if config.l2 > 0:
model = apply_kernel_regularizer(model,
tf.keras.regularizers.l2(config.l2))
model.compile(optimizer=opt, loss=custom_loss, metrics=[d_total, cos_sim])
model.summary()
""" DATA """
train_set = make_dataset(config, training=True)
test_set = make_dataset(config, training=False)
""" TRAINING """
callbacks = [
CSVLogger(NAME.replace('.h5', '.log'), append=True),
LearningRateScheduler(custom_scheduler(4096, TOTAL_EPOCH / 12)),
SWA(start_epoch=TOTAL_EPOCH // 2, swa_freq=2),
ModelCheckpoint(NAME,
monitor='val_d_total',
save_best_only=True,
verbose=1),
TerminateOnNaN()
]
model.fit(train_set,
epochs=TOTAL_EPOCH,
batch_size=BATCH_SIZE,
steps_per_epoch=config.steps_per_epoch,
validation_data=test_set,
validation_steps=12,
callbacks=callbacks)
model.save(NAME.replace('.h5', '_SWA.h5'))
mode='max',
save_best_only=True),
TensorBoard(log_dir=f'./logs/{NAME.replace(".h5", "")}',
histogram_freq=0,
profile_batch=2),
TerminateOnNaN()
]
model.fit(train_dataset,
epochs=TOTAL_EPOCH,
batch_size=BATCH_SIZE,
validation_data=(val_x, val_y),
steps_per_epoch=x.shape[0]//BATCH_SIZE,
class_weight=class_weight,
callbacks=callbacks)
result = model.evaluate(test_x, test_y, verbose=1)
with open(NAME.replace('.h5', '.log'), 'a') as f:
f.write(f'\n{result}\n')
# For SWA
repeat = x.shape[0] // BATCH_SIZE
for x, y in train_dataset:
model(x, training=True)
repeat -= 1
if repeat <= 0:
break
model.evaluate(test_x, test_y, verbose=1)
model.save(NAME.replace('.h5', "_SWA.h5"))
