In [3]:
#Baseline small CNN for project Milestone
import pandas as pd
import cv2
import os
import numpy as np
from tqdm import tqdm
import os
import gc
from glob import glob
from sklearn.metrics import fbeta_score
import sklearn.metrics
from sklearn.model_selection import train_test_split
import matplotlib.pyplot as plt
import csv
# Keras libraries
import keras as k
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras.layers.normalization import BatchNormalization
from keras.callbacks import Callback, EarlyStopping
from keras import backend
from keras.optimizers import Adam
from keras.callbacks import ModelCheckpoint
import h5py
In [4]:
df_train = pd.read_csv('/home/joerj/train_v2.csv')
In [5]:
# referred to https://www.kaggle.com/anokas/simple-keras-starter for help reading data and setting up basic Keras model
x = []
x_test = []
y = []
flatten = lambda l: [item for sublist in l for item in sublist]
labels = list(set(flatten([l.split(' ') for l in df_train['tags'].values])))
labels.sort()
label_map = {l: i for i, l in enumerate(labels)}
inv_label_map = {i: l for l, i in label_map.items()}
for f, tags in tqdm(df_train.values, miniters=1000):
img = cv2.imread('/home/joerj/train-jpg/train-jpg/{}.jpg'.format(f))
targets = np.zeros(17)
for t in tags.split(' '):
targets[label_map[t]] = 1
x.append(cv2.resize(img, (32, 32)))
y.append(targets)
100%|██████████| 40479/40479 [07:56<00:00, 84.87it/s]
In [7]:
split = 35000
x_train, x_valid, y_train, y_valid = x[:split], x[split:], y[:split], y[split:]
mean_image = np.mean(x_train, axis=0)
x_train -= mean_image
x_valid -= mean_image
x_train /= 128.
x_valid /= 128.
y_train = np.array(y_train, np.uint8)
x_train = np.array(x_train, np.float16)
y_valid = np.array(y_valid, np.uint8)
x_valid = np.array(x_valid, np.float16)
In [54]:
x = [1, 2, 3, 4, 5]
x[:4]
Out[54]:
[1, 2, 3, 4]
In [8]:
#Create model class - model outline sourced from here: https://github.com/EKami/planet-amazon-deforestation
class LossHistory(Callback):
def __init__(self):
super().__init__()
self.train_losses = []
self.val_losses = []
def on_epoch_end(self, epoch, logs={}):
self.train_losses.append(logs.get('loss'))
self.val_losses.append(logs.get('val_loss'))
class AmazonClassifier:
def __init__(self):
self.losses = []
self.classifier = Sequential()
def add_conv_layer_init(self, img_size=(32, 32), c=3, f = 32, p = .25):
self.classifier.add(BatchNormalization(input_shape=(*img_size, c)))
self.classifier.add(Conv2D(f, kernel_size=(3, 3),
padding = 'same',
activation='relu'))
self.classifier.add(Conv2D(f, (3, 3), activation='relu', padding = 'same'))
self.classifier.add(MaxPooling2D(pool_size=(2, 2)))
self.classifier.add(Dropout(p))
def add_conv_layer_mid(self, img_size=(32, 32), c=3, f = 32, p = .25):
self.classifier.add(Conv2D(f, kernel_size=(3, 3),
padding = 'same',
activation='relu'))
self.classifier.add(Conv2D(f, (3, 3), activation='relu', padding = 'same' ))
self.classifier.add(MaxPooling2D(pool_size=(2, 2)))
self.classifier.add(Dropout(p))
def _get_fbeta_score(self, classifier, X_valid, y_valid):
p_valid = classifier.predict(X_valid)
return fbeta_score(y_valid, np.array(p_valid) > 0.2, beta=2, average='samples')
def add_flatten_layer(self):
self.classifier.add(Flatten())
def add_dense_layer(self, output_size = 17, p = 0.5):
self.classifier.add(Dense(512, activation='relu'))
self.classifier.add(BatchNormalization())
self.classifier.add(Dropout(0.5))
self.classifier.add(Dense(output_size, activation='sigmoid'))
def train_model(self, x_train, y_train, learn_rate=0.001, epoch=5, batch_size=128, validation_split_size=0.2, train_callbacks=()):
history = LossHistory()
X_train, X_valid, y_train, y_valid = train_test_split(x_train, y_train,
test_size=validation_split_size, random_state = 1234)
opt = Adam(lr=learn_rate)
self.classifier.compile(loss='binary_crossentropy', optimizer=opt, metrics=['accuracy'])
# early stopping will auto-stop training process if model stops learning after 3 epochs
earlyStopping = EarlyStopping(monitor='val_loss', patience=3, verbose=0, mode='auto')
self.classifier.fit(X_train, y_train,
batch_size=batch_size,
epochs=epoch,
verbose=1,
validation_data=(X_valid, y_valid),
callbacks=[history, *train_callbacks, earlyStopping])
fbeta_score = self._get_fbeta_score(self.classifier, X_valid, y_valid)
return [history.train_losses, history.val_losses, fbeta_score]
def save_weights(self, weight_file_path):
self.classifier.save_weights(weight_file_path)
def load_weights(self, weight_file_path):
self.classifier.load_weights(weight_file_path)
def predict(self, x_test):
predictions = self.classifier.predict(x_test)
return predictions
def map_predictions(self, predictions, labels_map, thresholds):
"""
Return the predictions mapped to their labels
:param predictions: the predictions from the predict() method
:param labels_map: the map
:param thresholds: The threshold of each class to be considered as existing or not existing
:return: the predictions list mapped to their labels
"""
predictions_labels = []
for prediction in predictions:
labels = [labels_map[i] for i, value in enumerate(prediction) if value > thresholds[i]]
predictions_labels.append(labels)
return predictions_labels
def close(self):
backend.clear_session()
In [9]:
#Grid search from ekami - works pretty well
filepath="weights.best.hdf5"
checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True)
p_conv = 0.29592314649019363
p_all = 0.5619647177889551
scale = 1.134238146597395
#Use best parameters
batch_size = 128
validation_split_size = .20
classifier = AmazonClassifier()
classifier.add_conv_layer_init(f = 32, p = p_conv)
classifier.add_conv_layer_mid(f = 64, p = p_conv)
classifier.add_conv_layer_mid(f = 128, p = p_conv)
classifier.add_flatten_layer()
classifier.add_dense_layer(p = p_all)
train_losses, val_losses, scores_list = [], [], []
epochs_arr = [5, 10]
learn_rates = [0.001 * scale, 0.0001 * scale, 0.00001 * scale]
for learn_rate in learn_rates:
for epochs in epochs_arr:
tmp_train_losses, tmp_val_losses, score = classifier.train_model(x_train, y_train, learn_rate, epochs,
batch_size, validation_split_size=validation_split_size,
train_callbacks=[checkpoint])
train_losses += tmp_train_losses
val_losses += tmp_val_losses
scores_list.append(score)
Train on 28000 samples, validate on 7000 samples
Epoch 1/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.2924 - acc: 0.8963Epoch 00000: val_acc improved from -inf to 0.90408, saving model to weights.best.hdf5
28000/28000 [==============================] - 20s - loss: 0.2919 - acc: 0.8965 - val_loss: 0.2559 - val_acc: 0.9041
Epoch 2/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1620 - acc: 0.9361Epoch 00001: val_acc improved from 0.90408 to 0.91682, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1619 - acc: 0.9361 - val_loss: 0.2024 - val_acc: 0.9168
Epoch 3/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1507 - acc: 0.9400Epoch 00002: val_acc improved from 0.91682 to 0.93457, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1507 - acc: 0.9400 - val_loss: 0.1590 - val_acc: 0.9346
Epoch 4/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1452 - acc: 0.9424Epoch 00003: val_acc improved from 0.93457 to 0.94409, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1452 - acc: 0.9424 - val_loss: 0.1362 - val_acc: 0.9441
Epoch 5/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1384 - acc: 0.9451Epoch 00004: val_acc improved from 0.94409 to 0.94532, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1385 - acc: 0.9451 - val_loss: 0.1352 - val_acc: 0.9453
Train on 28000 samples, validate on 7000 samples
Epoch 1/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1358 - acc: 0.9464Epoch 00000: val_acc improved from 0.94532 to 0.95082, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1358 - acc: 0.9464 - val_loss: 0.1256 - val_acc: 0.9508
Epoch 2/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1317 - acc: 0.9480Epoch 00001: val_acc improved from 0.95082 to 0.95156, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1316 - acc: 0.9481 - val_loss: 0.1220 - val_acc: 0.9516
Epoch 3/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1294 - acc: 0.9495Epoch 00002: val_acc improved from 0.95156 to 0.95239, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1293 - acc: 0.9495 - val_loss: 0.1204 - val_acc: 0.9524
Epoch 4/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1272 - acc: 0.9502Epoch 00003: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1272 - acc: 0.9502 - val_loss: 0.1216 - val_acc: 0.9520
Epoch 5/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1250 - acc: 0.9511Epoch 00004: val_acc improved from 0.95239 to 0.95308, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1249 - acc: 0.9511 - val_loss: 0.1192 - val_acc: 0.9531
Epoch 6/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1234 - acc: 0.9519Epoch 00005: val_acc improved from 0.95308 to 0.95315, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1235 - acc: 0.9519 - val_loss: 0.1224 - val_acc: 0.9532
Epoch 7/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1230 - acc: 0.9520Epoch 00006: val_acc improved from 0.95315 to 0.95449, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1229 - acc: 0.9520 - val_loss: 0.1180 - val_acc: 0.9545
Epoch 8/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1205 - acc: 0.9526Epoch 00007: val_acc improved from 0.95449 to 0.95454, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1205 - acc: 0.9526 - val_loss: 0.1175 - val_acc: 0.9545
Epoch 9/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1192 - acc: 0.9533Epoch 00008: val_acc improved from 0.95454 to 0.95580, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1193 - acc: 0.9533 - val_loss: 0.1134 - val_acc: 0.9558
Epoch 10/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1183 - acc: 0.9537Epoch 00009: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1183 - acc: 0.9537 - val_loss: 0.1161 - val_acc: 0.9542
Train on 28000 samples, validate on 7000 samples
Epoch 1/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1115 - acc: 0.9564Epoch 00000: val_acc improved from 0.95580 to 0.95703, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1116 - acc: 0.9564 - val_loss: 0.1098 - val_acc: 0.9570
Epoch 2/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1093 - acc: 0.9569Epoch 00001: val_acc improved from 0.95703 to 0.95760, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1094 - acc: 0.9569 - val_loss: 0.1093 - val_acc: 0.9576
Epoch 3/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1091 - acc: 0.9571Epoch 00002: val_acc improved from 0.95760 to 0.95796, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1091 - acc: 0.9571 - val_loss: 0.1086 - val_acc: 0.9580
Epoch 4/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1081 - acc: 0.9574Epoch 00003: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1080 - acc: 0.9574 - val_loss: 0.1087 - val_acc: 0.9576
Epoch 5/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1075 - acc: 0.9576Epoch 00004: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1076 - acc: 0.9576 - val_loss: 0.1090 - val_acc: 0.9578
Train on 28000 samples, validate on 7000 samples
Epoch 1/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1071 - acc: 0.9579Epoch 00000: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1071 - acc: 0.9579 - val_loss: 0.1080 - val_acc: 0.9576
Epoch 2/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1064 - acc: 0.9584Epoch 00001: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1063 - acc: 0.9584 - val_loss: 0.1084 - val_acc: 0.9575
Epoch 3/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1068 - acc: 0.9581Epoch 00002: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1068 - acc: 0.9581 - val_loss: 0.1079 - val_acc: 0.9578
Epoch 4/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1059 - acc: 0.9585Epoch 00003: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1059 - acc: 0.9585 - val_loss: 0.1085 - val_acc: 0.9570
Epoch 5/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1056 - acc: 0.9582Epoch 00004: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1055 - acc: 0.9582 - val_loss: 0.1078 - val_acc: 0.9578
Epoch 6/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1048 - acc: 0.9585Epoch 00005: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1049 - acc: 0.9585 - val_loss: 0.1084 - val_acc: 0.9576
Epoch 7/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1051 - acc: 0.9590Epoch 00006: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1052 - acc: 0.9590 - val_loss: 0.1084 - val_acc: 0.9576
Epoch 8/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1046 - acc: 0.9591Epoch 00007: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1047 - acc: 0.9591 - val_loss: 0.1087 - val_acc: 0.9578
Epoch 9/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1040 - acc: 0.9592Epoch 00008: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1040 - acc: 0.9592 - val_loss: 0.1082 - val_acc: 0.9579
Train on 28000 samples, validate on 7000 samples
Epoch 1/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1037 - acc: 0.9593Epoch 00000: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1037 - acc: 0.9592 - val_loss: 0.1075 - val_acc: 0.9579
Epoch 2/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1033 - acc: 0.9594Epoch 00001: val_acc improved from 0.95796 to 0.95818, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1033 - acc: 0.9594 - val_loss: 0.1076 - val_acc: 0.9582
Epoch 3/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1027 - acc: 0.9597Epoch 00002: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1026 - acc: 0.9597 - val_loss: 0.1076 - val_acc: 0.9580
Epoch 4/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1029 - acc: 0.9596Epoch 00003: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1030 - acc: 0.9596 - val_loss: 0.1074 - val_acc: 0.9581
Epoch 5/5
27904/28000 [============================>.] - ETA: 0s - loss: 0.1027 - acc: 0.9595Epoch 00004: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1028 - acc: 0.9595 - val_loss: 0.1074 - val_acc: 0.9581
Train on 28000 samples, validate on 7000 samples
Epoch 1/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1026 - acc: 0.9597Epoch 00000: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1025 - acc: 0.9597 - val_loss: 0.1075 - val_acc: 0.9581
Epoch 2/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1033 - acc: 0.9593Epoch 00001: val_acc improved from 0.95818 to 0.95819, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1032 - acc: 0.9593 - val_loss: 0.1073 - val_acc: 0.9582
Epoch 3/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1027 - acc: 0.9599Epoch 00002: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1027 - acc: 0.9599 - val_loss: 0.1074 - val_acc: 0.9580
Epoch 4/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1024 - acc: 0.9597Epoch 00003: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1024 - acc: 0.9597 - val_loss: 0.1075 - val_acc: 0.9581
Epoch 5/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1025 - acc: 0.9595Epoch 00004: val_acc improved from 0.95819 to 0.95823, saving model to weights.best.hdf5
28000/28000 [==============================] - 12s - loss: 0.1025 - acc: 0.9595 - val_loss: 0.1074 - val_acc: 0.9582
Epoch 6/10
27904/28000 [============================>.] - ETA: 0s - loss: 0.1023 - acc: 0.9599Epoch 00005: val_acc did not improve
28000/28000 [==============================] - 12s - loss: 0.1023 - acc: 0.9599 - val_loss: 0.1074 - val_acc: 0.9582
In [10]:
#best: loss: 0.1026 - acc: 0.9598Epoch 00001: val_acc improved from 0.95800 to 0.9581
# best training: 0.9597
#save error over epochs
np.savetxt("error_CNN.csv", np.vstack((val_losses, train_losses)), fmt='%.18e', delimiter=',')
np.savetxt("scores_CNN.csv", scores_list, fmt='%.18e', delimiter=',')
In [11]:
# Save model predictions for ensemble with CNN-8
p_valid = classifier.predict(x_valid)
#np.save("CNN_predict.npy", p_valid, allow_pickle=True, fix_imports=True)
#np.save("target_validation.npy", y_valid, allow_pickle=True, fix_imports=True)
In [24]:
score = fbeta_score(y_valid, np.array(p_valid) > 0.2, beta=2, average='samples')
predicted = np.array(p_valid) > 0.2
Out[24]:
(5479, 17)
In [71]:
for i in range(46 ,50):
if sum(y_valid[i,:] != predicted[i,:]) > 3:
print(35000 + i, sum(y_valid[i,:] != predicted[i,:]))
print(y_valid[i,:])
print(predicted[i,:])
print(labels)
#35000 is labelled agriculture, clear, cultivation, and primary
#labelled - agriculture, cultivation, habitation, partly cloudy, primary, road
35049 4
[1 0 0 0 0 1 0 0 1 0 0 0 1 1 0 0 1]
[ True False False False False False False False True True False True
True True False False False]
['agriculture', 'artisinal_mine', 'bare_ground', 'blooming', 'blow_down', 'clear', 'cloudy', 'conventional_mine', 'cultivation', 'habitation', 'haze', 'partly_cloudy', 'primary', 'road', 'selective_logging', 'slash_burn', 'water']
In [42]:
for i in range(17):
print(labels[i], " & ", np.round_( np.sum(y_valid[:,i] == predicted[:,i]) / y_valid.shape[0], 3), "&",
np.round_( fbeta_score(y_valid[:, i], predicted[:,i], beta=2), 4), "\\\\")
agriculture & 0.862 & 0.8827 \\
artisinal_mine & 0.995 & 0.6904 \\
bare_ground & 0.969 & 0.3053 \\
blooming & 0.993 & 0.0633 \\
blow_down & 0.998 & 0.0 \\
clear & 0.939 & 0.975 \\
cloudy & 0.976 & 0.8705 \\
conventional_mine & 0.998 & 0.0 \\
cultivation & 0.873 & 0.6543 \\
habitation & 0.901 & 0.5886 \\
haze & 0.951 & 0.781 \\
partly_cloudy & 0.962 & 0.9293 \\
primary & 0.964 & 0.9903 \\
road & 0.86 & 0.7995 \\
selective_logging & 0.991 & 0.1269 \\
slash_burn & 0.993 & 0.0 \\
water & 0.865 & 0.7144 \\
/home/cs231n/myVE35/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.
'precision', 'predicted', average, warn_for)
In [51]:
#Predict on the test set
# Implemenet best threshold selection?
x = []
x_test = []
y = []
df_test = pd.read_csv('/home/joerj/sample_submission_v2.csv')
for f in tqdm(df_test.image_name, miniters=1000):
if('test' in f):
img = cv2.imread('/home/joerj/test-jpg/test-jpg/' + f + '.jpg')
else:
img = cv2.imread('/home/joerj/test-jpg-additional/test-jpg-additional/' + f + '.jpg')
x.append(cv2.resize(img, (32, 32)))
100%|██████████| 61191/61191 [08:17<00:00, 123.00it/s]
In [64]:
x_test = x[:30000] - mean_image
x_test /= 128
x_test = np.array(x_test, np.float16)
#Predict on the test set
p_test = classifier.predict(x_test)
test_pred = np.array(p_test) > 0.2
test_pred = pd.DataFrame(test_pred, columns = labels)
preds = []
for i in tqdm(range(test_pred.shape[0]), miniters=1000):
a = test_pred.ix[[i]]
a = a.transpose()
a = a.loc[a[i] == True]
' '.join(list(a.index))
preds.append(' '.join(list(a.index)))
0%| | 0/30000 [00:00<?, ?it/s]/home/cs231n/myVE35/lib/python3.5/site-packages/ipykernel/__main__.py:11: DeprecationWarning:
.ix is deprecated. Please use
.loc for label based indexing or
.iloc for positional indexing
See the documentation here:
http://pandas.pydata.org/pandas-docs/stable/indexing.html#deprecate_ix
100%|██████████| 30000/30000 [00:31<00:00, 939.88it/s]
In [65]:
x_test = x[30000:] - mean_image
x_test /= 128
x_test = np.array(x_test, np.float16)
#Predict on the test set
p_test = classifier.predict(x_test)
test_pred = np.array(p_test) > 0.2
test_pred = pd.DataFrame(test_pred, columns = labels)
for i in tqdm(range(test_pred.shape[0]), miniters=1000):
a = test_pred.ix[[i]]
a = a.transpose()
a = a.loc[a[i] == True]
' '.join(list(a.index))
preds.append(' '.join(list(a.index)))
0%| | 0/31191 [00:00<?, ?it/s]/home/cs231n/myVE35/lib/python3.5/site-packages/ipykernel/__main__.py:10: DeprecationWarning:
.ix is deprecated. Please use
.loc for label based indexing or
.iloc for positional indexing
See the documentation here:
http://pandas.pydata.org/pandas-docs/stable/indexing.html#deprecate_ix
100%|██████████| 31191/31191 [00:32<00:00, 946.24it/s]
In [66]:
df_test = pd.read_csv('/home/joerj/sample_submission_v2.csv')
df_test['tags'] = preds
df_test.to_csv('submission.csv', index=False)
In [15]:
# Try to improve via Random hyperparameter search
validation_split_size = 5000
num_experiments = 7
best_p_conv = -1
best_batch_size = -1
best_lr = -1
best_batch = -1
best_s = -1
for i in range(num_experiments):
p_conv = np.random.uniform(low = 0.2, high = 0.3)
p_all = np.random.uniform(low = 0.4, high = 0.6)
batch_size = np.random.choice((64, 128))
scale = np.random.uniform(low = 0.5, high = 1.5)
learn_rates = [0.001 * scale, 0.0001 * scale, 0.00001 * scale]
classifier = AmazonClassifier()
classifier.add_conv_layer_init(f = 32, p = p_conv)
classifier.add_conv_layer_mid(f = 64, p = p_conv)
classifier.add_conv_layer_mid(f = 128, p = p_conv)
classifier.add_flatten_layer()
classifier.add_dense_layer(p = p_all)
filepath="weights.best.hdf5"
checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True)
train_losses, val_losses, scores_list = [], [], []
epochs_arr = [5, 10]
for learn_rate in learn_rates:
for epochs in epochs_arr:
tmp_train_losses, tmp_val_losses, score = classifier.train_model(x_train, y_train, learn_rate, epochs,
batch_size, validation_split_size=validation_split_size,
train_callbacks=[checkpoint])
train_losses += tmp_train_losses
val_losses += tmp_val_losses
scores_list.append(score)
s = max(scores_list)
if(s > best_s):
best_p_conv = p_conv
best_p_all = p_all
best_lr_scale = scale
best_batch_size = batch_size
best_s = s
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.3179 - acc: 0.8906Epoch 00000: val_acc improved from -inf to 0.90341, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.3177 - acc: 0.8906 - val_loss: 0.2544 - val_acc: 0.9034
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1585 - acc: 0.9376Epoch 00001: val_acc improved from 0.90341 to 0.91965, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1584 - acc: 0.9376 - val_loss: 0.1938 - val_acc: 0.9196
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1469 - acc: 0.9419Epoch 00002: val_acc improved from 0.91965 to 0.94216, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1469 - acc: 0.9419 - val_loss: 0.1442 - val_acc: 0.9422
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1404 - acc: 0.9444Epoch 00003: val_acc improved from 0.94216 to 0.94801, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1404 - acc: 0.9444 - val_loss: 0.1298 - val_acc: 0.9480
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1346 - acc: 0.9469Epoch 00004: val_acc improved from 0.94801 to 0.95052, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1346 - acc: 0.9469 - val_loss: 0.1250 - val_acc: 0.9505
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1309 - acc: 0.9486Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1310 - acc: 0.9486 - val_loss: 0.1297 - val_acc: 0.9486
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1271 - acc: 0.9499Epoch 00001: val_acc improved from 0.95052 to 0.95146, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1271 - acc: 0.9499 - val_loss: 0.1232 - val_acc: 0.9515
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1237 - acc: 0.9516Epoch 00002: val_acc improved from 0.95146 to 0.95447, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1237 - acc: 0.9516 - val_loss: 0.1156 - val_acc: 0.9545
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1215 - acc: 0.9522Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1215 - acc: 0.9522 - val_loss: 0.1182 - val_acc: 0.9539
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1202 - acc: 0.9530Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1202 - acc: 0.9530 - val_loss: 0.1204 - val_acc: 0.9532
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1184 - acc: 0.9537Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1183 - acc: 0.9537 - val_loss: 0.1193 - val_acc: 0.9544
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1161 - acc: 0.9546Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1161 - acc: 0.9546 - val_loss: 0.1213 - val_acc: 0.9532
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1079 - acc: 0.9577Epoch 00000: val_acc improved from 0.95447 to 0.95764, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1079 - acc: 0.9577 - val_loss: 0.1097 - val_acc: 0.9576
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1060 - acc: 0.9585Epoch 00001: val_acc improved from 0.95764 to 0.95773, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1061 - acc: 0.9585 - val_loss: 0.1096 - val_acc: 0.9577
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1051 - acc: 0.9589Epoch 00002: val_acc improved from 0.95773 to 0.95788, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1051 - acc: 0.9589 - val_loss: 0.1092 - val_acc: 0.9579
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1044 - acc: 0.9593Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1044 - acc: 0.9593 - val_loss: 0.1094 - val_acc: 0.9576
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1031 - acc: 0.9595Epoch 00004: val_acc improved from 0.95788 to 0.95809, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1032 - acc: 0.9595 - val_loss: 0.1084 - val_acc: 0.9581
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1028 - acc: 0.9599Epoch 00000: val_acc improved from 0.95809 to 0.95848, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1028 - acc: 0.9599 - val_loss: 0.1087 - val_acc: 0.9585
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1024 - acc: 0.9599Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1024 - acc: 0.9599 - val_loss: 0.1091 - val_acc: 0.9578
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1013 - acc: 0.9602Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1013 - acc: 0.9602 - val_loss: 0.1095 - val_acc: 0.9576
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1011 - acc: 0.9602Epoch 00003: val_acc improved from 0.95848 to 0.95869, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1011 - acc: 0.9602 - val_loss: 0.1086 - val_acc: 0.9587
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1011 - acc: 0.9603Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1011 - acc: 0.9603 - val_loss: 0.1078 - val_acc: 0.9586
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1006 - acc: 0.9606Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1005 - acc: 0.9606 - val_loss: 0.1089 - val_acc: 0.9584
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1000 - acc: 0.9610Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1000 - acc: 0.9610 - val_loss: 0.1080 - val_acc: 0.9582
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0994 - acc: 0.9613Epoch 00007: val_acc improved from 0.95869 to 0.95885, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.0994 - acc: 0.9613 - val_loss: 0.1086 - val_acc: 0.9588
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0989 - acc: 0.9612Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0990 - acc: 0.9612 - val_loss: 0.1082 - val_acc: 0.9586
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0978 - acc: 0.9615Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.0978 - acc: 0.9615 - val_loss: 0.1081 - val_acc: 0.9588
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0979 - acc: 0.9617Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0979 - acc: 0.9617 - val_loss: 0.1079 - val_acc: 0.9588
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0977 - acc: 0.9615Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0977 - acc: 0.9615 - val_loss: 0.1081 - val_acc: 0.9587
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0978 - acc: 0.9617Epoch 00003: val_acc improved from 0.95885 to 0.95900, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.0977 - acc: 0.9617 - val_loss: 0.1078 - val_acc: 0.9590
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0979 - acc: 0.9615Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0979 - acc: 0.9615 - val_loss: 0.1079 - val_acc: 0.9589
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0976 - acc: 0.9618Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0975 - acc: 0.9618 - val_loss: 0.1079 - val_acc: 0.9588
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0977 - acc: 0.9615Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0976 - acc: 0.9615 - val_loss: 0.1079 - val_acc: 0.9588
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0976 - acc: 0.9618Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0976 - acc: 0.9618 - val_loss: 0.1078 - val_acc: 0.9586
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0976 - acc: 0.9620Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0976 - acc: 0.9620 - val_loss: 0.1080 - val_acc: 0.9588
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0971 - acc: 0.9621Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0971 - acc: 0.9621 - val_loss: 0.1079 - val_acc: 0.9587
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0976 - acc: 0.9615Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0976 - acc: 0.9615 - val_loss: 0.1077 - val_acc: 0.9588
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0971 - acc: 0.9620Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0971 - acc: 0.9621 - val_loss: 0.1078 - val_acc: 0.9588
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0968 - acc: 0.9619Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0968 - acc: 0.9619 - val_loss: 0.1080 - val_acc: 0.9587
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0965 - acc: 0.9624Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0965 - acc: 0.9624 - val_loss: 0.1079 - val_acc: 0.9588
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0971 - acc: 0.9617Epoch 00009: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0971 - acc: 0.9617 - val_loss: 0.1078 - val_acc: 0.9588
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.2413 - acc: 0.9106Epoch 00000: val_acc improved from -inf to 0.91462, saving model to weights.best.hdf5
30000/30000 [==============================] - 17s - loss: 0.2410 - acc: 0.9107 - val_loss: 0.2059 - val_acc: 0.9146
Epoch 2/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1556 - acc: 0.9378Epoch 00001: val_acc improved from 0.91462 to 0.94059, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1555 - acc: 0.9378 - val_loss: 0.1474 - val_acc: 0.9406
Epoch 3/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1462 - acc: 0.9416Epoch 00002: val_acc improved from 0.94059 to 0.94447, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1462 - acc: 0.9416 - val_loss: 0.1395 - val_acc: 0.9445
Epoch 4/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1396 - acc: 0.9443Epoch 00003: val_acc improved from 0.94447 to 0.94827, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1396 - acc: 0.9443 - val_loss: 0.1301 - val_acc: 0.9483
Epoch 5/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1345 - acc: 0.9467Epoch 00004: val_acc improved from 0.94827 to 0.95019, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1345 - acc: 0.9467 - val_loss: 0.1284 - val_acc: 0.9502
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1328 - acc: 0.9476Epoch 00000: val_acc improved from 0.95019 to 0.95213, saving model to weights.best.hdf5
30000/30000 [==============================] - 17s - loss: 0.1328 - acc: 0.9476 - val_loss: 0.1217 - val_acc: 0.9521
Epoch 2/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1290 - acc: 0.9494Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1289 - acc: 0.9495 - val_loss: 0.1239 - val_acc: 0.9516
Epoch 3/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1265 - acc: 0.9507Epoch 00002: val_acc improved from 0.95213 to 0.95261, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1265 - acc: 0.9508 - val_loss: 0.1218 - val_acc: 0.9526
Epoch 4/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1247 - acc: 0.9512Epoch 00003: val_acc improved from 0.95261 to 0.95325, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1246 - acc: 0.9512 - val_loss: 0.1189 - val_acc: 0.9532
Epoch 5/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1220 - acc: 0.9521Epoch 00004: val_acc improved from 0.95325 to 0.95525, saving model to weights.best.hdf5
30000/30000 [==============================] - 17s - loss: 0.1219 - acc: 0.9521 - val_loss: 0.1153 - val_acc: 0.9552
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1220 - acc: 0.9524Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1220 - acc: 0.9524 - val_loss: 0.1192 - val_acc: 0.9531
Epoch 7/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1193 - acc: 0.9534Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1192 - acc: 0.9534 - val_loss: 0.1193 - val_acc: 0.9542
Epoch 8/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1173 - acc: 0.9541Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1173 - acc: 0.9541 - val_loss: 0.1189 - val_acc: 0.9537
Epoch 9/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1174 - acc: 0.9541Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1174 - acc: 0.9541 - val_loss: 0.1206 - val_acc: 0.9515
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1084 - acc: 0.9575Epoch 00000: val_acc improved from 0.95525 to 0.95768, saving model to weights.best.hdf5
30000/30000 [==============================] - 17s - loss: 0.1084 - acc: 0.9575 - val_loss: 0.1101 - val_acc: 0.9577
Epoch 2/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1061 - acc: 0.9584Epoch 00001: val_acc improved from 0.95768 to 0.95818, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1060 - acc: 0.9584 - val_loss: 0.1089 - val_acc: 0.9582
Epoch 3/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1053 - acc: 0.9585Epoch 00002: val_acc improved from 0.95818 to 0.95833, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1053 - acc: 0.9585 - val_loss: 0.1092 - val_acc: 0.9583
Epoch 4/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1045 - acc: 0.9592Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1045 - acc: 0.9592 - val_loss: 0.1086 - val_acc: 0.9580
Epoch 5/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.1033 - acc: 0.9594Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1033 - acc: 0.9594 - val_loss: 0.1088 - val_acc: 0.9582
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1030 - acc: 0.9595Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1030 - acc: 0.9595 - val_loss: 0.1088 - val_acc: 0.9580
Epoch 2/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1024 - acc: 0.9600Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1023 - acc: 0.9600 - val_loss: 0.1090 - val_acc: 0.9580
Epoch 3/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1022 - acc: 0.9599Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1022 - acc: 0.9599 - val_loss: 0.1084 - val_acc: 0.9583
Epoch 4/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1010 - acc: 0.9606Epoch 00003: val_acc improved from 0.95833 to 0.95849, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1010 - acc: 0.9607 - val_loss: 0.1080 - val_acc: 0.9585
Epoch 5/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1013 - acc: 0.9603Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1012 - acc: 0.9603 - val_loss: 0.1090 - val_acc: 0.9584
Epoch 6/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.1007 - acc: 0.9607Epoch 00005: val_acc improved from 0.95849 to 0.95879, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1006 - acc: 0.9608 - val_loss: 0.1087 - val_acc: 0.9588
Epoch 7/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0996 - acc: 0.9610Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0996 - acc: 0.9610 - val_loss: 0.1086 - val_acc: 0.9585
Epoch 8/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0998 - acc: 0.9606Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0998 - acc: 0.9606 - val_loss: 0.1087 - val_acc: 0.9587
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.0988 - acc: 0.9611Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0988 - acc: 0.9611 - val_loss: 0.1084 - val_acc: 0.9587
Epoch 2/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.0984 - acc: 0.9616Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0983 - acc: 0.9616 - val_loss: 0.1085 - val_acc: 0.9587
Epoch 3/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.0983 - acc: 0.9613Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0984 - acc: 0.9613 - val_loss: 0.1084 - val_acc: 0.9587
Epoch 4/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.0982 - acc: 0.9612Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0983 - acc: 0.9612 - val_loss: 0.1083 - val_acc: 0.9588
Epoch 5/5
29888/30000 [============================>.] - ETA: 0s - loss: 0.0979 - acc: 0.9617Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0980 - acc: 0.9617 - val_loss: 0.1083 - val_acc: 0.9587
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0980 - acc: 0.9615Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0980 - acc: 0.9614 - val_loss: 0.1083 - val_acc: 0.9588
Epoch 2/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0982 - acc: 0.9614Epoch 00001: val_acc improved from 0.95879 to 0.95881, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.0982 - acc: 0.9614 - val_loss: 0.1083 - val_acc: 0.9588
Epoch 3/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0979 - acc: 0.9617Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0979 - acc: 0.9617 - val_loss: 0.1084 - val_acc: 0.9588
Epoch 4/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0982 - acc: 0.9615Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0981 - acc: 0.9615 - val_loss: 0.1082 - val_acc: 0.9586
Epoch 5/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0978 - acc: 0.9616Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0978 - acc: 0.9616 - val_loss: 0.1083 - val_acc: 0.9587
Epoch 6/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0979 - acc: 0.9619Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0979 - acc: 0.9619 - val_loss: 0.1084 - val_acc: 0.9586
Epoch 7/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0973 - acc: 0.9615Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0973 - acc: 0.9616 - val_loss: 0.1085 - val_acc: 0.9585
Epoch 8/10
29888/30000 [============================>.] - ETA: 0s - loss: 0.0978 - acc: 0.9618Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0978 - acc: 0.9617 - val_loss: 0.1084 - val_acc: 0.9588
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.2502 - acc: 0.9090Epoch 00000: val_acc improved from -inf to 0.90458, saving model to weights.best.hdf5
30000/30000 [==============================] - 15s - loss: 0.2500 - acc: 0.9091 - val_loss: 0.2535 - val_acc: 0.9046
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1584 - acc: 0.9373Epoch 00001: val_acc improved from 0.90458 to 0.91861, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1584 - acc: 0.9373 - val_loss: 0.1988 - val_acc: 0.9186
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1484 - acc: 0.9409Epoch 00002: val_acc improved from 0.91861 to 0.94138, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1483 - acc: 0.9409 - val_loss: 0.1457 - val_acc: 0.9414
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1421 - acc: 0.9436Epoch 00003: val_acc improved from 0.94138 to 0.94632, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1422 - acc: 0.9436 - val_loss: 0.1365 - val_acc: 0.9463
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1381 - acc: 0.9454Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1381 - acc: 0.9454 - val_loss: 0.1345 - val_acc: 0.9454
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1345 - acc: 0.9472Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.1345 - acc: 0.9472 - val_loss: 0.1351 - val_acc: 0.9453
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1317 - acc: 0.9481Epoch 00001: val_acc improved from 0.94632 to 0.94914, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1316 - acc: 0.9482 - val_loss: 0.1265 - val_acc: 0.9491
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1293 - acc: 0.9491Epoch 00002: val_acc improved from 0.94914 to 0.95041, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1293 - acc: 0.9491 - val_loss: 0.1257 - val_acc: 0.9504
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1276 - acc: 0.9496Epoch 00003: val_acc improved from 0.95041 to 0.95209, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1275 - acc: 0.9496 - val_loss: 0.1206 - val_acc: 0.9521
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1248 - acc: 0.9511Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1248 - acc: 0.9511 - val_loss: 0.1254 - val_acc: 0.9496
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1234 - acc: 0.9516Epoch 00005: val_acc improved from 0.95209 to 0.95375, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1234 - acc: 0.9516 - val_loss: 0.1193 - val_acc: 0.9538
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1223 - acc: 0.9519Epoch 00006: val_acc improved from 0.95375 to 0.95395, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1223 - acc: 0.9519 - val_loss: 0.1179 - val_acc: 0.9540
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1210 - acc: 0.9527Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1211 - acc: 0.9527 - val_loss: 0.1180 - val_acc: 0.9535
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1198 - acc: 0.9531Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1197 - acc: 0.9531 - val_loss: 0.1184 - val_acc: 0.9538
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1179 - acc: 0.9542Epoch 00009: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1179 - acc: 0.9542 - val_loss: 0.1188 - val_acc: 0.9539
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1113 - acc: 0.9566Epoch 00000: val_acc improved from 0.95395 to 0.95679, saving model to weights.best.hdf5
30000/30000 [==============================] - 15s - loss: 0.1113 - acc: 0.9566 - val_loss: 0.1118 - val_acc: 0.9568
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1095 - acc: 0.9572Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1095 - acc: 0.9572 - val_loss: 0.1115 - val_acc: 0.9564
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1082 - acc: 0.9576Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1082 - acc: 0.9576 - val_loss: 0.1124 - val_acc: 0.9562
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1077 - acc: 0.9580Epoch 00003: val_acc improved from 0.95679 to 0.95711, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1077 - acc: 0.9580 - val_loss: 0.1107 - val_acc: 0.9571
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1075 - acc: 0.9579Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1076 - acc: 0.9579 - val_loss: 0.1108 - val_acc: 0.9565
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1069 - acc: 0.9580Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.1069 - acc: 0.9580 - val_loss: 0.1107 - val_acc: 0.9569
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1066 - acc: 0.9582Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1066 - acc: 0.9582 - val_loss: 0.1106 - val_acc: 0.9570
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1061 - acc: 0.9587Epoch 00002: val_acc improved from 0.95711 to 0.95746, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1061 - acc: 0.9586 - val_loss: 0.1103 - val_acc: 0.9575
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1056 - acc: 0.9589Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1057 - acc: 0.9589 - val_loss: 0.1100 - val_acc: 0.9574
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1053 - acc: 0.9585Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1053 - acc: 0.9585 - val_loss: 0.1099 - val_acc: 0.9574
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1050 - acc: 0.9588Epoch 00005: val_acc improved from 0.95746 to 0.95752, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1050 - acc: 0.9588 - val_loss: 0.1096 - val_acc: 0.9575
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1045 - acc: 0.9589Epoch 00006: val_acc improved from 0.95752 to 0.95761, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1045 - acc: 0.9590 - val_loss: 0.1098 - val_acc: 0.9576
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1042 - acc: 0.9592Epoch 00007: val_acc improved from 0.95761 to 0.95798, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1042 - acc: 0.9592 - val_loss: 0.1095 - val_acc: 0.9580
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1041 - acc: 0.9593Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1041 - acc: 0.9593 - val_loss: 0.1098 - val_acc: 0.9579
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1042 - acc: 0.9592Epoch 00009: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1042 - acc: 0.9592 - val_loss: 0.1095 - val_acc: 0.9578
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1024 - acc: 0.9600Epoch 00000: val_acc improved from 0.95798 to 0.95799, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1024 - acc: 0.9599 - val_loss: 0.1090 - val_acc: 0.9580
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1029 - acc: 0.9598Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1028 - acc: 0.9598 - val_loss: 0.1089 - val_acc: 0.9579
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1025 - acc: 0.9600Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1025 - acc: 0.9600 - val_loss: 0.1089 - val_acc: 0.9579
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1022 - acc: 0.9600Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1022 - acc: 0.9600 - val_loss: 0.1091 - val_acc: 0.9578
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1020 - acc: 0.9601Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1020 - acc: 0.9602 - val_loss: 0.1090 - val_acc: 0.9579
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1021 - acc: 0.9600Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.1021 - acc: 0.9600 - val_loss: 0.1090 - val_acc: 0.9580
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1022 - acc: 0.9601Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1022 - acc: 0.9601 - val_loss: 0.1089 - val_acc: 0.9579
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1019 - acc: 0.9599Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1019 - acc: 0.9599 - val_loss: 0.1088 - val_acc: 0.9579
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1018 - acc: 0.9600Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1018 - acc: 0.9600 - val_loss: 0.1090 - val_acc: 0.9578
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1026 - acc: 0.9600Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1025 - acc: 0.9600 - val_loss: 0.1088 - val_acc: 0.9578
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1018 - acc: 0.9601Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1018 - acc: 0.9601 - val_loss: 0.1090 - val_acc: 0.9579
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1018 - acc: 0.9602Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1018 - acc: 0.9602 - val_loss: 0.1091 - val_acc: 0.9578
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.2764 - acc: 0.9024Epoch 00000: val_acc improved from -inf to 0.90418, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.2762 - acc: 0.9025 - val_loss: 0.2512 - val_acc: 0.9042
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1597 - acc: 0.9366Epoch 00001: val_acc improved from 0.90418 to 0.92222, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1597 - acc: 0.9366 - val_loss: 0.1932 - val_acc: 0.9222
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1512 - acc: 0.9398Epoch 00002: val_acc improved from 0.92222 to 0.93955, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1512 - acc: 0.9398 - val_loss: 0.1472 - val_acc: 0.9396
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1439 - acc: 0.9429Epoch 00003: val_acc improved from 0.93955 to 0.94398, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1439 - acc: 0.9429 - val_loss: 0.1410 - val_acc: 0.9440
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1387 - acc: 0.9451Epoch 00004: val_acc improved from 0.94398 to 0.94694, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1388 - acc: 0.9451 - val_loss: 0.1334 - val_acc: 0.9469
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1352 - acc: 0.9470Epoch 00000: val_acc improved from 0.94694 to 0.94720, saving model to weights.best.hdf5
30000/30000 [==============================] - 15s - loss: 0.1352 - acc: 0.9470 - val_loss: 0.1323 - val_acc: 0.9472
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1310 - acc: 0.9485Epoch 00001: val_acc improved from 0.94720 to 0.95193, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1311 - acc: 0.9484 - val_loss: 0.1222 - val_acc: 0.9519
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1277 - acc: 0.9499Epoch 00002: val_acc improved from 0.95193 to 0.95275, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1277 - acc: 0.9499 - val_loss: 0.1213 - val_acc: 0.9528
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1257 - acc: 0.9505Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1257 - acc: 0.9505 - val_loss: 0.1294 - val_acc: 0.9505
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1252 - acc: 0.9512Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1251 - acc: 0.9512 - val_loss: 0.1281 - val_acc: 0.9511
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1219 - acc: 0.9523Epoch 00005: val_acc improved from 0.95275 to 0.95361, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1220 - acc: 0.9523 - val_loss: 0.1191 - val_acc: 0.9536
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1212 - acc: 0.9524Epoch 00006: val_acc improved from 0.95361 to 0.95407, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1213 - acc: 0.9524 - val_loss: 0.1188 - val_acc: 0.9541
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1198 - acc: 0.9529Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1199 - acc: 0.9529 - val_loss: 0.1198 - val_acc: 0.9534
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1199 - acc: 0.9531Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1200 - acc: 0.9530 - val_loss: 0.1210 - val_acc: 0.9537
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1179 - acc: 0.9538Epoch 00009: val_acc improved from 0.95407 to 0.95465, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1180 - acc: 0.9538 - val_loss: 0.1148 - val_acc: 0.9546
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1104 - acc: 0.9568Epoch 00000: val_acc improved from 0.95465 to 0.95669, saving model to weights.best.hdf5
30000/30000 [==============================] - 15s - loss: 0.1104 - acc: 0.9568 - val_loss: 0.1108 - val_acc: 0.9567
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1090 - acc: 0.9572Epoch 00001: val_acc improved from 0.95669 to 0.95693, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1090 - acc: 0.9572 - val_loss: 0.1104 - val_acc: 0.9569
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1078 - acc: 0.9579Epoch 00002: val_acc improved from 0.95693 to 0.95694, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1078 - acc: 0.9579 - val_loss: 0.1102 - val_acc: 0.9569
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1072 - acc: 0.9580Epoch 00003: val_acc improved from 0.95694 to 0.95782, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1073 - acc: 0.9580 - val_loss: 0.1087 - val_acc: 0.9578
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1069 - acc: 0.9582Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1068 - acc: 0.9582 - val_loss: 0.1092 - val_acc: 0.9577
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1067 - acc: 0.9583Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.1067 - acc: 0.9583 - val_loss: 0.1100 - val_acc: 0.9575
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1062 - acc: 0.9584Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1062 - acc: 0.9584 - val_loss: 0.1087 - val_acc: 0.9576
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1052 - acc: 0.9586Epoch 00002: val_acc improved from 0.95782 to 0.95794, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1051 - acc: 0.9586 - val_loss: 0.1088 - val_acc: 0.9579
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1051 - acc: 0.9588Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1051 - acc: 0.9588 - val_loss: 0.1087 - val_acc: 0.9578
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1048 - acc: 0.9589Epoch 00004: val_acc improved from 0.95794 to 0.95799, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1047 - acc: 0.9590 - val_loss: 0.1085 - val_acc: 0.9580
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1042 - acc: 0.9591Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1042 - acc: 0.9591 - val_loss: 0.1088 - val_acc: 0.9577
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1042 - acc: 0.9591Epoch 00006: val_acc improved from 0.95799 to 0.95806, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1042 - acc: 0.9591 - val_loss: 0.1084 - val_acc: 0.9581
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1038 - acc: 0.9594Epoch 00007: val_acc improved from 0.95806 to 0.95816, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1038 - acc: 0.9594 - val_loss: 0.1084 - val_acc: 0.9582
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1030 - acc: 0.9595Epoch 00008: val_acc improved from 0.95816 to 0.95825, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1030 - acc: 0.9595 - val_loss: 0.1086 - val_acc: 0.9582
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1034 - acc: 0.9594Epoch 00009: val_acc improved from 0.95825 to 0.95840, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1035 - acc: 0.9593 - val_loss: 0.1083 - val_acc: 0.9584
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1023 - acc: 0.9600Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.1022 - acc: 0.9600 - val_loss: 0.1079 - val_acc: 0.9581
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1020 - acc: 0.9601Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1020 - acc: 0.9601 - val_loss: 0.1079 - val_acc: 0.9584
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1025 - acc: 0.9600Epoch 00002: val_acc improved from 0.95840 to 0.95844, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1024 - acc: 0.9600 - val_loss: 0.1079 - val_acc: 0.9584
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1024 - acc: 0.9596Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1024 - acc: 0.9596 - val_loss: 0.1078 - val_acc: 0.9583
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1014 - acc: 0.9601Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1014 - acc: 0.9601 - val_loss: 0.1078 - val_acc: 0.9583
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1015 - acc: 0.9602Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.1015 - acc: 0.9602 - val_loss: 0.1079 - val_acc: 0.9582
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1012 - acc: 0.9601Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1013 - acc: 0.9601 - val_loss: 0.1078 - val_acc: 0.9583
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1011 - acc: 0.9601Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1011 - acc: 0.9601 - val_loss: 0.1078 - val_acc: 0.9583
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1012 - acc: 0.9603Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1012 - acc: 0.9603 - val_loss: 0.1077 - val_acc: 0.9584
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1018 - acc: 0.9600Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1018 - acc: 0.9600 - val_loss: 0.1076 - val_acc: 0.9583
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1014 - acc: 0.9599Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1014 - acc: 0.9599 - val_loss: 0.1077 - val_acc: 0.9583
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1011 - acc: 0.9602Epoch 00006: val_acc improved from 0.95844 to 0.95846, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1011 - acc: 0.9602 - val_loss: 0.1075 - val_acc: 0.9585
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1013 - acc: 0.9604Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1012 - acc: 0.9604 - val_loss: 0.1079 - val_acc: 0.9582
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1013 - acc: 0.9601Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1014 - acc: 0.9601 - val_loss: 0.1078 - val_acc: 0.9582
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1012 - acc: 0.9604Epoch 00009: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1012 - acc: 0.9604 - val_loss: 0.1078 - val_acc: 0.9583
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.2547 - acc: 0.9063Epoch 00000: val_acc improved from -inf to 0.90360, saving model to weights.best.hdf5
30000/30000 [==============================] - 17s - loss: 0.2546 - acc: 0.9063 - val_loss: 0.2560 - val_acc: 0.9036
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1579 - acc: 0.9370Epoch 00001: val_acc improved from 0.90360 to 0.92431, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1578 - acc: 0.9370 - val_loss: 0.1938 - val_acc: 0.9243
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1482 - acc: 0.9408Epoch 00002: val_acc improved from 0.92431 to 0.94199, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1483 - acc: 0.9408 - val_loss: 0.1442 - val_acc: 0.9420
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1414 - acc: 0.9436Epoch 00003: val_acc improved from 0.94199 to 0.94384, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1414 - acc: 0.9436 - val_loss: 0.1373 - val_acc: 0.9438
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1361 - acc: 0.9460Epoch 00004: val_acc improved from 0.94384 to 0.94982, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1361 - acc: 0.9460 - val_loss: 0.1297 - val_acc: 0.9498
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1344 - acc: 0.9472Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 14s - loss: 0.1344 - acc: 0.9472 - val_loss: 0.1283 - val_acc: 0.9494
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1309 - acc: 0.9486Epoch 00001: val_acc improved from 0.94982 to 0.95012, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1309 - acc: 0.9486 - val_loss: 0.1256 - val_acc: 0.9501
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1277 - acc: 0.9498Epoch 00002: val_acc improved from 0.95012 to 0.95153, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1276 - acc: 0.9498 - val_loss: 0.1242 - val_acc: 0.9515
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1259 - acc: 0.9508Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1259 - acc: 0.9508 - val_loss: 0.1237 - val_acc: 0.9513
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1227 - acc: 0.9519Epoch 00004: val_acc improved from 0.95153 to 0.95167, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1227 - acc: 0.9519 - val_loss: 0.1283 - val_acc: 0.9517
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1215 - acc: 0.9523Epoch 00005: val_acc improved from 0.95167 to 0.95465, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1215 - acc: 0.9523 - val_loss: 0.1154 - val_acc: 0.9546
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1212 - acc: 0.9522Epoch 00006: val_acc improved from 0.95465 to 0.95482, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1213 - acc: 0.9521 - val_loss: 0.1158 - val_acc: 0.9548
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1193 - acc: 0.9534Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1192 - acc: 0.9534 - val_loss: 0.1199 - val_acc: 0.9534
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1170 - acc: 0.9543Epoch 00008: val_acc improved from 0.95482 to 0.95487, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1170 - acc: 0.9543 - val_loss: 0.1157 - val_acc: 0.9549
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1161 - acc: 0.9543Epoch 00009: val_acc improved from 0.95487 to 0.95504, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1161 - acc: 0.9543 - val_loss: 0.1158 - val_acc: 0.9550
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1086 - acc: 0.9575Epoch 00000: val_acc improved from 0.95504 to 0.95792, saving model to weights.best.hdf5
30000/30000 [==============================] - 15s - loss: 0.1086 - acc: 0.9576 - val_loss: 0.1095 - val_acc: 0.9579
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1069 - acc: 0.9584Epoch 00001: val_acc improved from 0.95792 to 0.95813, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1068 - acc: 0.9584 - val_loss: 0.1083 - val_acc: 0.9581
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1059 - acc: 0.9586Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1059 - acc: 0.9586 - val_loss: 0.1083 - val_acc: 0.9581
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1045 - acc: 0.9589Epoch 00003: val_acc improved from 0.95813 to 0.95841, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1045 - acc: 0.9589 - val_loss: 0.1079 - val_acc: 0.9584
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1048 - acc: 0.9588Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1047 - acc: 0.9588 - val_loss: 0.1083 - val_acc: 0.9580
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1040 - acc: 0.9593Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 15s - loss: 0.1040 - acc: 0.9593 - val_loss: 0.1079 - val_acc: 0.9582
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1035 - acc: 0.9595Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1035 - acc: 0.9595 - val_loss: 0.1075 - val_acc: 0.9582
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1031 - acc: 0.9598Epoch 00002: val_acc improved from 0.95841 to 0.95841, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1031 - acc: 0.9597 - val_loss: 0.1075 - val_acc: 0.9584
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1019 - acc: 0.9601Epoch 00003: val_acc improved from 0.95841 to 0.95842, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1019 - acc: 0.9601 - val_loss: 0.1074 - val_acc: 0.9584
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1020 - acc: 0.9599Epoch 00004: val_acc improved from 0.95842 to 0.95845, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1020 - acc: 0.9599 - val_loss: 0.1076 - val_acc: 0.9584
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1015 - acc: 0.9600Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1015 - acc: 0.9600 - val_loss: 0.1086 - val_acc: 0.9582
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1012 - acc: 0.9602Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1012 - acc: 0.9602 - val_loss: 0.1081 - val_acc: 0.9583
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1011 - acc: 0.9604Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1010 - acc: 0.9604 - val_loss: 0.1084 - val_acc: 0.9580
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1003 - acc: 0.9607Epoch 00000: val_acc improved from 0.95845 to 0.95872, saving model to weights.best.hdf5
30000/30000 [==============================] - 15s - loss: 0.1003 - acc: 0.9607 - val_loss: 0.1073 - val_acc: 0.9587
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0998 - acc: 0.9607Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0998 - acc: 0.9607 - val_loss: 0.1071 - val_acc: 0.9586
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0997 - acc: 0.9608Epoch 00002: val_acc improved from 0.95872 to 0.95878, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.0997 - acc: 0.9608 - val_loss: 0.1072 - val_acc: 0.9588
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0998 - acc: 0.9608Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0998 - acc: 0.9609 - val_loss: 0.1071 - val_acc: 0.9586
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0995 - acc: 0.9609Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0995 - acc: 0.9609 - val_loss: 0.1072 - val_acc: 0.9586
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0992 - acc: 0.9611Epoch 00000: val_acc improved from 0.95878 to 0.95882, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.0992 - acc: 0.9611 - val_loss: 0.1072 - val_acc: 0.9588
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0991 - acc: 0.9611Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0991 - acc: 0.9612 - val_loss: 0.1070 - val_acc: 0.9586
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0995 - acc: 0.9607Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0995 - acc: 0.9608 - val_loss: 0.1071 - val_acc: 0.9586
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0998 - acc: 0.9607Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0997 - acc: 0.9608 - val_loss: 0.1071 - val_acc: 0.9586
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0992 - acc: 0.9612Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0993 - acc: 0.9612 - val_loss: 0.1072 - val_acc: 0.9586
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0991 - acc: 0.9611Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0991 - acc: 0.9612 - val_loss: 0.1071 - val_acc: 0.9587
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.2659 - acc: 0.9018Epoch 00000: val_acc improved from -inf to 0.90325, saving model to weights.best.hdf5
30000/30000 [==============================] - 18s - loss: 0.2658 - acc: 0.9018 - val_loss: 0.2608 - val_acc: 0.9032
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1572 - acc: 0.9379Epoch 00001: val_acc improved from 0.90325 to 0.92476, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1572 - acc: 0.9379 - val_loss: 0.1999 - val_acc: 0.9248
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1479 - acc: 0.9412Epoch 00002: val_acc improved from 0.92476 to 0.94302, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1478 - acc: 0.9412 - val_loss: 0.1441 - val_acc: 0.9430
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1410 - acc: 0.9442Epoch 00003: val_acc improved from 0.94302 to 0.94662, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1410 - acc: 0.9442 - val_loss: 0.1333 - val_acc: 0.9466
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1371 - acc: 0.9459Epoch 00004: val_acc improved from 0.94662 to 0.94706, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1370 - acc: 0.9459 - val_loss: 0.1311 - val_acc: 0.9471
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1339 - acc: 0.9473Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 15s - loss: 0.1340 - acc: 0.9472 - val_loss: 0.1409 - val_acc: 0.9450
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1301 - acc: 0.9491Epoch 00001: val_acc improved from 0.94706 to 0.94952, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1301 - acc: 0.9491 - val_loss: 0.1326 - val_acc: 0.9495
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1279 - acc: 0.9499Epoch 00002: val_acc improved from 0.94952 to 0.95251, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1279 - acc: 0.9499 - val_loss: 0.1208 - val_acc: 0.9525
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1259 - acc: 0.9507Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1259 - acc: 0.9507 - val_loss: 0.1252 - val_acc: 0.9516
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1231 - acc: 0.9518Epoch 00004: val_acc improved from 0.95251 to 0.95494, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1231 - acc: 0.9518 - val_loss: 0.1170 - val_acc: 0.9549
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1219 - acc: 0.9524Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1219 - acc: 0.9524 - val_loss: 0.1190 - val_acc: 0.9532
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1197 - acc: 0.9531Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1197 - acc: 0.9531 - val_loss: 0.1345 - val_acc: 0.9485
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1198 - acc: 0.9534Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1199 - acc: 0.9534 - val_loss: 0.1172 - val_acc: 0.9539
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1180 - acc: 0.9540Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1180 - acc: 0.9541 - val_loss: 0.1184 - val_acc: 0.9540
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1096 - acc: 0.9571Epoch 00000: val_acc improved from 0.95494 to 0.95652, saving model to weights.best.hdf5
30000/30000 [==============================] - 16s - loss: 0.1096 - acc: 0.9571 - val_loss: 0.1104 - val_acc: 0.9565
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1081 - acc: 0.9578Epoch 00001: val_acc improved from 0.95652 to 0.95696, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1081 - acc: 0.9578 - val_loss: 0.1100 - val_acc: 0.9570
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1072 - acc: 0.9580Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1072 - acc: 0.9580 - val_loss: 0.1107 - val_acc: 0.9568
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1064 - acc: 0.9583Epoch 00003: val_acc improved from 0.95696 to 0.95751, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1064 - acc: 0.9584 - val_loss: 0.1094 - val_acc: 0.9575
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1054 - acc: 0.9586Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1056 - acc: 0.9586 - val_loss: 0.1090 - val_acc: 0.9574
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1053 - acc: 0.9589Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1053 - acc: 0.9589 - val_loss: 0.1097 - val_acc: 0.9573
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1053 - acc: 0.9585Epoch 00001: val_acc improved from 0.95751 to 0.95761, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1054 - acc: 0.9585 - val_loss: 0.1086 - val_acc: 0.9576
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1042 - acc: 0.9590Epoch 00002: val_acc improved from 0.95761 to 0.95767, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1043 - acc: 0.9590 - val_loss: 0.1089 - val_acc: 0.9577
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1039 - acc: 0.9595Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1039 - acc: 0.9595 - val_loss: 0.1093 - val_acc: 0.9575
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1034 - acc: 0.9592Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1034 - acc: 0.9592 - val_loss: 0.1097 - val_acc: 0.9575
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1030 - acc: 0.9596Epoch 00005: val_acc improved from 0.95767 to 0.95787, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1030 - acc: 0.9596 - val_loss: 0.1080 - val_acc: 0.9579
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1033 - acc: 0.9594Epoch 00006: val_acc improved from 0.95787 to 0.95796, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1032 - acc: 0.9594 - val_loss: 0.1079 - val_acc: 0.9580
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1019 - acc: 0.9600Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1019 - acc: 0.9600 - val_loss: 0.1082 - val_acc: 0.9579
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1016 - acc: 0.9600Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1016 - acc: 0.9600 - val_loss: 0.1094 - val_acc: 0.9577
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1015 - acc: 0.9601Epoch 00009: val_acc improved from 0.95796 to 0.95824, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1014 - acc: 0.9601 - val_loss: 0.1089 - val_acc: 0.9582
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1003 - acc: 0.9605Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 15s - loss: 0.1003 - acc: 0.9605 - val_loss: 0.1080 - val_acc: 0.9580
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1005 - acc: 0.9608Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1005 - acc: 0.9609 - val_loss: 0.1079 - val_acc: 0.9580
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1006 - acc: 0.9605Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1005 - acc: 0.9605 - val_loss: 0.1077 - val_acc: 0.9580
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0997 - acc: 0.9606Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0997 - acc: 0.9606 - val_loss: 0.1079 - val_acc: 0.9579
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0998 - acc: 0.9609Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0998 - acc: 0.9609 - val_loss: 0.1079 - val_acc: 0.9579
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1002 - acc: 0.9605Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 15s - loss: 0.1002 - acc: 0.9605 - val_loss: 0.1078 - val_acc: 0.9578
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1003 - acc: 0.9604Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1002 - acc: 0.9604 - val_loss: 0.1079 - val_acc: 0.9579
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0996 - acc: 0.9605Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0996 - acc: 0.9605 - val_loss: 0.1077 - val_acc: 0.9579
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1000 - acc: 0.9609Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1000 - acc: 0.9609 - val_loss: 0.1081 - val_acc: 0.9576
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0997 - acc: 0.9608Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0997 - acc: 0.9608 - val_loss: 0.1079 - val_acc: 0.9578
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0997 - acc: 0.9608Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0996 - acc: 0.9608 - val_loss: 0.1079 - val_acc: 0.9579
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0994 - acc: 0.9608Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0994 - acc: 0.9608 - val_loss: 0.1081 - val_acc: 0.9578
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.3347 - acc: 0.8876Epoch 00000: val_acc improved from -inf to 0.90322, saving model to weights.best.hdf5
30000/30000 [==============================] - 19s - loss: 0.3345 - acc: 0.8877 - val_loss: 0.2551 - val_acc: 0.9032
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1607 - acc: 0.9371Epoch 00001: val_acc improved from 0.90322 to 0.91632, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1607 - acc: 0.9371 - val_loss: 0.2033 - val_acc: 0.9163
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1489 - acc: 0.9411Epoch 00002: val_acc improved from 0.91632 to 0.93705, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1489 - acc: 0.9411 - val_loss: 0.1544 - val_acc: 0.9370
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1406 - acc: 0.9446Epoch 00003: val_acc improved from 0.93705 to 0.94322, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1406 - acc: 0.9446 - val_loss: 0.1421 - val_acc: 0.9432
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1356 - acc: 0.9466Epoch 00004: val_acc improved from 0.94322 to 0.94539, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1357 - acc: 0.9466 - val_loss: 0.1377 - val_acc: 0.9454
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1330 - acc: 0.9478Epoch 00000: val_acc improved from 0.94539 to 0.94722, saving model to weights.best.hdf5
30000/30000 [==============================] - 17s - loss: 0.1330 - acc: 0.9478 - val_loss: 0.1326 - val_acc: 0.9472
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1299 - acc: 0.9490Epoch 00001: val_acc improved from 0.94722 to 0.95175, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1298 - acc: 0.9490 - val_loss: 0.1232 - val_acc: 0.9518
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1255 - acc: 0.9506Epoch 00002: val_acc improved from 0.95175 to 0.95271, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1255 - acc: 0.9507 - val_loss: 0.1212 - val_acc: 0.9527
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1231 - acc: 0.9518Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1231 - acc: 0.9518 - val_loss: 0.1240 - val_acc: 0.9516
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1215 - acc: 0.9523Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1216 - acc: 0.9522 - val_loss: 0.1221 - val_acc: 0.9516
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1195 - acc: 0.9530Epoch 00005: val_acc improved from 0.95271 to 0.95418, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1196 - acc: 0.9530 - val_loss: 0.1178 - val_acc: 0.9542
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1185 - acc: 0.9535Epoch 00006: val_acc improved from 0.95418 to 0.95474, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1185 - acc: 0.9535 - val_loss: 0.1177 - val_acc: 0.9547
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1167 - acc: 0.9543Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1167 - acc: 0.9543 - val_loss: 0.1192 - val_acc: 0.9540
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1146 - acc: 0.9552Epoch 00008: val_acc improved from 0.95474 to 0.95613, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1146 - acc: 0.9552 - val_loss: 0.1133 - val_acc: 0.9561
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1135 - acc: 0.9556Epoch 00009: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1135 - acc: 0.9555 - val_loss: 0.1149 - val_acc: 0.9555
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1058 - acc: 0.9584Epoch 00000: val_acc improved from 0.95613 to 0.95738, saving model to weights.best.hdf5
30000/30000 [==============================] - 18s - loss: 0.1059 - acc: 0.9584 - val_loss: 0.1106 - val_acc: 0.9574
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1037 - acc: 0.9593Epoch 00001: val_acc improved from 0.95738 to 0.95741, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1037 - acc: 0.9593 - val_loss: 0.1097 - val_acc: 0.9574
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1028 - acc: 0.9594Epoch 00002: val_acc improved from 0.95741 to 0.95782, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1028 - acc: 0.9594 - val_loss: 0.1098 - val_acc: 0.9578
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1025 - acc: 0.9597Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1026 - acc: 0.9597 - val_loss: 0.1104 - val_acc: 0.9576
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.1019 - acc: 0.9600Epoch 00004: val_acc improved from 0.95782 to 0.95798, saving model to weights.best.hdf5
30000/30000 [==============================] - 13s - loss: 0.1018 - acc: 0.9600 - val_loss: 0.1091 - val_acc: 0.9580
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1016 - acc: 0.9601Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.1016 - acc: 0.9601 - val_loss: 0.1104 - val_acc: 0.9579
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1014 - acc: 0.9604Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1014 - acc: 0.9604 - val_loss: 0.1103 - val_acc: 0.9577
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1009 - acc: 0.9603Epoch 00002: val_acc improved from 0.95798 to 0.95806, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.1009 - acc: 0.9603 - val_loss: 0.1098 - val_acc: 0.9581
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.1005 - acc: 0.9604Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.1005 - acc: 0.9604 - val_loss: 0.1096 - val_acc: 0.9580
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0999 - acc: 0.9607Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0998 - acc: 0.9607 - val_loss: 0.1092 - val_acc: 0.9579
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0994 - acc: 0.9609Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0994 - acc: 0.9609 - val_loss: 0.1100 - val_acc: 0.9576
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0989 - acc: 0.9610Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0989 - acc: 0.9610 - val_loss: 0.1097 - val_acc: 0.9579
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0991 - acc: 0.9609Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0991 - acc: 0.9609 - val_loss: 0.1102 - val_acc: 0.9579
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0984 - acc: 0.9610Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0984 - acc: 0.9611 - val_loss: 0.1095 - val_acc: 0.9580
Train on 30000 samples, validate on 5000 samples
Epoch 1/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0976 - acc: 0.9614Epoch 00000: val_acc improved from 0.95806 to 0.95829, saving model to weights.best.hdf5
30000/30000 [==============================] - 17s - loss: 0.0976 - acc: 0.9614 - val_loss: 0.1090 - val_acc: 0.9583
Epoch 2/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0970 - acc: 0.9619Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0970 - acc: 0.9619 - val_loss: 0.1090 - val_acc: 0.9582
Epoch 3/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0975 - acc: 0.9615Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0975 - acc: 0.9614 - val_loss: 0.1090 - val_acc: 0.9582
Epoch 4/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0973 - acc: 0.9615Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0973 - acc: 0.9615 - val_loss: 0.1090 - val_acc: 0.9582
Epoch 5/5
29952/30000 [============================>.] - ETA: 0s - loss: 0.0973 - acc: 0.9620Epoch 00004: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0973 - acc: 0.9620 - val_loss: 0.1089 - val_acc: 0.9583
Train on 30000 samples, validate on 5000 samples
Epoch 1/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0972 - acc: 0.9614Epoch 00000: val_acc did not improve
30000/30000 [==============================] - 16s - loss: 0.0973 - acc: 0.9613 - val_loss: 0.1091 - val_acc: 0.9583
Epoch 2/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0978 - acc: 0.9618Epoch 00001: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0977 - acc: 0.9618 - val_loss: 0.1089 - val_acc: 0.9580
Epoch 3/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0971 - acc: 0.9618Epoch 00002: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0971 - acc: 0.9618 - val_loss: 0.1089 - val_acc: 0.9581
Epoch 4/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0973 - acc: 0.9616Epoch 00003: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0973 - acc: 0.9616 - val_loss: 0.1091 - val_acc: 0.9581
Epoch 5/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0968 - acc: 0.9619Epoch 00004: val_acc improved from 0.95829 to 0.95831, saving model to weights.best.hdf5
30000/30000 [==============================] - 14s - loss: 0.0969 - acc: 0.9619 - val_loss: 0.1090 - val_acc: 0.9583
Epoch 6/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0970 - acc: 0.9615Epoch 00005: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0970 - acc: 0.9615 - val_loss: 0.1090 - val_acc: 0.9582
Epoch 7/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0964 - acc: 0.9619Epoch 00006: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0963 - acc: 0.9619 - val_loss: 0.1089 - val_acc: 0.9583
Epoch 8/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0970 - acc: 0.9617Epoch 00007: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0971 - acc: 0.9616 - val_loss: 0.1091 - val_acc: 0.9582
Epoch 9/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0964 - acc: 0.9621Epoch 00008: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0964 - acc: 0.9621 - val_loss: 0.1090 - val_acc: 0.9583
Epoch 10/10
29952/30000 [============================>.] - ETA: 0s - loss: 0.0974 - acc: 0.9616Epoch 00009: val_acc did not improve
30000/30000 [==============================] - 13s - loss: 0.0974 - acc: 0.9616 - val_loss: 0.1090 - val_acc: 0.9582
In [17]:
print(best_s)
print(best_p_conv)
print(best_p_all )
print(best_lr_scale )
print(best_batch_size )
0.909633109922
0.29592314649019363
0.5619647177889551
1.134238146597395
128
Content source: Rohisha/classify-satellite-imagery
Similar notebooks: