如何在keras子类化模型中使用回调？

Question

I instantiate keras subclassed model, as tensorflow guide ,

To be able to use .fit to my model class, I need to create compute_output_shape . Nevertheless, using callbacks for fitted model throw NotImplementedError .

So, what can I do to use callbacks in keras subclassed model, such as tensorboard, checkpoints, etc ?

Answer 1

You can try this:

model = SubclassModel()
# Callbacks you can define
tensorboard_callback = tf.keras.callbacks.TensorBoard(log_dir='model')
# Add callbacks to fit parameter
model.fit(data, labels, batch_size=100, epochs=5, callbacks=[tensorboard_callback])