Best way to remove dropout layers for symbolic computation

[Isaac Gerg]

I am working off the conv_filter_visualization.py script and want to turn off my dropout layers.  I suppose I can recreate the network after the weights are loaded, but is there a preferred way?

Thanks,

Isaac

[Isaac Gerg]

I fixed it....

    for k in model.layers:

        if type(k) is keras.layers.Dropout:

            model.layers.remove(k)

[dmytro.p...@ria.com]

This realisation not correct. If you remove droupout layer loop miss one element.

My code:

def delete_layers_from_keras(model):
    i = 0
    while i < len(model.layers):
        k = model.layers[i]

        if type(k) is keras.layers.Dropout:
            model.layers.remove(k)

        elif type(k) is keras.layers.BatchNormalization:
            model.layers.remove(k)
        elif  type(k) is keras.engine.training.Model:
            delete_training_layers_from_keras(k)
            i += 1
        else:
            i += 1
    return model

вторник, 23 февраля 2016 г., 17:50:23 UTC+2 пользователь Isaac Gerg написал: