CODEWITHSUNDEEP
python-3.xneural-networkdeep-learningkeraskeras-layer
Eric
Eric

Reputation: 1158

Adding new nodes to output layer in Keras

I want to add new nodes to the output layer to train it later, i'm doing:

def add_outputs(self, n_new_outputs):
    out = self.model.get_layer('fc8').output
    last_layer = self.model.get_layer('fc7').output
    out2 = Dense(n_new_outputs, activation='softmax', name='fc9')(last_layer)
    output = merge([out, out2], mode='concat')
    self.model = Model(input=self.model.input, output=output)

where 'fc7'is the fully connected layer before the output layer 'fc8'. I exect to have just the last layer with out = self.model.get_layer('fc8').output but the output is all the model. Is there any way to take just a layer from a network? Maybe theres other easier way to do it....

Thanks!!!!

Upvotes: 5

Views: 1608

Answers (1)

Eric
Eric

Reputation: 1158

Finally i find a solution:

1) get the weights from the last layer

2) add zeros to the weights and random initialize it's connections

3) pop the output layer and create a new one

4) set new weights to the new layer

here the code:

 def add_outputs(self, n_new_outputs):
        #Increment the number of outputs
        self.n_outputs += n_new_outputs
        weights = self.model.get_layer('fc8').get_weights()
        #Adding new weights, weights will be 0 and the connections random
        shape = weights[0].shape[0]
        weights[1] = np.concatenate((weights[1], np.zeros(n_new_outputs)), axis=0)
        weights[0] = np.concatenate((weights[0], -0.0001 * np.random.random_sample((shape, n_new_outputs)) + 0.0001), axis=1)
        #Deleting the old output layer
        self.model.layers.pop()
        last_layer = self.model.get_layer('batchnormalization_1').output
        #New output layer
        out = Dense(self.n_outputs, activation='softmax', name='fc8')(last_layer)
        self.model = Model(input=self.model.input, output=out)
        #set weights to the layer
        self.model.get_layer('fc8').set_weights(weights)
        print(weights[0])

Upvotes: 3

Related Questions