To use your trained neural network model later, you need to save two things:

1. Model layout; usually in JSON or YAML format.
2. Model weights; usually in HDF5 format.

model_json = model.to_json()
with open('./model.json', 'w') as json_file:
    json_file.write(model_json)

model.save_weights('./model.h5')

You can later load the model with model_from_json.

from keras.models import model_from_json

with open('./model.json', 'r') as json_file:
    loaded_model_json = json_file.read()

loaded_model = model_from_json(loaded_model_json)

loaded_model.load_weights('./model.h5')

# Model is ready for use after it's compiled.
loaded_model.compile(loss='binary_crossentropy', optimizer='rmsprop')

You can also load partial weights with by_name.

model1 = Sequential()
model1.add(Dense(2, input_dim=3, name="dense_1"))
model1.add(Dense(3, name="dense_2"))
model.save_weights('./weights.h5')

model2 = Sequential()
model2.add(Dense(2, input_dim=3, name="dense_1"))
model2.add(Dense(10, name="new_dense"))
model2.load_weights('./weights.h5', by_name=True)

You can use ModelCheckpoint to save model from a specific epoch. This can be advantageous if val_acc is very spiky.

keras.callbacks.ModelCheckpoint(
	filepath,
	monitor='val_loss',
	verbose=0,
	save_best_only=True,
	save_weights_only=False,
	mode='auto',
	period=1
)

Sources

• Save and Load Your Keras Deep Learning Models