Unable to Get Layer Output if Model has Dropout - Keras 1.0

[brian8128]

In this example https://github.com/julienr/ipynb_playground/blob/master/keras/convmnist/keras_cnn_mnist_v1.ipynb the author is able to build a backend function to get the output of a specific layer of a NN. If I modify the example to include dropout, for instance making the following change to cell 7, it no longer works.

model = Sequential()

model.add(Convolution2D(32, 3, 3, border_mode='valid', input_shape=X_train.shape[1:]))
model.add(Dropout(0.01)) # This breaks it
convout1 = Activation('relu')
model.add(convout1)
model.add(Convolution2D(32, 3, 3))

I get the following error message:

MissingInputError: ("An input of the graph, used to compute DimShuffle{x,x,x,x}(keras_learning_phase), was not provided and not given a value.Use the Theano flag exception_verbosity='high',for more information on this error.", keras_learning_phase)

The error makes sense - the dropout layer needs to know whether it's training time or test time in order to decide whether or not to do dropout - but I don't know how to pass in the learning phase parameter.

[fchollet]

but I don't know how to pass in the learning phase parameter.

You add K.learning_phase() as an input to your function, and when calling the function you pass 1 or 0 as value for the learning phase. This is covered in the FAQ.