETA: 13s - loss: 0.6835 - accuracy: 0.5684
96/338 [=======>......................] - ETA: 13s - loss: 0.6835 - accuracy: 0.5693
97/338 [=======>......................] - ETA: 13s - loss: 0.6833 - accuracy: 0.5706
98/338 [=======>......................] - ETA: 13s - loss: 0.6832 - accuracy: 0.5721
99/338 [=======>......................] - ETA: 13s - loss: 0.6830 - accuracy: 0.5732
100/338 [=======>......................] - ETA: 13s - loss: 0.6830 - accuracy: 0.5741
101/338 [=======>......................] - ETA: 13s - loss: 0.6829 - accuracy: 0.5749
102/338 [========>.....................] - ETA: 13s - loss: 0.6829 - accuracy: 0.5751
103/338 [========>.....................] - ETA: 13s - loss: 0.6827 - accuracy: 0.5768
104/338 [========>.....................] - ETA: 12s - loss: 0.6827 - accuracy: 0.5775
105/338 [========>.....................] - ETA: 12s - loss: 0.6826 - accuracy: 0.5783
106/338 [========>.....................] - ETA: 12s - loss: 0.6826 - accuracy: 0.5781
107/338 [========>.....................] - ETA: 12s - loss: 0.6825 - accuracy: 0.5794
108/338 [========>.....................] - ETA: 12s - loss: 0.6824 - accuracy: 0.5802
109/338 [========>.....................] - ETA: 12s - loss: 0.6823 - accuracy: 0.5820
110/338 [========>.....................] - ETA: 12s - loss: 0.6820 - accuracy: 0.5844
111/338 [========>.....................] - ETA: 12s - loss: 0.6818 - accuracy: 0.5856
112/338 [========>.....................] - ETA: 12s - loss: 0.6819 - accuracy: 0.5862
113/338 [=========>....................] - ETA: 12s - loss: 0.6821 - accuracy: 0.5860
114/338 [=========>....................] - ETA: 12s - loss: 0.6822 - accuracy: 0.5855
115/338 [=========>....................] - ETA: 12s - loss: 0.6823 - accuracy: 0.5853
116/338 [=========>....................] - ETA: 12s - loss: 0.6822 - accuracy: 0.5862
117/338 [=========>....................] - ETA: 12s - loss: 0.6822 - accuracy: 0.5868
118/338 [=========>....................] - ETA: 12s - loss: 0.6822 - accuracy: 0.5869
119/338 [=========>....................] - ETA: 12s - loss: 0.6821 - accuracy: 0.5872
120/338 [=========>....................] - ETA: 12s - loss: 0.6820 - accuracy: 0.5883
121/338 [=========>....................] - ETA: 12s - loss: 0.6820 - accuracy: 0.5881
122/338 [=========>....................] - ETA: 11s - loss: 0.6820 - accuracy: 0.5879
123/338 [=========>....................] - ETA: 11s - loss: 0.6822 - accuracy: 0.5877
124/338 [==========>...................] - ETA: 11s - loss: 0.6823 - accuracy: 0.5872
125/338 [==========>...................] - ETA: 11s - loss: 0.6822 - accuracy: 0.5880
126/338 [==========>...................] - ETA: 11s - loss: 0.6821 - accuracy: 0.5890
127/338 [==========>...................] - ETA: 11s - loss: 0.6820 - accuracy: 0.5896
128/338 [==========>...................] - ETA: 11s - loss: 0.6820 - accuracy: 0.5896
129/338 [==========>...................] - ETA: 11s - loss: 0.6823 - accuracy: 0.5882
130/338 [==========>...................] - ETA: 11s - loss: 0.6824 - accuracy: 0.5882
131/338 [==========>...................] - ETA: 11s - loss: 0.6826 - accuracy: 0.5873
132/338 [==========>...................] - ETA: 11s - loss: 0.6825 - accuracy: 0.5881
133/338 [==========>...................] - ETA: 11s - loss: 0.6824 - accuracy: 0.5891
134/338 [==========>...................] - ETA: 11s - loss: 0.6825 - accuracy: 0.5886
135/338 [==========>...................] - ETA: 11s - loss: 0.6826 - accuracy: 0.5882
136/338 [===========>..................] - ETA: 11s - loss: 0.6826 - accuracy: 0.5880
137/338 [===========>..................] - ETA: 11s - loss: 0.6824 - accuracy: 0.5892
138/338 [===========>..................] - ETA: 11s - loss: 0.6825 - accuracy: 0.5885
139/338 [===========>..................] - ETA: 11s - loss: 0.6825 - accuracy: 0.5888
140/338 [===========>..................] - ETA: 10s - loss: 0.6824 - accuracy: 0.5897
141/338 [===========>..................] - ETA: 10s - loss: 0.6823 - accuracy: 0.5904
142/338 [===========>..................] - ETA: 10s - loss: 0.6822 - accuracy: 0.5918
143/338 [===========>..................] - ETA: 10s - loss: 0.6821 - accuracy: 0.5924
144/338 [===========>..................] - ETA: 10s - loss: 0.6821 - accuracy: 0.5918
145/338 [===========>..................] - ETA: 10s - loss: 0.6821 - accuracy: 0.5918
146/338 [===========>..................] - ETA: 10s - loss: 0.6821 - accuracy: 0.5920
147/338 [============>.................] - ETA: 10s - loss: 0.6822 - accuracy: 0.5914
148/338 [============>.................] - ETA: 10s - loss: 0.6821 - accuracy: 0.5927
149/338 [============>.................] - ETA: 10s - loss: 0.6820 - accuracy: 0.5929
150/338 [============>.................] - ETA: 10s - loss: 0.6821 - accuracy: 0.5929
151/338 [============>.................] - ETA: 10s - loss: 0.6822 - accuracy: 0.5925
152/338 [============>.................] - ETA: 10s - loss: 0.6821 - accuracy: 0.5929
153/338 [============>.................] - ETA: 10s - loss: 0.6820 - accuracy: 0.5933
154/338 [============>.................] - ETA: 10s - loss: 0.6818 - accuracy: 0.5938
155/338 [============>.................] - ETA: 10s - loss: 0.6818 - accuracy: 0.5944
156/338 [============>.................] - ETA: 10s - loss: 0.6817 - accuracy: 0.5946
157/338 [============>.................] - ETA: 10s - loss: 0.6816 - accuracy: 0.5955
158/338 [=============>................] - ETA: 9s - loss: 0.6815 - accuracy: 0.5963
159/338 [=============>................] - ETA: 9s - loss: 0.6816 - accuracy: 0.5963
160/338 [=============>................] - ETA: 9s - loss: 0.6815 - accuracy: 0.5969
161/338 [=============>................] - ETA: 9s - loss: 0.6815 - accuracy: 0.5967
162/338 [=============>................] - ETA: 9s - loss: 0.6814 - accuracy: 0.5970
163/338 [=============>................] - ETA: 9s - loss: 0.6816 - accuracy: 0.5962
164/338 [=============>................] - ETA: 9s - loss: 0.6816 - accuracy: 0.5966
165/338 [=============>................] - ETA: 9s - loss: 0.6816 - accuracy: 0.5970
166/338 [=============>................] - ETA: 9s - loss: 0.6815 - accuracy: 0.5971
167/338 [=============>................] - ETA: 9s - loss: 0.6814 - accuracy: 0.5982
168/338 [=============>................] - ETA: 9s - loss: 0.6814 - accuracy: 0.5988
169/338 [==============>...............] - ETA: 9s - loss: 0.6814 - accuracy: 0.5993
170/338 [==============>...............] - ETA: 9s - loss: 0.6814 - accuracy: 0.5996
171/338 [==============>...............] - ETA: 9s - loss: 0.6815 - accuracy: 0.5990
172/338 [==============>...............] - ETA: 9s - loss: 0.6814 - accuracy: 0.5996
173/338 [==============>...............] - ETA: 9s - loss: 0.6814 - accuracy: 0.6001
174/338 [==============>...............] - ETA: 9s - loss: 0.6813 - accuracy: 0.6006
175/338 [==============>...............] - ETA: 9s - loss: 0.6811 - accuracy: 0.6014
176/338 [==============>...............] - ETA: 8s - loss: 0.6810 - accuracy: 0.6021
177/338 [==============>...............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6028
178/338 [==============>...............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6027
179/338 [==============>...............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6027
180/338 [==============>...............] - ETA: 8s - loss: 0.6810 - accuracy: 0.6023
181/338 [===============>..............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6029
182/338 [===============>..............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6025
183/338 [===============>..............] - ETA: 8s - loss: 0.6808 - accuracy: 0.6023
184/338 [===============>..............] - ETA: 8s - loss: 0.6808 - accuracy: 0.6029
185/338 [===============>..............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6027
186/338 [===============>..............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6025
187/338 [===============>..............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6031
188/338 [===============>..............] - ETA: 8s - loss: 0.6808 - accuracy: 0.6032
189/338 [===============>..............] - ETA: 8s - loss: 0.6807 - accuracy: 0.6038
190/338 [===============>..............] - ETA: 8s - loss: 0.6806 - accuracy: 0.6043
191/338 [===============>..............] - ETA: 8s - loss: 0.6806 - accuracy: 0.6045
192/338 [================>.............] - ETA: 8s - loss: 0.6806 - accuracy: 0.6042
193/338 [================>.............] - ETA: 8s - loss: 0.6809 - accuracy: 0.6031
194/338 [================>.............] - ETA: 7s - loss: 0.6807 - accuracy: 0.6039
195/338 [================>.............] - ETA: 7s - loss: 0.6808 - accuracy: 0.6038
196/338 [================>.............] - ETA: 7s - loss: 0.6807 - accuracy: 0.6041
197/338 [================>.............] - ETA: 7s - loss: 0.6806 - accuracy: 0.6044
198/338 [================>.............] - ETA: 7s - loss: 0.6806 - accuracy: 0.6046
199/338 [================>.............] - ETA: 7s - loss: 0.6807 - accuracy: 0.6047
200/338 [================>.............] - ETA: 7s - loss: 0.6808 - accuracy: 0.6041
201/338 [================>.............] - ETA: 7s - loss: 0.6807 - accuracy: 0.6045
202/338 [================>.............] - ETA: 7s - loss: 0.6807 - accuracy: 0.6046
203/338 [=================>............] - ETA: 7s - loss: 0.6806 - accuracy: 0.6053
204/338 [=================>............] - ETA: 7s - loss: 0.6805 - accuracy: 0.6062
205/338 [=================>............] - ETA: 7s - loss: 0.6805 - accuracy: 0.6059
206/338 [=================>............] - ETA: 7s - loss: 0.6805 - accuracy: 0.6062
207/338 [=================>............] - ETA: 7s - loss: 0.6804 - accuracy: 0.6066
208/338 [=================>............] - ETA: 7s - loss: 0.6804 - accuracy: 0.6067
209/338 [=================>............] - ETA: 7s - loss: 0.6805 - accuracy: 0.6062
210/338 [=================>............] - ETA: 7s - loss: 0.6804 - accuracy: 0.6065
211/338 [=================>............] - ETA: 7s - loss: 0.6804 - accuracy: 0.6069
212/338 [=================>............] - ETA: 6s - loss: 0.6805 - accuracy: 0.6066
213/338 [=================>............] - ETA: 6s - loss: 0.6805 - accuracy: 0.6064
214/338 [=================>............] - ETA: 6s - loss: 0.6804 - accuracy: 0.6067
215/338 [==================>...........] - ETA: 6s - loss: 0.6804 - accuracy: 0.6070
216/338 [==================>...........] - ETA: 6s - loss: 0.6803 - accuracy: 0.6072
217/338 [==================>...........] - ETA: 6s - loss: 0.6803 - accuracy: 0.6071
218/338 [==================>...........] - ETA: 6s - loss: 0.6802 - accuracy: 0.6074
219/338 [==================>...........] - ETA: 6s - loss: 0.6801 - accuracy: 0.6079
220/338 [==================>...........] - ETA: 6s - loss: 0.6801 - accuracy: 0.6081
221/338 [==================>...........] - ETA: 6s - loss: 0.6801 - accuracy: 0.6079
222/338 [==================>...........] - ETA: 6s - loss: 0.6802 - accuracy: 0.6071
223/338 [==================>...........] - ETA: 6s - loss: 0.6803 - accuracy: 0.6071
224/338 [==================>...........] - ETA: 6s - loss: 0.6802 - accuracy: 0.6076
225/338 [==================>...........] - ETA: 6s - loss: 0.6803 - accuracy: 0.6072
226/338 [===================>..........] - ETA: 6s - loss: 0.6804 - accuracy: 0.6065
227/338 [===================>..........] - ETA: 6s - loss: 0.6805 - accuracy: 0.6063
228/338 [===================>..........] - ETA: 6s - loss: 0.6804 - accuracy: 0.6069
229/338 [===================>..........] - ETA: 6s - loss: 0.6804 - accuracy: 0.6067
230/338 [===================>..........] - ETA: 5s - loss: 0.6805 - accuracy: 0.6064
231/338 [===================>..........] - ETA: 5s - loss: 0.6806 - accuracy: 0.6055
232/338 [===================>..........] - ETA: 5s - loss: 0.6804 - accuracy: 0.6059
233/338 [===================>..........] - ETA: 5s - loss: 0.6806 - accuracy: 0.6053
234/338 [===================>..........] - ETA: 5s - loss: 0.6805 - accuracy: 0.6060
235/338 [===================>..........] - ETA: 5s - loss: 0.6805 - accuracy: 0.6060
236/338 [===================>..........] - ETA: 5s - loss: 0.6804 - accuracy: 0.6061
237/338 [====================>.........] - ETA: 5s - loss: 0.6804 - accuracy: 0.6060
238/338 [====================>.........] - ETA: 5s - loss: 0.6804 - accuracy: 0.6065
239/338 [====================>.........] - ETA: 5s - loss: 0.6803 - accuracy: 0.6071
240/338 [====================>.........] - ETA: 5s - loss: 0.6804 - accuracy: 0.6068
241/338 [====================>.........] - ETA: 5s - loss: 0.6804 - accuracy: 0.6062
242/338 [====================>.........] - ETA: 5s - loss: 0.6803 - accuracy: 0.6067
243/338 [====================>.........] - ETA: 5s - loss: 0.6803 - accuracy: 0.6071
244/338 [====================>.........] - ETA: 5s - loss: 0.6804 - accuracy: 0.6069
245/338 [====================>.........] - ETA: 5s - loss: 0.6803 - accuracy: 0.6075
246/338 [====================>.........] - ETA: 5s - loss: 0.6802 - accuracy: 0.6081
247/338 [====================>.........] - ETA: 5s - loss: 0.6801 - accuracy: 0.6084
248/338 [=====================>........] - ETA: 4s - loss: 0.6800 - accuracy: 0.6092
249/338 [=====================>........] - ETA: 4s - loss: 0.6798 - accuracy: 0.6099
250/338 [=====================>........] - ETA: 4s - loss: 0.6797 - accuracy: 0.6105
251/338 [=====================>........] - ETA: 4s - loss: 0.6797 - accuracy: 0.6104
252/338 [=====================>........] - ETA: 4s - loss: 0.6796 - accuracy: 0.6105
253/338 [=====================>........] - ETA: 4s - loss: 0.6795 - accuracy: 0.6113
254/338 [=====================>........] - ETA: 4s - loss: 0.6795 - accuracy: 0.6116
255/338 [=====================>........] - ETA: 4s - loss: 0.6794 - accuracy: 0.6120
256/338 [=====================>........] - ETA: 4s - loss: 0.6794 - accuracy: 0.6119
257/338 [=====================>........] - ETA: 4s - loss: 0.6794 - accuracy: 0.6122
258/338 [=====================>........] - ETA: 4s - loss: 0.6794 - accuracy: 0.6119
259/338 [=====================>........] - ETA: 4s - loss: 0.6794 - accuracy: 0.6122
260/338 [======================>.......] - ETA: 4s - loss: 0.6793 - accuracy: 0.6125
261/338 [======================>.......] - ETA: 4s - loss: 0.6793 - accuracy: 0.6128
262/338 [======================>.......] - ETA: 4s - loss: 0.6793 - accuracy: 0.6128
263/338 [======================>.......] - ETA: 4s - loss: 0.6792 - accuracy: 0.6131
264/338 [======================>.......] - ETA: 4s - loss: 0.6793 - accuracy: 0.6126
265/338 [======================>.......] - ETA: 4s - loss: 0.6792 - accuracy: 0.6132
266/338 [======================>.......] - ETA: 3s - loss: 0.6791 - accuracy: 0.6134
267/338 [======================>.......] - ETA: 3s - loss: 0.6792 - accuracy: 0.6133
268/338 [======================>.......] - ETA: 3s - loss: 0.6791 - accuracy: 0.6137
269/338 [======================>.......] - ETA: 3s - loss: 0.6791 - accuracy: 0.6137
270/338 [======================>.......] - ETA: 3s - loss: 0.6790 - accuracy: 0.6140
271/338 [=======================>......] - ETA: 3s - loss: 0.6790 - accuracy: 0.6136
272/338 [=======================>......] - ETA: 3s - loss: 0.6790 - accuracy: 0.6139
273/338 [=======================>......] - ETA: 3s - loss: 0.6790 - accuracy: 0.6138
274/338 [=======================>......] - ETA: 3s - loss: 0.6790 - accuracy: 0.6139
275/338 [=======================>......] - ETA: 3s - loss: 0.6790 - accuracy: 0.6142
276/338 [=======================>......] - ETA: 3s - loss: 0.6789 - accuracy: 0.6144
277/338 [=======================>......] - ETA: 3s - loss: 0.6789 - accuracy: 0.6142
278/338 [=======================>......] - ETA: 3s - loss: 0.6789 - accuracy: 0.6142
279/338 [=======================>......] - ETA: 3s - loss: 0.6789 - accuracy: 0.6145
280/338 [=======================>......] - ETA: 3s - loss: 0.6789 - accuracy: 0.6145
281/338 [=======================>......] - ETA: 3s - loss: 0.6789 - accuracy: 0.6145
282/338 [========================>.....] - ETA: 3s - loss: 0.6789 - accuracy: 0.6145
283/338 [========================>.....] - ETA: 3s - loss: 0.6789 - accuracy: 0.6143
284/338 [========================>.....] - ETA: 2s - loss: 0.6788 - accuracy: 0.6148
285/338 [========================>.....] - ETA: 2s - loss: 0.6789 - accuracy: 0.6143
286/338 [========================>.....] - ETA: 2s - loss: 0.6789 - accuracy: 0.6144
287/338 [========================>.....] - ETA: 2s - loss: 0.6788 - accuracy: 0.6147
288/338 [========================>.....] - ETA: 2s - loss: 0.6788 - accuracy: 0.6148
289/338 [========================>.....] - ETA: 2s - loss: 0.6787 - accuracy: 0.6152
290/338 [========================>.....] - ETA: 2s - loss: 0.6786 - accuracy: 0.6155
291/338 [========================>.....] - ETA: 2s - loss: 0.6785 - accuracy: 0.6158
292/338 [========================>.....] - ETA: 2s - loss: 0.6785 - accuracy: 0.6159
293/338 [=========================>....] - ETA: 2s - loss: 0.6785 - accuracy: 0.6163
294/338 [=========================>....] - ETA: 2s - loss: 0.6785 - accuracy: 0.6164
295/338 [=========================>....] - ETA: 2s - loss: 0.6785 - accuracy: 0.6161
296/338 [=========================>....] - ETA: 2s - loss: 0.6785 - accuracy: 0.6159
297/338 [=========================>....] - ETA: 2s - loss: 0.6786 - accuracy: 0.6154
298/338 [=========================>....] - ETA: 2s - loss: 0.6786 - accuracy: 0.6156
299/338 [=========================>....] - ETA: 2s - loss: 0.6786 - accuracy: 0.6155
300/338 [=========================>....] - ETA: 2s - loss: 0.6786 - accuracy: 0.6156
301/338 [=========================>....] - ETA: 2s - loss: 0.6786 - accuracy: 0.6156
302/338 [=========================>....] - ETA: 1s - loss: 0.6785 - accuracy: 0.6158
303/338 [=========================>....] - ETA: 1s - loss: 0.6785 - accuracy: 0.6156
304/338 [=========================>....] - ETA: 1s - loss: 0.6785 - accuracy: 0.6157
305/338 [==========================>...] - ETA: 1s - loss: 0.6785 - accuracy: 0.6158
306/338 [==========================>...] - ETA: 1s - loss: 0.6785 - accuracy: 0.6156
307/338 [==========================>...] - ETA: 1s - loss: 0.6784 - accuracy: 0.6159
308/338 [==========================>...] - ETA: 1s - loss: 0.6783 - accuracy: 0.6164
309/338 [==========================>...] - ETA: 1s - loss: 0.6784 - accuracy: 0.6159
310/338 [==========================>...] - ETA: 1s - loss: 0.6783 - accuracy: 0.6164
311/338 [==========================>...] - ETA: 1s - loss: 0.6783 - accuracy: 0.6164
312/338 [==========================>...] - ETA: 1s - loss: 0.6784 - accuracy: 0.6161
313/338 [==========================>...] - ETA: 1s - loss: 0.6783 - accuracy: 0.6163
314/338 [==========================>...] - ETA: 1s - loss: 0.6783 - accuracy: 0.6163
315/338 [==========================>...] - ETA: 1s - loss: 0.6782 - accuracy: 0.6166
316/338 [===========================>..] - ETA: 1s - loss: 0.6782 - accuracy: 0.6164
317/338 [===========================>..] - ETA: 1s - loss: 0.6782 - accuracy: 0.6163
318/338 [===========================>..] - ETA: 1s - loss: 0.6782 - accuracy: 0.6162
319/338 [===========================>..] - ETA: 1s - loss: 0.6782 - accuracy: 0.6164
320/338 [===========================>..] - ETA: 0s - loss: 0.6782 - accuracy: 0.6166
321/338 [===========================>..] - ETA: 0s - loss: 0.6781 - accuracy: 0.6166
322/338 [===========================>..] - ETA: 0s - loss: 0.6780 - accuracy: 0.6169
323/338 [===========================>..] - ETA: 0s - loss: 0.6780 - accuracy: 0.6173
324/338 [===========================>..] - ETA: 0s - loss: 0.6780 - accuracy: 0.6173
325/338 [===========================>..] - ETA: 0s - loss: 0.6780 - accuracy: 0.6176
326/338 [===========================>..] - ETA: 0s - loss: 0.6780 - accuracy: 0.6170
327/338 [============================>.] - ETA: 0s - loss: 0.6780 - accuracy: 0.6170
328/338 [============================>.] - ETA: 0s - loss: 0.6780 - accuracy: 0.6171
329/338 [============================>.] - ETA: 0s - loss: 0.6780 - accuracy: 0.6170
330/338 [============================>.] - ETA: 0s - loss: 0.6780 - accuracy: 0.6170
331/338 [============================>.] - ETA: 0s - loss: 0.6779 - accuracy: 0.6174
332/338 [============================>.] - ETA: 0s - loss: 0.6779 - accuracy: 0.6173
333/338 [============================>.] - ETA: 0s - loss: 0.6779 - accuracy: 0.6174
334/338 [============================>.] - ETA: 0s - loss: 0.6778 - accuracy: 0.6176
335/338 [============================>.] - ETA: 0s - loss: 0.6779 - accuracy: 0.6176
336/338 [============================>.] - ETA: 0s - loss: 0.6779 - accuracy: 0.6176
337/338 [============================>.] - ETA: 0s - loss: 0.6778 - accuracy: 0.6179
338/338 [==============================] - 20s 60ms/step - loss: 0.6778 - accuracy: 0.6178 - val_loss: 0.6953 - val_accuracy: 0.5135
Epoch 3/5
1/338 [..............................] - ETA: 18s - loss: 0.6356 - accuracy: 0.7812
2/338 [..............................] - ETA: 18s - loss: 0.6423 - accuracy: 0.7969
3/338 [..............................] - ETA: 18s - loss: 0.6508 - accuracy: 0.7500
4/338 [..............................] - ETA: 18s - loss: 0.6572 - accuracy: 0.7344
5/338 [..............................] - ETA: 18s - loss: 0.6610 - accuracy: 0.7063
6/338 [..............................] - ETA: 18s - loss: 0.6605 - accuracy: 0.7083
7/338 [..............................] - ETA: 18s - loss: 0.6629 - accuracy: 0.6964
8/338 [..............................] - ETA: 18s - loss: 0.6667 - accuracy: 0.6797
9/338 [..............................] - ETA: 18s - loss: 0.6695 - accuracy: 0.6632
10/338 [..............................] - ETA: 18s - loss: 0.6662 - accuracy: 0.6750
11/338 [..............................] - ETA: 18s - loss: 0.6676 - accuracy: 0.6648
12/338 [>.............................] - ETA: 18s - loss: 0.6687 - accuracy: 0.6562
13/338 [>.............................] - ETA: 18s - loss: 0.6695 - accuracy: 0.6490
14/338 [>.............................] - ETA: 18s - loss: 0.6705 - accuracy: 0.6473
15/338 [>.............................] - ETA: 17s - loss: 0.6720 - accuracy: 0.6417
16/338 [>.............................] - ETA: 17s - loss: 0.6701 - accuracy: 0.6484
17/338 [>.............................] - ETA: 17s - loss: 0.6687 - accuracy: 0.6526
18/338 [>.............................] - ETA: 17s - loss: 0.6700 - accuracy: 0.6458
19/338 [>.............................] - ETA: 17s - loss: 0.6698 - accuracy: 0.6464
20/338 [>.............................] - ETA: 17s - loss: 0.6701 - accuracy: 0.6469
21/338 [>.............................] - ETA: 17s - loss: 0.6700 - accuracy: 0.6488
22/338 [>.............................] - ETA: 17s - loss: 0.6709 - accuracy: 0.6435
23/338 [=>............................] - ETA: 17s - loss: 0.6728 - accuracy: 0.6345
24/338 [=>............................] - ETA: 17s - loss: 0.6727 - accuracy: 0.6354
25/338 [=>............................] - ETA: 17s - loss: 0.6722 - accuracy: 0.6363
26/338 [=>............................] - ETA: 17s - loss: 0.6725 - accuracy: 0.6346
27/338 [=>............................] - ETA: 17s - loss: 0.6723 - accuracy: 0.6343
28/338 [=>............................] - ETA: 17s - loss: 0.6717 - accuracy: 0.6362
29/338 [=>............................] - ETA: 17s - loss: 0.6728 - accuracy: 0.6315
30/338 [=>............................] - ETA: 17s - loss: 0.6726 - accuracy: 0.6333
31/338 [=>............................] - ETA: 17s - loss: 0.6723 - accuracy: 0.6341
32/338 [=>............................] - ETA: 16s - loss: 0.6717 - accuracy: 0.6377
33/338 [=>............................] - ETA: 16s - loss: 0.6716 - accuracy: 0.6373
34/338 [==>...........................] - ETA: 16s - loss: 0.6712 - accuracy: 0.6397
35/338 [==>...........................] - ETA: 16s - loss: 0.6707 - accuracy: 0.6420
36/338 [==>...........................] - ETA: 16s - loss: 0.6703 - accuracy: 0.6432
37/338 [==>...........................] - ETA: 16s - loss: 0.6695 - accuracy: 0.6470
38/338 [==>...........................] - ETA: 16s - loss: 0.6695 - accuracy: 0.6456
39/338 [==>...........................] - ETA: 16s - loss: 0.6698 - accuracy: 0.6450
40/338 [==>...........................] - ETA: 16s - loss: 0.6705 - accuracy: 0.6422
41/338 [==>...........................] - ETA: 16s - loss: 0.6703 - accuracy: 0.6425
42/338 [==>...........................] - ETA: 16s - loss: 0.6697 - accuracy: 0.6436
43/338 [==>...........................] - ETA: 16s - loss: 0.6699 - accuracy: 0.6432
44/338 [==>...........................] - ETA: 16s - loss: 0.6697 - accuracy: 0.6428
45/338 [==>...........................] - ETA: 16s - loss: 0.6694 - accuracy: 0.6438
46/338 [===>..........................] - ETA: 16s - loss: 0.6693 - accuracy: 0.6447
47/338 [===>..........................] - ETA: 16s - loss: 0.6691 - accuracy: 0.6449
48/338 [===>..........................] - ETA: 16s - loss: 0.6690 - accuracy: 0.6452
49/338 [===>..........................] - ETA: 16s - loss: 0.6682 - accuracy: 0.6486
50/338 [===>..........................] - ETA: 16s - loss: 0.6684 - accuracy: 0.6463
51/338 [===>..........................] - ETA: 15s - loss: 0.6687 - accuracy: 0.6440
52/338 [===>..........................] - ETA: 15s - loss: 0.6687 - accuracy: 0.6442
53/338 [===>..........................] - ETA: 15s - loss: 0.6691 - accuracy: 0.6427
54/338 [===>..........................] - ETA: 15s - loss: 0.6689 - accuracy: 0.6447
55/338 [===>..........................] - ETA: 15s - loss: 0.6689 - accuracy: 0.6455
56/338 [===>..........................] - ETA: 15s - loss: 0.6681 - accuracy: 0.6484
57/338 [====>.........................] - ETA: 15s - loss: 0.6682 - accuracy: 0.6480
58/338 [====>.........................] - ETA: 15s - loss: 0.6680 - accuracy: 0.6487
59/338 [====>.........................] - ETA: 15s - loss: 0.6687 - accuracy: 0.6451
60/338 [====>.........................] - ETA: 15s - loss: 0.6685 - accuracy: 0.6458
61/338 [====>.........................] - ETA: 15s - loss: 0.6683 - accuracy: 0.6465
62/338 [====>.........................] - ETA: 15s - loss: 0.6687 - accuracy: 0.6447
63/338 [====>.........................] - ETA: 15s - loss: 0.6686 - accuracy: 0.6448
64/338 [====>.........................] - ETA: 15s - loss: 0.6682 - accuracy: 0.6465
65/338 [====>.........................] - ETA: 15s - loss: 0.6679 - accuracy: 0.6481
66/338 [====>.........................] - ETA: 15s - loss: 0.6678 - accuracy: 0.6482
67/338 [====>.........................] - ETA: 15s - loss: 0.6681 - accuracy: 0.6465
68/338 [=====>........................] - ETA: 15s - loss: 0.6685 - accuracy: 0.6448
69/338 [=====>........................] - ETA: 14s - loss: 0.6683 - accuracy: 0.6458
70/338 [=====>........................] - ETA: 14s - loss: 0.6684 - accuracy: 0.6451
71/338 [=====>........................] - ETA: 14s - loss: 0.6677 - accuracy: 0.6474
72/338 [=====>........................] - ETA: 14s - loss: 0.6680 - accuracy: 0.6458
73/338 [=====>........................] - ETA: 14s - loss: 0.6679 - accuracy: 0.6464
74/338 [=====>........................] - ETA: 14s - loss: 0.6678 - accuracy: 0.6465
75/338 [=====>........................] - ETA: 14s - loss: 0.6681 - accuracy: 0.6454
76/338 [=====>........................] - ETA: 14s - loss: 0.6683 - accuracy: 0.6443
77/338 [=====>........................] - ETA: 14s - loss: 0.6683 - accuracy: 0.6445
78/338 [=====>........................] - ETA: 14s - loss: 0.6678 - accuracy: 0.6462
79/338 [======>.......................] - ETA: 14s - loss: 0.6680 - accuracy: 0.6452
80/338 [======>.......................] - ETA: 14s - loss: 0.6682 - accuracy: 0.6449
81/338 [======>.......................] - ETA: 14s - loss: 0.6676 - accuracy: 0.6470
82/338 [======>.......................] - ETA: 14s - loss: 0.6674 - accuracy: 0.6479
83/338 [======>.......................] - ETA: 14s - loss: 0.6674 - accuracy: 0.6476
84/338 [======>.......................] - ETA: 14s - loss: 0.6675 - accuracy: 0.6473
85/338 [======>.......................] - ETA: 14s - loss: 0.6673 - accuracy: 0.6478
86/338 [======>.......................] - ETA: 13s - loss: 0.6679 - accuracy: 0.6457
87/338 [======>.......................] - ETA: 13s - loss: 0.6681 - accuracy: 0.6448
88/338 [======>.......................] - ETA: 13s - loss: 0.6688 - accuracy: 0.6424
89/338 [======>.......................] - ETA: 13s - loss: 0.6690 - accuracy: 0.6422
90/338 [======>.......................] - ETA: 13s - loss: 0.6692 - accuracy: 0.6413
91/338 [=======>......................] - ETA: 13s - loss: 0.6694 - accuracy: 0.6405
92/338 [=======>......................] - ETA: 13s - loss: 0.6687 - accuracy: 0.6427
93/338 [=======>......................] - ETA: 13s - loss: 0.6685 - accuracy: 0.6435
94/338 [=======>......................] - ETA: 13s - loss: 0.6682 - accuracy: 0.6443
95/338 [=======>......................] - ETA: 13s - loss: 0.6678 - accuracy: 0.6457
96/338 [=======>......................] - ETA: 13s - loss: 0.6677 - accuracy: 0.6462
97/338 [=======>......................] - ETA: 13s - loss: 0.6675 - accuracy: 0.6466
98/338 [=======>......................] - ETA: 13s - loss: 0.6677 - accuracy: 0.6460
99/338 [=======>......................] - ETA: 13s - loss: 0.6677 - accuracy: 0.6458
100/338 [=======>......................] - ETA: 13s - loss: 0.6678 - accuracy: 0.6456
101/338 [=======>......................] - ETA: 13s - loss: 0.6671 - accuracy: 0.6482
102/338 [========>.....................] - ETA: 13s - loss: 0.6670 - accuracy: 0.6483
103/338 [========>.....................] - ETA: 13s - loss: 0.6668 - accuracy: 0.6487
104/338 [========>.....................] - ETA: 13s - loss: 0.6668 - accuracy: 0.6484
105/338 [========>.....................] - ETA: 12s - loss: 0.6668 - accuracy: 0.6485
106/338 [========>.....................] - ETA: 12s - loss: 0.6668 - accuracy: 0.6483
107/338 [========>.....................] - ETA: 12s - loss: 0.6668 - accuracy: 0.6484
108/338 [========>.....................] - ETA: 12s - loss: 0.6665 - accuracy: 0.6496
109/338 [========>.....................] - ETA: 12s - loss: 0.6666 - accuracy: 0.6494
110/338 [========>.....................] - ETA: 12s - loss: 0.6669 - accuracy: 0.6486
111/338 [========>.....................] - ETA: 12s - loss: 0.6671 - accuracy: 0.6478
112/338 [========>.....................] - ETA: 12s - loss: 0.6669 - accuracy: 0.6482
113/338 [=========>....................] - ETA: 12s - loss: 0.6671 - accuracy: 0.6474
114/338 [=========>....................] - ETA: 12s - loss: 0.6671 - accuracy: 0.6472
115/338 [=========>....................] - ETA: 12s - loss: 0.6673 - accuracy: 0.6462
116/338 [=========>....................] - ETA: 12s - loss: 0.6669 - accuracy: 0.6474
117/338 [=========>....................] - ETA: 12s - loss: 0.6669 - accuracy: 0.6474
118/338 [=========>....................] - ETA: 12s - loss: 0.6666 - accuracy: 0.6486
119/338 [=========>....................] - ETA: 12s - loss: 0.6667 - accuracy: 0.6481
120/338 [=========>....................] - ETA: 12s - loss: 0.6666 - accuracy: 0.6487
121/338 [=========>....................] - ETA: 12s - loss: 0.6663 - accuracy: 0.6493
122/338 [=========>....................] - ETA: 11s - loss: 0.6667 - accuracy: 0.6483
123/338 [=========>....................] - ETA: 11s - loss: 0.6670 - accuracy: 0.6474
124/338 [==========>...................] - ETA: 11s - loss: 0.6671 - accuracy: 0.6467
125/338 [==========>...................] - ETA: 11s - loss: 0.6671 - accuracy: 0.6465
126/338 [==========>...................] - ETA: 11s - loss: 0.6670 - accuracy: 0.6468
127/338 [==========>...................] - ETA: 11s - loss: 0.6667 - accuracy: 0.6479
128/338 [==========>...................] - ETA: 11s - loss: 0.6665 - accuracy: 0.6487
129/338 [==========>...................] - ETA: 11s - loss: 0.6664 - accuracy: 0.6487
130/338 [==========>...................] - ETA: 11s - loss: 0.6663 - accuracy: 0.6493
131/338 [==========>...................] - ETA: 11s - loss: 0.6661 - accuracy: 0.6498
132/338 [==========>...................] - ETA: 11s - loss: 0.6663 - accuracy: 0.6494
133/338 [==========>...................] - ETA: 11s - loss: 0.6660 - accuracy: 0.6501
134/338 [==========>...................] - ETA: 11s - loss: 0.6665 - accuracy: 0.6486
135/338 [==========>...................] - ETA: 11s - loss: 0.6665 - accuracy: 0.6486
136/338 [===========>..................] - ETA: 11s - loss: 0.6666 - accuracy: 0.6482
137/338 [===========>..................] - ETA: 11s - loss: 0.6665 - accuracy: 0.6485
138/338 [===========>..................] - ETA: 11s - loss: 0.6661 - accuracy: 0.6497
139/338 [===========>..................] - ETA: 11s - loss: 0.6661 - accuracy: 0.6500
140/338 [===========>..................] - ETA: 10s - loss: 0.6660 - accuracy: 0.6502
141/338 [===========>..................] - ETA: 10s - loss: 0.6661 - accuracy: 0.6498
142/338 [===========>..................] - ETA: 10s - loss: 0.6659 - accuracy: 0.6507
143/338 [===========>..................] - ETA: 10s - loss: 0.6658 - accuracy: 0.6506
144/338 [===========>..................] - ETA: 10s - loss: 0.6657 - accuracy: 0.6508
145/338 [===========>..................] - ETA: 10s - loss: 0.6657 - accuracy: 0.6509
146/338 [===========>..................] - ETA: 10s - loss: 0.6658 - accuracy: 0.6505
147/338 [============>.................] - ETA: 10s - loss: 0.6658 - accuracy: 0.6503
148/338 [============>.................] - ETA: 10s - loss: 0.6658 - accuracy: 0.6503
149/338 [============>.................] - ETA: 10s - loss: 0.6660 - accuracy: 0.6497
150/338 [============>.................] - ETA: 10s - loss: 0.6658 - accuracy: 0.6502
151/338 [============>.................] - ETA: 10s - loss: 0.6656 - accuracy: 0.6507
152/338 [============>.................] - ETA: 10s - loss: 0.6659 - accuracy: 0.6499
153/338 [============>.................] - ETA: 10s - loss: 0.6659 - accuracy: 0.6497
154/338 [============>.................] - ETA: 10s - loss: 0.6661 - accuracy: 0.6489
155/338 [============>.................] - ETA: 10s - loss: 0.6663 - accuracy: 0.6482
156/338 [============>.................] - ETA: 10s - loss: 0.6664 - accuracy: 0.6474
157/338 [============>.................] - ETA: 10s - loss: 0.6661 - accuracy: 0.6483
158/338 [=============>................] - ETA: 9s - loss: 0.6661 - accuracy: 0.6483
159/338 [=============>................] - ETA: 9s - loss: 0.6661 - accuracy: 0.6482
160/338 [=============>................] - ETA: 9s - loss: 0.6663 - accuracy: 0.6475
161/338 [=============>................] - ETA: 9s - loss: 0.6662 - accuracy: 0.6477
162/338 [=============>................] - ETA: 9s - loss: 0.6662 - accuracy: 0.6478
163/338 [=============>................] - ETA: 9s - loss: 0.6662 - accuracy: 0.6478
164/338 [=============>................] - ETA: 9s - loss: 0.6662 - accuracy: 0.6479
165/338 [=============>................] - ETA: 9s - loss: 0.6661 - accuracy: 0.6481
166/338 [=============>................] - ETA: 9s - loss: 0.6665 - accuracy: 0.6466
167/338 [=============>................] - ETA: 9s - loss: 0.6664 - accuracy: 0.6471
168/338 [=============>................] - ETA: 9s - loss: 0.6663 - accuracy: 0.6471
169/338 [==============>...............] - ETA: 9s - loss: 0.6662 - accuracy: 0.6474
170/338 [==============>...............] - ETA: 9s - loss: 0.6664 - accuracy: 0.6465
171/338 [==============>...............] - ETA: 9s - loss: 0.6664 - accuracy: 0.6466
172/338 [==============>...............] - ETA: 9s - loss: 0.6667 - accuracy: 0.6455
173/338 [==============>...............] - ETA: 9s - loss: 0.6668 - accuracy: 0.6452
174/338 [==============>...............] - ETA: 9s - loss: 0.6668 - accuracy: 0.6451
175/338 [==============>...............] - ETA: 8s - loss: 0.6668 - accuracy: 0.6452
176/338 [==============>...............] - ETA: 8s - loss: 0.6668 - accuracy: 0.6452
177/338 [==============>...............] - ETA: 8s - loss: 0.6665 - accuracy: 0.6462
178/338 [==============>...............] - ETA: 8s - loss: 0.6663 - accuracy: 0.6466
179/338 [==============>...............] - ETA: 8s - loss: 0.6666 - accuracy: 0.6456
180/338 [==============>...............] - ETA: 8s - loss: 0.6665 - accuracy: 0.6460
181/338 [===============>..............] - ETA: 8s - loss: 0.6664 - accuracy: 0.6462
182/338 [===============>..............] - ETA: 8s - loss: 0.6663 - accuracy: 0.6465
183/338 [===============>..............] - ETA: 8s - loss: 0.6663 - accuracy: 0.6465
184/338 [===============>..............] - ETA: 8s - loss: 0.6662 - accuracy: 0.6464
185/338 [===============>..............] - ETA: 8s - loss: 0.6659 - accuracy: 0.6471
186/338 [===============>..............] - ETA: 8s - loss: 0.6662 - accuracy: 0.6463
187/338 [===============>..............] - ETA: 8s - loss: 0.6664 - accuracy: 0.6454
188/338 [===============>..............] - ETA: 8s - loss: 0.6664 - accuracy: 0.6451
189/338 [===============>..............] - ETA: 8s - loss: 0.6665 - accuracy: 0.6448
190/338 [===============>..............] - ETA: 8s - loss: 0.6666 - accuracy: 0.6444
191/338 [===============>..............] - ETA: 8s - loss: 0.6668 - accuracy: 0.6437
192/338 [================>.............] - ETA: 8s - loss: 0.6668 - accuracy: 0.6436
193/338 [================>.............] - ETA: 8s - loss: 0.6670 - accuracy: 0.6430
194/338 [================>.............] - ETA: 7s - loss: 0.6669 - accuracy: 0.6432
195/338 [================>.............] - ETA: 7s - loss: 0.6669 - accuracy: 0.6431
196/338 [================>.............] - ETA: 7s - loss: 0.6669 - accuracy: 0.6430
197/338 [================>.............] - ETA: 7s - loss: 0.6669 - accuracy: 0.6429
198/338 [================>.............] - ETA: 7s - loss: 0.6669 - accuracy: 0.6430
199/338 [================>.............] - ETA: 7s - loss: 0.6668 - accuracy: 0.6431
200/338 [================>.............] - ETA: 7s - loss: 0.6670 - accuracy: 0.6425
201/338 [================>.............] - ETA: 7s - loss: 0.6669 - accuracy: 0.6426
202/338 [================>.............] - ETA: 7s - loss: 0.6668 - accuracy: 0.6428
203/338 [=================>............] - ETA: 7s - loss: 0.6666 - accuracy: 0.6433
204/338 [=================>............] - ETA: 7s - loss: 0.6666 - accuracy: 0.6434
205/338 [=================>............] - ETA: 7s - loss: 0.6667 - accuracy: 0.6428
206/338 [=================>............] - ETA: 7s - loss: 0.6667 - accuracy: 0.6431
207/338 [=================>............] - ETA: 7s - loss: 0.6667 - accuracy: 0.6430
208/338 [=================>............] - ETA: 7s - loss: 0.6666 - accuracy: 0.6432
209/338 [=================>............] - ETA: 7s - loss: 0.6665 - accuracy: 0.6434
210/338 [=================>............] - ETA: 7s - loss: 0.6665 - accuracy: 0.6433
211/338 [=================>............] - ETA: 7s - loss: 0.6663 - accuracy: 0.6438
212/338 [=================>............] - ETA: 6s - loss: 0.6661 - accuracy: 0.6442
213/338 [=================>............] - ETA: 6s - loss: 0.6661 - accuracy: 0.6444
214/338 [=================>............] - ETA: 6s - loss: 0.6661 - accuracy: 0.6443
215/338 [==================>...........] - ETA: 6s - loss: 0.6660 - accuracy: 0.6443
216/338 [==================>...........] - ETA: 6s - loss: 0.6659 - accuracy: 0.6448
217/338 [==================>...........] - ETA: 6s - loss: 0.6658 - accuracy: 0.6449
218/338 [==================>...........] - ETA: 6s - loss: 0.6658 - accuracy: 0.6448
219/338 [==================>...........] - ETA: 6s - loss: 0.6659 - accuracy: 0.6444
220/338 [==================>...........] - ETA: 6s - loss: 0.6658 - accuracy: 0.6447
221/338 [==================>...........] - ETA: 6s - loss: 0.6657 - accuracy: 0.6447
222/338 [==================>...........] - ETA: 6s - loss: 0.6656 - accuracy: 0.6448
223/338 [==================>...........] - ETA: 6s - loss: 0.6657 - accuracy: 0.6448
224/338 [==================>...........] - ETA: 6s - loss: 0.6656 - accuracy: 0.6449
225/338 [==================>...........] - ETA: 6s - loss: 0.6657 - accuracy: 0.6446
226/338 [===================>..........] - ETA: 6s - loss: 0.6658 - accuracy: 0.6441
227/338 [===================>..........] - ETA: 6s - loss: 0.6658 - accuracy: 0.6440
228/338 [===================>..........] - ETA: 6s - loss: 0.6659 - accuracy: 0.6436
229/338 [===================>..........] - ETA: 6s - loss: 0.6660 - accuracy: 0.6433
230/338 [===================>..........] - ETA: 5s - loss: 0.6658 - accuracy: 0.6439
231/338 [===================>..........] - ETA: 5s - loss: 0.6657 - accuracy: 0.6441
232/338 [===================>..........] - ETA: 5s - loss: 0.6659 - accuracy: 0.6433
233/338 [===================>..........] - ETA: 5s - loss: 0.6659 - accuracy: 0.6432
234/338 [===================>..........] - ETA: 5s - loss: 0.6660 - accuracy: 0.6430
235/338 [===================>..........] - ETA: 5s - loss: 0.6659 - accuracy: 0.6435
236/338 [===================>..........] - ETA: 5s - loss: 0.6660 - accuracy: 0.6430
237/338 [====================>.........] - ETA: 5s - loss: 0.6660 - accuracy: 0.6431
238/338 [====================>.........] - ETA: 5s - loss: 0.6661 - accuracy: 0.6429
239/338 [====================>.........] - ETA: 5s - loss: 0.6661 - accuracy: 0.6425
240/338 [====================>.........] - ETA: 5s - loss: 0.6660 - accuracy: 0.6427
241/338 [====================>.........] - ETA: 5s - loss: 0.6661 - accuracy: 0.6425
242/338 [====================>.........] - ETA: 5s - loss: 0.6660 - accuracy: 0.6427
243/338 [====================>.........] - ETA: 5s - loss: 0.6660 - accuracy: 0.6426
244/338 [====================>.........] - ETA: 5s - loss: 0.6660 - accuracy: 0.6424
245/338 [====================>.........] - ETA: 5s - loss: 0.6659 - accuracy: 0.6427
246/338 [====================>.........] - ETA: 5s - loss: 0.6659 - accuracy: 0.6425
247/338 [====================>.........] - ETA: 5s - loss: 0.6658 - accuracy: 0.6427
248/338 [=====================>........] - ETA: 4s - loss: 0.6660 - accuracy: 0.6423
249/338 [=====================>........] - ETA: 4s - loss: 0.6659 - accuracy: 0.6423
250/338 [=====================>........] - ETA: 4s - loss: 0.6660 - accuracy: 0.6420
251/338 [=====================>........] - ETA: 4s - loss: 0.6661 - accuracy: 0.6419
252/338 [=====================>........] - ETA: 4s - loss: 0.6659 - accuracy: 0.6424
253/338 [=====================>........] - ETA: 4s - loss: 0.6660 - accuracy: 0.6420
254/338 [=====================>........] - ETA: 4s - loss: 0.6661 - accuracy: 0.6419
255/338 [=====================>........] - ETA: 4s - loss: 0.6660 - accuracy: 0.6419
256/338 [=====================>........] - ETA: 4s - loss: 0.6661 - accuracy: 0.6415
257/338 [=====================>........] - ETA: 4s - loss: 0.6661 - accuracy: 0.6417
258/338 [=====================>........] - ETA: 4s - loss: 0.6662 - accuracy: 0.6414
259/338 [=====================>........] - ETA: 4s - loss: 0.6661 - accuracy: 0.6413
260/338 [======================>.......] - ETA: 4s - loss: 0.6662 - accuracy: 0.6411
261/338 [======================>.......] - ETA: 4s - loss: 0.6661 - accuracy: 0.6413
262/338 [======================>.......] - ETA: 4s - loss: 0.6663 - accuracy: 0.6407
263/338 [======================>.......] - ETA: 4s - loss: 0.6664 - accuracy: 0.6403
264/338 [======================>.......] - ETA: 4s - loss: 0.6662 - accuracy: 0.6409
265/338 [======================>.......] - ETA: 4s - loss: 0.6663 - accuracy: 0.6407
266/338 [======================>.......] - ETA: 3s - loss: 0.6662 - accuracy: 0.6410
267/338 [======================>.......] - ETA: 3s - loss: 0.6661 - accuracy: 0.6413
268/338 [======================>.......] - ETA: 3s - loss: 0.6660 - accuracy: 0.6417
269/338 [======================>.......] - ETA: 3s - loss: 0.6658 - accuracy: 0.6422
270/338 [======================>.......] - ETA: 3s - loss: 0.6656 - accuracy: 0.6426
271/338 [=======================>......] - ETA: 3s - loss: 0.6656 - accuracy: 0.6424
272/338 [=======================>......] - ETA: 3s - loss: 0.6656 - accuracy: 0.6423
273/338 [=======================>......] - ETA: 3s - loss: 0.6655 - accuracy: 0.6425
274/338 [=======================>......] - ETA: 3s - loss: 0.6655 - accuracy: 0.6424
275/338 [=======================>......] - ETA: 3s - loss: 0.6655 - accuracy: 0.6425
276/338 [=======================>......] - ETA: 3s - loss: 0.6655 - accuracy: 0.6424
277/338 [=======================>......] - ETA: 3s - loss: 0.6653 - accuracy: 0.6431
278/338 [=======================>......] - ETA: 3s - loss: 0.6654 - accuracy: 0.6425
279/338 [=======================>......] - ETA: 3s - loss: 0.6653 - accuracy: 0.6428
280/338 [=======================>......] - ETA: 3s - loss: 0.6654 - accuracy: 0.6426
281/338 [=======================>......] - ETA: 3s - loss: 0.6656 - accuracy: 0.6421
282/338 [========================>.....] - ETA: 3s - loss: 0.6656 - accuracy: 0.6420
283/338 [========================>.....] - ETA: 3s - loss: 0.6657 - accuracy: 0.6416
284/338 [========================>.....] - ETA: 2s - loss: 0.6658 - accuracy: 0.6412
285/338 [========================>.....] - ETA: 2s - loss: 0.6657 - accuracy: 0.6414
286/338 [========================>.....] - ETA: 2s - loss: 0.6658 - accuracy: 0.6413
287/338 [========================>.....] - ETA: 2s - loss: 0.6659 - accuracy: 0.6409
288/338 [========================>.....] - ETA: 2s - loss: 0.6658 - accuracy: 0.6410
289/338 [========================>.....] - ETA: 2s - loss: 0.6657 - accuracy: 0.6413
290/338 [========================>.....] - ETA: 2s - loss: 0.6655 - accuracy: 0.6417
291/338 [========================>.....] - ETA: 2s - loss: 0.6654 - accuracy: 0.6422
292/338 [========================>.....] - ETA: 2s - loss: 0.6651 - accuracy: 0.6429
293/338 [=========================>....] - ETA: 2s - loss: 0.6652 - accuracy: 0.6427
294/338 [=========================>....] - ETA: 2s - loss: 0.6651 - accuracy: 0.6428
295/338 [=========================>....] - ETA: 2s - loss: 0.6651 - accuracy: 0.6429
296/338 [=========================>....] - ETA: 2s - loss: 0.6651 - accuracy: 0.6426
297/338 [=========================>....] - ETA: 2s - loss: 0.6652 - accuracy: 0.6425
298/338 [=========================>....] - ETA: 2s - loss: 0.6651 - accuracy: 0.6428
299/338 [=========================>....] - ETA: 2s - loss: 0.6651 - accuracy: 0.6427
300/338 [=========================>....] - ETA: 2s - loss: 0.6651 - accuracy: 0.6427
301/338 [=========================>....] - ETA: 2s - loss: 0.6652 - accuracy: 0.6425
302/338 [=========================>....] - ETA: 1s - loss: 0.6653 - accuracy: 0.6422
303/338 [=========================>....] - ETA: 1s - loss: 0.6652 - accuracy: 0.6423
304/338 [=========================>....] - ETA: 1s - loss: 0.6651 - accuracy: 0.6426
305/338 [==========================>...] - ETA: 1s - loss: 0.6652 - accuracy: 0.6423
306/338 [==========================>...] - ETA: 1s - loss: 0.6652 - accuracy: 0.6423
307/338 [==========================>...] - ETA: 1s - loss: 0.6652 - accuracy: 0.6421
308/338 [==========================>...] - ETA: 1s - loss: 0.6652 - accuracy: 0.6419
309/338 [==========================>...] - ETA: 1s - loss: 0.6652 - accuracy: 0.6419
310/338 [==========================>...] - ETA: 1s - loss: 0.6651 - accuracy: 0.6421
311/338 [==========================>...] - ETA: 1s - loss: 0.6652 - accuracy: 0.6421
312/338 [==========================>...] - ETA: 1s - loss: 0.6651 - accuracy: 0.6422
313/338 [==========================>...] - ETA: 1s - loss: 0.6652 - accuracy: 0.6418
314/338 [==========================>...] - ETA: 1s - loss: 0.6651 - accuracy: 0.6418
315/338 [==========================>...] - ETA: 1s - loss: 0.6651 - accuracy: 0.6421
316/338 [===========================>..] - ETA: 1s - loss: 0.6650 - accuracy: 0.6422
317/338 [===========================>..] - ETA: 1s - loss: 0.6650 - accuracy: 0.6423
318/338 [===========================>..] - ETA: 1s - loss: 0.6650 - accuracy: 0.6421
319/338 [===========================>..] - ETA: 1s - loss: 0.6649 - accuracy: 0.6424
320/338 [===========================>..] - ETA: 0s - loss: 0.6647 - accuracy: 0.6430
321/338 [===========================>..] - ETA: 0s - loss: 0.6647 - accuracy: 0.6428
322/338 [===========================>..] - ETA: 0s - loss: 0.6646 - accuracy: 0.6431
323/338 [===========================>..] - ETA: 0s - loss: 0.6645 - accuracy: 0.6433
324/338 [===========================>..] - ETA: 0s - loss: 0.6645 - accuracy: 0.6433
325/338 [===========================>..] - ETA: 0s - loss: 0.6643 - accuracy: 0.6437
326/338 [===========================>..] - ETA: 0s - loss: 0.6644 - accuracy: 0.6435
327/338 [============================>.] - ETA: 0s - loss: 0.6643 - accuracy: 0.6436
328/338 [============================>.] - ETA: 0s - loss: 0.6645 - accuracy: 0.6433
329/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6436
330/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6436
331/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6436
332/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6435
333/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6435
334/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6434
335/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6435
336/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6435
337/338 [============================>.] - ETA: 0s - loss: 0.6644 - accuracy: 0.6434
338/338 [==============================] - 20s 60ms/step - loss: 0.6645 - accuracy: 0.6433 - val_loss: 0.6822 - val_accuracy: 0.5774
Epoch 4/5
1/338 [..............................] - ETA: 18s - loss: 0.6555 - accuracy: 0.6250
2/338 [..............................] - ETA: 17s - loss: 0.6554 - accuracy: 0.6406
3/338 [..............................] - ETA: 18s - loss: 0.6540 - accuracy: 0.6562
4/338 [..............................] - ETA: 18s - loss: 0.6511 - accuracy: 0.6562
5/338 [..............................] - ETA: 18s - loss: 0.6516 - accuracy: 0.6562
6/338 [..............................] - ETA: 18s - loss: 0.6559 - accuracy: 0.6406
7/338 [..............................] - ETA: 17s - loss: 0.6517 - accuracy: 0.6562
8/338 [..............................] - ETA: 17s - loss: 0.6521 - accuracy: 0.6562
9/338 [..............................] - ETA: 17s - loss: 0.6490 - accuracy: 0.6667
10/338 [..............................] - ETA: 17s - loss: 0.6571 - accuracy: 0.6469
11/338 [..............................] - ETA: 17s - loss: 0.6593 - accuracy: 0.6420
12/338 [>.............................] - ETA: 17s - loss: 0.6572 - accuracy: 0.6484
13/338 [>.............................] - ETA: 17s - loss: 0.6544 - accuracy: 0.6562
14/338 [>.............................] - ETA: 17s - loss: 0.6531 - accuracy: 0.6607
15/338 [>.............................] - ETA: 17s - loss: 0.6543 - accuracy: 0.6583
16/338 [>.............................] - ETA: 17s - loss: 0.6563 - accuracy: 0.6523
17/338 [>.............................] - ETA: 17s - loss: 0.6568 - accuracy: 0.6507
18/338 [>.............................] - ETA: 17s - loss: 0.6577 - accuracy: 0.6493
19/338 [>.............................] - ETA: 17s - loss: 0.6611 - accuracy: 0.6398
20/338 [>.............................] - ETA: 17s - loss: 0.6601 - accuracy: 0.6406
21/338 [>.............................] - ETA: 17s - loss: 0.6591 - accuracy: 0.6429
22/338 [>.............................] - ETA: 17s - loss: 0.6597 - accuracy: 0.6406
23/338 [=>............................] - ETA: 17s - loss: 0.6601 - accuracy: 0.6399
24/338 [=>............................] - ETA: 17s - loss: 0.6596 - accuracy: 0.6393
25/338 [=>............................] - ETA: 17s - loss: 0.6590 - accuracy: 0.6400
26/338 [=>............................] - ETA: 17s - loss: 0.6602 - accuracy: 0.6382
27/338 [=>............................] - ETA: 17s - loss: 0.6616 - accuracy: 0.6343
28/338 [=>............................] - ETA: 16s - loss: 0.6619 - accuracy: 0.6339
29/338 [=>............................] - ETA: 16s - loss: 0.6624 - accuracy: 0.6325
30/338 [=>............................] - ETA: 16s - loss: 0.6623 - accuracy: 0.6333
31/338 [=>............................] - ETA: 16s - loss: 0.6615 - accuracy: 0.6351
32/338 [=>............................] - ETA: 16s - loss: 0.6617 - accuracy: 0.6348
33/338 [=>............................] - ETA: 16s - loss: 0.6628 - accuracy: 0.6326
34/338 [==>...........................] - ETA: 16s - loss: 0.6632 - accuracy: 0.6314
35/338 [==>...........................] - ETA: 16s - loss: 0.6621 - accuracy: 0.6339
36/338 [==>...........................] - ETA: 16s - loss: 0.6626 - accuracy: 0.6328
37/338 [==>...........................] - ETA: 16s - loss: 0.6642 - accuracy: 0.6292
38/338 [==>...........................] - ETA: 16s - loss: 0.6642 - accuracy: 0.6291
39/338 [==>...........................] - ETA: 16s - loss: 0.6637 - accuracy: 0.6306
40/338 [==>...........................] - ETA: 16s - loss: 0.6627 - accuracy: 0.6328
41/338 [==>...........................] - ETA: 16s - loss: 0.6616 - accuracy: 0.6364
42/338 [==>...........................] - ETA: 16s - loss: 0.6608 - accuracy: 0.6384
43/338 [==>...........................] - ETA: 16s - loss: 0.6610 - accuracy: 0.6381
44/338 [==>...........................] - ETA: 16s - loss: 0.6608 - accuracy: 0.6392
45/338 [==>...........................] - ETA: 16s - loss: 0.6615 - accuracy: 0.6375
46/338 [===>..........................] - ETA: 15s - loss: 0.6602 - accuracy: 0.6399
47/338 [===>..........................] - ETA: 15s - loss: 0.6597 - accuracy: 0.6416
48/338 [===>..........................] - ETA: 15s - loss: 0.6596 - accuracy: 0.6413
49/338 [===>..........................] - ETA: 15s - loss: 0.6595 - accuracy: 0.6422
50/338 [===>..........................] - ETA: 15s - loss: 0.6594 - accuracy: 0.6431
51/338 [===>..........................] - ETA: 15s - loss: 0.6602 - accuracy: 0.6409
52/338 [===>..........................] - ETA: 15s - loss: 0.6606 - accuracy: 0.6400
53/338 [===>..........................] - ETA: 15s - loss: 0.6602 - accuracy: 0.6409
54/338 [===>..........................] - ETA: 15s - loss: 0.6593 - accuracy: 0.6429
55/338 [===>..........................] - ETA: 15s - loss: 0.6586 - accuracy: 0.6443
56/338 [===>..........................] - ETA: 15s - loss: 0.6585 - accuracy: 0.6445
57/338 [====>.........................] - ETA: 15s - loss: 0.6579 - accuracy: 0.6458
58/338 [====>.........................] - ETA: 15s - loss: 0.6585 - accuracy: 0.6439
59/338 [====>.........................] - ETA: 15s - loss: 0.6590 - accuracy: 0.6425
60/338 [====>.........................] - ETA: 15s - loss: 0.6586 - accuracy: 0.6427
61/338 [====>.........................] - ETA: 15s - loss: 0.6593 - accuracy: 0.6409
62/338 [====>.........................] - ETA: 15s - loss: 0.6594 - accuracy: 0.6401
63/338 [====>.........................] - ETA: 15s - loss: 0.6595 - accuracy: 0.6394
64/338 [====>.........................] - ETA: 15s - loss: 0.6591 - accuracy: 0.6401
65/338 [====>.........................] - ETA: 15s - loss: 0.6591 - accuracy: 0.6399
66/338 [====>.........................] - ETA: 15s - loss: 0.6588 - accuracy: 0.6402
67/338 [====>.........................] - ETA: 14s - loss: 0.6584 - accuracy: 0.6413
68/338 [=====>........................] - ETA: 14s - loss: 0.6585 - accuracy: 0.6415
69/338 [=====>........................] - ETA: 14s - loss: 0.6588 - accuracy: 0.6399
70/338 [=====>........................] - ETA: 14s - loss: 0.6588 - accuracy: 0.6402
71/338 [=====>........................] - ETA: 14s - loss: 0.6586 - accuracy: 0.6408
72/338 [=====>........................] - ETA: 14s - loss: 0.6584 - accuracy: 0.6415
73/338 [=====>........................] - ETA: 14s - loss: 0.6578 - accuracy: 0.6430
74/338 [=====>........................] - ETA: 14s - loss: 0.6586 - accuracy: 0.6410
75/338 [=====>........................] - ETA: 14s - loss: 0.6590 - accuracy: 0.6400
76/338 [=====>........................] - ETA: 14s - loss: 0.6584 - accuracy: 0.6419
77/338 [=====>........................] - ETA: 14s - loss: 0.6577 - accuracy: 0.6437
78/338 [=====>........................] - ETA: 14s - loss: 0.6578 - accuracy: 0.6438
79/338 [======>.......................] - ETA: 14s - loss: 0.6585 - accuracy: 0.6420
80/338 [======>.......................] - ETA: 14s - loss: 0.6585 - accuracy: 0.6426
81/338 [======>.......................] - ETA: 14s - loss: 0.6588 - accuracy: 0.6416
82/338 [======>.......................] - ETA: 14s - loss: 0.6595 - accuracy: 0.6402
83/338 [======>.......................] - ETA: 14s - loss: 0.6593 - accuracy: 0.6408
84/338 [======>.......................] - ETA: 13s - loss: 0.6589 - accuracy: 0.6414
85/338 [======>.......................] - ETA: 13s - loss: 0.6584 - accuracy: 0.6426
86/338 [======>.......................] - ETA: 13s - loss: 0.6577 - accuracy: 0.6443
87/338 [======>.......................] - ETA: 13s - loss: 0.6582 - accuracy: 0.6437
88/338 [======>.......................] - ETA: 13s - loss: 0.6586 - accuracy: 0.6428
89/338 [======>.......................] - ETA: 13s - loss: 0.6588 - accuracy: 0.6422
90/338 [======>.......................] - ETA: 13s - loss: 0.6589 - accuracy: 0.6420
91/338 [=======>......................] - ETA: 13s - loss: 0.6594 - accuracy: 0.6408
92/338 [=======>......................] - ETA: 13s - loss: 0.6595 - accuracy: 0.6406
93/338 [=======>......................] - ETA: 13s - loss: 0.6593 - accuracy: 0.6411
94/338 [=======>......................] - ETA: 13s - loss: 0.6590 - accuracy: 0.6420
95/338 [=======>......................] - ETA: 13s - loss: 0.6588 - accuracy: 0.6421
96/338 [=======>......................] - ETA: 13s - loss: 0.6585 - accuracy: 0.6429
97/338 [=======>......................] - ETA: 13s - loss: 0.6584 - accuracy: 0.6430
98/338 [=======>......................] - ETA: 13s - loss: 0.6585 - accuracy: 0.6429
99/338 [=======>......................] - ETA: 13s - loss: 0.6582 - accuracy: 0.6436
100/338 [=======>......................] - ETA: 13s - loss: 0.6585 - accuracy: 0.6428
101/338 [=======>......................] - ETA: 13s - loss: 0.6586 - accuracy: 0.6426
102/338 [========>.....................] - ETA: 12s - loss: 0.6582 - accuracy: 0.6437
103/338 [========>.....................] - ETA: 12s - loss: 0.6583 - accuracy: 0.6435
104/338 [========>.....................] - ETA: 12s - loss: 0.6582 - accuracy: 0.6436
105/338 [========>.....................] - ETA: 12s - loss: 0.6581 - accuracy: 0.6438
106/338 [========>.....................] - ETA: 12s - loss: 0.6581 - accuracy: 0.6439
107/338 [========>.....................] - ETA: 12s - loss: 0.6582 - accuracy: 0.6437
108/338 [========>.....................] - ETA: 12s - loss: 0.6581 - accuracy: 0.6438
109/338 [========>.....................] - ETA: 12s - loss: 0.6581 - accuracy: 0.6436
110/338 [========>.....................] - ETA: 12s - loss: 0.6577 - accuracy: 0.6446
111/338 [========>.....................] - ETA: 12s - loss: 0.6577 - accuracy: 0.6444
112/338 [========>.....................] - ETA: 12s - loss: 0.6576 - accuracy: 0.6448
113/338 [=========>....................] - ETA: 12s - loss: 0.6575 - accuracy: 0.6452
114/338 [=========>....................] - ETA: 12s - loss: 0.6570 - accuracy: 0.6461
115/338 [=========>....................] - ETA: 12s - loss: 0.6571 - accuracy: 0.6457
116/338 [=========>....................] - ETA: 12s - loss: 0.6578 - accuracy: 0.6444
117/338 [=========>....................] - ETA: 12s - loss: 0.6576 - accuracy: 0.6448
118/338 [=========>....................] - ETA: 12s - loss: 0.6578 - accuracy: 0.6443
119/338 [=========>....................] - ETA: 12s - loss: 0.6579 - accuracy: 0.6442
120/338 [=========>....................] - ETA: 11s - loss: 0.6582 - accuracy: 0.6432
121/338 [=========>....................] - ETA: 11s - loss: 0.6581 - accuracy: 0.6433
122/338 [=========>....................] - ETA: 11s - loss: 0.6581 - accuracy: 0.6432
123/338 [=========>....................] - ETA: 11s - loss: 0.6579 - accuracy: 0.6438
124/338 [==========>...................] - ETA: 11s - loss: 0.6578 - accuracy: 0.6442
125/338 [==========>...................] - ETA: 11s - loss: 0.6576 - accuracy: 0.6447
126/338 [==========>...................] - ETA: 11s - loss: 0.6579 - accuracy: 0.6436
127/338 [==========>...................] - ETA: 11s - loss: 0.6576 - accuracy: 0.6437
128/338 [==========>...................] - ETA: 11s - loss: 0.6575 - accuracy: 0.6443
129/338 [==========>...................] - ETA: 11s - loss: 0.6570 - accuracy: 0.6456
130/338 [==========>...................] - ETA: 11s - loss: 0.6570 - accuracy: 0.6459
131/338 [==========>...................] - ETA: 11s - loss: 0.6571 - accuracy: 0.6460
132/338 [==========>...................] - ETA: 11s - loss: 0.6567 - accuracy: 0.6463
133/338 [==========>...................] - ETA: 11s - loss: 0.6570 - accuracy: 0.6461
134/338 [==========>...................] - ETA: 11s - loss: 0.6570 - accuracy: 0.6462
135/338 [==========>...................] - ETA: 11s - loss: 0.6570 - accuracy: 0.6463
136/338 [===========>..................] - ETA: 11s - loss: 0.6572 - accuracy: 0.6455
137/338 [===========>..................] - ETA: 11s - loss: 0.6573 - accuracy: 0.6451
138/338 [===========>..................] - ETA: 11s - loss: 0.6575 - accuracy: 0.6445
139/338 [===========>..................] - ETA: 10s - loss: 0.6578 - accuracy: 0.6437
140/338 [===========>..................] - ETA: 10s - loss: 0.6576 - accuracy: 0.6433
141/338 [===========>..................] - ETA: 10s - loss: 0.6579 - accuracy: 0.6421
142/338 [===========>..................] - ETA: 10s - loss: 0.6576 - accuracy: 0.6426
143/338 [===========>..................] - ETA: 10s - loss: 0.6577 - accuracy: 0.6425
144/338 [===========>..................] - ETA: 10s - loss: 0.6578 - accuracy: 0.6417
145/338 [===========>..................] - ETA: 10s - loss: 0.6578 - accuracy: 0.6414
146/338 [===========>..................] - ETA: 10s - loss: 0.6578 - accuracy: 0.6417
147/338 [============>.................] - ETA: 10s - loss: 0.6576 - accuracy: 0.6429
148/338 [============>.................] - ETA: 10s - loss: 0.6576 - accuracy: 0.6429
149/338 [============>.................] - ETA: 10s - loss: 0.6575 - accuracy: 0.6428
150/338 [============>.................] - ETA: 10s - loss: 0.6574 - accuracy: 0.6431
151/338 [============>.................] - ETA: 10s - loss: 0.6574 - accuracy: 0.6426
152/338 [============>.................] - ETA: 10s - loss: 0.6576 - accuracy: 0.6421
153/338 [============>.................] - ETA: 10s - loss: 0.6575 - accuracy: 0.6422
154/338 [============>.................] - ETA: 10s - loss: 0.6570 - accuracy: 0.6431
155/338 [============>.................] - ETA: 10s - loss: 0.6570 - accuracy: 0.6435
156/338 [============>.................] - ETA: 10s - loss: 0.6568 - accuracy: 0.6438
157/338 [============>.................] - ETA: 9s - loss: 0.6568 - accuracy: 0.6439
158/338 [=============>................] - ETA: 9s - loss: 0.6574 - accuracy: 0.6426
159/338 [=============>................] - ETA: 9s - loss: 0.6576 - accuracy: 0.6421
160/338 [=============>................] - ETA: 9s - loss: 0.6575 - accuracy: 0.6422
161/338 [=============>................] - ETA: 9s - loss: 0.6570 - accuracy: 0.6434
162/338 [=============>................] - ETA: 9s - loss: 0.6567 - accuracy: 0.6439
163/338 [=============>................] - ETA: 9s - loss: 0.6565 - accuracy: 0.6444
164/338 [=============>................] - ETA: 9s - loss: 0.6563 - accuracy: 0.6448
165/338 [=============>................] - ETA: 9s - loss: 0.6563 - accuracy: 0.6449
166/338 [=============>................] - ETA: 9s - loss: 0.6563 - accuracy: 0.6448
167/338 [=============>................] - ETA: 9s - loss: 0.6563 - accuracy: 0.6448
168/338 [=============>................] - ETA: 9s - loss: 0.6562 - accuracy: 0.6445
169/338 [==============>...............] - ETA: 9s - loss: 0.6562 - accuracy: 0.6444
170/338 [==============>...............] - ETA: 9s - loss: 0.6562 - accuracy: 0.6445
171/338 [==============>...............] - ETA: 9s - loss: 0.6563 - accuracy: 0.6440
172/338 [==============>...............] - ETA: 9s - loss: 0.6563 - accuracy: 0.6441
173/338 [==============>...............] - ETA: 9s - loss: 0.6564 - accuracy: 0.6440
174/338 [==============>...............] - ETA: 9s - loss: 0.6564 - accuracy: 0.6439
175/338 [==============>...............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6436
176/338 [==============>...............] - ETA: 8s - loss: 0.6566 - accuracy: 0.6436
177/338 [==============>...............] - ETA: 8s - loss: 0.6564 - accuracy: 0.6439
178/338 [==============>...............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6434
179/338 [==============>...............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6435
180/338 [==============>...............] - ETA: 8s - loss: 0.6566 - accuracy: 0.6432
181/338 [===============>..............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6433
182/338 [===============>..............] - ETA: 8s - loss: 0.6564 - accuracy: 0.6434
183/338 [===============>..............] - ETA: 8s - loss: 0.6564 - accuracy: 0.6438
184/338 [===============>..............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6432
185/338 [===============>..............] - ETA: 8s - loss: 0.6566 - accuracy: 0.6431
186/338 [===============>..............] - ETA: 8s - loss: 0.6566 - accuracy: 0.6431
187/338 [===============>..............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6434
188/338 [===============>..............] - ETA: 8s - loss: 0.6567 - accuracy: 0.6431
189/338 [===============>..............] - ETA: 8s - loss: 0.6566 - accuracy: 0.6434
190/338 [===============>..............] - ETA: 8s - loss: 0.6567 - accuracy: 0.6433
191/338 [===============>..............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6435
192/338 [================>.............] - ETA: 8s - loss: 0.6565 - accuracy: 0.6436
193/338 [================>.............] - ETA: 8s - loss: 0.6564 - accuracy: 0.6436
194/338 [================>.............] - ETA: 7s - loss: 0.6567 - accuracy: 0.6430
195/338 [================>.............] - ETA: 7s - loss: 0.6565 - accuracy: 0.6434
196/338 [================>.............] - ETA: 7s - loss: 0.6567 - accuracy: 0.6429
197/338 [================>.............] - ETA: 7s - loss: 0.6566 - accuracy: 0.6431
198/338 [================>.............] - ETA: 7s - loss: 0.6564 - accuracy: 0.6436
199/338 [================>.............] - ETA: 7s - loss: 0.6564 - accuracy: 0.6437
200/338 [================>.............] - ETA: 7s - loss: 0.6567 - accuracy: 0.6430
201/338 [================>.............] - ETA: 7s - loss: 0.6565 - accuracy: 0.6435
202/338 [================>.............] - ETA: 7s - loss: 0.6566 - accuracy: 0.6431
203/338 [=================>............] - ETA: 7s - loss: 0.6567 - accuracy: 0.6432
204/338 [=================>............] - ETA: 7s - loss: 0.6567 - accuracy: 0.6432
205/338 [=================>............] - ETA: 7s - loss: 0.6566 - accuracy: 0.6433
206/338 [=================>............] - ETA: 7s - loss: 0.6564 - accuracy: 0.6437
207/338 [=================>............] - ETA: 7s - loss: 0.6566 - accuracy: 0.6431
208/338 [=================>............] - ETA: 7s - loss: 0.6564 - accuracy: 0.6435
209/338 [=================>............] - ETA: 7s - loss: 0.6562 - accuracy: 0.6437
210/338 [=================>............] - ETA: 7s - loss: 0.6563 - accuracy: 0.6435
211/338 [=================>............] - ETA: 7s - loss: 0.6562 - accuracy: 0.6437
212/338 [=================>............] - ETA: 6s - loss: 0.6559 - accuracy: 0.6445
213/338 [=================>............] - ETA: 6s - loss: 0.6556 - accuracy: 0.6450
214/338 [=================>............] - ETA: 6s - loss: 0.6556 - accuracy: 0.6449
215/338 [==================>...........] - ETA: 6s - loss: 0.6555 - accuracy: 0.6452
216/338 [==================>...........] - ETA: 6s - loss: 0.6555 - accuracy: 0.6450
217/338 [==================>...........] - ETA: 6s - loss: 0.6552 - accuracy: 0.6456
218/338 [==================>...........] - ETA: 6s - loss: 0.6554 - accuracy: 0.6451
219/338 [==================>...........] - ETA: 6s - loss: 0.6555 - accuracy: 0.6450
220/338 [==================>...........] - ETA: 6s - loss: 0.6553 - accuracy: 0.6455
221/338 [==================>...........] - ETA: 6s - loss: 0.6553 - accuracy: 0.6456
222/338 [==================>...........] - ETA: 6s - loss: 0.6552 - accuracy: 0.6457
223/338 [==================>...........] - ETA: 6s - loss: 0.6554 - accuracy: 0.6453
224/338 [==================>...........] - ETA: 6s - loss: 0.6554 - accuracy: 0.6452
225/338 [==================>...........] - ETA: 6s - loss: 0.6551 - accuracy: 0.6460
226/338 [===================>..........] - ETA: 6s - loss: 0.6552 - accuracy: 0.6457
227/338 [===================>..........] - ETA: 6s - loss: 0.6551 - accuracy: 0.6459
228/338 [===================>..........] - ETA: 6s - loss: 0.6551 - accuracy: 0.6460
229/338 [===================>..........] - ETA: 6s - loss: 0.6550 - accuracy: 0.6464
230/338 [===================>..........] - ETA: 5s - loss: 0.6551 - accuracy: 0.6462
231/338 [===================>..........] - ETA: 5s - loss: 0.6550 - accuracy: 0.6464
232/338 [===================>..........] - ETA: 5s - loss: 0.6553 - accuracy: 0.6460
233/338 [===================>..........] - ETA: 5s - loss: 0.6554 - accuracy: 0.6455
234/338 [===================>..........] - ETA: 5s - loss: 0.6550 - accuracy: 0.6465
235/338 [===================>..........] - ETA: 5s - loss: 0.6549 - accuracy: 0.6464
236/338 [===================>..........] - ETA: 5s - loss: 0.6548 - accuracy: 0.6467
237/338 [====================>.........] - ETA: 5s - loss: 0.6551 - accuracy: 0.6460
238/338 [====================>.........] - ETA: 5s - loss: 0.6553 - accuracy: 0.6457
239/338 [====================>.........] - ETA: 5s - loss: 0.6551 - accuracy: 0.6461
240/338 [====================>.........] - ETA: 5s - loss: 0.6550 - accuracy: 0.6465
241/338 [====================>.........] - ETA: 5s - loss: 0.6549 - accuracy: 0.6467
242/338 [====================>.........] - ETA: 5s - loss: 0.6550 - accuracy: 0.6466
243/338 [====================>.........] - ETA: 5s - loss: 0.6550 - accuracy: 0.6463
244/338 [====================>.........] - ETA: 5s - loss: 0.6550 - accuracy: 0.6466
245/338 [====================>.........] - ETA: 5s - loss: 0.6548 - accuracy: 0.6471
246/338 [====================>.........] - ETA: 5s - loss: 0.6548 - accuracy: 0.6471
247/338 [====================>.........] - ETA: 5s - loss: 0.6548 - accuracy: 0.6471
248/338 [=====================>........] - ETA: 4s - loss: 0.6548 - accuracy: 0.6471
249/338 [=====================>........] - ETA: 4s - loss: 0.6548 - accuracy: 0.6472
250/338 [=====================>........] - ETA: 4s - loss: 0.6548 - accuracy: 0.6472
251/338 [=====================>........] - ETA: 4s - loss: 0.6547 - accuracy: 0.6474
252/338 [=====================>........] - ETA: 4s - loss: 0.6547 - accuracy: 0.6473
253/338 [=====================>........] - ETA: 4s - loss: 0.6546 - accuracy: 0.6476
254/338 [=====================>........] - ETA: 4s - loss: 0.6545 - accuracy: 0.6480
255/338 [=====================>........] - ETA: 4s - loss: 0.6546 - accuracy: 0.6477
256/338 [=====================>........] - ETA: 4s - loss: 0.6544 - accuracy: 0.6479
257/338 [=====================>........] - ETA: 4s - loss: 0.6544 - accuracy: 0.6480
258/338 [=====================>........] - ETA: 4s - loss: 0.6544 - accuracy: 0.6479
259/338 [=====================>........] - ETA: 4s - loss: 0.6544 - accuracy: 0.6478
260/338 [======================>.......] - ETA: 4s - loss: 0.6544 - accuracy: 0.6478
261/338 [======================>.......] - ETA: 4s - loss: 0.6544 - accuracy: 0.6476
262/338 [======================>.......] - ETA: 4s - loss: 0.6546 - accuracy: 0.6471
263/338 [======================>.......] - ETA: 4s - loss: 0.6545 - accuracy: 0.6473
264/338 [======================>.......] - ETA: 4s - loss: 0.6545 - accuracy: 0.6471
265/338 [======================>.......] - ETA: 4s - loss: 0.6546 - accuracy: 0.6468
266/338 [======================>.......] - ETA: 3s - loss: 0.6547 - accuracy: 0.6470
267/338 [======================>.......] - ETA: 3s - loss: 0.6548 - accuracy: 0.6467
268/338 [======================>.......] - ETA: 3s - loss: 0.6546 - accuracy: 0.6472
269/338 [======================>.......] - ETA: 3s - loss: 0.6545 - accuracy: 0.6475
270/338 [======================>.......] - ETA: 3s - loss: 0.6545 - accuracy: 0.6475
271/338 [=======================>......] - ETA: 3s - loss: 0.6546 - accuracy: 0.6473
272/338 [=======================>......] - ETA: 3s - loss: 0.6546 - accuracy: 0.6469
273/338 [=======================>......] - ETA: 3s - loss: 0.6547 - accuracy: 0.6467
274/338 [=======================>......] - ETA: 3s - loss: 0.6546 - accuracy: 0.6471
275/338 [=======================>......] - ETA: 3s - loss: 0.6546 - accuracy: 0.6472
276/338 [=======================>......] - ETA: 3s - loss: 0.6545 - accuracy: 0.6473
277/338 [=======================>......] - ETA: 3s - loss: 0.6545 - accuracy: 0.6473
278/338 [=======================>......] - ETA: 3s - loss: 0.6543 - accuracy: 0.6475
279/338 [=======================>......] - ETA: 3s - loss: 0.6543 - accuracy: 0.6474
280/338 [=======================>......] - ETA: 3s - loss: 0.6542 - accuracy: 0.6477
281/338 [=======================>......] - ETA: 3s - loss: 0.6541 - accuracy: 0.6479
282/338 [========================>.....] - ETA: 3s - loss: 0.6539 - accuracy: 0.6484
283/338 [========================>.....] - ETA: 3s - loss: 0.6539 - accuracy: 0.6484
284/338 [========================>.....] - ETA: 2s - loss: 0.6538 - accuracy: 0.6485
285/338 [========================>.....] - ETA: 2s - loss: 0.6539 - accuracy: 0.6485
286/338 [========================>.....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6475
287/338 [========================>.....] - ETA: 2s - loss: 0.6544 - accuracy: 0.6472
288/338 [========================>.....] - ETA: 2s - loss: 0.6544 - accuracy: 0.6472
289/338 [========================>.....] - ETA: 2s - loss: 0.6545 - accuracy: 0.6473
290/338 [========================>.....] - ETA: 2s - loss: 0.6544 - accuracy: 0.6473
291/338 [========================>.....] - ETA: 2s - loss: 0.6546 - accuracy: 0.6470
292/338 [========================>.....] - ETA: 2s - loss: 0.6545 - accuracy: 0.6474
293/338 [=========================>....] - ETA: 2s - loss: 0.6545 - accuracy: 0.6473
294/338 [=========================>....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6476
295/338 [=========================>....] - ETA: 2s - loss: 0.6544 - accuracy: 0.6475
296/338 [=========================>....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6478
297/338 [=========================>....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6476
298/338 [=========================>....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6475
299/338 [=========================>....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6475
300/338 [=========================>....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6477
301/338 [=========================>....] - ETA: 2s - loss: 0.6543 - accuracy: 0.6474
302/338 [=========================>....] - ETA: 1s - loss: 0.6543 - accuracy: 0.6472
303/338 [=========================>....] - ETA: 1s - loss: 0.6543 - accuracy: 0.6474
304/338 [=========================>....] - ETA: 1s - loss: 0.6543 - accuracy: 0.6473
305/338 [==========================>...] - ETA: 1s - loss: 0.6543 - accuracy: 0.6470
306/338 [==========================>...] - ETA: 1s - loss: 0.6545 - accuracy: 0.6468
307/338 [==========================>...] - ETA: 1s - loss: 0.6546 - accuracy: 0.6463
308/338 [==========================>...] - ETA: 1s - loss: 0.6545 - accuracy: 0.6467
309/338 [==========================>...] - ETA: 1s - loss: 0.6545 - accuracy: 0.6466
310/338 [==========================>...] - ETA: 1s - loss: 0.6545 - accuracy: 0.6467
311/338 [==========================>...] - ETA: 1s - loss: 0.6544 - accuracy: 0.6468
312/338 [==========================>...] - ETA: 1s - loss: 0.6545 - accuracy: 0.6465
313/338 [==========================>...] - ETA: 1s - loss: 0.6546 - accuracy: 0.6465
314/338 [==========================>...] - ETA: 1s - loss: 0.6547 - accuracy: 0.6464
315/338 [==========================>...] - ETA: 1s - loss: 0.6547 - accuracy: 0.6464
316/338 [===========================>..] - ETA: 1s - loss: 0.6548 - accuracy: 0.6462
317/338 [===========================>..] - ETA: 1s - loss: 0.6549 - accuracy: 0.6460
318/338 [===========================>..] - ETA: 1s - loss: 0.6550 - accuracy: 0.6458
319/338 [===========================>..] - ETA: 1s - loss: 0.6548 - accuracy: 0.6461
320/338 [===========================>..] - ETA: 0s - loss: 0.6549 - accuracy: 0.6458
321/338 [===========================>..] - ETA: 0s - loss: 0.6547 - accuracy: 0.6462
322/338 [===========================>..] - ETA: 0s - loss: 0.6549 - accuracy: 0.6459
323/338 [===========================>..] - ETA: 0s - loss: 0.6549 - accuracy: 0.6458
324/338 [===========================>..] - ETA: 0s - loss: 0.6549 - accuracy: 0.6459
325/338 [===========================>..] - ETA: 0s - loss: 0.6548 - accuracy: 0.6464
326/338 [===========================>..] - ETA: 0s - loss: 0.6547 - accuracy: 0.6463
327/338 [============================>.] - ETA: 0s - loss: 0.6547 - accuracy: 0.6464
328/338 [============================>.] - ETA: 0s - loss: 0.6550 - accuracy: 0.6460
329/338 [============================>.] - ETA: 0s - loss: 0.6550 - accuracy: 0.6459
330/338 [============================>.] - ETA: 0s - loss: 0.6551 - accuracy: 0.6456
331/338 [============================>.] - ETA: 0s - loss: 0.6550 - accuracy: 0.6460
332/338 [============================>.] - ETA: 0s - loss: 0.6548 - accuracy: 0.6464
333/338 [============================>.] - ETA: 0s - loss: 0.6545 - accuracy: 0.6471
334/338 [============================>.] - ETA: 0s - loss: 0.6544 - accuracy: 0.6471
335/338 [============================>.] - ETA: 0s - loss: 0.6544 - accuracy: 0.6470
336/338 [============================>.] - ETA: 0s - loss: 0.6545 - accuracy: 0.6468
337/338 [============================>.] - ETA: 0s - loss: 0.6545 - accuracy: 0.6468
338/338 [==============================] - 20s 59ms/step - loss: 0.6546 - accuracy: 0.6465 - val_loss: 0.6755 - val_accuracy: 0.5881
Epoch 5/5
1/338 [..............................] - ETA: 18s - loss: 0.6381 - accuracy: 0.6250
2/338 [..............................] - ETA: 17s - loss: 0.6495 - accuracy: 0.6250
3/338 [..............................] - ETA: 18s - loss: 0.6589 - accuracy: 0.6354
4/338 [..............................] - ETA: 18s - loss: 0.6448 - accuracy: 0.6562
5/338 [..............................] - ETA: 18s - loss: 0.6379 - accuracy: 0.6562
6/338 [..............................] - ETA: 18s - loss: 0.6466 - accuracy: 0.6406
7/338 [..............................] - ETA: 18s - loss: 0.6521 - accuracy: 0.6339
8/338 [..............................] - ETA: 18s - loss: 0.6526 - accuracy: 0.6328
9/338 [..............................] - ETA: 18s - loss: 0.6548 - accuracy: 0.6285
10/338 [..............................] - ETA: 18s - loss: 0.6558 - accuracy: 0.6281
11/338 [..............................] - ETA: 18s - loss: 0.6554 - accuracy: 0.6335
12/338 [>.............................] - ETA: 18s - loss: 0.6521 - accuracy: 0.6406
13/338 [>.............................] - ETA: 17s - loss: 0.6528 - accuracy: 0.6418
14/338 [>.............................] - ETA: 17s - loss: 0.6486 - accuracy: 0.6562
15/338 [>.............................] - ETA: 17s - loss: 0.6473 - accuracy: 0.6583
16/338 [>.............................] - ETA: 17s - loss: 0.6487 - accuracy: 0.6543
17/338 [>.............................] - ETA: 17s - loss: 0.6492 - accuracy: 0.6507
18/338 [>.............................] - ETA: 17s - loss: 0.6494 - accuracy: 0.6545
19/338 [>.............................] - ETA: 17s - loss: 0.6476 - accuracy: 0.6579
20/338 [>.............................] - ETA: 17s - loss: 0.6489 - accuracy: 0.6547
21/338 [>.............................] - ETA: 17s - loss: 0.6474 - accuracy: 0.6592
22/338 [>.............................] - ETA: 17s - loss: 0.6468 - accuracy: 0.6591
23/338 [=>............................] - ETA: 17s - loss: 0.6456 - accuracy: 0.6603
24/338 [=>............................] - ETA: 17s - loss: 0.6449 - accuracy: 0.6615
25/338 [=>............................] - ETA: 17s - loss: 0.6439 - accuracy: 0.6650
26/338 [=>............................] - ETA: 17s - loss: 0.6437 - accuracy: 0.6647
27/338 [=>............................] - ETA: 17s - loss: 0.6434 - accuracy: 0.6678
28/338 [=>............................] - ETA: 17s - loss: 0.6436 - accuracy: 0.6685
29/338 [=>............................] - ETA: 17s - loss: 0.6427 - accuracy: 0.6713
30/338 [=>............................] - ETA: 16s - loss: 0.6424 - accuracy: 0.6708
31/338 [=>............................] - ETA: 16s - loss: 0.6428 - accuracy: 0.6694
32/338 [=>............................] - ETA: 16s - loss: 0.6414 - accuracy: 0.6729
33/338 [=>............................] - ETA: 16s - loss: 0.6434 - accuracy: 0.6676
34/338 [==>...........................] - ETA: 16s - loss: 0.6443 - accuracy: 0.6664
35/338 [==>...........................] - ETA: 16s - loss: 0.6426 - accuracy: 0.6687
36/338 [==>...........................] - ETA: 16s - loss: 0.6431 - accuracy: 0.6675
37/338 [==>...........................] - ETA: 16s - loss: 0.6445 - accuracy: 0.6647
38/338 [==>...........................] - ETA: 16s - loss: 0.6446 - accuracy: 0.6653
39/338 [==>...........................] - ETA: 16s - loss: 0.6445 - accuracy: 0.6659
40/338 [==>...........................] - ETA: 16s - loss: 0.6433 - accuracy: 0.6680
41/338 [==>...........................] - ETA: 16s - loss: 0.6416 - accuracy: 0.6707
42/338 [==>...........................] - ETA: 16s - loss: 0.6410 - accuracy: 0.6711
43/338 [==>...........................] - ETA: 16s - loss: 0.6413 - accuracy: 0.6715
44/338 [==>...........................] - ETA: 16s - loss: 0.6407 - accuracy: 0.6726
45/338 [==>...........................] - ETA: 16s - loss: 0.6399 - accuracy: 0.6736
46/338 [===>..........................] - ETA: 16s - loss: 0.6406 - accuracy: 0.6726
47/338 [===>..........................] - ETA: 15s - loss: 0.6411 - accuracy: 0.6722
48/338 [===>..........................] - ETA: 15s - loss: 0.6413 - accuracy: 0.6719
49/338 [===>..........................] - ETA: 15s - loss: 0.6403 - accuracy: 0.6741
50/338 [===>..........................] - ETA: 15s - loss: 0.6411 - accuracy: 0.6731
51/338 [===>..........................] - ETA: 15s - loss: 0.6422 - accuracy: 0.6710
52/338 [===>..........................] - ETA: 15s - loss: 0.6424 - accuracy: 0.6701
53/338 [===>..........................] - ETA: 15s - loss: 0.6415 - accuracy: 0.6716
54/338 [===>..........................] - ETA: 15s - loss: 0.6413 - accuracy: 0.6725
55/338 [===>..........................] - ETA: 15s - loss: 0.6403 - accuracy: 0.6744
56/338 [===>..........................] - ETA: 15s - loss: 0.6391 - accuracy: 0.6769
57/338 [====>.........................] - ETA: 15s - loss: 0.6396 - accuracy: 0.6754
58/338 [====>.........................] - ETA: 15s - loss: 0.6402 - accuracy: 0.6735
59/338 [====>.........................] - ETA: 15s - loss: 0.6400 - accuracy: 0.6737
60/338 [====>.........................] - ETA: 15s - loss: 0.6414 - accuracy: 0.6714
61/338 [====>.........................] - ETA: 15s - loss: 0.6422 - accuracy: 0.6696
62/338 [====>.........................] - ETA: 15s - loss: 0.6416 - accuracy: 0.6709
63/338 [====>.........................] - ETA: 15s - loss: 0.6414 - accuracy: 0.6711
64/338 [====>.........................] - ETA: 14s - loss: 0.6408 - accuracy: 0.6724
65/338 [====>.........................] - ETA: 14s - loss: 0.6403 - accuracy: 0.6721
66/338 [====>.........................] - ETA: 14s - loss: 0.6412 - accuracy: 0.6705
67/338 [====>.........................] - ETA: 14s - loss: 0.6418 - accuracy: 0.6698
68/338 [=====>........................] - ETA: 14s - loss: 0.6420 - accuracy: 0.6691
69/338 [=====>........................] - ETA: 14s - loss: 0.6422 - accuracy: 0.6689
70/338 [=====>........................] - ETA: 14s - loss: 0.6423 - accuracy: 0.6679
71/338 [=====>........................] - ETA: 14s - loss: 0.6418 - accuracy: 0.6686
72/338 [=====>........................] - ETA: 14s - loss: 0.6418 - accuracy: 0.6688
73/338 [=====>........................] - ETA: 14s - loss: 0.6420 - accuracy: 0.6682
74/338 [=====>........................] - ETA: 14s - loss: 0.6424 - accuracy: 0.6672
75/338 [=====>........................] - ETA: 14s - loss: 0.6438 - accuracy: 0.6642
76/338 [=====>........................] - ETA: 14s - loss: 0.6441 - accuracy: 0.6637
77/338 [=====>........................] - ETA: 14s - loss: 0.6444 - accuracy: 0.6627
78/338 [=====>........................] - ETA: 14s - loss: 0.6445 - accuracy: 0.6619
79/338 [======>.......................] - ETA: 14s - loss: 0.6447 - accuracy: 0.6610
80/338 [======>.......................] - ETA: 14s - loss: 0.6446 - accuracy: 0.6609
81/338 [======>.......................] - ETA: 14s - loss: 0.6444 - accuracy: 0.6613
82/338 [======>.......................] - ETA: 14s - loss: 0.6449 - accuracy: 0.6597
83/338 [======>.......................] - ETA: 13s - loss: 0.6439 - accuracy: 0.6627
84/338 [======>.......................] - ETA: 13s - loss: 0.6437 - accuracy: 0.6629
85/338 [======>.......................] - ETA: 13s - loss: 0.6435 - accuracy: 0.6625
86/338 [======>.......................] - ETA: 13s - loss: 0.6439 - accuracy: 0.6621
87/338 [======>.......................] - ETA: 13s - loss: 0.6443 - accuracy: 0.6613
88/338 [======>.......................] - ETA: 13s - loss: 0.6445 - accuracy: 0.6609
89/338 [======>.......................] - ETA: 13s - loss: 0.6449 - accuracy: 0.6605
90/338 [======>.......................] - ETA: 13s - loss: 0.6454 - accuracy: 0.6601
91/338 [=======>......................] - ETA: 13s - loss: 0.6449 - accuracy: 0.6611
92/338 [=======>......................] - ETA: 13s - loss: 0.6454 - accuracy: 0.6603
93/338 [=======>......................] - ETA: 13s - loss: 0.6451 - accuracy: 0.6593
94/338 [=======>......................] - ETA: 13s - loss: 0.6449 - accuracy: 0.6592
95/338 [=======>......................] - ETA: 13s - loss: 0.6449 - accuracy: 0.6589
96/338 [=======>......................] - ETA: 13s - loss: 0.6445 - accuracy: 0.6595
97/338 [=======>......................] - ETA: 13s - loss: 0.6440 - accuracy: 0.6608
98/338 [=======>......................] - ETA: 13s - loss: 0.6439 - accuracy: 0.6617
99/338 [=======>......................] - ETA: 13s - loss: 0.6441 - accuracy: 0.6613
100/338 [=======>......................] - ETA: 13s - loss: 0.6447 - accuracy: 0.6606
101/338 [=======>......................] - ETA: 12s - loss: 0.6449 - accuracy: 0.6600
102/338 [========>.....................] - ETA: 12s - loss: 0.6448 - accuracy: 0.6602
103/338 [========>.....................] - ETA: 12s - loss: 0.6448 - accuracy: 0.6596
104/338 [========>.....................] - ETA: 12s - loss: 0.6443 - accuracy: 0.6605
105/338 [========>.....................] - ETA: 12s - loss: 0.6442 - accuracy: 0.6610
106/338 [========>.....................] - ETA: 12s - loss: 0.6443 - accuracy: 0.6613
107/338 [========>.....................] - ETA: 12s - loss: 0.6442 - accuracy: 0.6609
108/338 [========>.....................] - ETA: 12s - loss: 0.6447 - accuracy: 0.6594
109/338 [========>.....................] - ETA: 12s - loss: 0.6444 - accuracy: 0.6600
110/338 [========>.....................] - ETA: 12s - loss: 0.6442 - accuracy: 0.6605
111/338 [========>.....................] - ETA: 12s - loss: 0.6445 - accuracy: 0.6599
112/338 [========>.....................] - ETA: 12s - loss: 0.6448 - accuracy: 0.6590
113/338 [=========>....................] - ETA: 12s - loss: 0.6443 - accuracy: 0.6598
114/338 [=========>....................] - ETA: 12s - loss: 0.6439 - accuracy: 0.6604
115/338 [=========>....................] - ETA: 12s - loss: 0.6439 - accuracy: 0.6603
116/338 [=========>....................] - ETA: 12s - loss: 0.6436 - accuracy: 0.6611
117/338 [=========>....................] - ETA: 12s - loss: 0.6432 - accuracy: 0.6619
118/338 [=========>....................] - ETA: 12s - loss: 0.6428 - accuracy: 0.6623
119/338 [=========>....................] - ETA: 11s - loss: 0.6426 - accuracy: 0.6623
120/338 [=========>....................] - ETA: 11s - loss: 0.6429 - accuracy: 0.6615
121/338 [=========>....................] - ETA: 11s - loss: 0.6425 - accuracy: 0.6624
122/338 [=========>....................] - ETA: 11s - loss: 0.6423 - accuracy: 0.6632
123/338 [=========>....................] - ETA: 11s - loss: 0.6423 - accuracy: 0.6626
124/338 [==========>...................] - ETA: 11s - loss: 0.6422 - accuracy: 0.6626
125/338 [==========>...................] - ETA: 11s - loss: 0.6424 - accuracy: 0.6620
126/338 [==========>...................] - ETA: 11s - loss: 0.6420 - accuracy: 0.6629
127/338 [==========>...................] - ETA: 11s - loss: 0.6426 - accuracy: 0.6619
128/338 [==========>...................] - ETA: 11s - loss: 0.6429 - accuracy: 0.6614
129/338 [==========>...................] - ETA: 11s - loss: 0.6423 - accuracy: 0.6625
130/338 [==========>...................] - ETA: 11s - loss: 0.6425 - accuracy: 0.6620
131/338 [==========>...................] - ETA: 11s - loss: 0.6422 - accuracy: 0.6625
132/338 [==========>...................] - ETA: 11s - loss: 0.6419 - accuracy: 0.6629
133/338 [==========>...................] - ETA: 11s - loss: 0.6417 - accuracy: 0.6633
134/338 [==========>...................] - ETA: 11s - loss: 0.6419 - accuracy: 0.6630
135/338 [==========>...................] - ETA: 11s - loss: 0.6420 - accuracy: 0.6627
136/338 [===========>..................] - ETA: 11s - loss: 0.6415 - accuracy: 0.6636
137/338 [===========>..................] - ETA: 11s - loss: 0.6416 - accuracy: 0.6631
138/338 [===========>..................] - ETA: 10s - loss: 0.6418 - accuracy: 0.6626
139/338 [===========>..................] - ETA: 10s - loss: 0.6421 - accuracy: 0.6623
140/338 [===========>..................] - ETA: 10s - loss: 0.6422 - accuracy: 0.6621
141/338 [===========>..................] - ETA: 10s - loss: 0.6421 - accuracy: 0.6620
142/338 [===========>..................] - ETA: 10s - loss: 0.6421 - accuracy: 0.6622
143/338 [===========>..................] - ETA: 10s - loss: 0.6421 - accuracy: 0.6624
144/338 [===========>..................] - ETA: 10s - loss: 0.6422 - accuracy: 0.6621
145/338 [===========>..................] - ETA: 10s - loss: 0.6427 - accuracy: 0.6614
146/338 [===========>..................] - ETA: 10s - loss: 0.6425 - accuracy: 0.6618
147/338 [============>.................] - ETA: 10s - loss: 0.6425 - accuracy: 0.6620
148/338 [============>.................] - ETA: 10s - loss: 0.6423 - accuracy: 0.6622
149/338 [============>.................] - ETA: 10s - loss: 0.6422 - accuracy: 0.6621
150/338 [============>.................] - ETA: 10s - loss: 0.6416 - accuracy: 0.6631
151/338 [============>.................] - ETA: 10s - loss: 0.6417 - accuracy: 0.6629
152/338 [============>.................] - ETA: 10s - loss: 0.6418 - accuracy: 0.6628
153/338 [============>.................] - ETA: 10s - loss: 0.6419 - accuracy: 0.6622
154/338 [============>.................] - ETA: 10s - loss: 0.6418 - accuracy: 0.6625
155/338 [============>.................] - ETA: 10s - loss: 0.6418 - accuracy: 0.6625
156/338 [============>.................] - ETA: 9s - loss: 0.6416 - accuracy: 0.6633
157/338 [============>.................] - ETA: 9s - loss: 0.6412 - accuracy: 0.6638
158/338 [=============>................] - ETA: 9s - loss: 0.6413 - accuracy: 0.6636
159/338 [=============>................] - ETA: 9s - loss: 0.6414 - accuracy: 0.6633
160/338 [=============>................] - ETA: 9s - loss: 0.6414 - accuracy: 0.6631
161/338 [=============>................] - ETA: 9s - loss: 0.6415 - accuracy: 0.6630
162/338 [=============>................] - ETA: 9s - loss: 0.6416 - accuracy: 0.6632
163/338 [=============>................] - ETA: 9s - loss: 0.6417 - accuracy: 0.6630
164/338 [=============>................] - ETA: 9s - loss: 0.6418 - accuracy: 0.6629
165/338 [=============>................] - ETA: 9s - loss: 0.6421 - accuracy: 0.6623
166/338 [=============>................] - ETA: 9s - loss: 0.6421 - accuracy: 0.6623
167/338 [=============>................] - ETA: 9s - loss: 0.6417 - accuracy: 0.6628
168/338 [=============>................] - ETA: 9s - loss: 0.6421 - accuracy: 0.6622
169/338 [==============>...............] - ETA: 9s - loss: 0.6420 - accuracy: 0.6622
170/338 [==============>...............] - ETA: 9s - loss: 0.6418 - accuracy: 0.6627
171/338 [==============>...............] - ETA: 9s - loss: 0.6421 - accuracy: 0.6623
172/338 [==============>...............] - ETA: 9s - loss: 0.6421 - accuracy: 0.6621
173/338 [==============>...............] - ETA: 9s - loss: 0.6419 - accuracy: 0.6622
174/338 [==============>...............] - ETA: 9s - loss: 0.6415 - accuracy: 0.6625
175/338 [==============>...............] - ETA: 8s - loss: 0.6414 - accuracy: 0.6629
176/338 [==============>...............] - ETA: 8s - loss: 0.6416 - accuracy: 0.6625
177/338 [==============>...............] - ETA: 8s - loss: 0.6419 - accuracy: 0.6623
178/338 [==============>...............] - ETA: 8s - loss: 0.6418 - accuracy: 0.6624
179/338 [==============>...............] - ETA: 8s - loss: 0.6414 - accuracy: 0.6632
180/338 [==============>...............] - ETA: 8s - loss: 0.6414 - accuracy: 0.6632
181/338 [===============>..............] - ETA: 8s - loss: 0.6417 - accuracy: 0.6626
182/338 [===============>..............] - ETA: 8s - loss: 0.6417 - accuracy: 0.6628
183/338 [===============>..............] - ETA: 8s - loss: 0.6415 - accuracy: 0.6633
184/338 [===============>..............] - ETA: 8s - loss: 0.6413 - accuracy: 0.6637
185/338 [===============>..............] - ETA: 8s - loss: 0.6412 - accuracy: 0.6640
186/338 [===============>..............] - ETA: 8s - loss: 0.6414 - accuracy: 0.6636
187/338 [===============>..............] - ETA: 8s - loss: 0.6412 - accuracy: 0.6641
188/338 [===============>..............] - ETA: 8s - loss: 0.6415 - accuracy: 0.6634
189/338 [===============>..............] - ETA: 8s - loss: 0.6414 - accuracy: 0.6637
190/338 [===============>..............] - ETA: 8s - loss: 0.6419 - accuracy: 0.6628
191/338 [===============>..............] - ETA: 8s - loss: 0.6416 - accuracy: 0.6633
192/338 [================>.............] - ETA: 8s - loss: 0.6417 - accuracy: 0.6631
193/338 [================>.............] - ETA: 7s - loss: 0.6414 - accuracy: 0.6635
194/338 [================>.............] - ETA: 7s - loss: 0.6412 - accuracy: 0.6638
195/338 [================>.............] - ETA: 7s - loss: 0.6413 - accuracy: 0.6638
196/338 [================>.............] - ETA: 7s - loss: 0.6412 - accuracy: 0.6637
197/338 [================>.............] - ETA: 7s - loss: 0.6412 - accuracy: 0.6640
198/338 [================>.............] - ETA: 7s - loss: 0.6409 - accuracy: 0.6645
199/338 [================>.............] - ETA: 7s - loss: 0.6410 - accuracy: 0.6643
200/338 [================>.............] - ETA: 7s - loss: 0.6408 - accuracy: 0.6645
201/338 [================>.............] - ETA: 7s - loss: 0.6409 - accuracy: 0.6643
202/338 [================>.............] - ETA: 7s - loss: 0.6410 - accuracy: 0.6643
203/338 [=================>............] - ETA: 7s - loss: 0.6407 - accuracy: 0.6649
204/338 [=================>............] - ETA: 7s - loss: 0.6404 - accuracy: 0.6651
205/338 [=================>............] - ETA: 7s - loss: 0.6405 - accuracy: 0.6651
206/338 [=================>............] - ETA: 7s - loss: 0.6407 - accuracy: 0.6647
207/338 [=================>............] - ETA: 7s - loss: 0.6409 - accuracy: 0.6644
208/338 [=================>............] - ETA: 7s - loss: 0.6410 - accuracy: 0.6641
209/338 [=================>............] - ETA: 7s - loss: 0.6412 - accuracy: 0.6639
210/338 [=================>............] - ETA: 7s - loss: 0.6412 - accuracy: 0.6641
211/338 [=================>............] - ETA: 6s - loss: 0.6411 - accuracy: 0.6641
212/338 [=================>............] - ETA: 6s - loss: 0.6410 - accuracy: 0.6642
213/338 [=================>............] - ETA: 6s - loss: 0.6406 - accuracy: 0.6651
214/338 [=================>............] - ETA: 6s - loss: 0.6407 - accuracy: 0.6652
215/338 [==================>...........] - ETA: 6s - loss: 0.6406 - accuracy: 0.6654
216/338 [==================>...........] - ETA: 6s - loss: 0.6408 - accuracy: 0.6648
217/338 [==================>...........] - ETA: 6s - loss: 0.6408 - accuracy: 0.6646
218/338 [==================>...........] - ETA: 6s - loss: 0.6409 - accuracy: 0.6640
219/338 [==================>...........] - ETA: 6s - loss: 0.6411 - accuracy: 0.6638
220/338 [==================>...........] - ETA: 6s - loss: 0.6408 - accuracy: 0.6643
221/338 [==================>...........] - ETA: 6s - loss: 0.6406 - accuracy: 0.6645
222/338 [==================>...........] - ETA: 6s - loss: 0.6409 - accuracy: 0.6640
223/338 [==================>...........] - ETA: 6s - loss: 0.6410 - accuracy: 0.6638
224/338 [==================>...........] - ETA: 6s - loss: 0.6413 - accuracy: 0.6636
225/338 [==================>...........] - ETA: 6s - loss: 0.6415 - accuracy: 0.6635
226/338 [===================>..........] - ETA: 6s - loss: 0.6413 - accuracy: 0.6641
227/338 [===================>..........] - ETA: 6s - loss: 0.6412 - accuracy: 0.6644
228/338 [===================>..........] - ETA: 6s - loss: 0.6412 - accuracy: 0.6643
229/338 [===================>..........] - ETA: 5s - loss: 0.6413 - accuracy: 0.6638
230/338 [===================>..........] - ETA: 5s - loss: 0.6411 - accuracy: 0.6640
231/338 [===================>..........] - ETA: 5s - loss: 0.6409 - accuracy: 0.6642
232/338 [===================>..........] - ETA: 5s - loss: 0.6407 - accuracy: 0.6646
233/338 [===================>..........] - ETA: 5s - loss: 0.6414 - accuracy: 0.6634
234/338 [===================>..........] - ETA: 5s - loss: 0.6415 - accuracy: 0.6633
235/338 [===================>..........] - ETA: 5s - loss: 0.6415 - accuracy: 0.6636
236/338 [===================>..........] - ETA: 5s - loss: 0.6415 - accuracy: 0.6635
237/338 [====================>.........] - ETA: 5s - loss: 0.6417 - accuracy: 0.6632
238/338 [====================>.........] - ETA: 5s - loss: 0.6421 - accuracy: 0.6627
239/338 [====================>.........] - ETA: 5s - loss: 0.6417 - accuracy: 0.6632
240/338 [====================>.........] - ETA: 5s - loss: 0.6418 - accuracy: 0.6630
241/338 [====================>.........] - ETA: 5s - loss: 0.6415 - accuracy: 0.6635
242/338 [====================>.........] - ETA: 5s - loss: 0.6413 - accuracy: 0.6637
243/338 [====================>.........] - ETA: 5s - loss: 0.6414 - accuracy: 0.6636
244/338 [====================>.........] - ETA: 5s - loss: 0.6415 - accuracy: 0.6633
245/338 [====================>.........] - ETA: 5s - loss: 0.6415 - accuracy: 0.6635
246/338 [====================>.........] - ETA: 5s - loss: 0.6417 - accuracy: 0.6631
247/338 [====================>.........] - ETA: 4s - loss: 0.6414 - accuracy: 0.6636
248/338 [=====================>........] - ETA: 4s - loss: 0.6413 - accuracy: 0.6638
249/338 [=====================>........] - ETA: 4s - loss: 0.6415 - accuracy: 0.6633
250/338 [=====================>........] - ETA: 4s - loss: 0.6415 - accuracy: 0.6633
251/338 [=====================>........] - ETA: 4s - loss: 0.6415 - accuracy: 0.6632
252/338 [=====================>........] - ETA: 4s - loss: 0.6413 - accuracy: 0.6634
253/338 [=====================>........] - ETA: 4s - loss: 0.6414 - accuracy: 0.6630
254/338 [=====================>........] - ETA: 4s - loss: 0.6413 - accuracy: 0.6629
255/338 [=====================>........] - ETA: 4s - loss: 0.6412 - accuracy: 0.6630
256/338 [=====================>........] - ETA: 4s - loss: 0.6411 - accuracy: 0.6630
257/338 [=====================>........] - ETA: 4s - loss: 0.6411 - accuracy: 0.6629
258/338 [=====================>........] - ETA: 4s - loss: 0.6411 - accuracy: 0.6627
259/338 [=====================>........] - ETA: 4s - loss: 0.6414 - accuracy: 0.6625
260/338 [======================>.......] - ETA: 4s - loss: 0.6414 - accuracy: 0.6624
261/338 [======================>.......] - ETA: 4s - loss: 0.6414 - accuracy: 0.6626
262/338 [======================>.......] - ETA: 4s - loss: 0.6414 - accuracy: 0.6627
263/338 [======================>.......] - ETA: 4s - loss: 0.6415 - accuracy: 0.6623
264/338 [======================>.......] - ETA: 4s - loss: 0.6419 - accuracy: 0.6615
265/338 [======================>.......] - ETA: 4s - loss: 0.6420 - accuracy: 0.6613
266/338 [======================>.......] - ETA: 3s - loss: 0.6423 - accuracy: 0.6607
267/338 [======================>.......] - ETA: 3s - loss: 0.6421 - accuracy: 0.6612
268/338 [======================>.......] - ETA: 3s - loss: 0.6423 - accuracy: 0.6607
269/338 [======================>.......] - ETA: 3s - loss: 0.6423 - accuracy: 0.6605
270/338 [======================>.......] - ETA: 3s - loss: 0.6424 - accuracy: 0.6603
271/338 [=======================>......] - ETA: 3s - loss: 0.6424 - accuracy: 0.6604
272/338 [=======================>......] - ETA: 3s - loss: 0.6422 - accuracy: 0.6608
273/338 [=======================>......] - ETA: 3s - loss: 0.6423 - accuracy: 0.6606
274/338 [=======================>......] - ETA: 3s - loss: 0.6423 - accuracy: 0.6606
275/338 [=======================>......] - ETA: 3s - loss: 0.6420 - accuracy: 0.6614
276/338 [=======================>......] - ETA: 3s - loss: 0.6420 - accuracy: 0.6615
277/338 [=======================>......] - ETA: 3s - loss: 0.6423 - accuracy: 0.6610
278/338 [=======================>......] - ETA: 3s - loss: 0.6422 - accuracy: 0.6609
279/338 [=======================>......] - ETA: 3s - loss: 0.6424 - accuracy: 0.6607
280/338 [=======================>......] - ETA: 3s - loss: 0.6425 - accuracy: 0.6605
281/338 [=======================>......] - ETA: 3s - loss: 0.6425 - accuracy: 0.6606
282/338 [========================>.....] - ETA: 3s - loss: 0.6425 - accuracy: 0.6605
283/338 [========================>.....] - ETA: 3s - loss: 0.6425 - accuracy: 0.6607
284/338 [========================>.....] - ETA: 2s - loss: 0.6425 - accuracy: 0.6603
285/338 [========================>.....] - ETA: 2s - loss: 0.6426 - accuracy: 0.6603
286/338 [========================>.....] - ETA: 2s - loss: 0.6429 - accuracy: 0.6600
287/338 [========================>.....] - ETA: 2s - loss: 0.6429 - accuracy: 0.6601
288/338 [========================>.....] - ETA: 2s - loss: 0.6427 - accuracy: 0.6603
289/338 [========================>.....] - ETA: 2s - loss: 0.6424 - accuracy: 0.6607
290/338 [========================>.....] - ETA: 2s - loss: 0.6424 - accuracy: 0.6608
291/338 [========================>.....] - ETA: 2s - loss: 0.6421 - accuracy: 0.6613
292/338 [========================>.....] - ETA: 2s - loss: 0.6421 - accuracy: 0.6614
293/338 [=========================>....] - ETA: 2s - loss: 0.6420 - accuracy: 0.6614
294/338 [=========================>....] - ETA: 2s - loss: 0.6419 - accuracy: 0.6616
295/338 [=========================>....] - ETA: 2s - loss: 0.6419 - accuracy: 0.6617
296/338 [=========================>....] - ETA: 2s - loss: 0.6419 - accuracy: 0.6617
297/338 [=========================>....] - ETA: 2s - loss: 0.6417 - accuracy: 0.6618
298/338 [=========================>....] - ETA: 2s - loss: 0.6417 - accuracy: 0.6617
299/338 [=========================>....] - ETA: 2s - loss: 0.6414 - accuracy: 0.6621
300/338 [=========================>....] - ETA: 2s - loss: 0.6416 - accuracy: 0.6617
301/338 [=========================>....] - ETA: 2s - loss: 0.6415 - accuracy: 0.6618
302/338 [=========================>....] - ETA: 1s - loss: 0.6415 - accuracy: 0.6615
303/338 [=========================>....] - ETA: 1s - loss: 0.6416 - accuracy: 0.6615
304/338 [=========================>....] - ETA: 1s - loss: 0.6415 - accuracy: 0.6617
305/338 [==========================>...] - ETA: 1s - loss: 0.6414 - accuracy: 0.6620
306/338 [==========================>...] - ETA: 1s - loss: 0.6415 - accuracy: 0.6618
307/338 [==========================>...] - ETA: 1s - loss: 0.6414 - accuracy: 0.6617
308/338 [==========================>...] - ETA: 1s - loss: 0.6417 - accuracy: 0.6613
309/338 [==========================>...] - ETA: 1s - loss: 0.6416 - accuracy: 0.6615
310/338 [==========================>...] - ETA: 1s - loss: 0.6416 - accuracy: 0.6612
311/338 [==========================>...] - ETA: 1s - loss: 0.6416 - accuracy: 0.6614
312/338 [==========================>...] - ETA: 1s - loss: 0.6416 - accuracy: 0.6613
313/338 [==========================>...] - ETA: 1s - loss: 0.6416 - accuracy: 0.6611
314/338 [==========================>...] - ETA: 1s - loss: 0.6416 - accuracy: 0.6613
315/338 [==========================>...] - ETA: 1s - loss: 0.6414 - accuracy: 0.6613
316/338 [===========================>..] - ETA: 1s - loss: 0.6416 - accuracy: 0.6611
317/338 [===========================>..] - ETA: 1s - loss: 0.6417 - accuracy: 0.6608
318/338 [===========================>..] - ETA: 1s - loss: 0.6419 - accuracy: 0.6603
319/338 [===========================>..] - ETA: 1s - loss: 0.6418 - accuracy: 0.6607
320/338 [===========================>..] - ETA: 0s - loss: 0.6419 - accuracy: 0.6604
321/338 [===========================>..] - ETA: 0s - loss: 0.6421 - accuracy: 0.6602
322/338 [===========================>..] - ETA: 0s - loss: 0.6420 - accuracy: 0.6604
323/338 [===========================>..] - ETA: 0s - loss: 0.6419 - accuracy: 0.6604
324/338 [===========================>..] - ETA: 0s - loss: 0.6420 - accuracy: 0.6604
325/338 [===========================>..] - ETA: 0s - loss: 0.6418 - accuracy: 0.6608
326/338 [===========================>..] - ETA: 0s - loss: 0.6419 - accuracy: 0.6608
327/338 [============================>.] - ETA: 0s - loss: 0.6419 - accuracy: 0.6609
328/338 [============================>.] - ETA: 0s - loss: 0.6418 - accuracy: 0.6609
329/338 [============================>.] - ETA: 0s - loss: 0.6420 - accuracy: 0.6608
330/338 [============================>.] - ETA: 0s - loss: 0.6418 - accuracy: 0.6612
331/338 [============================>.] - ETA: 0s - loss: 0.6418 - accuracy: 0.6613
332/338 [============================>.] - ETA: 0s - loss: 0.6418 - accuracy: 0.6612
333/338 [============================>.] - ETA: 0s - loss: 0.6421 - accuracy: 0.6608
334/338 [============================>.] - ETA: 0s - loss: 0.6421 - accuracy: 0.6609
335/338 [============================>.] - ETA: 0s - loss: 0.6419 - accuracy: 0.6612
336/338 [============================>.] - ETA: 0s - loss: 0.6420 - accuracy: 0.6612
337/338 [============================>.] - ETA: 0s - loss: 0.6420 - accuracy: 0.6610
338/338 [==============================] - 20s 59ms/step - loss: 0.6421 - accuracy: 0.6607 - val_loss: 0.6564 - val_accuracy: 0.6429
1/97 [..............................] - ETA: 2s - loss: 0.6222 - accuracy: 0.6875
5/97 [>.............................] - ETA: 1s - loss: 0.6673 - accuracy: 0.6062
9/97 [=>............................] - ETA: 1s - loss: 0.6800 - accuracy: 0.5938
13/97 [===>..........................] - ETA: 1s - loss: 0.6656 - accuracy: 0.6226
17/97 [====>.........................] - ETA: 1s - loss: 0.6661 - accuracy: 0.6250
21/97 [=====>........................] - ETA: 1s - loss: 0.6644 - accuracy: 0.6235
25/97 [======>.......................] - ETA: 1s - loss: 0.6616 - accuracy: 0.6263
29/97 [=======>......................] - ETA: 1s - loss: 0.6644 - accuracy: 0.6239
33/97 [=========>....................] - ETA: 0s - loss: 0.6625 - accuracy: 0.6316
37/97 [==========>...................] - ETA: 0s - loss: 0.6599 - accuracy: 0.6343
41/97 [===========>..................] - ETA: 0s - loss: 0.6621 - accuracy: 0.6288
45/97 [============>.................] - ETA: 0s - loss: 0.6630 - accuracy: 0.6292
49/97 [==============>...............] - ETA: 0s - loss: 0.6617 - accuracy: 0.6327
53/97 [===============>..............] - ETA: 0s - loss: 0.6599 - accuracy: 0.6350
57/97 [================>.............] - ETA: 0s - loss: 0.6594 - accuracy: 0.6349
61/97 [=================>............] - ETA: 0s - loss: 0.6600 - accuracy: 0.6337
65/97 [===================>..........] - ETA: 0s - loss: 0.6593 - accuracy: 0.6346
69/97 [====================>.........] - ETA: 0s - loss: 0.6597 - accuracy: 0.6341
73/97 [=====================>........] - ETA: 0s - loss: 0.6599 - accuracy: 0.6361
77/97 [======================>.......] - ETA: 0s - loss: 0.6600 - accuracy: 0.6368
81/97 [========================>.....] - ETA: 0s - loss: 0.6579 - accuracy: 0.6389
85/97 [=========================>....] - ETA: 0s - loss: 0.6582 - accuracy: 0.6371
89/97 [==========================>...] - ETA: 0s - loss: 0.6585 - accuracy: 0.6373
93/97 [===========================>..] - ETA: 0s - loss: 0.6578 - accuracy: 0.6401
97/97 [==============================] - ETA: 0s - loss: 0.6564 - accuracy: 0.6429
97/97 [==============================] - 2s 16ms/step - loss: 0.6564 - accuracy: 0.6429
Epoch 1/5
2024-06-04 17:41:52.632537: W external/local_tsl/tsl/framework/cpu_allocator_impl.cc:83] Allocation of 728865000 exceeds 10% of free system memory.
1/338 [..............................] - ETA: 5:34 - loss: 0.7081 - accuracy: 0.6562
3/338 [..............................] - ETA: 11s - loss: 0.7093 - accuracy: 0.5208
5/338 [..............................] - ETA: 11s - loss: 0.7092 - accuracy: 0.4875
7/338 [..............................] - ETA: 11s - loss: 0.7084 - accuracy: 0.5045
9/338 [..............................] - ETA: 11s - loss: 0.7078 - accuracy: 0.5104
11/338 [..............................] - ETA: 11s - loss: 0.7073 - accuracy: 0.5028
13/338 [>.............................] - ETA: 11s - loss: 0.7064 - accuracy: 0.5288
15/338 [>.............................] - ETA: 10s - loss: 0.7060 - accuracy: 0.5229
17/338 [>.............................] - ETA: 10s - loss: 0.7059 - accuracy: 0.5074
19/338 [>.............................] - ETA: 10s - loss: 0.7055 - accuracy: 0.5049
21/338 [>.............................] - ETA: 10s - loss: 0.7051 - accuracy: 0.5015
23/338 [=>............................] - ETA: 10s - loss: 0.7046 - accuracy: 0.5095
25/338 [=>............................] - ETA: 10s - loss: 0.7043 - accuracy: 0.5088
27/338 [=>............................] - ETA: 10s - loss: 0.7039 - accuracy: 0.5116
29/338 [=>............................] - ETA: 10s - loss: 0.7034 - accuracy: 0.5194
31/338 [=>............................] - ETA: 10s - loss: 0.7030 - accuracy: 0.5232
33/338 [=>............................] - ETA: 10s - loss: 0.7026 - accuracy: 0.5256
35/338 [==>...........................] - ETA: 10s - loss: 0.7023 - accuracy: 0.5268
37/338 [==>...........................] - ETA: 10s - loss: 0.7022 - accuracy: 0.5211
39/338 [==>...........................] - ETA: 10s - loss: 0.7021 - accuracy: 0.5160
41/338 [==>...........................] - ETA: 9s - loss: 0.7018 - accuracy: 0.5152
43/338 [==>...........................] - ETA: 9s - loss: 0.7016 - accuracy: 0.5153
45/338 [==>...........................] - ETA: 9s - loss: 0.7013 - accuracy: 0.5181
47/338 [===>..........................] - ETA: 9s - loss: 0.7011 - accuracy: 0.5206
49/338 [===>..........................] - ETA: 9s - loss: 0.7010 - accuracy: 0.5166
51/338 [===>..........................] - ETA: 9s - loss: 0.7007 - accuracy: 0.5202
53/338 [===>..........................] - ETA: 9s - loss: 0.7005 - accuracy: 0.5195
55/338 [===>..........................] - ETA: 9s - loss: 0.7003 - accuracy: 0.5199
57/338 [====>.........................] - ETA: 9s - loss: 0.7002 - accuracy: 0.5175
59/338 [====>.........................] - ETA: 9s - loss: 0.7002 - accuracy: 0.5143
61/338 [====>.........................] - ETA: 9s - loss: 0.7000 - accuracy: 0.5159
63/338 [====>.........................] - ETA: 9s - loss: 0.6999 - accuracy: 0.5159
65/338 [====>.........................] - ETA: 9s - loss: 0.6998 - accuracy: 0.5135
67/338 [====>.........................] - ETA: 9s - loss: 0.6997 - accuracy: 0.5131
69/338 [=====>........................] - ETA: 9s - loss: 0.6996 - accuracy: 0.5131
71/338 [=====>........................] - ETA: 8s - loss: 0.6995 - accuracy: 0.5123
73/338 [=====>........................] - ETA: 8s - loss: 0.6993 - accuracy: 0.5124
75/338 [=====>........................] - ETA: 8s - loss: 0.6992 - accuracy: 0.5129
77/338 [=====>........................] - ETA: 8s - loss: 0.6992 - accuracy: 0.5114
79/338 [======>.......................] - ETA: 8s - loss: 0.6990 - accuracy: 0.5127
81/338 [======>.......................] - ETA: 8s - loss: 0.6989 - accuracy: 0.5143
83/338 [======>.......................] - ETA: 8s - loss: 0.6988 - accuracy: 0.5139
85/338 [======>.......................] - ETA: 8s - loss: 0.6987 - accuracy: 0.5140
87/338 [======>.......................] - ETA: 8s - loss: 0.6986 - accuracy: 0.5162
89/338 [======>.......................] - ETA: 8s - loss: 0.6985 - accuracy: 0.5158
91/338 [=======>......................] - ETA: 8s - loss: 0.6985 - accuracy: 0.5144
93/338 [=======>......................] - ETA: 8s - loss: 0.6984 - accuracy: 0.5144
95/338 [=======>......................] - ETA: 8s - loss: 0.6983 - accuracy: 0.5145
97/338 [=======>......................] - ETA: 8s - loss: 0.6982 - accuracy: 0.5151
99/338 [=======>......................] - ETA: 7s - loss: 0.6981 - accuracy: 0.5148
101/338 [=======>......................] - ETA: 7s - loss: 0.6980 - accuracy: 0.5152
103/338 [========>.....................] - ETA: 7s - loss: 0.6979 - accuracy: 0.5170
105/338 [========>.....................] - ETA: 7s - loss: 0.6978 - accuracy: 0.5182
107/338 [========>.....................] - ETA: 7s - loss: 0.6976 - accuracy: 0.5196
109/338 [========>.....................] - ETA: 7s - loss: 0.6976 - accuracy: 0.5189
111/338 [========>.....................] - ETA: 7s - loss: 0.6975 - accuracy: 0.5191
113/338 [=========>....................] - ETA: 7s - loss: 0.6974 - accuracy: 0.5194
115/338 [=========>....................] - ETA: 7s - loss: 0.6974 - accuracy: 0.5185
117/338 [=========>....................] - ETA: 7s - loss: 0.6973 - accuracy: 0.5200
119/338 [=========>....................] - ETA: 7s - loss: 0.6973 - accuracy: 0.5194
121/338 [=========>....................] - ETA: 7s - loss: 0.6972 - accuracy: 0.5204
123/338 [=========>....................] - ETA: 7s - loss: 0.6971 - accuracy: 0.5206
125/338 [==========>...................] - ETA: 7s - loss: 0.6971 - accuracy: 0.5192
127/338 [==========>...................] - ETA: 7s - loss: 0.6972 - accuracy: 0.5180
129/338 [==========>...................] - ETA: 6s - loss: 0.6970 - accuracy: 0.5191
131/338 [==========>...................] - ETA: 6s - loss: 0.6971 - accuracy: 0.5174
133/338 [==========>...................] - ETA: 6s - loss: 0.6969 - accuracy: 0.5197
135/338 [==========>...................] - ETA: 6s - loss: 0.6968 - accuracy: 0.5206
137/338 [===========>..................] - ETA: 6s - loss: 0.6967 - accuracy: 0.5219
139/338 [===========>..................] - ETA: 6s - loss: 0.6965 - accuracy: 0.5234
141/338 [===========>..................] - ETA: 6s - loss: 0.6964 - accuracy: 0.5239
143/338 [===========>..................] - ETA: 6s - loss: 0.6963 - accuracy: 0.5251
145/338 [===========>..................] - ETA: 6s - loss: 0.6962 - accuracy: 0.5254
147/338 [============>.................] - ETA: 6s - loss: 0.6961 - accuracy: 0.5264
149/338 [============>.................] - ETA: 6s - loss: 0.6962 - accuracy: 0.5254
151/338 [============>.................] - ETA: 6s - loss: 0.6960 - accuracy: 0.5267
153/338 [============>.................] - ETA: 6s - loss: 0.6960 - accuracy: 0.5259
155/338 [============>.................] - ETA: 6s - loss: 0.6960 - accuracy: 0.5256
157/338 [============>.................] - ETA: 6s - loss: 0.6960 - accuracy: 0.5259
159/338 [=============>................] - ETA: 5s - loss: 0.6961 - accuracy: 0.5242
161/338 [=============>................] - ETA: 5s - loss: 0.6959 - accuracy: 0.5252
163/338 [=============>................] - ETA: 5s - loss: 0.6958 - accuracy: 0.5261
165/338 [=============>................] - ETA: 5s - loss: 0.6958 - accuracy: 0.5258
167/338 [=============>................] - ETA: 5s - loss: 0.6958 - accuracy: 0.5254
169/338 [==============>...............] - ETA: 5s - loss: 0.6958 - accuracy: 0.5253
171/338 [==============>...............] - ETA: 5s - loss: 0.6956 - accuracy: 0.5263
173/338 [==============>...............] - ETA: 5s - loss: 0.6955 - accuracy: 0.5269
175/338 [==============>...............] - ETA: 5s - loss: 0.6954 - accuracy: 0.5277
177/338 [==============>...............] - ETA: 5s - loss: 0.6953 - accuracy: 0.5282
179/338 [==============>...............] - ETA: 5s - loss: 0.6953 - accuracy: 0.5285
181/338 [===============>..............] - ETA: 5s - loss: 0.6952 - accuracy: 0.5292
183/338 [===============>..............] - ETA: 5s - loss: 0.6951 - accuracy: 0.5294
185/338 [===============>..............] - ETA: 5s - loss: 0.6951 - accuracy: 0.5294
187/338 [===============>..............] - ETA: 5s - loss: 0.6950 - accuracy: 0.5299
189/338 [===============>..............] - ETA: 4s - loss: 0.6950 - accuracy: 0.5289
191/338 [===============>..............] - ETA: 4s - loss: 0.6950 - accuracy: 0.5290
193/338 [================>.............] - ETA: 4s - loss: 0.6948 - accuracy: 0.5300
195/338 [================>.............] - ETA: 4s - loss: 0.6948 - accuracy: 0.5304
197/338 [================>.............] - ETA: 4s - loss: 0.6946 - accuracy: 0.5316
199/338 [================>.............] - ETA: 4s - loss: 0.6946 - accuracy: 0.5316
201/338 [================>.............] - ETA: 4s - loss: 0.6945 - accuracy: 0.5320
203/338 [=================>............] - ETA: 4s - loss: 0.6943 - accuracy: 0.5333
205/338 [=================>............] - ETA: 4s - loss: 0.6944 - accuracy: 0.5328
207/338 [=================>............] - ETA: 4s - loss: 0.6944 - accuracy: 0.5319
209/338 [=================>............] - ETA: 4s - loss: 0.6945 - accuracy: 0.5312
211/338 [=================>............] - ETA: 4s - loss: 0.6945 - accuracy: 0.5307
213/338 [=================>............] - ETA: 4s - loss: 0.6944 - accuracy: 0.5314
215/338 [==================>...........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5305
217/338 [==================>...........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5301
219/338 [==================>...........] - ETA: 3s - loss: 0.6945 - accuracy: 0.5300
221/338 [==================>...........] - ETA: 3s - loss: 0.6945 - accuracy: 0.5300
223/338 [==================>...........] - ETA: 3s - loss: 0.6944 - accuracy: 0.5304
225/338 [==================>...........] - ETA: 3s - loss: 0.6943 - accuracy: 0.5307
227/338 [===================>..........] - ETA: 3s - loss: 0.6943 - accuracy: 0.5310
229/338 [===================>..........] - ETA: 3s - loss: 0.6943 - accuracy: 0.5306
231/338 [===================>..........] - ETA: 3s - loss: 0.6942 - accuracy: 0.5311
233/338 [===================>..........] - ETA: 3s - loss: 0.6942 - accuracy: 0.5307
235/338 [===================>..........] - ETA: 3s - loss: 0.6941 - accuracy: 0.5316
237/338 [====================>.........] - ETA: 3s - loss: 0.6940 - accuracy: 0.5326
239/338 [====================>.........] - ETA: 3s - loss: 0.6939 - accuracy: 0.5327
241/338 [====================>.........] - ETA: 3s - loss: 0.6938 - accuracy: 0.5332
243/338 [====================>.........] - ETA: 3s - loss: 0.6938 - accuracy: 0.5329
245/338 [====================>.........] - ETA: 3s - loss: 0.6937 - accuracy: 0.5335
247/338 [====================>.........] - ETA: 3s - loss: 0.6937 - accuracy: 0.5331
249/338 [=====================>........] - ETA: 2s - loss: 0.6937 - accuracy: 0.5334
251/338 [=====================>........] - ETA: 2s - loss: 0.6936 - accuracy: 0.5335
253/338 [=====================>........] - ETA: 2s - loss: 0.6937 - accuracy: 0.5331
255/338 [=====================>........] - ETA: 2s - loss: 0.6937 - accuracy: 0.5325
257/338 [=====================>........] - ETA: 2s - loss: 0.6938 - accuracy: 0.5319
259/338 [=====================>........] - ETA: 2s - loss: 0.6937 - accuracy: 0.5320
261/338 [======================>.......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5314
263/338 [======================>.......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5312
265/338 [======================>.......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5311
267/338 [======================>.......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5309
269/338 [======================>.......] - ETA: 2s - loss: 0.6937 - accuracy: 0.5311
271/338 [=======================>......] - ETA: 2s - loss: 0.6937 - accuracy: 0.5309
273/338 [=======================>......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5301
275/338 [=======================>......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5297
277/338 [=======================>......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5302
279/338 [=======================>......] - ETA: 1s - loss: 0.6938 - accuracy: 0.5302
281/338 [=======================>......] - ETA: 1s - loss: 0.6937 - accuracy: 0.5304
283/338 [========================>.....] - ETA: 1s - loss: 0.6938 - accuracy: 0.5298
285/338 [========================>.....] - ETA: 1s - loss: 0.6938 - accuracy: 0.5297
287/338 [========================>.....] - ETA: 1s - loss: 0.6937 - accuracy: 0.5295
289/338 [========================>.....] - ETA: 1s - loss: 0.6937 - accuracy: 0.5297
291/338 [========================>.....] - ETA: 1s - loss: 0.6938 - accuracy: 0.5290
293/338 [=========================>....] - ETA: 1s - loss: 0.6936 - accuracy: 0.5301
295/338 [=========================>....] - ETA: 1s - loss: 0.6937 - accuracy: 0.5291
297/338 [=========================>....] - ETA: 1s - loss: 0.6937 - accuracy: 0.5291
299/338 [=========================>....] - ETA: 1s - loss: 0.6937 - accuracy: 0.5283
301/338 [=========================>....] - ETA: 1s - loss: 0.6937 - accuracy: 0.5284
303/338 [=========================>....] - ETA: 1s - loss: 0.6937 - accuracy: 0.5284
305/338 [==========================>...] - ETA: 1s - loss: 0.6937 - accuracy: 0.5286
307/338 [==========================>...] - ETA: 1s - loss: 0.6936 - accuracy: 0.5290
309/338 [==========================>...] - ETA: 0s - loss: 0.6936 - accuracy: 0.5290
311/338 [==========================>...] - ETA: 0s - loss: 0.6936 - accuracy: 0.5283
313/338 [==========================>...] - ETA: 0s - loss: 0.6936 - accuracy: 0.5290
315/338 [==========================>...] - ETA: 0s - loss: 0.6936 - accuracy: 0.5289
317/338 [===========================>..] - ETA: 0s - loss: 0.6935 - accuracy: 0.5287
319/338 [===========================>..] - ETA: 0s - loss: 0.6936 - accuracy: 0.5285
321/338 [===========================>..] - ETA: 0s - loss: 0.6936 - accuracy: 0.5280
323/338 [===========================>..] - ETA: 0s - loss: 0.6935 - accuracy: 0.5284
325/338 [===========================>..] - ETA: 0s - loss: 0.6936 - accuracy: 0.5279
327/338 [============================>.] - ETA: 0s - loss: 0.6935 - accuracy: 0.5288
329/338 [============================>.] - ETA: 0s - loss: 0.6935 - accuracy: 0.5288
331/338 [============================>.] - ETA: 0s - loss: 0.6935 - accuracy: 0.5286
333/338 [============================>.] - ETA: 0s - loss: 0.6935 - accuracy: 0.5279
335/338 [============================>.] - ETA: 0s - loss: 0.6935 - accuracy: 0.5273
337/338 [============================>.] - ETA: 0s - loss: 0.6935 - accuracy: 0.5274
338/338 [==============================] - 13s 36ms/step - loss: 0.6935 - accuracy: 0.5271 - val_loss: 0.6919 - val_accuracy: 0.5235
Epoch 2/5
1/338 [..............................] - ETA: 11s - loss: 0.6857 - accuracy: 0.5938
3/338 [..............................] - ETA: 11s - loss: 0.6883 - accuracy: 0.5729
5/338 [..............................] - ETA: 10s - loss: 0.6912 - accuracy: 0.5312
7/338 [..............................] - ETA: 11s - loss: 0.6895 - accuracy: 0.5491
9/338 [..............................] - ETA: 10s - loss: 0.6887 - accuracy: 0.5521
11/338 [..............................] - ETA: 10s - loss: 0.6889 - accuracy: 0.5511
13/338 [>.............................] - ETA: 10s - loss: 0.6881 - accuracy: 0.5529
15/338 [>.............................] - ETA: 10s - loss: 0.6877 - accuracy: 0.5542
17/338 [>.............................] - ETA: 10s - loss: 0.6879 - accuracy: 0.5496
19/338 [>.............................] - ETA: 10s - loss: 0.6879 - accuracy: 0.5493
21/338 [>.............................] - ETA: 10s - loss: 0.6893 - accuracy: 0.5387
23/338 [=>............................] - ETA: 10s - loss: 0.6897 - accuracy: 0.5340
25/338 [=>............................] - ETA: 10s - loss: 0.6893 - accuracy: 0.5387
27/338 [=>............................] - ETA: 10s - loss: 0.6896 - accuracy: 0.5359
29/338 [=>............................] - ETA: 10s - loss: 0.6895 - accuracy: 0.5356
31/338 [=>............................] - ETA: 10s - loss: 0.6894 - accuracy: 0.5363
33/338 [=>............................] - ETA: 10s - loss: 0.6898 - accuracy: 0.5303
35/338 [==>...........................] - ETA: 10s - loss: 0.6902 - accuracy: 0.5259
37/338 [==>...........................] - ETA: 9s - loss: 0.6901 - accuracy: 0.5262
39/338 [==>...........................] - ETA: 9s - loss: 0.6903 - accuracy: 0.5248
41/338 [==>...........................] - ETA: 9s - loss: 0.6903 - accuracy: 0.5252
43/338 [==>...........................] - ETA: 9s - loss: 0.6905 - accuracy: 0.5218
45/338 [==>...........................] - ETA: 9s - loss: 0.6905 - accuracy: 0.5236
47/338 [===>..........................] - ETA: 9s - loss: 0.6904 - accuracy: 0.5266
49/338 [===>..........................] - ETA: 9s - loss: 0.6904 - accuracy: 0.5261
51/338 [===>..........................] - ETA: 9s - loss: 0.6901 - accuracy: 0.5288
53/338 [===>..........................] - ETA: 9s - loss: 0.6899 - accuracy: 0.5312
55/338 [===>..........................] - ETA: 9s - loss: 0.6901 - accuracy: 0.5307
57/338 [====>.........................] - ETA: 9s - loss: 0.6899 - accuracy: 0.5323
59/338 [====>.........................] - ETA: 9s - loss: 0.6902 - accuracy: 0.5312
61/338 [====>.........................] - ETA: 9s - loss: 0.6900 - accuracy: 0.5318
63/338 [====>.........................] - ETA: 9s - loss: 0.6898 - accuracy: 0.5332
65/338 [====>.........................] - ETA: 9s - loss: 0.6900 - accuracy: 0.5317
67/338 [====>.........................] - ETA: 8s - loss: 0.6902 - accuracy: 0.5303
69/338 [=====>........................] - ETA: 8s - loss: 0.6902 - accuracy: 0.5294
71/338 [=====>........................] - ETA: 8s - loss: 0.6900 - accuracy: 0.5299
73/338 [=====>........................] - ETA: 8s - loss: 0.6898 - accuracy: 0.5317
75/338 [=====>........................] - ETA: 8s - loss: 0.6892 - accuracy: 0.5358
77/338 [=====>........................] - ETA: 8s - loss: 0.6894 - accuracy: 0.5345
79/338 [======>.......................] - ETA: 8s - loss: 0.6896 - accuracy: 0.5332
81/338 [======>.......................] - ETA: 8s - loss: 0.6896 - accuracy: 0.5328
83/338 [======>.......................] - ETA: 8s - loss: 0.6898 - accuracy: 0.5316
85/338 [======>.......................] - ETA: 8s - loss: 0.6896 - accuracy: 0.5327
87/338 [======>.......................] - ETA: 8s - loss: 0.6896 - accuracy: 0.5320
89/338 [======>.......................] - ETA: 8s - loss: 0.6898 - accuracy: 0.5309
91/338 [=======>......................] - ETA: 8s - loss: 0.6899 - accuracy: 0.5292
93/338 [=======>......................] - ETA: 8s - loss: 0.6898 - accuracy: 0.5292
95/338 [=======>......................] - ETA: 8s - loss: 0.6899 - accuracy: 0.5289
97/338 [=======>......................] - ETA: 7s - loss: 0.6898 - accuracy: 0.5296
99/338 [=======>......................] - ETA: 7s - loss: 0.6899 - accuracy: 0.5284
101/338 [=======>......................] - ETA: 7s - loss: 0.6897 - accuracy: 0.5303
103/338 [========>.....................] - ETA: 7s - loss: 0.6895 - accuracy: 0.5316
105/338 [========>.....................] - ETA: 7s - loss: 0.6896 - accuracy: 0.5301
107/338 [========>.....................] - ETA: 7s - loss: 0.6897 - accuracy: 0.5298
109/338 [========>.....................] - ETA: 7s - loss: 0.6899 - accuracy: 0.5278
111/338 [========>.....................] - ETA: 7s - loss: 0.6899 - accuracy: 0.5279
113/338 [=========>....................] - ETA: 7s - loss: 0.6898 - accuracy: 0.5282
115/338 [=========>....................] - ETA: 7s - loss: 0.6900 - accuracy: 0.5266
117/338 [=========>....................] - ETA: 7s - loss: 0.6899 - accuracy: 0.5264
119/338 [=========>....................] - ETA: 7s - loss: 0.6899 - accuracy: 0.5273
121/338 [=========>....................] - ETA: 7s - loss: 0.6898 - accuracy: 0.5276
123/338 [=========>....................] - ETA: 7s - loss: 0.6898 - accuracy: 0.5285
125/338 [==========>...................] - ETA: 7s - loss: 0.6898 - accuracy: 0.5278
127/338 [==========>...................] - ETA: 6s - loss: 0.6897 - accuracy: 0.5281
129/338 [==========>...................] - ETA: 6s - loss: 0.6896 - accuracy: 0.5283
131/338 [==========>...................] - ETA: 6s - loss: 0.6896 - accuracy: 0.5284
133/338 [==========>...................] - ETA: 6s - loss: 0.6895 - accuracy: 0.5291
135/338 [==========>...................] - ETA: 6s - loss: 0.6893 - accuracy: 0.5310
137/338 [===========>..................] - ETA: 6s - loss: 0.6892 - accuracy: 0.5310
139/338 [===========>..................] - ETA: 6s - loss: 0.6891 - accuracy: 0.5328
141/338 [===========>..................] - ETA: 6s - loss: 0.6891 - accuracy: 0.5332
143/338 [===========>..................] - ETA: 6s - loss: 0.6890 - accuracy: 0.5337
145/338 [===========>..................] - ETA: 6s - loss: 0.6890 - accuracy: 0.5330
147/338 [============>.................] - ETA: 6s - loss: 0.6891 - accuracy: 0.5327
149/338 [============>.................] - ETA: 6s - loss: 0.6889 - accuracy: 0.5340
151/338 [============>.................] - ETA: 6s - loss: 0.6887 - accuracy: 0.5348
153/338 [============>.................] - ETA: 6s - loss: 0.6888 - accuracy: 0.5343
155/338 [============>.................] - ETA: 6s - loss: 0.6890 - accuracy: 0.5323
157/338 [============>.................] - ETA: 6s - loss: 0.6889 - accuracy: 0.5330
159/338 [=============>................] - ETA: 5s - loss: 0.6888 - accuracy: 0.5332
161/338 [=============>................] - ETA: 5s - loss: 0.6889 - accuracy: 0.5326
163/338 [=============>................] - ETA: 5s - loss: 0.6888 - accuracy: 0.5330
165/338 [=============>................] - ETA: 5s - loss: 0.6887 - accuracy: 0.5341
167/338 [=============>................] - ETA: 5s - loss: 0.6887 - accuracy: 0.5342
169/338 [==============>...............] - ETA: 5s - loss: 0.6886 - accuracy: 0.5344
171/338 [==============>...............] - ETA: 5s - loss: 0.6887 - accuracy: 0.5331
173/338 [==============>...............] - ETA: 5s - loss: 0.6886 - accuracy: 0.5327
175/338 [==============>...............] - ETA: 5s - loss: 0.6887 - accuracy: 0.5323
177/338 [==============>...............] - ETA: 5s - loss: 0.6888 - accuracy: 0.5320
179/338 [==============>...............] - ETA: 5s - loss: 0.6888 - accuracy: 0.5312
181/338 [===============>..............] - ETA: 5s - loss: 0.6889 - accuracy: 0.5304
183/338 [===============>..............] - ETA: 5s - loss: 0.6888 - accuracy: 0.5302
185/338 [===============>..............] - ETA: 5s - loss: 0.6888 - accuracy: 0.5306
187/338 [===============>..............] - ETA: 5s - loss: 0.6888 - accuracy: 0.5306
189/338 [===============>..............] - ETA: 4s - loss: 0.6887 - accuracy: 0.5306
191/338 [===============>..............] - ETA: 4s - loss: 0.6887 - accuracy: 0.5308
193/338 [================>.............] - ETA: 4s - loss: 0.6888 - accuracy: 0.5309
195/338 [================>.............] - ETA: 4s - loss: 0.6888 - accuracy: 0.5312
197/338 [================>.............] - ETA: 4s - loss: 0.6888 - accuracy: 0.5308
199/338 [================>.............] - ETA: 4s - loss: 0.6887 - accuracy: 0.5311
201/338 [================>.............] - ETA: 4s - loss: 0.6888 - accuracy: 0.5302
203/338 [=================>............] - ETA: 4s - loss: 0.6889 - accuracy: 0.5296
205/338 [=================>............] - ETA: 4s - loss: 0.6887 - accuracy: 0.5309
207/338 [=================>............] - ETA: 4s - loss: 0.6888 - accuracy: 0.5300
209/338 [=================>............] - ETA: 4s - loss: 0.6887 - accuracy: 0.5299
211/338 [=================>............] - ETA: 4s - loss: 0.6888 - accuracy: 0.5298
213/338 [=================>............] - ETA: 4s - loss: 0.6888 - accuracy: 0.5295
215/338 [==================>...........] - ETA: 4s - loss: 0.6888 - accuracy: 0.5288
217/338 [==================>...........] - ETA: 4s - loss: 0.6889 - accuracy: 0.5287
219/338 [==================>...........] - ETA: 3s - loss: 0.6889 - accuracy: 0.5285
221/338 [==================>...........] - ETA: 3s - loss: 0.6888 - accuracy: 0.5287
223/338 [==================>...........] - ETA: 3s - loss: 0.6888 - accuracy: 0.5284
225/338 [==================>...........] - ETA: 3s - loss: 0.6888 - accuracy: 0.5289
227/338 [===================>..........] - ETA: 3s - loss: 0.6888 - accuracy: 0.5290
229/338 [===================>..........] - ETA: 3s - loss: 0.6886 - accuracy: 0.5296
231/338 [===================>..........] - ETA: 3s - loss: 0.6885 - accuracy: 0.5303
233/338 [===================>..........] - ETA: 3s - loss: 0.6883 - accuracy: 0.5311
235/338 [===================>..........] - ETA: 3s - loss: 0.6883 - accuracy: 0.5309
237/338 [====================>.........] - ETA: 3s - loss: 0.6886 - accuracy: 0.5295
239/338 [====================>.........] - ETA: 3s - loss: 0.6887 - accuracy: 0.5292
241/338 [====================>.........] - ETA: 3s - loss: 0.6888 - accuracy: 0.5283
243/338 [====================>.........] - ETA: 3s - loss: 0.6889 - accuracy: 0.5285
245/338 [====================>.........] - ETA: 3s - loss: 0.6889 - accuracy: 0.5283
247/338 [====================>.........] - ETA: 3s - loss: 0.6888 - accuracy: 0.5287
249/338 [=====================>........] - ETA: 2s - loss: 0.6888 - accuracy: 0.5291
251/338 [=====================>........] - ETA: 2s - loss: 0.6888 - accuracy: 0.5289
253/338 [=====================>........] - ETA: 2s - loss: 0.6888 - accuracy: 0.5293
255/338 [=====================>........] - ETA: 2s - loss: 0.6888 - accuracy: 0.5289
257/338 [=====================>........] - ETA: 2s - loss: 0.6889 - accuracy: 0.5283
259/338 [=====================>........] - ETA: 2s - loss: 0.6889 - accuracy: 0.5279
261/338 [======================>.......] - ETA: 2s - loss: 0.6889 - accuracy: 0.5279
263/338 [======================>.......] - ETA: 2s - loss: 0.6889 - accuracy: 0.5283
265/338 [======================>.......] - ETA: 2s - loss: 0.6889 - accuracy: 0.5289
267/338 [======================>.......] - ETA: 2s - loss: 0.6888 - accuracy: 0.5290
269/338 [======================>.......] - ETA: 2s - loss: 0.6888 - accuracy: 0.5285
271/338 [=======================>......] - ETA: 2s - loss: 0.6887 - accuracy: 0.5295
273/338 [=======================>......] - ETA: 2s - loss: 0.6888 - accuracy: 0.5290
275/338 [=======================>......] - ETA: 2s - loss: 0.6888 - accuracy: 0.5284
277/338 [=======================>......] - ETA: 2s - loss: 0.6887 - accuracy: 0.5282
279/338 [=======================>......] - ETA: 1s - loss: 0.6888 - accuracy: 0.5277
281/338 [=======================>......] - ETA: 1s - loss: 0.6887 - accuracy: 0.5278
283/338 [========================>.....] - ETA: 1s - loss: 0.6888 - accuracy: 0.5269
285/338 [========================>.....] - ETA: 1s - loss: 0.6888 - accuracy: 0.5277
287/338 [========================>.....] - ETA: 1s - loss: 0.6887 - accuracy: 0.5275
289/338 [========================>.....] - ETA: 1s - loss: 0.6887 - accuracy: 0.5275
291/338 [========================>.....] - ETA: 1s - loss: 0.6887 - accuracy: 0.5271
293/338 [=========================>....] - ETA: 1s - loss: 0.6887 - accuracy: 0.5266
295/338 [=========================>....] - ETA: 1s - loss: 0.6887 - accuracy: 0.5264
297/338 [=========================>....] - ETA: 1s - loss: 0.6887 - accuracy: 0.5265
299/338 [=========================>....] - ETA: 1s - loss: 0.6886 - accuracy: 0.5268
301/338 [=========================>....] - ETA: 1s - loss: 0.6887 - accuracy: 0.5265
303/338 [=========================>....] - ETA: 1s - loss: 0.6886 - accuracy: 0.5265
305/338 [==========================>...] - ETA: 1s - loss: 0.6886 - accuracy: 0.5272
307/338 [==========================>...] - ETA: 1s - loss: 0.6885 - accuracy: 0.5272
309/338 [==========================>...] - ETA: 0s - loss: 0.6885 - accuracy: 0.5270
311/338 [==========================>...] - ETA: 0s - loss: 0.6884 - accuracy: 0.5272
313/338 [==========================>...] - ETA: 0s - loss: 0.6884 - accuracy: 0.5266
315/338 [==========================>...] - ETA: 0s - loss: 0.6884 - accuracy: 0.5265
317/338 [===========================>..] - ETA: 0s - loss: 0.6884 - accuracy: 0.5266
319/338 [===========================>..] - ETA: 0s - loss: 0.6884 - accuracy: 0.5264
321/338 [===========================>..] - ETA: 0s - loss: 0.6884 - accuracy: 0.5267
323/338 [===========================>..] - ETA: 0s - loss: 0.6884 - accuracy: 0.5269
325/338 [===========================>..] - ETA: 0s - loss: 0.6884 - accuracy: 0.5267
327/338 [============================>.] - ETA: 0s - loss: 0.6883 - accuracy: 0.5274
329/338 [============================>.] - ETA: 0s - loss: 0.6883 - accuracy: 0.5279
331/338 [============================>.] - ETA: 0s - loss: 0.6884 - accuracy: 0.5279
333/338 [============================>.] - ETA: 0s - loss: 0.6884 - accuracy: 0.5280
335/338 [============================>.] - ETA: 0s - loss: 0.6883 - accuracy: 0.5288
337/338 [============================>.] - ETA: 0s - loss: 0.6883 - accuracy: 0.5292
338/338 [==============================] - 12s 35ms/step - loss: 0.6883 - accuracy: 0.5292 - val_loss: 0.6859 - val_accuracy: 0.5913
Epoch 3/5
1/338 [..............................] - ETA: 13s - loss: 0.6761 - accuracy: 0.6562
3/338 [..............................] - ETA: 13s - loss: 0.6712 - accuracy: 0.6354
5/338 [..............................] - ETA: 12s - loss: 0.6749 - accuracy: 0.6000
7/338 [..............................] - ETA: 11s - loss: 0.6817 - accuracy: 0.5580
9/338 [..............................] - ETA: 11s - loss: 0.6803 - accuracy: 0.5694
11/338 [..............................] - ETA: 11s - loss: 0.6812 - accuracy: 0.5739
13/338 [>.............................] - ETA: 11s - loss: 0.6812 - accuracy: 0.5745
15/338 [>.............................] - ETA: 10s - loss: 0.6813 - accuracy: 0.5750
17/338 [>.............................] - ETA: 10s - loss: 0.6816 - accuracy: 0.5699
19/338 [>.............................] - ETA: 10s - loss: 0.6827 - accuracy: 0.5658
21/338 [>.............................] - ETA: 10s - loss: 0.6818 - accuracy: 0.5670
23/338 [=>............................] - ETA: 10s - loss: 0.6827 - accuracy: 0.5707
25/338 [=>............................] - ETA: 10s - loss: 0.6829 - accuracy: 0.5750
27/338 [=>............................] - ETA: 10s - loss: 0.6828 - accuracy: 0.5810
29/338 [=>............................] - ETA: 10s - loss: 0.6829 - accuracy: 0.5841
31/338 [=>............................] - ETA: 10s - loss: 0.6831 - accuracy: 0.5817
33/338 [=>............................] - ETA: 10s - loss: 0.6840 - accuracy: 0.5777
35/338 [==>...........................] - ETA: 10s - loss: 0.6845 - accuracy: 0.5795
37/338 [==>...........................] - ETA: 10s - loss: 0.6847 - accuracy: 0.5802
39/338 [==>...........................] - ETA: 9s - loss: 0.6839 - accuracy: 0.5849
41/338 [==>...........................] - ETA: 9s - loss: 0.6838 - accuracy: 0.5854
43/338 [==>...........................] - ETA: 9s - loss: 0.6841 - accuracy: 0.5879
45/338 [==>...........................] - ETA: 9s - loss: 0.6841 - accuracy: 0.5924
47/338 [===>..........................] - ETA: 9s - loss: 0.6839 - accuracy: 0.5951
49/338 [===>..........................] - ETA: 9s - loss: 0.6839 - accuracy: 0.5995
51/338 [===>..........................] - ETA: 9s - loss: 0.6839 - accuracy: 0.6005
53/338 [===>..........................] - ETA: 9s - loss: 0.6841 - accuracy: 0.5985
55/338 [===>..........................] - ETA: 9s - loss: 0.6836 - accuracy: 0.6023
57/338 [====>.........................] - ETA: 9s - loss: 0.6837 - accuracy: 0.6009
59/338 [====>.........................] - ETA: 9s - loss: 0.6839 - accuracy: 0.5975
61/338 [====>.........................] - ETA: 9s - loss: 0.6840 - accuracy: 0.5963
63/338 [====>.........................] - ETA: 9s - loss: 0.6844 - accuracy: 0.5938
65/338 [====>.........................] - ETA: 9s - loss: 0.6844 - accuracy: 0.5947
67/338 [====>.........................] - ETA: 9s - loss: 0.6844 - accuracy: 0.5942
69/338 [=====>........................] - ETA: 8s - loss: 0.6843 - accuracy: 0.5951
71/338 [=====>........................] - ETA: 8s - loss: 0.6843 - accuracy: 0.5946
73/338 [=====>........................] - ETA: 8s - loss: 0.6844 - accuracy: 0.5959
75/338 [=====>........................] - ETA: 8s - loss: 0.6845 - accuracy: 0.5946
77/338 [=====>........................] - ETA: 8s - loss: 0.6845 - accuracy: 0.5938
79/338 [======>.......................] - ETA: 8s - loss: 0.6843 - accuracy: 0.5934
81/338 [======>.......................] - ETA: 8s - loss: 0.6845 - accuracy: 0.5918
83/338 [======>.......................] - ETA: 8s - loss: 0.6846 - accuracy: 0.5907
85/338 [======>.......................] - ETA: 8s - loss: 0.6843 - accuracy: 0.5930
87/338 [======>.......................] - ETA: 8s - loss: 0.6843 - accuracy: 0.5927
89/338 [======>.......................] - ETA: 8s - loss: 0.6844 - accuracy: 0.5934
91/338 [=======>......................] - ETA: 8s - loss: 0.6843 - accuracy: 0.5951
93/338 [=======>......................] - ETA: 8s - loss: 0.6840 - accuracy: 0.5964
95/338 [=======>......................] - ETA: 8s - loss: 0.6839 - accuracy: 0.5967
97/338 [=======>......................] - ETA: 8s - loss: 0.6839 - accuracy: 0.5960
99/338 [=======>......................] - ETA: 7s - loss: 0.6842 - accuracy: 0.5950
101/338 [=======>......................] - ETA: 7s - loss: 0.6839 - accuracy: 0.5965
103/338 [========>.....................] - ETA: 7s - loss: 0.6837 - accuracy: 0.5971
105/338 [========>.....................] - ETA: 7s - loss: 0.6836 - accuracy: 0.5970
107/338 [========>.....................] - ETA: 7s - loss: 0.6835 - accuracy: 0.5964
109/338 [========>.....................] - ETA: 7s - loss: 0.6837 - accuracy: 0.5949
111/338 [========>.....................] - ETA: 7s - loss: 0.6837 - accuracy: 0.5949
113/338 [=========>....................] - ETA: 7s - loss: 0.6831 - accuracy: 0.5979
115/338 [=========>....................] - ETA: 7s - loss: 0.6833 - accuracy: 0.5967
117/338 [=========>....................] - ETA: 7s - loss: 0.6833 - accuracy: 0.5959
119/338 [=========>....................] - ETA: 7s - loss: 0.6835 - accuracy: 0.5951
121/338 [=========>....................] - ETA: 7s - loss: 0.6834 - accuracy: 0.5963
123/338 [=========>....................] - ETA: 7s - loss: 0.6833 - accuracy: 0.5963
125/338 [==========>...................] - ETA: 7s - loss: 0.6832 - accuracy: 0.5972
127/338 [==========>...................] - ETA: 7s - loss: 0.6832 - accuracy: 0.5984
129/338 [==========>...................] - ETA: 6s - loss: 0.6833 - accuracy: 0.5979
131/338 [==========>...................] - ETA: 6s - loss: 0.6830 - accuracy: 0.5997
133/338 [==========>...................] - ETA: 6s - loss: 0.6826 - accuracy: 0.6010
135/338 [==========>...................] - ETA: 6s - loss: 0.6825 - accuracy: 0.6009
137/338 [===========>..................] - ETA: 6s - loss: 0.6825 - accuracy: 0.6015
139/338 [===========>..................] - ETA: 6s - loss: 0.6826 - accuracy: 0.6021
141/338 [===========>..................] - ETA: 6s - loss: 0.6828 - accuracy: 0.6008
143/338 [===========>..................] - ETA: 6s - loss: 0.6826 - accuracy: 0.6012
145/338 [===========>..................] - ETA: 6s - loss: 0.6826 - accuracy: 0.6013
147/338 [============>.................] - ETA: 6s - loss: 0.6828 - accuracy: 0.6010
149/338 [============>.................] - ETA: 6s - loss: 0.6826 - accuracy: 0.6019
151/338 [============>.................] - ETA: 6s - loss: 0.6826 - accuracy: 0.6020
153/338 [============>.................] - ETA: 6s - loss: 0.6825 - accuracy: 0.6025
155/338 [============>.................] - ETA: 6s - loss: 0.6825 - accuracy: 0.6038
157/338 [============>.................] - ETA: 6s - loss: 0.6826 - accuracy: 0.6037
159/338 [=============>................] - ETA: 5s - loss: 0.6824 - accuracy: 0.6053
161/338 [=============>................] - ETA: 5s - loss: 0.6824 - accuracy: 0.6048
163/338 [=============>................] - ETA: 5s - loss: 0.6823 - accuracy: 0.6051
165/338 [=============>................] - ETA: 5s - loss: 0.6823 - accuracy: 0.6053
167/338 [=============>................] - ETA: 5s - loss: 0.6822 - accuracy: 0.6065
169/338 [==============>...............] - ETA: 5s - loss: 0.6824 - accuracy: 0.6060
171/338 [==============>...............] - ETA: 5s - loss: 0.6824 - accuracy: 0.6058
173/338 [==============>...............] - ETA: 5s - loss: 0.6822 - accuracy: 0.6069
175/338 [==============>...............] - ETA: 5s - loss: 0.6822 - accuracy: 0.6070
177/338 [==============>...............] - ETA: 5s - loss: 0.6821 - accuracy: 0.6075
179/338 [==============>...............] - ETA: 5s - loss: 0.6821 - accuracy: 0.6072
181/338 [===============>..............] - ETA: 5s - loss: 0.6822 - accuracy: 0.6067
183/338 [===============>..............] - ETA: 5s - loss: 0.6821 - accuracy: 0.6072
185/338 [===============>..............] - ETA: 5s - loss: 0.6820 - accuracy: 0.6074
187/338 [===============>..............] - ETA: 4s - loss: 0.6819 - accuracy: 0.6085
189/338 [===============>..............] - ETA: 4s - loss: 0.6818 - accuracy: 0.6088
191/338 [===============>..............] - ETA: 4s - loss: 0.6819 - accuracy: 0.6086
193/338 [================>.............] - ETA: 4s - loss: 0.6818 - accuracy: 0.6095
195/338 [================>.............] - ETA: 4s - loss: 0.6815 - accuracy: 0.6109
197/338 [================>.............] - ETA: 4s - loss: 0.6813 - accuracy: 0.6120
199/338 [================>.............] - ETA: 4s - loss: 0.6812 - accuracy: 0.6123
201/338 [================>.............] - ETA: 4s - loss: 0.6813 - accuracy: 0.6116
203/338 [=================>............] - ETA: 4s - loss: 0.6816 - accuracy: 0.6105
205/338 [=================>............] - ETA: 4s - loss: 0.6815 - accuracy: 0.6108
207/338 [=================>............] - ETA: 4s - loss: 0.6815 - accuracy: 0.6108
209/338 [=================>............] - ETA: 4s - loss: 0.6816 - accuracy: 0.6102
211/338 [=================>............] - ETA: 4s - loss: 0.6814 - accuracy: 0.6108
213/338 [=================>............] - ETA: 4s - loss: 0.6815 - accuracy: 0.6109
215/338 [==================>...........] - ETA: 4s - loss: 0.6815 - accuracy: 0.6109
217/338 [==================>...........] - ETA: 4s - loss: 0.6817 - accuracy: 0.6097
219/338 [==================>...........] - ETA: 3s - loss: 0.6816 - accuracy: 0.6104
221/338 [==================>...........] - ETA: 3s - loss: 0.6814 - accuracy: 0.6111
223/338 [==================>...........] - ETA: 3s - loss: 0.6814 - accuracy: 0.6110
225/338 [==================>...........] - ETA: 3s - loss: 0.6816 - accuracy: 0.6101
227/338 [===================>..........] - ETA: 3s - loss: 0.6813 - accuracy: 0.6111
229/338 [===================>..........] - ETA: 3s - loss: 0.6813 - accuracy: 0.6109
231/338 [===================>..........] - ETA: 3s - loss: 0.6816 - accuracy: 0.6097
233/338 [===================>..........] - ETA: 3s - loss: 0.6814 - accuracy: 0.6105
235/338 [===================>..........] - ETA: 3s - loss: 0.6814 - accuracy: 0.6104
237/338 [====================>.........] - ETA: 3s - loss: 0.6811 - accuracy: 0.6118
239/338 [====================>.........] - ETA: 3s - loss: 0.6810 - accuracy: 0.6121
241/338 [====================>.........] - ETA: 3s - loss: 0.6811 - accuracy: 0.6116
243/338 [====================>.........] - ETA: 3s - loss: 0.6811 - accuracy: 0.6118
245/338 [====================>.........] - ETA: 3s - loss: 0.6810 - accuracy: 0.6122
247/338 [====================>.........] - ETA: 3s - loss: 0.6809 - accuracy: 0.6125
249/338 [=====================>........] - ETA: 2s - loss: 0.6805 - accuracy: 0.6140
251/338 [=====================>........] - ETA: 2s - loss: 0.6805 - accuracy: 0.6142
253/338 [=====================>........] - ETA: 2s - loss: 0.6805 - accuracy: 0.6145
255/338 [=====================>........] - ETA: 2s - loss: 0.6804 - accuracy: 0.6145
257/338 [=====================>........] - ETA: 2s - loss: 0.6804 - accuracy: 0.6141
259/338 [=====================>........] - ETA: 2s - loss: 0.6804 - accuracy: 0.6140
261/338 [======================>.......] - ETA: 2s - loss: 0.6804 - accuracy: 0.6139
263/338 [======================>.......] - ETA: 2s - loss: 0.6804 - accuracy: 0.6139
265/338 [======================>.......] - ETA: 2s - loss: 0.6804 - accuracy: 0.6136
267/338 [======================>.......] - ETA: 2s - loss: 0.6805 - accuracy: 0.6129
269/338 [======================>.......] - ETA: 2s - loss: 0.6806 - accuracy: 0.6128
271/338 [=======================>......] - ETA: 2s - loss: 0.6806 - accuracy: 0.6121
273/338 [=======================>......] - ETA: 2s - loss: 0.6806 - accuracy: 0.6124
275/338 [=======================>......] - ETA: 2s - loss: 0.6805 - accuracy: 0.6127
277/338 [=======================>......] - ETA: 2s - loss: 0.6805 - accuracy: 0.6128
279/338 [=======================>......] - ETA: 1s - loss: 0.6804 - accuracy: 0.6131
281/338 [=======================>......] - ETA: 1s - loss: 0.6803 - accuracy: 0.6137
283/338 [========================>.....] - ETA: 1s - loss: 0.6803 - accuracy: 0.6132
285/338 [========================>.....] - ETA: 1s - loss: 0.6803 - accuracy: 0.6133
287/338 [========================>.....] - ETA: 1s - loss: 0.6802 - accuracy: 0.6136
289/338 [========================>.....] - ETA: 1s - loss: 0.6803 - accuracy: 0.6132
291/338 [========================>.....] - ETA: 1s - loss: 0.6803 - accuracy: 0.6124
293/338 [=========================>....] - ETA: 1s - loss: 0.6803 - accuracy: 0.6124
295/338 [=========================>....] - ETA: 1s - loss: 0.6803 - accuracy: 0.6123
297/338 [=========================>....] - ETA: 1s - loss: 0.6802 - accuracy: 0.6129
299/338 [=========================>....] - ETA: 1s - loss: 0.6801 - accuracy: 0.6133
301/338 [=========================>....] - ETA: 1s - loss: 0.6801 - accuracy: 0.6131
303/338 [=========================>....] - ETA: 1s - loss: 0.6801 - accuracy: 0.6133
305/338 [==========================>...] - ETA: 1s - loss: 0.6799 - accuracy: 0.6139
307/338 [==========================>...] - ETA: 1s - loss: 0.6798 - accuracy: 0.6142
309/338 [==========================>...] - ETA: 0s - loss: 0.6799 - accuracy: 0.6142
311/338 [==========================>...] - ETA: 0s - loss: 0.6798 - accuracy: 0.6145
313/338 [==========================>...] - ETA: 0s - loss: 0.6797 - accuracy: 0.6146
315/338 [==========================>...] - ETA: 0s - loss: 0.6796 - accuracy: 0.6153
317/338 [===========================>..] - ETA: 0s - loss: 0.6796 - accuracy: 0.6150
319/338 [===========================>..] - ETA: 0s - loss: 0.6797 - accuracy: 0.6142
321/338 [===========================>..] - ETA: 0s - loss: 0.6798 - accuracy: 0.6138
323/338 [===========================>..] - ETA: 0s - loss: 0.6798 - accuracy: 0.6138
325/338 [===========================>..] - ETA: 0s - loss: 0.6796 - accuracy: 0.6144
327/338 [============================>.] - ETA: 0s - loss: 0.6797 - accuracy: 0.6142
329/338 [============================>.] - ETA: 0s - loss: 0.6796 - accuracy: 0.6143
331/338 [============================>.] - ETA: 0s - loss: 0.6796 - accuracy: 0.6142
333/338 [============================>.] - ETA: 0s - loss: 0.6796 - accuracy: 0.6146
335/338 [============================>.] - ETA: 0s - loss: 0.6796 - accuracy: 0.6145
337/338 [============================>.] - ETA: 0s - loss: 0.6795 - accuracy: 0.6148
338/338 [==============================] - 12s 35ms/step - loss: 0.6795 - accuracy: 0.6148 - val_loss: 0.6732 - val_accuracy: 0.6276
Epoch 4/5
1/338 [..............................] - ETA: 11s - loss: 0.6431 - accuracy: 0.7500
3/338 [..............................] - ETA: 11s - loss: 0.6609 - accuracy: 0.7188
5/338 [..............................] - ETA: 10s - loss: 0.6727 - accuracy: 0.6687
7/338 [..............................] - ETA: 10s - loss: 0.6684 - accuracy: 0.6741
9/338 [..............................] - ETA: 11s - loss: 0.6728 - accuracy: 0.6528
11/338 [..............................] - ETA: 10s - loss: 0.6762 - accuracy: 0.6392
13/338 [>.............................] - ETA: 10s - loss: 0.6764 - accuracy: 0.6298
15/338 [>.............................] - ETA: 10s - loss: 0.6775 - accuracy: 0.6271
17/338 [>.............................] - ETA: 10s - loss: 0.6741 - accuracy: 0.6397
19/338 [>.............................] - ETA: 10s - loss: 0.6717 - accuracy: 0.6497
21/338 [>.............................] - ETA: 10s - loss: 0.6725 - accuracy: 0.6443
23/338 [=>............................] - ETA: 10s - loss: 0.6690 - accuracy: 0.6562
25/338 [=>............................] - ETA: 10s - loss: 0.6698 - accuracy: 0.6513
27/338 [=>............................] - ETA: 10s - loss: 0.6696 - accuracy: 0.6493
29/338 [=>............................] - ETA: 10s - loss: 0.6705 - accuracy: 0.6433
31/338 [=>............................] - ETA: 10s - loss: 0.6711 - accuracy: 0.6401
33/338 [=>............................] - ETA: 10s - loss: 0.6706 - accuracy: 0.6430
35/338 [==>...........................] - ETA: 10s - loss: 0.6713 - accuracy: 0.6411
37/338 [==>...........................] - ETA: 9s - loss: 0.6708 - accuracy: 0.6419
39/338 [==>...........................] - ETA: 9s - loss: 0.6725 - accuracy: 0.6346
41/338 [==>...........................] - ETA: 9s - loss: 0.6725 - accuracy: 0.6326
43/338 [==>...........................] - ETA: 9s - loss: 0.6728 - accuracy: 0.6315
45/338 [==>...........................] - ETA: 9s - loss: 0.6731 - accuracy: 0.6299
47/338 [===>..........................] - ETA: 9s - loss: 0.6722 - accuracy: 0.6336
49/338 [===>..........................] - ETA: 9s - loss: 0.6721 - accuracy: 0.6346
51/338 [===>..........................] - ETA: 9s - loss: 0.6713 - accuracy: 0.6366
53/338 [===>..........................] - ETA: 9s - loss: 0.6723 - accuracy: 0.6327
55/338 [===>..........................] - ETA: 9s - loss: 0.6724 - accuracy: 0.6324
57/338 [====>.........................] - ETA: 9s - loss: 0.6726 - accuracy: 0.6321
59/338 [====>.........................] - ETA: 9s - loss: 0.6730 - accuracy: 0.6314
61/338 [====>.........................] - ETA: 9s - loss: 0.6722 - accuracy: 0.6337
63/338 [====>.........................] - ETA: 9s - loss: 0.6718 - accuracy: 0.6349
65/338 [====>.........................] - ETA: 9s - loss: 0.6714 - accuracy: 0.6361
67/338 [====>.........................] - ETA: 8s - loss: 0.6719 - accuracy: 0.6343
69/338 [=====>........................] - ETA: 8s - loss: 0.6716 - accuracy: 0.6359
71/338 [=====>........................] - ETA: 8s - loss: 0.6718 - accuracy: 0.6351
73/338 [=====>........................] - ETA: 8s - loss: 0.6721 - accuracy: 0.6340
75/338 [=====>........................] - ETA: 8s - loss: 0.6717 - accuracy: 0.6354
77/338 [=====>........................] - ETA: 8s - loss: 0.6708 - accuracy: 0.6384
79/338 [======>.......................] - ETA: 8s - loss: 0.6704 - accuracy: 0.6396
81/338 [======>.......................] - ETA: 8s - loss: 0.6709 - accuracy: 0.6385
83/338 [======>.......................] - ETA: 8s - loss: 0.6705 - accuracy: 0.6397
85/338 [======>.......................] - ETA: 8s - loss: 0.6699 - accuracy: 0.6404
87/338 [======>.......................] - ETA: 8s - loss: 0.6695 - accuracy: 0.6419
89/338 [======>.......................] - ETA: 8s - loss: 0.6694 - accuracy: 0.6419
91/338 [=======>......................] - ETA: 8s - loss: 0.6697 - accuracy: 0.6408
93/338 [=======>......................] - ETA: 8s - loss: 0.6701 - accuracy: 0.6394
95/338 [=======>......................] - ETA: 8s - loss: 0.6701 - accuracy: 0.6391
97/338 [=======>......................] - ETA: 7s - loss: 0.6701 - accuracy: 0.6385
99/338 [=======>......................] - ETA: 7s - loss: 0.6696 - accuracy: 0.6398
101/338 [=======>......................] - ETA: 7s - loss: 0.6698 - accuracy: 0.6392
103/338 [========>.....................] - ETA: 7s - loss: 0.6697 - accuracy: 0.6393
105/338 [========>.....................] - ETA: 7s - loss: 0.6694 - accuracy: 0.6402
107/338 [========>.....................] - ETA: 7s - loss: 0.6693 - accuracy: 0.6405
109/338 [========>.....................] - ETA: 7s - loss: 0.6690 - accuracy: 0.6413
111/338 [========>.....................] - ETA: 7s - loss: 0.6688 - accuracy: 0.6413
113/338 [=========>....................] - ETA: 7s - loss: 0.6684 - accuracy: 0.6421
115/338 [=========>....................] - ETA: 7s - loss: 0.6682 - accuracy: 0.6427
117/338 [=========>....................] - ETA: 7s - loss: 0.6678 - accuracy: 0.6440
119/338 [=========>....................] - ETA: 7s - loss: 0.6678 - accuracy: 0.6436
121/338 [=========>....................] - ETA: 7s - loss: 0.6677 - accuracy: 0.6436
123/338 [=========>....................] - ETA: 7s - loss: 0.6678 - accuracy: 0.6428
125/338 [==========>...................] - ETA: 7s - loss: 0.6678 - accuracy: 0.6428
127/338 [==========>...................] - ETA: 6s - loss: 0.6677 - accuracy: 0.6430
129/338 [==========>...................] - ETA: 6s - loss: 0.6676 - accuracy: 0.6432
131/338 [==========>...................] - ETA: 6s - loss: 0.6677 - accuracy: 0.6424
133/338 [==========>...................] - ETA: 6s - loss: 0.6676 - accuracy: 0.6424
135/338 [==========>...................] - ETA: 6s - loss: 0.6677 - accuracy: 0.6426
137/338 [===========>..................] - ETA: 6s - loss: 0.6675 - accuracy: 0.6435
139/338 [===========>..................] - ETA: 6s - loss: 0.6676 - accuracy: 0.6428
141/338 [===========>..................] - ETA: 6s - loss: 0.6678 - accuracy: 0.6416
143/338 [===========>..................] - ETA: 6s - loss: 0.6675 - accuracy: 0.6425
145/338 [===========>..................] - ETA: 6s - loss: 0.6674 - accuracy: 0.6427
147/338 [============>.................] - ETA: 6s - loss: 0.6672 - accuracy: 0.6426
149/338 [============>.................] - ETA: 6s - loss: 0.6672 - accuracy: 0.6426
151/338 [============>.................] - ETA: 6s - loss: 0.6675 - accuracy: 0.6416
153/338 [============>.................] - ETA: 6s - loss: 0.6677 - accuracy: 0.6403
155/338 [============>.................] - ETA: 6s - loss: 0.6678 - accuracy: 0.6397
157/338 [============>.................] - ETA: 5s - loss: 0.6680 - accuracy: 0.6395
159/338 [=============>................] - ETA: 5s - loss: 0.6678 - accuracy: 0.6403
161/338 [=============>................] - ETA: 5s - loss: 0.6677 - accuracy: 0.6405
163/338 [=============>................] - ETA: 5s - loss: 0.6675 - accuracy: 0.6411
165/338 [=============>................] - ETA: 5s - loss: 0.6674 - accuracy: 0.6413
167/338 [=============>................] - ETA: 5s - loss: 0.6672 - accuracy: 0.6418
169/338 [==============>...............] - ETA: 5s - loss: 0.6674 - accuracy: 0.6409
171/338 [==============>...............] - ETA: 5s - loss: 0.6672 - accuracy: 0.6413
173/338 [==============>...............] - ETA: 5s - loss: 0.6673 - accuracy: 0.6407
175/338 [==============>...............] - ETA: 5s - loss: 0.6671 - accuracy: 0.6411
177/338 [==============>...............] - ETA: 5s - loss: 0.6673 - accuracy: 0.6404
179/338 [==============>...............] - ETA: 5s - loss: 0.6674 - accuracy: 0.6400
181/338 [===============>..............] - ETA: 5s - loss: 0.6675 - accuracy: 0.6395
183/338 [===============>..............] - ETA: 5s - loss: 0.6675 - accuracy: 0.6395
185/338 [===============>..............] - ETA: 5s - loss: 0.6675 - accuracy: 0.6392
187/338 [===============>..............] - ETA: 4s - loss: 0.6674 - accuracy: 0.6395
189/338 [===============>..............] - ETA: 4s - loss: 0.6673 - accuracy: 0.6396
191/338 [===============>..............] - ETA: 4s - loss: 0.6674 - accuracy: 0.6394
193/338 [================>.............] - ETA: 4s - loss: 0.6674 - accuracy: 0.6396
195/338 [================>.............] - ETA: 4s - loss: 0.6677 - accuracy: 0.6386
197/338 [================>.............] - ETA: 4s - loss: 0.6677 - accuracy: 0.6386
199/338 [================>.............] - ETA: 4s - loss: 0.6677 - accuracy: 0.6385
201/338 [================>.............] - ETA: 4s - loss: 0.6677 - accuracy: 0.6387
203/338 [=================>............] - ETA: 4s - loss: 0.6676 - accuracy: 0.6390
205/338 [=================>............] - ETA: 4s - loss: 0.6673 - accuracy: 0.6396
207/338 [=================>............] - ETA: 4s - loss: 0.6672 - accuracy: 0.6402
209/338 [=================>............] - ETA: 4s - loss: 0.6671 - accuracy: 0.6403
211/338 [=================>............] - ETA: 4s - loss: 0.6671 - accuracy: 0.6403
213/338 [=================>............] - ETA: 4s - loss: 0.6672 - accuracy: 0.6401
215/338 [==================>...........] - ETA: 4s - loss: 0.6670 - accuracy: 0.6407
217/338 [==================>...........] - ETA: 3s - loss: 0.6671 - accuracy: 0.6401
219/338 [==================>...........] - ETA: 3s - loss: 0.6673 - accuracy: 0.6397
221/338 [==================>...........] - ETA: 3s - loss: 0.6671 - accuracy: 0.6401
223/338 [==================>...........] - ETA: 3s - loss: 0.6673 - accuracy: 0.6396
225/338 [==================>...........] - ETA: 3s - loss: 0.6670 - accuracy: 0.6403
227/338 [===================>..........] - ETA: 3s - loss: 0.6668 - accuracy: 0.6408
229/338 [===================>..........] - ETA: 3s - loss: 0.6667 - accuracy: 0.6410
231/338 [===================>..........] - ETA: 3s - loss: 0.6666 - accuracy: 0.6412
233/338 [===================>..........] - ETA: 3s - loss: 0.6665 - accuracy: 0.6414
235/338 [===================>..........] - ETA: 3s - loss: 0.6665 - accuracy: 0.6412
237/338 [====================>.........] - ETA: 3s - loss: 0.6665 - accuracy: 0.6414
239/338 [====================>.........] - ETA: 3s - loss: 0.6664 - accuracy: 0.6415
241/338 [====================>.........] - ETA: 3s - loss: 0.6663 - accuracy: 0.6419
243/338 [====================>.........] - ETA: 3s - loss: 0.6665 - accuracy: 0.6412
245/338 [====================>.........] - ETA: 3s - loss: 0.6663 - accuracy: 0.6415
247/338 [====================>.........] - ETA: 3s - loss: 0.6662 - accuracy: 0.6416
249/338 [=====================>........] - ETA: 2s - loss: 0.6662 - accuracy: 0.6416
251/338 [=====================>........] - ETA: 2s - loss: 0.6664 - accuracy: 0.6409
253/338 [=====================>........] - ETA: 2s - loss: 0.6664 - accuracy: 0.6409
255/338 [=====================>........] - ETA: 2s - loss: 0.6666 - accuracy: 0.6402
257/338 [=====================>........] - ETA: 2s - loss: 0.6668 - accuracy: 0.6396
259/338 [=====================>........] - ETA: 2s - loss: 0.6667 - accuracy: 0.6401
261/338 [======================>.......] - ETA: 2s - loss: 0.6669 - accuracy: 0.6392
263/338 [======================>.......] - ETA: 2s - loss: 0.6668 - accuracy: 0.6393
265/338 [======================>.......] - ETA: 2s - loss: 0.6672 - accuracy: 0.6380
267/338 [======================>.......] - ETA: 2s - loss: 0.6671 - accuracy: 0.6382
269/338 [======================>.......] - ETA: 2s - loss: 0.6672 - accuracy: 0.6382
271/338 [=======================>......] - ETA: 2s - loss: 0.6669 - accuracy: 0.6388
273/338 [=======================>......] - ETA: 2s - loss: 0.6670 - accuracy: 0.6387
275/338 [=======================>......] - ETA: 2s - loss: 0.6669 - accuracy: 0.6392
277/338 [=======================>......] - ETA: 2s - loss: 0.6672 - accuracy: 0.6382
279/338 [=======================>......] - ETA: 1s - loss: 0.6673 - accuracy: 0.6377
281/338 [=======================>......] - ETA: 1s - loss: 0.6672 - accuracy: 0.6377
283/338 [========================>.....] - ETA: 1s - loss: 0.6672 - accuracy: 0.6375
285/338 [========================>.....] - ETA: 1s - loss: 0.6673 - accuracy: 0.6372
287/338 [========================>.....] - ETA: 1s - loss: 0.6671 - accuracy: 0.6378
289/338 [========================>.....] - ETA: 1s - loss: 0.6671 - accuracy: 0.6382
291/338 [========================>.....] - ETA: 1s - loss: 0.6670 - accuracy: 0.6384
293/338 [=========================>....] - ETA: 1s - loss: 0.6669 - accuracy: 0.6388
295/338 [=========================>....] - ETA: 1s - loss: 0.6669 - accuracy: 0.6386
297/338 [=========================>....] - ETA: 1s - loss: 0.6667 - accuracy: 0.6391
299/338 [=========================>....] - ETA: 1s - loss: 0.6666 - accuracy: 0.6392
301/338 [=========================>....] - ETA: 1s - loss: 0.6664 - accuracy: 0.6398
303/338 [=========================>....] - ETA: 1s - loss: 0.6665 - accuracy: 0.6393
305/338 [==========================>...] - ETA: 1s - loss: 0.6666 - accuracy: 0.6389
307/338 [==========================>...] - ETA: 1s - loss: 0.6664 - accuracy: 0.6393
309/338 [==========================>...] - ETA: 0s - loss: 0.6664 - accuracy: 0.6392
311/338 [==========================>...] - ETA: 0s - loss: 0.6663 - accuracy: 0.6393
313/338 [==========================>...] - ETA: 0s - loss: 0.6663 - accuracy: 0.6395
315/338 [==========================>...] - ETA: 0s - loss: 0.6661 - accuracy: 0.6399
317/338 [===========================>..] - ETA: 0s - loss: 0.6661 - accuracy: 0.6396
319/338 [===========================>..] - ETA: 0s - loss: 0.6661 - accuracy: 0.6396
321/338 [===========================>..] - ETA: 0s - loss: 0.6661 - accuracy: 0.6394
323/338 [===========================>..] - ETA: 0s - loss: 0.6660 - accuracy: 0.6394
325/338 [===========================>..] - ETA: 0s - loss: 0.6661 - accuracy: 0.6391
327/338 [============================>.] - ETA: 0s - loss: 0.6662 - accuracy: 0.6390
329/338 [============================>.] - ETA: 0s - loss: 0.6662 - accuracy: 0.6390
331/338 [============================>.] - ETA: 0s - loss: 0.6660 - accuracy: 0.6394
333/338 [============================>.] - ETA: 0s - loss: 0.6662 - accuracy: 0.6389
335/338 [============================>.] - ETA: 0s - loss: 0.6664 - accuracy: 0.6384
337/338 [============================>.] - ETA: 0s - loss: 0.6666 - accuracy: 0.6376
338/338 [==============================] - 12s 35ms/step - loss: 0.6666 - accuracy: 0.6374 - val_loss: 0.6612 - val_accuracy: 0.6341
Epoch 5/5
1/338 [..............................] - ETA: 11s - loss: 0.6941 - accuracy: 0.5312
3/338 [..............................] - ETA: 10s - loss: 0.6688 - accuracy: 0.6146
5/338 [..............................] - ETA: 10s - loss: 0.6594 - accuracy: 0.6438
7/338 [..............................] - ETA: 10s - loss: 0.6623 - accuracy: 0.6339
9/338 [..............................] - ETA: 10s - loss: 0.6657 - accuracy: 0.6181
11/338 [..............................] - ETA: 10s - loss: 0.6679 - accuracy: 0.6193
13/338 [>.............................] - ETA: 10s - loss: 0.6728 - accuracy: 0.6010
15/338 [>.............................] - ETA: 10s - loss: 0.6701 - accuracy: 0.6083
17/338 [>.............................] - ETA: 10s - loss: 0.6706 - accuracy: 0.6103
19/338 [>.............................] - ETA: 10s - loss: 0.6687 - accuracy: 0.6151
21/338 [>.............................] - ETA: 10s - loss: 0.6665 - accuracy: 0.6205
23/338 [=>............................] - ETA: 10s - loss: 0.6627 - accuracy: 0.6304
25/338 [=>............................] - ETA: 10s - loss: 0.6672 - accuracy: 0.6200
27/338 [=>............................] - ETA: 10s - loss: 0.6693 - accuracy: 0.6157
29/338 [=>............................] - ETA: 10s - loss: 0.6660 - accuracy: 0.6239
31/338 [=>............................] - ETA: 10s - loss: 0.6664 - accuracy: 0.6230
33/338 [=>............................] - ETA: 10s - loss: 0.6658 - accuracy: 0.6241
35/338 [==>...........................] - ETA: 9s - loss: 0.6656 - accuracy: 0.6268
37/338 [==>...........................] - ETA: 9s - loss: 0.6666 - accuracy: 0.6242
39/338 [==>...........................] - ETA: 9s - loss: 0.6653 - accuracy: 0.6274
41/338 [==>...........................] - ETA: 9s - loss: 0.6657 - accuracy: 0.6258
43/338 [==>...........................] - ETA: 9s - loss: 0.6635 - accuracy: 0.6315
45/338 [==>...........................] - ETA: 9s - loss: 0.6634 - accuracy: 0.6319
47/338 [===>..........................] - ETA: 9s - loss: 0.6637 - accuracy: 0.6316
49/338 [===>..........................] - ETA: 9s - loss: 0.6636 - accuracy: 0.6320
51/338 [===>..........................] - ETA: 9s - loss: 0.6662 - accuracy: 0.6275
53/338 [===>..........................] - ETA: 9s - loss: 0.6669 - accuracy: 0.6262
55/338 [===>..........................] - ETA: 9s - loss: 0.6676 - accuracy: 0.6256
57/338 [====>.........................] - ETA: 9s - loss: 0.6673 - accuracy: 0.6255
59/338 [====>.........................] - ETA: 9s - loss: 0.6664 - accuracy: 0.6276
61/338 [====>.........................] - ETA: 9s - loss: 0.6654 - accuracy: 0.6296
63/338 [====>.........................] - ETA: 9s - loss: 0.6631 - accuracy: 0.6354
65/338 [====>.........................] - ETA: 9s - loss: 0.6630 - accuracy: 0.6351
67/338 [====>.........................] - ETA: 8s - loss: 0.6623 - accuracy: 0.6367
69/338 [=====>........................] - ETA: 8s - loss: 0.6632 - accuracy: 0.6341
71/338 [=====>........................] - ETA: 8s - loss: 0.6632 - accuracy: 0.6338
73/338 [=====>........................] - ETA: 8s - loss: 0.6633 - accuracy: 0.6340
75/338 [=====>........................] - ETA: 8s - loss: 0.6645 - accuracy: 0.6317
77/338 [=====>........................] - ETA: 8s - loss: 0.6638 - accuracy: 0.6335
79/338 [======>.......................] - ETA: 8s - loss: 0.6636 - accuracy: 0.6337
81/338 [======>.......................] - ETA: 8s - loss: 0.6626 - accuracy: 0.6358
83/338 [======>.......................] - ETA: 8s - loss: 0.6629 - accuracy: 0.6352
85/338 [======>.......................] - ETA: 8s - loss: 0.6620 - accuracy: 0.6371
87/338 [======>.......................] - ETA: 8s - loss: 0.6619 - accuracy: 0.6376
89/338 [======>.......................] - ETA: 8s - loss: 0.6625 - accuracy: 0.6366
91/338 [=======>......................] - ETA: 8s - loss: 0.6627 - accuracy: 0.6360
93/338 [=======>......................] - ETA: 8s - loss: 0.6621 - accuracy: 0.6371
95/338 [=======>......................] - ETA: 8s - loss: 0.6620 - accuracy: 0.6368
97/338 [=======>......................] - ETA: 7s - loss: 0.6619 - accuracy: 0.6366
99/338 [=======>......................] - ETA: 7s - loss: 0.6615 - accuracy: 0.6376
101/338 [=======>......................] - ETA: 7s - loss: 0.6612 - accuracy: 0.6380
103/338 [========>.....................] - ETA: 7s - loss: 0.6614 - accuracy: 0.6374
105/338 [========>.....................] - ETA: 7s - loss: 0.6611 - accuracy: 0.6378
107/338 [========>.....................] - ETA: 7s - loss: 0.6611 - accuracy: 0.6384
109/338 [========>.....................] - ETA: 7s - loss: 0.6610 - accuracy: 0.6388
111/338 [========>.....................] - ETA: 7s - loss: 0.6600 - accuracy: 0.6416
113/338 [=========>....................] - ETA: 7s - loss: 0.6602 - accuracy: 0.6413
115/338 [=========>....................] - ETA: 7s - loss: 0.6601 - accuracy: 0.6416
117/338 [=========>....................] - ETA: 7s - loss: 0.6603 - accuracy: 0.6416
119/338 [=========>....................] - ETA: 7s - loss: 0.6595 - accuracy: 0.6434
121/338 [=========>....................] - ETA: 7s - loss: 0.6600 - accuracy: 0.6426
123/338 [=========>....................] - ETA: 7s - loss: 0.6610 - accuracy: 0.6402
125/338 [==========>...................] - ETA: 7s - loss: 0.6611 - accuracy: 0.6400
127/338 [==========>...................] - ETA: 6s - loss: 0.6612 - accuracy: 0.6398
129/338 [==========>...................] - ETA: 6s - loss: 0.6615 - accuracy: 0.6388
131/338 [==========>...................] - ETA: 6s - loss: 0.6613 - accuracy: 0.6393
133/338 [==========>...................] - ETA: 6s - loss: 0.6607 - accuracy: 0.6407
135/338 [==========>...................] - ETA: 6s - loss: 0.6610 - accuracy: 0.6400
137/338 [===========>..................] - ETA: 6s - loss: 0.6610 - accuracy: 0.6398
139/338 [===========>..................] - ETA: 6s - loss: 0.6613 - accuracy: 0.6392
141/338 [===========>..................] - ETA: 6s - loss: 0.6609 - accuracy: 0.6398
143/338 [===========>..................] - ETA: 6s - loss: 0.6610 - accuracy: 0.6394
145/338 [===========>..................] - ETA: 6s - loss: 0.6611 - accuracy: 0.6397
147/338 [============>.................] - ETA: 6s - loss: 0.6614 - accuracy: 0.6388
149/338 [============>.................] - ETA: 6s - loss: 0.6610 - accuracy: 0.6399
151/338 [============>.................] - ETA: 6s - loss: 0.6611 - accuracy: 0.6395
153/338 [============>.................] - ETA: 6s - loss: 0.6614 - accuracy: 0.6391
155/338 [============>.................] - ETA: 6s - loss: 0.6617 - accuracy: 0.6381
157/338 [============>.................] - ETA: 5s - loss: 0.6617 - accuracy: 0.6377
159/338 [=============>................] - ETA: 5s - loss: 0.6618 - accuracy: 0.6378
161/338 [=============>................] - ETA: 5s - loss: 0.6618 - accuracy: 0.6378
163/338 [=============>................] - ETA: 5s - loss: 0.6619 - accuracy: 0.6378
165/338 [=============>................] - ETA: 5s - loss: 0.6617 - accuracy: 0.6383
167/338 [=============>................] - ETA: 5s - loss: 0.6620 - accuracy: 0.6375
169/338 [==============>...............] - ETA: 5s - loss: 0.6627 - accuracy: 0.6361
171/338 [==============>...............] - ETA: 5s - loss: 0.6625 - accuracy: 0.6363
173/338 [==============>...............] - ETA: 5s - loss: 0.6625 - accuracy: 0.6364
175/338 [==============>...............] - ETA: 5s - loss: 0.6622 - accuracy: 0.6370
177/338 [==============>...............] - ETA: 5s - loss: 0.6623 - accuracy: 0.6368
179/338 [==============>...............] - ETA: 5s - loss: 0.6623 - accuracy: 0.6369
181/338 [===============>..............] - ETA: 5s - loss: 0.6621 - accuracy: 0.6376
183/338 [===============>..............] - ETA: 5s - loss: 0.6618 - accuracy: 0.6381
185/338 [===============>..............] - ETA: 5s - loss: 0.6620 - accuracy: 0.6378
187/338 [===============>..............] - ETA: 4s - loss: 0.6615 - accuracy: 0.6385
189/338 [===============>..............] - ETA: 4s - loss: 0.6616 - accuracy: 0.6387
191/338 [===============>..............] - ETA: 4s - loss: 0.6616 - accuracy: 0.6383
193/338 [================>.............] - ETA: 4s - loss: 0.6616 - accuracy: 0.6381
195/338 [================>.............] - ETA: 4s - loss: 0.6608 - accuracy: 0.6399
197/338 [================>.............] - ETA: 4s - loss: 0.6610 - accuracy: 0.6398
199/338 [================>.............] - ETA: 4s - loss: 0.6606 - accuracy: 0.6405
201/338 [================>.............] - ETA: 4s - loss: 0.6609 - accuracy: 0.6398
203/338 [=================>............] - ETA: 4s - loss: 0.6610 - accuracy: 0.6398
205/338 [=================>............] - ETA: 4s - loss: 0.6608 - accuracy: 0.6401
207/338 [=================>............] - ETA: 4s - loss: 0.6606 - accuracy: 0.6404
209/338 [=================>............] - ETA: 4s - loss: 0.6605 - accuracy: 0.6408
211/338 [=================>............] - ETA: 4s - loss: 0.6606 - accuracy: 0.6408
213/338 [=================>............] - ETA: 4s - loss: 0.6602 - accuracy: 0.6417
215/338 [==================>...........] - ETA: 4s - loss: 0.6608 - accuracy: 0.6403
217/338 [==================>...........] - ETA: 3s - loss: 0.6607 - accuracy: 0.6406
219/338 [==================>...........] - ETA: 3s - loss: 0.6606 - accuracy: 0.6408
221/338 [==================>...........] - ETA: 3s - loss: 0.6604 - accuracy: 0.6413
223/338 [==================>...........] - ETA: 3s - loss: 0.6603 - accuracy: 0.6413
225/338 [==================>...........] - ETA: 3s - loss: 0.6599 - accuracy: 0.6419
227/338 [===================>..........] - ETA: 3s - loss: 0.6602 - accuracy: 0.6414
229/338 [===================>..........] - ETA: 3s - loss: 0.6602 - accuracy: 0.6414
231/338 [===================>..........] - ETA: 3s - loss: 0.6601 - accuracy: 0.6416
233/338 [===================>..........] - ETA: 3s - loss: 0.6599 - accuracy: 0.6423
235/338 [===================>..........] - ETA: 3s - loss: 0.6600 - accuracy: 0.6419
237/338 [====================>.........] - ETA: 3s - loss: 0.6598 - accuracy: 0.6421
239/338 [====================>.........] - ETA: 3s - loss: 0.6598 - accuracy: 0.6421
241/338 [====================>.........] - ETA: 3s - loss: 0.6595 - accuracy: 0.6428
243/338 [====================>.........] - ETA: 3s - loss: 0.6592 - accuracy: 0.6434
245/338 [====================>.........] - ETA: 3s - loss: 0.6591 - accuracy: 0.6435
247/338 [====================>.........] - ETA: 3s - loss: 0.6592 - accuracy: 0.6433
249/338 [=====================>........] - ETA: 2s - loss: 0.6590 - accuracy: 0.6436
251/338 [=====================>........] - ETA: 2s - loss: 0.6593 - accuracy: 0.6428
253/338 [=====================>........] - ETA: 2s - loss: 0.6592 - accuracy: 0.6430
255/338 [=====================>........] - ETA: 2s - loss: 0.6593 - accuracy: 0.6426
257/338 [=====================>........] - ETA: 2s - loss: 0.6592 - accuracy: 0.6428
259/338 [=====================>........] - ETA: 2s - loss: 0.6596 - accuracy: 0.6420
261/338 [======================>.......] - ETA: 2s - loss: 0.6593 - accuracy: 0.6426
263/338 [======================>.......] - ETA: 2s - loss: 0.6594 - accuracy: 0.6423
265/338 [======================>.......] - ETA: 2s - loss: 0.6591 - accuracy: 0.6432
267/338 [======================>.......] - ETA: 2s - loss: 0.6593 - accuracy: 0.6427
269/338 [======================>.......] - ETA: 2s - loss: 0.6591 - accuracy: 0.6430
271/338 [=======================>......] - ETA: 2s - loss: 0.6593 - accuracy: 0.6426
273/338 [=======================>......] - ETA: 2s - loss: 0.6594 - accuracy: 0.6424
275/338 [=======================>......] - ETA: 2s - loss: 0.6596 - accuracy: 0.6418
277/338 [=======================>......] - ETA: 2s - loss: 0.6595 - accuracy: 0.6421
279/338 [=======================>......] - ETA: 1s - loss: 0.6592 - accuracy: 0.6426
281/338 [=======================>......] - ETA: 1s - loss: 0.6591 - accuracy: 0.6428
283/338 [========================>.....] - ETA: 1s - loss: 0.6589 - accuracy: 0.6432
285/338 [========================>.....] - ETA: 1s - loss: 0.6589 - accuracy: 0.6433
287/338 [========================>.....] - ETA: 1s - loss: 0.6593 - accuracy: 0.6422
289/338 [========================>.....] - ETA: 1s - loss: 0.6593 - accuracy: 0.6422
291/338 [========================>.....] - ETA: 1s - loss: 0.6594 - accuracy: 0.6421
293/338 [=========================>....] - ETA: 1s - loss: 0.6590 - accuracy: 0.6428
295/338 [=========================>....] - ETA: 1s - loss: 0.6591 - accuracy: 0.6426
297/338 [=========================>....] - ETA: 1s - loss: 0.6592 - accuracy: 0.6425
299/338 [=========================>....] - ETA: 1s - loss: 0.6592 - accuracy: 0.6425
301/338 [=========================>....] - ETA: 1s - loss: 0.6594 - accuracy: 0.6420
303/338 [=========================>....] - ETA: 1s - loss: 0.6593 - accuracy: 0.6420
305/338 [==========================>...] - ETA: 1s - loss: 0.6594 - accuracy: 0.6419
307/338 [==========================>...] - ETA: 1s - loss: 0.6596 - accuracy: 0.6415
309/338 [==========================>...] - ETA: 0s - loss: 0.6598 - accuracy: 0.6412
311/338 [==========================>...] - ETA: 0s - loss: 0.6598 - accuracy: 0.6412
313/338 [==========================>...] - ETA: 0s - loss: 0.6597 - accuracy: 0.6414
315/338 [==========================>...] - ETA: 0s - loss: 0.6598 - accuracy: 0.6412
317/338 [===========================>..] - ETA: 0s - loss: 0.6599 - accuracy: 0.6410
319/338 [===========================>..] - ETA: 0s - loss: 0.6601 - accuracy: 0.6405
321/338 [===========================>..] - ETA: 0s - loss: 0.6602 - accuracy: 0.6403
323/338 [===========================>..] - ETA: 0s - loss: 0.6600 - accuracy: 0.6408
325/338 [===========================>..] - ETA: 0s - loss: 0.6599 - accuracy: 0.6410
327/338 [============================>.] - ETA: 0s - loss: 0.6599 - accuracy: 0.6409
329/338 [============================>.] - ETA: 0s - loss: 0.6600 - accuracy: 0.6406
331/338 [============================>.] - ETA: 0s - loss: 0.6600 - accuracy: 0.6406
333/338 [============================>.] - ETA: 0s - loss: 0.6601 - accuracy: 0.6404
335/338 [============================>.] - ETA: 0s - loss: 0.6601 - accuracy: 0.6403
337/338 [============================>.] - ETA: 0s - loss: 0.6603 - accuracy: 0.6397
338/338 [==============================] - 12s 35ms/step - loss: 0.6602 - accuracy: 0.6400 - val_loss: 0.6557 - val_accuracy: 0.6442
1/97 [..............................] - ETA: 1s - loss: 0.6494 - accuracy: 0.6562
9/97 [=>............................] - ETA: 0s - loss: 0.6896 - accuracy: 0.5694
17/97 [====>.........................] - ETA: 0s - loss: 0.6747 - accuracy: 0.6011
25/97 [======>.......................] - ETA: 0s - loss: 0.6658 - accuracy: 0.6187
33/97 [=========>....................] - ETA: 0s - loss: 0.6648 - accuracy: 0.6203
40/97 [===========>..................] - ETA: 0s - loss: 0.6631 - accuracy: 0.6258
48/97 [=============>................] - ETA: 0s - loss: 0.6617 - accuracy: 0.6283
56/97 [================>.............] - ETA: 0s - loss: 0.6604 - accuracy: 0.6339
64/97 [==================>...........] - ETA: 0s - loss: 0.6590 - accuracy: 0.6377
72/97 [=====================>........] - ETA: 0s - loss: 0.6576 - accuracy: 0.6402
80/97 [=======================>......] - ETA: 0s - loss: 0.6568 - accuracy: 0.6414
88/97 [==========================>...] - ETA: 0s - loss: 0.6576 - accuracy: 0.6399
96/97 [============================>.] - ETA: 0s - loss: 0.6555 - accuracy: 0.6445
97/97 [==============================] - 1s 7ms/step - loss: 0.6557 - accuracy: 0.6442
Performance drop after ablating c0: -0.0012974143028259277
Epoch 1/5
2024-06-04 17:42:53.893611: W external/local_tsl/tsl/framework/cpu_allocator_impl.cc:83] Allocation of 728865000 exceeds 10% of free system memory.
1/338 [..............................] - ETA: 6:39 - loss: 0.7129 - accuracy: 0.4062
3/338 [..............................] - ETA: 16s - loss: 0.7118 - accuracy: 0.4167
5/338 [..............................] - ETA: 17s - loss: 0.7117 - accuracy: 0.3875
6/338 [..............................] - ETA: 17s - loss: 0.7105 - accuracy: 0.4271
8/338 [..............................] - ETA: 16s - loss: 0.7098 - accuracy: 0.4297
10/338 [..............................] - ETA: 16s - loss: 0.7087 - accuracy: 0.4625
12/338 [>.............................] - ETA: 16s - loss: 0.7082 - accuracy: 0.4557
13/338 [>.............................] - ETA: 16s - loss: 0.7078 - accuracy: 0.4712
15/338 [>.............................] - ETA: 15s - loss: 0.7070 - accuracy: 0.4854
17/338 [>.............................] - ETA: 15s - loss: 0.7064 - accuracy: 0.4945
19/338 [>.............................] - ETA: 15s - loss: 0.7059 - accuracy: 0.4934
21/338 [>.............................] - ETA: 15s - loss: 0.7056 - accuracy: 0.4821
23/338 [=>............................] - ETA: 15s - loss: 0.7052 - accuracy: 0.4810
25/338 [=>............................] - ETA: 15s - loss: 0.7049 - accuracy: 0.4750
27/338 [=>............................] - ETA: 15s - loss: 0.7046 - accuracy: 0.4676
29/338 [=>............................] - ETA: 15s - loss: 0.7042 - accuracy: 0.4731
31/338 [=>............................] - ETA: 14s - loss: 0.7038 - accuracy: 0.4798
33/338 [=>............................] - ETA: 14s - loss: 0.7035 - accuracy: 0.4830
35/338 [==>...........................] - ETA: 14s - loss: 0.7032 - accuracy: 0.4955
37/338 [==>...........................] - ETA: 14s - loss: 0.7028 - accuracy: 0.5008
39/338 [==>...........................] - ETA: 14s - loss: 0.7026 - accuracy: 0.5016
41/338 [==>...........................] - ETA: 14s - loss: 0.7023 - accuracy: 0.5030
43/338 [==>...........................] - ETA: 14s - loss: 0.7021 - accuracy: 0.5036
45/338 [==>...........................] - ETA: 14s - loss: 0.7019 - accuracy: 0.5014
47/338 [===>..........................] - ETA: 14s - loss: 0.7016 - accuracy: 0.5060
48/338 [===>..........................] - ETA: 14s - loss: 0.7015 - accuracy: 0.5026
50/338 [===>..........................] - ETA: 13s - loss: 0.7013 - accuracy: 0.5056
52/338 [===>..........................] - ETA: 13s - loss: 0.7010 - accuracy: 0.5078
54/338 [===>..........................] - ETA: 13s - loss: 0.7008 - accuracy: 0.5081
56/338 [===>..........................] - ETA: 13s - loss: 0.7005 - accuracy: 0.5134
58/338 [====>.........................] - ETA: 13s - loss: 0.7004 - accuracy: 0.5124
60/338 [====>.........................] - ETA: 13s - loss: 0.7003 - accuracy: 0.5125
62/338 [====>.........................] - ETA: 13s - loss: 0.7000 - accuracy: 0.5166
64/338 [====>.........................] - ETA: 13s - loss: 0.6997 - accuracy: 0.5195
66/338 [====>.........................] - ETA: 13s - loss: 0.6997 - accuracy: 0.5175
68/338 [=====>........................] - ETA: 13s - loss: 0.6996 - accuracy: 0.5156
70/338 [=====>........................] - ETA: 12s - loss: 0.6994 - accuracy: 0.5170
72/338 [=====>........................] - ETA: 12s - loss: 0.6993 - accuracy: 0.5165
74/338 [=====>........................] - ETA: 12s - loss: 0.6993 - accuracy: 0.5144
76/338 [=====>........................] - ETA: 12s - loss: 0.6992 - accuracy: 0.5127
78/338 [=====>........................] - ETA: 12s - loss: 0.6992 - accuracy: 0.5108
80/338 [======>.......................] - ETA: 12s - loss: 0.6993 - accuracy: 0.5066
82/338 [======>.......................] - ETA: 12s - loss: 0.6992 - accuracy: 0.5061
84/338 [======>.......................] - ETA: 12s - loss: 0.6991 - accuracy: 0.5063
85/338 [======>.......................] - ETA: 12s - loss: 0.6990 - accuracy: 0.5066
87/338 [======>.......................] - ETA: 12s - loss: 0.6989 - accuracy: 0.5083
89/338 [======>.......................] - ETA: 12s - loss: 0.6988 - accuracy: 0.5070
91/338 [=======>......................] - ETA: 11s - loss: 0.6988 - accuracy: 0.5058
93/338 [=======>......................] - ETA: 11s - loss: 0.6986 - accuracy: 0.5064
95/338 [=======>......................] - ETA: 11s - loss: 0.6985 - accuracy: 0.5079
97/338 [=======>......................] - ETA: 11s - loss: 0.6984 - accuracy: 0.5103
99/338 [=======>......................] - ETA: 11s - loss: 0.6983 - accuracy: 0.5110
101/338 [=======>......................] - ETA: 11s - loss: 0.6983 - accuracy: 0.5090
103/338 [========>.....................] - ETA: 11s - loss: 0.6983 - accuracy: 0.5058
105/338 [========>.....................] - ETA: 11s - loss: 0.6983 - accuracy: 0.5042
107/338 [========>.....................] - ETA: 11s - loss: 0.6983 - accuracy: 0.5041
109/338 [========>.....................] - ETA: 11s - loss: 0.6982 - accuracy: 0.5032
111/338 [========>.....................] - ETA: 10s - loss: 0.6981 - accuracy: 0.5045
112/338 [========>.....................] - ETA: 10s - loss: 0.6981 - accuracy: 0.5045
114/338 [=========>....................] - ETA: 10s - loss: 0.6979 - accuracy: 0.5071
116/338 [=========>....................] - ETA: 10s - loss: 0.6979 - accuracy: 0.5054
118/338 [=========>....................] - ETA: 10s - loss: 0.6978 - accuracy: 0.5082
120/338 [=========>....................] - ETA: 10s - loss: 0.6977 - accuracy: 0.5076
122/338 [=========>....................] - ETA: 10s - loss: 0.6977 - accuracy: 0.5061
124/338 [==========>...................] - ETA: 10s - loss: 0.6976 - accuracy: 0.5076
126/338 [==========>...................] - ETA: 10s - loss: 0.6975 - accuracy: 0.5097
127/338 [==========>...................] - ETA: 10s - loss: 0.6974 - accuracy: 0.5108
129/338 [==========>...................] - ETA: 10s - loss: 0.6974 - accuracy: 0.5114
131/338 [==========>...................] - ETA: 9s - loss: 0.6973 - accuracy: 0.5122
133/338 [==========>...................] - ETA: 9s - loss: 0.6972 - accuracy: 0.5122
134/338 [==========>...................] - ETA: 9s - loss: 0.6971 - accuracy: 0.5128
136/338 [===========>..................] - ETA: 9s - loss: 0.6970 - accuracy: 0.5145
138/338 [===========>..................] - ETA: 9s - loss: 0.6969 - accuracy: 0.5161
139/338 [===========>..................] - ETA: 9s - loss: 0.6969 - accuracy: 0.5164
141/338 [===========>..................] - ETA: 9s - loss: 0.6968 - accuracy: 0.5173
143/338 [===========>..................] - ETA: 9s - loss: 0.6968 - accuracy: 0.5160
145/338 [===========>..................] - ETA: 9s - loss: 0.6967 - accuracy: 0.5168
147/338 [============>.................] - ETA: 9s - loss: 0.6966 - accuracy: 0.5168
149/338 [============>.................] - ETA: 9s - loss: 0.6966 - accuracy: 0.5161
151/338 [============>.................] - ETA: 9s - loss: 0.6965 - accuracy: 0.5174
153/338 [============>.................] - ETA: 8s - loss: 0.6964 - accuracy: 0.5182
155/338 [============>.................] - ETA: 8s - loss: 0.6963 - accuracy: 0.5190
157/338 [============>.................] - ETA: 8s - loss: 0.6963 - accuracy: 0.5189
159/338 [=============>................] - ETA: 8s - loss: 0.6963 - accuracy: 0.5175
161/338 [=============>................] - ETA: 8s - loss: 0.6962 - accuracy: 0.5192
163/338 [=============>................] - ETA: 8s - loss: 0.6961 - accuracy: 0.5190
165/338 [=============>................] - ETA: 8s - loss: 0.6961 - accuracy: 0.5188
167/338 [=============>................] - ETA: 8s - loss: 0.6961 - accuracy: 0.5187
169/338 [==============>...............] - ETA: 8s - loss: 0.6960 - accuracy: 0.5189
171/338 [==============>...............] - ETA: 8s - loss: 0.6960 - accuracy: 0.5186
173/338 [==============>...............] - ETA: 7s - loss: 0.6959 - accuracy: 0.5190
175/338 [==============>...............] - ETA: 7s - loss: 0.6959 - accuracy: 0.5188
176/338 [==============>...............] - ETA: 7s - loss: 0.6959 - accuracy: 0.5181
178/338 [==============>...............] - ETA: 7s - loss: 0.6959 - accuracy: 0.5181
180/338 [==============>...............] - ETA: 7s - loss: 0.6958 - accuracy: 0.5182
182/338 [===============>..............] - ETA: 7s - loss: 0.6958 - accuracy: 0.5179
184/338 [===============>..............] - ETA: 7s - loss: 0.6958 - accuracy: 0.5185
186/338 [===============>..............] - ETA: 7s - loss: 0.6957 - accuracy: 0.5185
187/338 [===============>..............] - ETA: 7s - loss: 0.6957 - accuracy: 0.5189
188/338 [===============>..............] - ETA: 7s - loss: 0.6957 - accuracy: 0.5185
190/338 [===============>..............] - ETA: 7s - loss: 0.6956 - accuracy: 0.5191
192/338 [================>.............] - ETA: 7s - loss: 0.6956 - accuracy: 0.5197
194/338 [================>.............] - ETA: 6s - loss: 0.6955 - accuracy: 0.5195
196/338 [================>.............] - ETA: 6s - loss: 0.6954 - accuracy: 0.5206
197/338 [================>.............] - ETA: 6s - loss: 0.6954 - accuracy: 0.5208
198/338 [================>.............] - ETA: 6s - loss: 0.6954 - accuracy: 0.5210
200/338 [================>.............] - ETA: 6s - loss: 0.6952 - accuracy: 0.5220
202/338 [================>.............] - ETA: 6s - loss: 0.6952 - accuracy: 0.5221
204/338 [=================>............] - ETA: 6s - loss: 0.6952 - accuracy: 0.5218
206/338 [=================>............] - ETA: 6s - loss: 0.6953 - accuracy: 0.5206
208/338 [=================>............] - ETA: 6s - loss: 0.6952 - accuracy: 0.5209
210/338 [=================>............] - ETA: 6s - loss: 0.6952 - accuracy: 0.5204
212/338 [=================>............] - ETA: 6s - loss: 0.6952 - accuracy: 0.5203
214/338 [=================>............] - ETA: 5s - loss: 0.6952 - accuracy: 0.5203
216/338 [==================>...........] - ETA: 5s - loss: 0.6952 - accuracy: 0.5201
218/338 [==================>...........] - ETA: 5s - loss: 0.6952 - accuracy: 0.5194
220/338 [==================>...........] - ETA: 5s - loss: 0.6951 - accuracy: 0.5197
222/338 [==================>...........] - ETA: 5s - loss: 0.6951 - accuracy: 0.5198
224/338 [==================>...........] - ETA: 5s - loss: 0.6950 - accuracy: 0.5201
226/338 [===================>..........] - ETA: 5s - loss: 0.6950 - accuracy: 0.5203
228/338 [===================>..........] - ETA: 5s - loss: 0.6950 - accuracy: 0.5199
229/338 [===================>..........] - ETA: 5s - loss: 0.6949 - accuracy: 0.5201
231/338 [===================>..........] - ETA: 5s - loss: 0.6948 - accuracy: 0.5210
233/338 [===================>..........] - ETA: 5s - loss: 0.6948 - accuracy: 0.5215
235/338 [===================>..........] - ETA: 4s - loss: 0.6947 - accuracy: 0.5221
236/338 [===================>..........] - ETA: 4s - loss: 0.6946 - accuracy: 0.5224
238/338 [====================>.........] - ETA: 4s - loss: 0.6946 - accuracy: 0.5226
239/338 [====================>.........] - ETA: 4s - loss: 0.6946 - accuracy: 0.5225
240/338 [====================>.........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5234
242/338 [====================>.........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5234
243/338 [====================>.........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5229
245/338 [====================>.........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5221
246/338 [====================>.........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5220
248/338 [=====================>........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5217
249/338 [=====================>........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5212
251/338 [=====================>........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5210
253/338 [=====================>........] - ETA: 4s - loss: 0.6945 - accuracy: 0.5204
255/338 [=====================>........] - ETA: 4s - loss: 0.6944 - accuracy: 0.5214
257/338 [=====================>........] - ETA: 3s - loss: 0.6943 - accuracy: 0.5218
259/338 [=====================>........] - ETA: 3s - loss: 0.6943 - accuracy: 0.5218
260/338 [======================>.......] - ETA: 3s - loss: 0.6943 - accuracy: 0.5220
261/338 [======================>.......] - ETA: 3s - loss: 0.6942 - accuracy: 0.5230
262/338 [======================>.......] - ETA: 3s - loss: 0.6942 - accuracy: 0.5225
264/338 [======================>.......] - ETA: 3s - loss: 0.6942 - accuracy: 0.5219
266/338 [======================>.......] - ETA: 3s - loss: 0.6941 - accuracy: 0.5223
268/338 [======================>.......] - ETA: 3s - loss: 0.6940 - accuracy: 0.5231
270/338 [======================>.......] - ETA: 3s - loss: 0.6941 - accuracy: 0.5227
272/338 [=======================>......] - ETA: 3s - loss: 0.6941 - accuracy: 0.5224
274/338 [=======================>......] - ETA: 3s - loss: 0.6940 - accuracy: 0.5229
275/338 [=======================>......] - ETA: 3s - loss: 0.6939 - accuracy: 0.5235
277/338 [=======================>......] - ETA: 2s - loss: 0.6939 - accuracy: 0.5236
279/338 [=======================>......] - ETA: 2s - loss: 0.6939 - accuracy: 0.5235
281/338 [=======================>......] - ETA: 2s - loss: 0.6938 - accuracy: 0.5236
282/338 [========================>.....] - ETA: 2s - loss: 0.6938 - accuracy: 0.5239
284/338 [========================>.....] - ETA: 2s - loss: 0.6938 - accuracy: 0.5234
286/338 [========================>.....] - ETA: 2s - loss: 0.6937 - accuracy: 0.5238
288/338 [========================>.....] - ETA: 2s - loss: 0.6937 - accuracy: 0.5240
290/338 [========================>.....] - ETA: 2s - loss: 0.6937 - accuracy: 0.5236
292/338 [========================>.....] - ETA: 2s - loss: 0.6936 - accuracy: 0.5240
294/338 [=========================>....] - ETA: 2s - loss: 0.6937 - accuracy: 0.5235
296/338 [=========================>....] - ETA: 2s - loss: 0.6935 - accuracy: 0.5242
298/338 [=========================>....] - ETA: 1s - loss: 0.6935 - accuracy: 0.5237
300/338 [=========================>....] - ETA: 1s - loss: 0.6934 - accuracy: 0.5241
302/338 [=========================>....] - ETA: 1s - loss: 0.6934 - accuracy: 0.5237
303/338 [=========================>....] - ETA: 1s - loss: 0.6934 - accuracy: 0.5236
304/338 [=========================>....] - ETA: 1s - loss: 0.6934 - accuracy: 0.5235
306/338 [==========================>...] - ETA: 1s - loss: 0.6934 - accuracy: 0.5236
308/338 [==========================>...] - ETA: 1s - loss: 0.6934 - accuracy: 0.5234
310/338 [==========================>...] - ETA: 1s - loss: 0.6933 - accuracy: 0.5237
311/338 [==========================>...] - ETA: 1s - loss: 0.6933 - accuracy: 0.5243
313/338 [==========================>...] - ETA: 1s - loss: 0.6932 - accuracy: 0.5249
314/338 [==========================>...] - ETA: 1s - loss: 0.6932 - accuracy: 0.5245
316/338 [===========================>..] - ETA: 1s - loss: 0.6931 - accuracy: 0.5245
317/338 [===========================>..] - ETA: 1s - loss: 0.6931 - accuracy: 0.5246
319/338 [===========================>..] - ETA: 0s - loss: 0.6931 - accuracy: 0.5246
321/338 [===========================>..] - ETA: 0s - loss: 0.6931 - accuracy: 0.5246
323/338 [===========================>..] - ETA: 0s - loss: 0.6931 - accuracy: 0.5238
324/338 [===========================>..] - ETA: 0s - loss: 0.6931 - accuracy: 0.5237
326/338 [===========================>..] - ETA: 0s - loss: 0.6931 - accuracy: 0.5234
328/338 [============================>.] - ETA: 0s - loss: 0.6931 - accuracy: 0.5227
330/338 [============================>.] - ETA: 0s - loss: 0.6931 - accuracy: 0.5221
332/338 [============================>.] - ETA: 0s - loss: 0.6931 - accuracy: 0.5223
334/338 [============================>.] - ETA: 0s - loss: 0.6931 - accuracy: 0.5222
336/338 [============================>.] - ETA: 0s - loss: 0.6930 - accuracy: 0.5228
338/338 [==============================] - ETA: 0s - loss: 0.6930 - accuracy: 0.5230
338/338 [==============================] - 19s 53ms/step - loss: 0.6930 - accuracy: 0.5230 - val_loss: 0.6913 - val_accuracy: 0.5235
Epoch 2/5
1/338 [..............................] - ETA: 15s - loss: 0.6786 - accuracy: 0.6562
3/338 [..............................] - ETA: 15s - loss: 0.6910 - accuracy: 0.5208
5/338 [..............................] - ETA: 15s - loss: 0.6897 - accuracy: 0.5188
7/338 [..............................] - ETA: 15s - loss: 0.6911 - accuracy: 0.5000
9/338 [..............................] - ETA: 15s - loss: 0.6893 - accuracy: 0.5104
11/338 [..............................] - ETA: 15s - loss: 0.6899 - accuracy: 0.5085
13/338 [>.............................] - ETA: 15s - loss: 0.6884 - accuracy: 0.5240
15/338 [>.............................] - ETA: 15s - loss: 0.6898 - accuracy: 0.5104
17/338 [>.............................] - ETA: 15s - loss: 0.6905 - accuracy: 0.5000
19/338 [>.............................] - ETA: 15s - loss: 0.6906 - accuracy: 0.5033
21/338 [>.............................] - ETA: 15s - loss: 0.6903 - accuracy: 0.5060
23/338 [=>............................] - ETA: 15s - loss: 0.6894 - accuracy: 0.5122
25/338 [=>............................] - ETA: 15s - loss: 0.6893 - accuracy: 0.5163
27/338 [=>............................] - ETA: 15s - loss: 0.6891 - accuracy: 0.5162
29/338 [=>............................] - ETA: 14s - loss: 0.6889 - accuracy: 0.5183
31/338 [=>............................] - ETA: 14s - loss: 0.6885 - accuracy: 0.5222
32/338 [=>............................] - ETA: 14s - loss: 0.6887 - accuracy: 0.5215
34/338 [==>...........................] - ETA: 14s - loss: 0.6897 - accuracy: 0.5110
36/338 [==>...........................] - ETA: 14s - loss: 0.6894 - accuracy: 0.5122
38/338 [==>...........................] - ETA: 14s - loss: 0.6888 - accuracy: 0.5156
40/338 [==>...........................] - ETA: 14s - loss: 0.6890 - accuracy: 0.5117
42/338 [==>...........................] - ETA: 14s - loss: 0.6887 - accuracy: 0.5149
44/338 [==>...........................] - ETA: 14s - loss: 0.6878 - accuracy: 0.5199
46/338 [===>..........................] - ETA: 14s - loss: 0.6875 - accuracy: 0.5217
48/338 [===>..........................] - ETA: 14s - loss: 0.6874 - accuracy: 0.5221
50/338 [===>..........................] - ETA: 13s - loss: 0.6871 - accuracy: 0.5250
52/338 [===>..........................] - ETA: 13s - loss: 0.6871 - accuracy: 0.5240
54/338 [===>..........................] - ETA: 13s - loss: 0.6871 - accuracy: 0.5255
56/338 [===>..........................] - ETA: 13s - loss: 0.6868 - accuracy: 0.5262
58/338 [====>.........................] - ETA: 13s - loss: 0.6871 - accuracy: 0.5237
60/338 [====>.........................] - ETA: 13s - loss: 0.6873 - accuracy: 0.5224
61/338 [====>.........................] - ETA: 13s - loss: 0.6870 - accuracy: 0.5241
62/338 [====>.........................] - ETA: 13s - loss: 0.6871 - accuracy: 0.5247
64/338 [====>.........................] - ETA: 13s - loss: 0.6873 - accuracy: 0.5234
66/338 [====>.........................] - ETA: 13s - loss: 0.6874 - accuracy: 0.5227
68/338 [=====>........................] - ETA: 13s - loss: 0.6874 - accuracy: 0.5216
69/338 [=====>........................] - ETA: 13s - loss: 0.6871 - accuracy: 0.5254
71/338 [=====>........................] - ETA: 12s - loss: 0.6872 - accuracy: 0.5251
73/338 [=====>........................] - ETA: 12s - loss: 0.6870 - accuracy: 0.5278
75/338 [=====>........................] - ETA: 12s - loss: 0.6870 - accuracy: 0.5283
77/338 [=====>........................] - ETA: 12s - loss: 0.6867 - accuracy: 0.5296
79/338 [======>.......................] - ETA: 12s - loss: 0.6867 - accuracy: 0.5297
81/338 [======>.......................] - ETA: 12s - loss: 0.6868 - accuracy: 0.5297
83/338 [======>.......................] - ETA: 12s - loss: 0.6869 - accuracy: 0.5286
84/338 [======>.......................] - ETA: 12s - loss: 0.6868 - accuracy: 0.5298
85/338 [======>.......................] - ETA: 12s - loss: 0.6867 - accuracy: 0.5298
87/338 [======>.......................] - ETA: 12s - loss: 0.6868 - accuracy: 0.5295
89/338 [======>.......................] - ETA: 12s - loss: 0.6866 - accuracy: 0.5302
91/338 [=======>......................] - ETA: 11s - loss: 0.6866 - accuracy: 0.5306
92/338 [=======>......................] - ETA: 11s - loss: 0.6865 - accuracy: 0.5306
94/338 [=======>......................] - ETA: 11s - loss: 0.6863 - accuracy: 0.5326
95/338 [=======>......................] - ETA: 11s - loss: 0.6863 - accuracy: 0.5322
96/338 [=======>......................] - ETA: 11s - loss: 0.6864 - accuracy: 0.5316
98/338 [=======>......................] - ETA: 11s - loss: 0.6865 - accuracy: 0.5316
100/338 [=======>......................] - ETA: 11s - loss: 0.6868 - accuracy: 0.5297
102/338 [========>.....................] - ETA: 11s - loss: 0.6870 - accuracy: 0.5273
103/338 [========>.....................] - ETA: 11s - loss: 0.6869 - accuracy: 0.5279
105/338 [========>.....................] - ETA: 11s - loss: 0.6867 - accuracy: 0.5295
107/338 [========>.....................] - ETA: 11s - loss: 0.6869 - accuracy: 0.5275
109/338 [========>.....................] - ETA: 11s - loss: 0.6867 - accuracy: 0.5292
111/338 [========>.....................] - ETA: 11s - loss: 0.6864 - accuracy: 0.5327
112/338 [========>.....................] - ETA: 11s - loss: 0.6864 - accuracy: 0.5321
114/338 [=========>....................] - ETA: 10s - loss: 0.6863 - accuracy: 0.5326
116/338 [=========>....................] - ETA: 10s - loss: 0.6862 - accuracy: 0.5315
118/338 [=========>....................] - ETA: 10s - loss: 0.6861 - accuracy: 0.5320
119/338 [=========>....................] - ETA: 10s - loss: 0.6862 - accuracy: 0.5312
121/338 [=========>....................] - ETA: 10s - loss: 0.6863 - accuracy: 0.5312
123/338 [=========>....................] - ETA: 10s - loss: 0.6866 - accuracy: 0.5290
124/338 [==========>...................] - ETA: 10s - loss: 0.6865 - accuracy: 0.5295
126/338 [==========>...................] - ETA: 10s - loss: 0.6862 - accuracy: 0.5310
128/338 [==========>...................] - ETA: 10s - loss: 0.6862 - accuracy: 0.5308
130/338 [==========>...................] - ETA: 10s - loss: 0.6862 - accuracy: 0.5303
132/338 [==========>...................] - ETA: 10s - loss: 0.6861 - accuracy: 0.5315
134/338 [==========>...................] - ETA: 9s - loss: 0.6861 - accuracy: 0.5315
135/338 [==========>...................] - ETA: 9s - loss: 0.6860 - accuracy: 0.5324
136/338 [===========>..................] - ETA: 9s - loss: 0.6860 - accuracy: 0.5319
138/338 [===========>..................] - ETA: 9s - loss: 0.6860 - accuracy: 0.5312
139/338 [===========>..................] - ETA: 9s - loss: 0.6860 - accuracy: 0.5317
141/338 [===========>..................] - ETA: 9s - loss: 0.6859 - accuracy: 0.5312
143/338 [===========>..................] - ETA: 9s - loss: 0.6858 - accuracy: 0.5312
145/338 [===========>..................] - ETA: 9s - loss: 0.6857 - accuracy: 0.5315
146/338 [===========>..................] - ETA: 9s - loss: 0.6857 - accuracy: 0.5315
148/338 [============>.................] - ETA: 9s - loss: 0.6856 - accuracy: 0.5319
149/338 [============>.................] - ETA: 9s - loss: 0.6856 - accuracy: 0.5321
151/338 [============>.................] - ETA: 9s - loss: 0.6856 - accuracy: 0.5315
153/338 [============>.................] - ETA: 9s - loss: 0.6856 - accuracy: 0.5312
155/338 [============>.................] - ETA: 8s - loss: 0.6856 - accuracy: 0.5302
157/338 [============>.................] - ETA: 8s - loss: 0.6855 - accuracy: 0.5305
159/338 [=============>................] - ETA: 8s - loss: 0.6854 - accuracy: 0.5312
161/338 [=============>................] - ETA: 8s - loss: 0.6853 - accuracy: 0.5311
163/338 [=============>................] - ETA: 8s - loss: 0.6853 - accuracy: 0.5312
165/338 [=============>................] - ETA: 8s - loss: 0.6854 - accuracy: 0.5307
166/338 [=============>................] - ETA: 8s - loss: 0.6853 - accuracy: 0.5312
167/338 [=============>................] - ETA: 8s - loss: 0.6853 - accuracy: 0.5312
169/338 [==============>...............] - ETA: 8s - loss: 0.6854 - accuracy: 0.5307
171/338 [==============>...............] - ETA: 8s - loss: 0.6854 - accuracy: 0.5303
173/338 [==============>...............] - ETA: 8s - loss: 0.6854 - accuracy: 0.5303
175/338 [==============>...............] - ETA: 7s - loss: 0.6856 - accuracy: 0.5291
177/338 [==============>...............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5312
178/338 [==============>...............] - ETA: 7s - loss: 0.6852 - accuracy: 0.5318
180/338 [==============>...............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5314
182/338 [===============>..............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5314
184/338 [===============>..............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5316
185/338 [===============>..............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5316
187/338 [===============>..............] - ETA: 7s - loss: 0.6852 - accuracy: 0.5314
188/338 [===============>..............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5309
190/338 [===============>..............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5306
192/338 [================>.............] - ETA: 7s - loss: 0.6853 - accuracy: 0.5309
194/338 [================>.............] - ETA: 7s - loss: 0.6852 - accuracy: 0.5304
196/338 [================>.............] - ETA: 6s - loss: 0.6850 - accuracy: 0.5309
197/338 [================>.............] - ETA: 6s - loss: 0.6851 - accuracy: 0.5303
199/338 [================>.............] - ETA: 6s - loss: 0.6852 - accuracy: 0.5298
200/338 [================>.............] - ETA: 6s - loss: 0.6852 - accuracy: 0.5298
202/338 [================>.............] - ETA: 6s - loss: 0.6854 - accuracy: 0.5291
204/338 [=================>............] - ETA: 6s - loss: 0.6853 - accuracy: 0.5290
206/338 [=================>............] - ETA: 6s - loss: 0.6854 - accuracy: 0.5291
208/338 [=================>............] - ETA: 6s - loss: 0.6854 - accuracy: 0.5303
209/338 [=================>............] - ETA: 6s - loss: 0.6854 - accuracy: 0.5301
211/338 [=================>............] - ETA: 6s - loss: 0.6854 - accuracy: 0.5308
213/338 [=================>............] - ETA: 6s - loss: 0.6854 - accuracy: 0.5317
215/338 [==================>...........] - ETA: 6s - loss: 0.6852 - accuracy: 0.5331
217/338 [==================>...........] - ETA: 5s - loss: 0.6851 - accuracy: 0.5341
219/338 [==================>...........] - ETA: 5s - loss: 0.6851 - accuracy: 0.5345
221/338 [==================>...........] - ETA: 5s - loss: 0.6850 - accuracy: 0.5345
223/338 [==================>...........] - ETA: 5s - loss: 0.6851 - accuracy: 0.5348
225/338 [==================>...........] - ETA: 5s - loss: 0.6851 - accuracy: 0.5354
227/338 [===================>..........] - ETA: 5s - loss: 0.6851 - accuracy: 0.5363
229/338 [===================>..........] - ETA: 5s - loss: 0.6852 - accuracy: 0.5356
230/338 [===================>..........] - ETA: 5s - loss: 0.6852 - accuracy: 0.5357
232/338 [===================>..........] - ETA: 5s - loss: 0.6852 - accuracy: 0.5362
233/338 [===================>..........] - ETA: 5s - loss: 0.6851 - accuracy: 0.5373
235/338 [===================>..........] - ETA: 5s - loss: 0.6851 - accuracy: 0.5376
237/338 [====================>.........] - ETA: 4s - loss: 0.6852 - accuracy: 0.5378
239/338 [====================>.........] - ETA: 4s - loss: 0.6853 - accuracy: 0.5375
241/338 [====================>.........] - ETA: 4s - loss: 0.6853 - accuracy: 0.5381
243/338 [====================>.........] - ETA: 4s - loss: 0.6853 - accuracy: 0.5387
245/338 [====================>.........] - ETA: 4s - loss: 0.6853 - accuracy: 0.5392
247/338 [====================>.........] - ETA: 4s - loss: 0.6853 - accuracy: 0.5397
249/338 [=====================>........] - ETA: 4s - loss: 0.6853 - accuracy: 0.5394
251/338 [=====================>........] - ETA: 4s - loss: 0.6852 - accuracy: 0.5403
252/338 [=====================>........] - ETA: 4s - loss: 0.6852 - accuracy: 0.5407
254/338 [=====================>........] - ETA: 4s - loss: 0.6852 - accuracy: 0.5411
256/338 [=====================>........] - ETA: 4s - loss: 0.6852 - accuracy: 0.5416
257/338 [=====================>........] - ETA: 3s - loss: 0.6852 - accuracy: 0.5423
259/338 [=====================>........] - ETA: 3s - loss: 0.6851 - accuracy: 0.5431
261/338 [======================>.......] - ETA: 3s - loss: 0.6850 - accuracy: 0.5435
263/338 [======================>.......] - ETA: 3s - loss: 0.6852 - accuracy: 0.5429
265/338 [======================>.......] - ETA: 3s - loss: 0.6853 - accuracy: 0.5427
267/338 [======================>.......] - ETA: 3s - loss: 0.6851 - accuracy: 0.5441
268/338 [======================>.......] - ETA: 3s - loss: 0.6851 - accuracy: 0.5445
270/338 [======================>.......] - ETA: 3s - loss: 0.6852 - accuracy: 0.5447
272/338 [=======================>......] - ETA: 3s - loss: 0.6852 - accuracy: 0.5449
274/338 [=======================>......] - ETA: 3s - loss: 0.6851 - accuracy: 0.5457
276/338 [=======================>......] - ETA: 3s - loss: 0.6852 - accuracy: 0.5462
277/338 [=======================>......] - ETA: 2s - loss: 0.6850 - accuracy: 0.5472
278/338 [=======================>......] - ETA: 2s - loss: 0.6850 - accuracy: 0.5477
280/338 [=======================>......] - ETA: 2s - loss: 0.6848 - accuracy: 0.5488
282/338 [========================>.....] - ETA: 2s - loss: 0.6847 - accuracy: 0.5495
284/338 [========================>.....] - ETA: 2s - loss: 0.6847 - accuracy: 0.5500
285/338 [========================>.....] - ETA: 2s - loss: 0.6846 - accuracy: 0.5504
287/338 [========================>.....] - ETA: 2s - loss: 0.6846 - accuracy: 0.5507
289/338 [========================>.....] - ETA: 2s - loss: 0.6847 - accuracy: 0.5506
291/338 [========================>.....] - ETA: 2s - loss: 0.6847 - accuracy: 0.5505
292/338 [========================>.....] - ETA: 2s - loss: 0.6847 - accuracy: 0.5510
293/338 [=========================>....] - ETA: 2s - loss: 0.6846 - accuracy: 0.5516
294/338 [=========================>....] - ETA: 2s - loss: 0.6846 - accuracy: 0.5517
296/338 [=========================>....] - ETA: 2s - loss: 0.6846 - accuracy: 0.5516
298/338 [=========================>....] - ETA: 1s - loss: 0.6846 - accuracy: 0.5526
300/338 [=========================>....] - ETA: 1s - loss: 0.6845 - accuracy: 0.5537
302/338 [=========================>....] - ETA: 1s - loss: 0.6844 - accuracy: 0.5540
304/338 [=========================>....] - ETA: 1s - loss: 0.6845 - accuracy: 0.5541
306/338 [==========================>...] - ETA: 1s - loss: 0.6845 - accuracy: 0.5544
308/338 [==========================>...] - ETA: 1s - loss: 0.6844 - accuracy: 0.5549
310/338 [==========================>...] - ETA: 1s - loss: 0.6844 - accuracy: 0.5549
312/338 [==========================>...] - ETA: 1s - loss: 0.6844 - accuracy: 0.5559
314/338 [==========================>...] - ETA: 1s - loss: 0.6844 - accuracy: 0.5563
316/338 [===========================>..] - ETA: 1s - loss: 0.6842 - accuracy: 0.5575
318/338 [===========================>..] - ETA: 0s - loss: 0.6842 - accuracy: 0.5580
320/338 [===========================>..] - ETA: 0s - loss: 0.6843 - accuracy: 0.5576
322/338 [===========================>..] - ETA: 0s - loss: 0.6843 - accuracy: 0.5573
324/338 [===========================>..] - ETA: 0s - loss: 0.6842 - accuracy: 0.5582
326/338 [===========================>..] - ETA: 0s - loss: 0.6841 - accuracy: 0.5588
328/338 [============================>.] - ETA: 0s - loss: 0.6842 - accuracy: 0.5582
330/338 [============================>.] - ETA: 0s - loss: 0.6842 - accuracy: 0.5585
332/338 [============================>.] - ETA: 0s - loss: 0.6842 - accuracy: 0.5591
334/338 [============================>.] - ETA: 0s - loss: 0.6841 - accuracy: 0.5593
335/338 [============================>.] - ETA: 0s - loss: 0.6841 - accuracy: 0.5596
337/338 [============================>.] - ETA: 0s - loss: 0.6841 - accuracy: 0.5599
338/338 [==============================] - 18s 53ms/step - loss: 0.6842 - accuracy: 0.5599 - val_loss: 0.6834 - val_accuracy: 0.5910
Epoch 3/5
1/338 [..............................] - ETA: 16s - loss: 0.6924 - accuracy: 0.5625
2/338 [..............................] - ETA: 17s - loss: 0.6781 - accuracy: 0.6250
4/338 [..............................] - ETA: 16s - loss: 0.6738 - accuracy: 0.6562
6/338 [..............................] - ETA: 16s - loss: 0.6747 - accuracy: 0.6562
8/338 [..............................] - ETA: 15s - loss: 0.6767 - accuracy: 0.6484
10/338 [..............................] - ETA: 15s - loss: 0.6764 - accuracy: 0.6562
12/338 [>.............................] - ETA: 16s - loss: 0.6728 - accuracy: 0.6745
14/338 [>.............................] - ETA: 15s - loss: 0.6743 - accuracy: 0.6540
16/338 [>.............................] - ETA: 15s - loss: 0.6739 - accuracy: 0.6543
18/338 [>.............................] - ETA: 15s - loss: 0.6716 - accuracy: 0.6632
20/338 [>.............................] - ETA: 15s - loss: 0.6721 - accuracy: 0.6562
22/338 [>.............................] - ETA: 15s - loss: 0.6725 - accuracy: 0.6577
24/338 [=>............................] - ETA: 15s - loss: 0.6731 - accuracy: 0.6549
26/338 [=>............................] - ETA: 15s - loss: 0.6735 - accuracy: 0.6526
28/338 [=>............................] - ETA: 15s - loss: 0.6721 - accuracy: 0.6618
30/338 [=>............................] - ETA: 14s - loss: 0.6719 - accuracy: 0.6625
32/338 [=>............................] - ETA: 14s - loss: 0.6725 - accuracy: 0.6602
33/338 [=>............................] - ETA: 14s - loss: 0.6719 - accuracy: 0.6610
35/338 [==>...........................] - ETA: 14s - loss: 0.6732 - accuracy: 0.6554
37/338 [==>...........................] - ETA: 14s - loss: 0.6740 - accuracy: 0.6495
38/338 [==>...........................] - ETA: 14s - loss: 0.6738 - accuracy: 0.6538
40/338 [==>...........................] - ETA: 14s - loss: 0.6741 - accuracy: 0.6531
42/338 [==>...........................] - ETA: 14s - loss: 0.6742 - accuracy: 0.6503
44/338 [==>...........................] - ETA: 14s - loss: 0.6740 - accuracy: 0.6491
45/338 [==>...........................] - ETA: 14s - loss: 0.6745 - accuracy: 0.6451
47/338 [===>..........................] - ETA: 14s - loss: 0.6747 - accuracy: 0.6410
49/338 [===>..........................] - ETA: 14s - loss: 0.6755 - accuracy: 0.6352
51/338 [===>..........................] - ETA: 13s - loss: 0.6750 - accuracy: 0.6354
53/338 [===>..........................] - ETA: 13s - loss: 0.6749 - accuracy: 0.6380
54/338 [===>..........................] - ETA: 13s - loss: 0.6749 - accuracy: 0.6383
56/338 [===>..........................] - ETA: 13s - loss: 0.6754 - accuracy: 0.6334
58/338 [====>.........................] - ETA: 13s - loss: 0.6753 - accuracy: 0.6358
60/338 [====>.........................] - ETA: 13s - loss: 0.6750 - accuracy: 0.6375
62/338 [====>.........................] - ETA: 13s - loss: 0.6755 - accuracy: 0.6351
64/338 [====>.........................] - ETA: 13s - loss: 0.6756 - accuracy: 0.6328
66/338 [====>.........................] - ETA: 13s - loss: 0.6754 - accuracy: 0.6330
68/338 [=====>........................] - ETA: 13s - loss: 0.6754 - accuracy: 0.6319
70/338 [=====>........................] - ETA: 12s - loss: 0.6756 - accuracy: 0.6299
71/338 [=====>........................] - ETA: 12s - loss: 0.6755 - accuracy: 0.6312
73/338 [=====>........................] - ETA: 12s - loss: 0.6755 - accuracy: 0.6297
75/338 [=====>........................] - ETA: 12s - loss: 0.6760 - accuracy: 0.6275
76/338 [=====>........................] - ETA: 12s - loss: 0.6757 - accuracy: 0.6291
78/338 [=====>........................] - ETA: 12s - loss: 0.6754 - accuracy: 0.6294
80/338 [======>.......................] - ETA: 12s - loss: 0.6756 - accuracy: 0.6285
82/338 [======>.......................] - ETA: 12s - loss: 0.6756 - accuracy: 0.6284
84/338 [======>.......................] - ETA: 12s - loss: 0.6753 - accuracy: 0.6306
86/338 [======>.......................] - ETA: 12s - loss: 0.6754 - accuracy: 0.6312
88/338 [======>.......................] - ETA: 12s - loss: 0.6752 - accuracy: 0.6321
90/338 [======>.......................] - ETA: 12s - loss: 0.6752 - accuracy: 0.6326
92/338 [=======>......................] - ETA: 11s - loss: 0.6752 - accuracy: 0.6315
94/338 [=======>......................] - ETA: 11s - loss: 0.6756 - accuracy: 0.6300
96/338 [=======>......................] - ETA: 11s - loss: 0.6755 - accuracy: 0.6302
97/338 [=======>......................] - ETA: 11s - loss: 0.6756 - accuracy: 0.6302
99/338 [=======>......................] - ETA: 11s - loss: 0.6760 - accuracy: 0.6291
101/338 [=======>......................] - ETA: 11s - loss: 0.6757 - accuracy: 0.6290
103/338 [========>.....................] - ETA: 11s - loss: 0.6756 - accuracy: 0.6289
105/338 [========>.....................] - ETA: 11s - loss: 0.6755 - accuracy: 0.6292
107/338 [========>.....................] - ETA: 11s - loss: 0.6756 - accuracy: 0.6282
109/338 [========>.....................] - ETA: 11s - loss: 0.6754 - accuracy: 0.6284
111/338 [========>.....................] - ETA: 11s - loss: 0.6755 - accuracy: 0.6270
113/338 [=========>....................] - ETA: 10s - loss: 0.6755 - accuracy: 0.6280
115/338 [=========>....................] - ETA: 10s - loss: 0.6754 - accuracy: 0.6283
116/338 [=========>....................] - ETA: 10s - loss: 0.6756 - accuracy: 0.6277
118/338 [=========>....................] - ETA: 10s - loss: 0.6756 - accuracy: 0.6266
120/338 [=========>....................] - ETA: 10s - loss: 0.6759 - accuracy: 0.6247
122/338 [=========>....................] - ETA: 10s - loss: 0.6757 - accuracy: 0.6258
124/338 [==========>...................] - ETA: 10s - loss: 0.6760 - accuracy: 0.6250
125/338 [==========>...................] - ETA: 10s - loss: 0.6760 - accuracy: 0.6242
127/338 [==========>...................] - ETA: 10s - loss: 0.6756 - accuracy: 0.6257
129/338 [==========>...................] - ETA: 10s - loss: 0.6757 - accuracy: 0.6252
131/338 [==========>...................] - ETA: 10s - loss: 0.6758 - accuracy: 0.6248
133/338 [==========>...................] - ETA: 9s - loss: 0.6762 - accuracy: 0.6245
135/338 [==========>...................] - ETA: 9s - loss: 0.6763 - accuracy: 0.6243
137/338 [===========>..................] - ETA: 9s - loss: 0.6760 - accuracy: 0.6250
139/338 [===========>..................] - ETA: 9s - loss: 0.6761 - accuracy: 0.6246
141/338 [===========>..................] - ETA: 9s - loss: 0.6761 - accuracy: 0.6250
143/338 [===========>..................] - ETA: 9s - loss: 0.6762 - accuracy: 0.6246
144/338 [===========>..................] - ETA: 9s - loss: 0.6764 - accuracy: 0.6237
146/338 [===========>..................] - ETA: 9s - loss: 0.6761 - accuracy: 0.6244
148/338 [============>.................] - ETA: 9s - loss: 0.6760 - accuracy: 0.6250
150/338 [============>.................] - ETA: 9s - loss: 0.6757 - accuracy: 0.6267
152/338 [============>.................] - ETA: 9s - loss: 0.6755 - accuracy: 0.6281
154/338 [============>.................] - ETA: 8s - loss: 0.6758 - accuracy: 0.6268
156/338 [============>.................] - ETA: 8s - loss: 0.6759 - accuracy: 0.6258
158/338 [=============>................] - ETA: 8s - loss: 0.6759 - accuracy: 0.6258
160/338 [=============>................] - ETA: 8s - loss: 0.6756 - accuracy: 0.6270
162/338 [=============>................] - ETA: 8s - loss: 0.6753 - accuracy: 0.6279
163/338 [=============>................] - ETA: 8s - loss: 0.6751 - accuracy: 0.6285
165/338 [=============>................] - ETA: 8s - loss: 0.6752 - accuracy: 0.6275
166/338 [=============>................] - ETA: 8s - loss: 0.6753 - accuracy: 0.6273
168/338 [=============>................] - ETA: 8s - loss: 0.6752 - accuracy: 0.6278
170/338 [==============>...............] - ETA: 8s - loss: 0.6752 - accuracy: 0.6279
172/338 [==============>...............] - ETA: 8s - loss: 0.6752 - accuracy: 0.6279
174/338 [==============>...............] - ETA: 7s - loss: 0.6751 - accuracy: 0.6282
176/338 [==============>...............] - ETA: 7s - loss: 0.6751 - accuracy: 0.6280
178/338 [==============>...............] - ETA: 7s - loss: 0.6751 - accuracy: 0.6278
180/338 [==============>...............] - ETA: 7s - loss: 0.6749 - accuracy: 0.6283
181/338 [===============>..............] - ETA: 7s - loss: 0.6749 - accuracy: 0.6276
183/338 [===============>..............] - ETA: 7s - loss: 0.6751 - accuracy: 0.6264
185/338 [===============>..............] - ETA: 7s - loss: 0.6751 - accuracy: 0.6258
187/338 [===============>..............] - ETA: 7s - loss: 0.6755 - accuracy: 0.6245
189/338 [===============>..............] - ETA: 7s - loss: 0.6754 - accuracy: 0.6247
190/338 [===============>..............] - ETA: 7s - loss: 0.6754 - accuracy: 0.6247
192/338 [================>.............] - ETA: 7s - loss: 0.6754 - accuracy: 0.6245
194/338 [================>.............] - ETA: 6s - loss: 0.6751 - accuracy: 0.6260
196/338 [================>.............] - ETA: 6s - loss: 0.6750 - accuracy: 0.6260
197/338 [================>.............] - ETA: 6s - loss: 0.6749 - accuracy: 0.6264
199/338 [================>.............] - ETA: 6s - loss: 0.6749 - accuracy: 0.6261
201/338 [================>.............] - ETA: 6s - loss: 0.6749 - accuracy: 0.6258
202/338 [================>.............] - ETA: 6s - loss: 0.6748 - accuracy: 0.6267
204/338 [=================>............] - ETA: 6s - loss: 0.6746 - accuracy: 0.6273
206/338 [=================>............] - ETA: 6s - loss: 0.6747 - accuracy: 0.6265
208/338 [=================>............] - ETA: 6s - loss: 0.6747 - accuracy: 0.6270
210/338 [=================>............] - ETA: 6s - loss: 0.6744 - accuracy: 0.6271
212/338 [=================>............] - ETA: 6s - loss: 0.6747 - accuracy: 0.6263
214/338 [=================>............] - ETA: 6s - loss: 0.6746 - accuracy: 0.6265
216/338 [==================>...........] - ETA: 5s - loss: 0.6746 - accuracy: 0.6260
218/338 [==================>...........] - ETA: 5s - loss: 0.6745 - accuracy: 0.6263
220/338 [==================>...........] - ETA: 5s - loss: 0.6744 - accuracy: 0.6266
221/338 [==================>...........] - ETA: 5s - loss: 0.6745 - accuracy: 0.6264
223/338 [==================>...........] - ETA: 5s - loss: 0.6745 - accuracy: 0.6267
225/338 [==================>...........] - ETA: 5s - loss: 0.6745 - accuracy: 0.6264
227/338 [===================>..........] - ETA: 5s - loss: 0.6747 - accuracy: 0.6253
229/338 [===================>..........] - ETA: 5s - loss: 0.6748 - accuracy: 0.6246
231/338 [===================>..........] - ETA: 5s - loss: 0.6747 - accuracy: 0.6250
232/338 [===================>..........] - ETA: 5s - loss: 0.6745 - accuracy: 0.6257
234/338 [===================>..........] - ETA: 5s - loss: 0.6743 - accuracy: 0.6263
236/338 [===================>..........] - ETA: 4s - loss: 0.6744 - accuracy: 0.6259
238/338 [====================>.........] - ETA: 4s - loss: 0.6745 - accuracy: 0.6253
240/338 [====================>.........] - ETA: 4s - loss: 0.6744 - accuracy: 0.6254
242/338 [====================>.........] - ETA: 4s - loss: 0.6746 - accuracy: 0.6246
244/338 [====================>.........] - ETA: 4s - loss: 0.6747 - accuracy: 0.6242
246/338 [====================>.........] - ETA: 4s - loss: 0.6748 - accuracy: 0.6240
248/338 [=====================>........] - ETA: 4s - loss: 0.6748 - accuracy: 0.6236
250/338 [=====================>........] - ETA: 4s - loss: 0.6748 - accuracy: 0.6235
251/338 [=====================>........] - ETA: 4s - loss: 0.6747 - accuracy: 0.6238
253/338 [=====================>........] - ETA: 4s - loss: 0.6748 - accuracy: 0.6230
255/338 [=====================>........] - ETA: 4s - loss: 0.6747 - accuracy: 0.6234
257/338 [=====================>........] - ETA: 3s - loss: 0.6747 - accuracy: 0.6232
259/338 [=====================>........] - ETA: 3s - loss: 0.6746 - accuracy: 0.6234
261/338 [======================>.......] - ETA: 3s - loss: 0.6746 - accuracy: 0.6233
263/338 [======================>.......] - ETA: 3s - loss: 0.6746 - accuracy: 0.6236
265/338 [======================>.......] - ETA: 3s - loss: 0.6747 - accuracy: 0.6231
266/338 [======================>.......] - ETA: 3s - loss: 0.6748 - accuracy: 0.6228
267/338 [======================>.......] - ETA: 3s - loss: 0.6746 - accuracy: 0.6234
269/338 [======================>.......] - ETA: 3s - loss: 0.6744 - accuracy: 0.6241
271/338 [=======================>......] - ETA: 3s - loss: 0.6745 - accuracy: 0.6238
272/338 [=======================>......] - ETA: 3s - loss: 0.6744 - accuracy: 0.6243
274/338 [=======================>......] - ETA: 3s - loss: 0.6744 - accuracy: 0.6245
276/338 [=======================>......] - ETA: 3s - loss: 0.6744 - accuracy: 0.6245
278/338 [=======================>......] - ETA: 2s - loss: 0.6745 - accuracy: 0.6246
280/338 [=======================>......] - ETA: 2s - loss: 0.6744 - accuracy: 0.6247
282/338 [========================>.....] - ETA: 2s - loss: 0.6744 - accuracy: 0.6247
284/338 [========================>.....] - ETA: 2s - loss: 0.6745 - accuracy: 0.6243
286/338 [========================>.....] - ETA: 2s - loss: 0.6747 - accuracy: 0.6237
288/338 [========================>.....] - ETA: 2s - loss: 0.6746 - accuracy: 0.6241
290/338 [========================>.....] - ETA: 2s - loss: 0.6747 - accuracy: 0.6236
292/338 [========================>.....] - ETA: 2s - loss: 0.6746 - accuracy: 0.6235
294/338 [=========================>....] - ETA: 2s - loss: 0.6746 - accuracy: 0.6235
296/338 [=========================>....] - ETA: 2s - loss: 0.6747 - accuracy: 0.6229
298/338 [=========================>....] - ETA: 1s - loss: 0.6747 - accuracy: 0.6229
300/338 [=========================>....] - ETA: 1s - loss: 0.6746 - accuracy: 0.6230
302/338 [=========================>....] - ETA: 1s - loss: 0.6746 - accuracy: 0.6229
304/338 [=========================>....] - ETA: 1s - loss: 0.6747 - accuracy: 0.6227
306/338 [==========================>...] - ETA: 1s - loss: 0.6746 - accuracy: 0.6228
307/338 [==========================>...] - ETA: 1s - loss: 0.6745 - accuracy: 0.6233
308/338 [==========================>...] - ETA: 1s - loss: 0.6745 - accuracy: 0.6234
310/338 [==========================>...] - ETA: 1s - loss: 0.6744 - accuracy: 0.6240
312/338 [==========================>...] - ETA: 1s - loss: 0.6741 - accuracy: 0.6250
314/338 [==========================>...] - ETA: 1s - loss: 0.6741 - accuracy: 0.6249
316/338 [===========================>..] - ETA: 1s - loss: 0.6740 - accuracy: 0.6252
318/338 [===========================>..] - ETA: 0s - loss: 0.6741 - accuracy: 0.6246
320/338 [===========================>..] - ETA: 0s - loss: 0.6742 - accuracy: 0.6246
322/338 [===========================>..] - ETA: 0s - loss: 0.6741 - accuracy: 0.6249
324/338 [===========================>..] - ETA: 0s - loss: 0.6739 - accuracy: 0.6253
326/338 [===========================>..] - ETA: 0s - loss: 0.6739 - accuracy: 0.6255
328/338 [============================>.] - ETA: 0s - loss: 0.6737 - accuracy: 0.6260
329/338 [============================>.] - ETA: 0s - loss: 0.6737 - accuracy: 0.6258
331/338 [============================>.] - ETA: 0s - loss: 0.6736 - accuracy: 0.6260
333/338 [============================>.] - ETA: 0s - loss: 0.6735 - accuracy: 0.6263
335/338 [============================>.] - ETA: 0s - loss: 0.6736 - accuracy: 0.6264
337/338 [============================>.] - ETA: 0s - loss: 0.6735 - accuracy: 0.6269
338/338 [==============================] - 18s 52ms/step - loss: 0.6735 - accuracy: 0.6270 - val_loss: 0.6701 - val_accuracy: 0.6182
Epoch 4/5
1/338 [..............................] - ETA: 16s - loss: 0.6510 - accuracy: 0.6875
3/338 [..............................] - ETA: 15s - loss: 0.6626 - accuracy: 0.6146
5/338 [..............................] - ETA: 15s - loss: 0.6623 - accuracy: 0.5875
6/338 [..............................] - ETA: 16s - loss: 0.6599 - accuracy: 0.6094
8/338 [..............................] - ETA: 16s - loss: 0.6642 - accuracy: 0.6055
10/338 [..............................] - ETA: 15s - loss: 0.6638 - accuracy: 0.6062
12/338 [>.............................] - ETA: 15s - loss: 0.6634 - accuracy: 0.6094
14/338 [>.............................] - ETA: 15s - loss: 0.6702 - accuracy: 0.5982
16/338 [>.............................] - ETA: 15s - loss: 0.6700 - accuracy: 0.5996
18/338 [>.............................] - ETA: 15s - loss: 0.6738 - accuracy: 0.5920
20/338 [>.............................] - ETA: 15s - loss: 0.6709 - accuracy: 0.6016
21/338 [>.............................] - ETA: 15s - loss: 0.6695 - accuracy: 0.6071
23/338 [=>............................] - ETA: 15s - loss: 0.6694 - accuracy: 0.6114
25/338 [=>............................] - ETA: 15s - loss: 0.6667 - accuracy: 0.6225
27/338 [=>............................] - ETA: 15s - loss: 0.6667 - accuracy: 0.6204
29/338 [=>............................] - ETA: 15s - loss: 0.6666 - accuracy: 0.6218
31/338 [=>............................] - ETA: 14s - loss: 0.6675 - accuracy: 0.6179
33/338 [=>............................] - ETA: 14s - loss: 0.6678 - accuracy: 0.6203
35/338 [==>...........................] - ETA: 14s - loss: 0.6667 - accuracy: 0.6223
37/338 [==>...........................] - ETA: 14s - loss: 0.6680 - accuracy: 0.6208
39/338 [==>...........................] - ETA: 14s - loss: 0.6679 - accuracy: 0.6226
41/338 [==>...........................] - ETA: 14s - loss: 0.6674 - accuracy: 0.6235
43/338 [==>...........................] - ETA: 14s - loss: 0.6683 - accuracy: 0.6214
45/338 [==>...........................] - ETA: 14s - loss: 0.6671 - accuracy: 0.6264
47/338 [===>..........................] - ETA: 14s - loss: 0.6656 - accuracy: 0.6323
48/338 [===>..........................] - ETA: 14s - loss: 0.6664 - accuracy: 0.6302
49/338 [===>..........................] - ETA: 13s - loss: 0.6654 - accuracy: 0.6333
51/338 [===>..........................] - ETA: 13s - loss: 0.6651 - accuracy: 0.6354
53/338 [===>..........................] - ETA: 13s - loss: 0.6658 - accuracy: 0.6344
55/338 [===>..........................] - ETA: 13s - loss: 0.6660 - accuracy: 0.6352
57/338 [====>.........................] - ETA: 13s - loss: 0.6655 - accuracy: 0.6382
59/338 [====>.........................] - ETA: 13s - loss: 0.6655 - accuracy: 0.6393
61/338 [====>.........................] - ETA: 13s - loss: 0.6655 - accuracy: 0.6378
63/338 [====>.........................] - ETA: 13s - loss: 0.6654 - accuracy: 0.6369
65/338 [====>.........................] - ETA: 13s - loss: 0.6651 - accuracy: 0.6385
67/338 [====>.........................] - ETA: 13s - loss: 0.6652 - accuracy: 0.6367
69/338 [=====>........................] - ETA: 13s - loss: 0.6650 - accuracy: 0.6368
70/338 [=====>........................] - ETA: 13s - loss: 0.6648 - accuracy: 0.6371
72/338 [=====>........................] - ETA: 12s - loss: 0.6651 - accuracy: 0.6363
74/338 [=====>........................] - ETA: 12s - loss: 0.6644 - accuracy: 0.6385
76/338 [=====>........................] - ETA: 12s - loss: 0.6645 - accuracy: 0.6386
77/338 [=====>........................] - ETA: 12s - loss: 0.6644 - accuracy: 0.6384
79/338 [======>.......................] - ETA: 12s - loss: 0.6644 - accuracy: 0.6381
81/338 [======>.......................] - ETA: 12s - loss: 0.6644 - accuracy: 0.6370
83/338 [======>.......................] - ETA: 12s - loss: 0.6648 - accuracy: 0.6352
85/338 [======>.......................] - ETA: 12s - loss: 0.6651 - accuracy: 0.6338
87/338 [======>.......................] - ETA: 12s - loss: 0.6646 - accuracy: 0.6361
88/338 [======>.......................] - ETA: 12s - loss: 0.6647 - accuracy: 0.6357
90/338 [======>.......................] - ETA: 12s - loss: 0.6648 - accuracy: 0.6344
91/338 [=======>......................] - ETA: 12s - loss: 0.6654 - accuracy: 0.6319
93/338 [=======>......................] - ETA: 11s - loss: 0.6645 - accuracy: 0.6344
95/338 [=======>......................] - ETA: 11s - loss: 0.6644 - accuracy: 0.6339
97/338 [=======>......................] - ETA: 11s - loss: 0.6643 - accuracy: 0.6340
99/338 [=======>......................] - ETA: 11s - loss: 0.6642 - accuracy: 0.6345
101/338 [=======>......................] - ETA: 11s - loss: 0.6638 - accuracy: 0.6368
102/338 [========>.....................] - ETA: 11s - loss: 0.6632 - accuracy: 0.6388
104/338 [========>.....................] - ETA: 11s - loss: 0.6633 - accuracy: 0.6385
106/338 [========>.....................] - ETA: 11s - loss: 0.6631 - accuracy: 0.6389
108/338 [========>.....................] - ETA: 11s - loss: 0.6634 - accuracy: 0.6383
110/338 [========>.....................] - ETA: 11s - loss: 0.6630 - accuracy: 0.6386
111/338 [========>.....................] - ETA: 11s - loss: 0.6631 - accuracy: 0.6385
112/338 [========>.....................] - ETA: 10s - loss: 0.6632 - accuracy: 0.6381
114/338 [