Day 8 - Saving and Restoring Models

• Models created using the sequential or functional API can be saved using the save() method.

model = keras.layers.Sequential([...])
model.compile(...)
model.fit(...)
model.save("my_model.h5")

• Keras uses the HDF5 format to save:
  • model's architecture (all layers and their hyperparameters)
  • values of all parameters (every layer's weights and biases)
  • the optimizer and its hyperparameters and state

• The model can be loaded as follows:

model = keras.models.load_model("my_model.h5")

• When using model subclassing, this approach can not be used. In this case, the model parameters can be saved and restored using the save_weights() and load_weights() methods.

• When training for long hours, it is a good practice to save model at regular intervals during training instead of just saving at the end of training. This can be done using callbacks.