model = model.model(training=True)
model.summary()
model.compile(optimizer=optimizer,
loss=tf.keras.losses.BinaryCrossentropy(),
metrics=["accuracy"])
# model.load_weights( MODEL_CHECK_POINT_PATH + 'cp{}.ckpt'.format(50))
train_dataset = DataGenerator(TRAIN_SUMMARY_PATH)
test_dataset = DataGenerator(TEST_SUMMARY_PATH)
steps_per_epoch = train_dataset.number_of_batches
with open(LOG_PATH, 'w') as wf:
wf.write('{}\t{}\t{}\t{}\t{}\n'.format('step', 'train_loss',
'train_acc', 'test_loss',
'test_acc'))
for i in range(20, 50):
train_data = train_dataset.data_generator(training=True,
img_augmentation=True)
test_data = test_dataset.data_generator(training=False)
model.optimizer.lr = 0.0001
history = model.fit_generator(train_data,
steps_per_epoch=steps_per_epoch,
epochs=1)
train_loss = (history.history['loss'])[0]
train_acc = (history.history['accuracy'])[0]
test_loss = []
test_acc = []
for x_test, y_test in test_data:
loss, acc = model.evaluate(x_test, y_test, verbose=False)
test_loss.append(loss)
test_acc.append(acc)
# break
test_loss = np.mean(test_loss)
optimizer = model.optimizer()
model = model.model(training=True)
model.compile(optimizer=optimizer,
loss=tf.keras.losses.BinaryCrossentropy(),
metrics=["accuracy"])
# model.load_weights('check_point/cp{}.h5'.format(50))
train_data = DataGenerator(TRAIN_SUMMARY)
steps_per_epoch = train_data.number_of_training_batches
test_data = DataGenerator(TEST_SUMMARY)
with open(LOG_PATH, 'w') as wf:
wf.write('{}\t{}\t{}\t{}\t{}\n'.format('step', 'train_loss',
'train_acc', 'test_loss',
'test_acc'))
for i in range(50):
train = train_data.data_generator(training=True)
test = test_data.data_generator(training=False)
history = model.fit_generator(train,
steps_per_epoch=steps_per_epoch,
epochs=1)
train_loss = (history.history['loss'])[0]
train_acc = (history.history['accuracy'])[0]
test_loss = []
test_acc = []
for x_test, y_test in test:
loss, acc = model.evaluate(x_test, y_test, verbose=False)
test_loss.append(loss)
test_acc.append(acc)
