tf.keras.layers.pop（）不起作用，但是tf.keras._layers.pop（）没有

Question

I want to pop the last layer of the model. So I use the tf.keras.layers.pop() , but it doesn't work.

base_model.summary()

base_model.layers.pop()

base_model.summary()

When I use tf.keras._layers.pop() , it works.

base_model.summary()

base_model._layers.pop()
base_model.summary()

I don't find docs about this usage. Could someone help explain this?

Answer 1

I agree this is confusing. The reason is that model.layers returns a shallow copy of the layers list so:

The tldr is dont use model.layers.pop() to remove the last layer. Instead we should create a new model with all but the last layer. Perhaps something like this:

new_model = tf.keras.models.Sequential(base_model.layers[:-1])

Checkout this github issue for more details

Answer 2

@Stewart_R clearly shows the solution to the problem :)

Let me just put a simple code with the solution.

loaded_model = keras.models.load_model(fname)

    # remove the last 2 layers
    sliced_loaded_model = Sequential(loaded_model.layers[:-2])

    # set trainable=Fasle for the layers from loaded_model
    for layer in sliced_loaded_model.layers:
        layer.trainable = False

    # add new layers
    sliced_loaded_model.add(Dense(32, activation='relu'))  # trainable=True is default
    sliced_loaded_model.add(Dense(1))

    # compile
    sliced_loaded_model.compile(loss='mse', optimizer='adam', metrics=[])

    # fit
    ...