Called set_weights(weights) on layer "i2h" with a weight list of length 3418, but the layer was expecting 0 weights

Question

I am trying to implement the word2vec algorithm with Keras but I am getting

ValueError: You called `set_weights(weights)` on layer "i2h" with a  weight list of length 3418, but the layer was expecting 2 weights. Provided weights: [[ 0.07142857  0.07142857  0.07142857 ...,  0.0714...

as I try to set the weights for the shared matrix from the input to the hidden layer i2h:

class Word2Vec:

    def __init__(self, window_size, word_vectors):

        vocab_size = word_vectors.shape[0]
        embedding_size = word_vectors.shape[1]

        i2h = Dense(embedding_size, activation='linear', name='i2h')

        inputs = list()
        h_activations = list()

        for i in range(window_size):

            in_x = Input(shape=(vocab_size, 1), name='in_{:d}'.format(i))
            inputs.append(in_x)
            h_activation = i2h(in_x)
            h_activations.append(h_activation)

        i2h.set_weights(word_vectors)

        h = merge(h_activations, mode='ave')

        h2out = Dense(vocab_size, activation='softmax', name='out')(h)

        self.model = Model(input=inputs, output=[h2out])
        self.model.compile(optimizer='adam', loss='mse')

I don't quite understand how I can set this weight matrix.

I have also tried to use the Dense() layer as input

i2h = Dense(embedding_size, input_dim=vocab_size, activation='linear', name='i2h')
i2h.set_weights(word_vectors)

but I am getting the same error.

How can I set the shared weights in this case?

Answer 1

I have faced a similar problem and found the solution is to add the layer to an existing model first, and then invoke set_weights. So for your example I propose to move the line i2h.set_weights(word_vectors) to be after the line self.model = Model(input=inputs, output=[h2out])