将权重设置为 model 从另一个带有附加层的

Question

I was trying to update weights in Keras model(let it be model A).

Weights are given from another model (model B- it's an extension of model A, two extra layers at the end) which is in training loop process(train_on_batch), im updating weights using method:

modelA.set_weights(modelB.get_weights())

And surprisingly it worked (the process run), even if network's architectures are different. How's that possible? Is set_weights() automatically cutting of additional part of weights? Or something is wrong and updated weight are mixed up?

Answer 1

It seems the set_weights method loads a list of float regarding of the layers' name.

What you might be looking for is the load_weights method with argument by_name=True ( documentation )