Keras saves the best model in the training process

With xDeepFM For example

Save model parameters

from deepctr.models import xDeepFM
model = xDeepFM(...)
model.compile(...)
model.fit(...)
# save_weights
model.save_weights('... .h5')

Read h5 Model

model = xDeepFM(...)
# load_weights
model.load_weights('... .h5')

because h5 The file saves the model parameters , Therefore, the model structure needs to be built manually , If you need to save the model structure , Sure Use json File save model structure .

Save the optimal model weight

The above method saves all epoch The last result after training , But not necessarily the optimal value , So you can use checkpoint Save the optimal weight .

model = xDeepFM(...)
model.compile(...)
# checkpoint
checkpoint = tf.keras.callbacks.ModelCheckpoint(filepath='... .h5', monitor='val_loss', verbose=1, save_best_only=True, mode = 'min')
callback_list = [checkpoint]
model.fit(...,callbacks = callback_list )

ModelCheckpoint The parameters of the ,filepath I.e. preserved h5 file ,monitor It is the indicator of monitoring , It's usually val_loss,val_acc etc. ,mode Specify whether to save the maximum value or the minimum value , For specific explanation, please refer to Official documents .