Keras: How to get output of intermediate layers when the input is multiple ?

[Tristacheng]

My model has 14 inputs, finally concatenated into one, I want to get the output of 'merged' and 'dense1', but when I run my code ,it reports error like below:

This is my model:

    sub_model = Sequential()
    sub_model.add(Convolution3D(number_filter,size_filter,size_filter,size_filter, input_shape=(50,4, 3,3),activation='tanh'))
    sub_model.add(Dropout(0.1))
    sub_model.add(Flatten())
    sub_model.add(Dense(sub_fc,W_regularizer = l1l2(l1 = 0.01,l2 = 0.2),activation='tanh'))
    numIn = en_number
    Inputi = ['']*numIn
    sub_outi = ['']*numIn
    for i in range(numIn):
        Inputi[i] = Input(shape =(50, 4, 3, 3))
        sub_outi[i] = sub_model(Inputi[i])
    merged = merge(sub_outi,mode = 'concat')
    dense1 = Dense(number_fc,W_regularizer = l1l2(l1 = 0.01,l2 = 0.2),activation='tanh')(merged)
    dense1 = Dropout(0.5)(dense1)
    output = Dense(2, activation='softmax')(dense1)
    model = Model(input=Inputi, output=output)
    model.summary()  

and this is my code to extract the feature:

###        define theano funtion to get output of FC layer
        get_feature = K.function([model.layers[0].input,K.learning_phase()],[model.layers[-2].output]) 

        FC_train_feature = get_feature([X_train,1])[0]
        FC_test_feature = get_feature([X_test,0])[0]
        FC_val_feature = get_feature([X_val,0])[0]

The model works well, but the feature cannot be extracted. Can you help me? I am sorry to disturb you.

@fchollet

@bmabey

@jfsantos

[jmhessel]

This is probably better asked in the google group in the future, because it's not a bug.

• You might want to look into the TimeDistributed wrapper instead of doing a giant loop. Stack all your inputs into a single tensor, and then use TimeDistributed.

• To answer your specific question, K.function takes two arguments: a list of inputs and a list of outputs. Your output (the Dropout layer) depends on all of your inputs. So, you need to feed the function all of your inputs.

[Tristacheng]

Thank you for your reply. I'll have a try using Time Distributed wrapper.
But my input length is 14 the same as my model defined, I think I had feed all of inputs, so I don't know why the error occurs.

@jmhessel
Thank you for your help.

[Tristacheng]

This model can be realized by TimeDistributed wrapper in this form.
Thanks for your advices. @jmhessel

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs, but feel free to re-open it if needed.

[alyato]

@Tristacheng @jmhessel
Thanks. how to change the code for image, not text.
My model has two inputs and one outputs.
The backend is tensorflow
the code is

def getFeatureMap(model,layer,X_batch):
featureMap=K.function([model.layers[0].input,model.layers[0].input],[model.layers[layer].output])
gmp = featureMap([X_batch,0])
return gmp

but it show me an error.
I also use the following code.

model_extractfeatures = Model(input=[model.layers[0].input,model.layers[0].input], output=model.layers[-2].output)
everyFm = model_extractfeatures.predict(li[i])

It doesn't work.
how can i rewrite the code? Thanks.