How to merge two LSTM layers in Keras

Question

I’m working with Keras on a sentence similarity task (using the STS dataset) and am having problems merging the layers. The data consists of 1184 sentence pairs each scored between 0 and 5. Below are the shapes of my numpy arrays. I’ve padded each of the sentences to 50 words and run them through and embedding layer, using the glove embedding’s with 100 dimensions. When merging the two networks I'm getting an error..

Exception: Error when checking model input: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 1 arrays but instead got the following list of 2 arrays:

Here is what my code looks like

total training data = 1184
X1.shape = (1184, 50)
X2.shape = (1184, 50)
Y.shape = (1184, 1)


embedding_matrix = np.zeros((len(word_index) + 1, EMBEDDING_DIM))
for word, i in word_index.items():
    embedding_vector = embeddings_index.get(word)
    if embedding_vector is not None:
        # words not found in embedding index will be all-zeros.
        embedding_matrix[i] = embedding_vector

embedding_layer = Embedding(len(word_index) + 1,
                            EMBEDDING_DIM,
                            weights=[embedding_matrix],
                            input_length=50,
                            trainable=False)

s1rnn = Sequential()
s1rnn.add(embedding_layer)
s1rnn.add(LSTM(128, input_shape=(100, 1)))
s1rnn.add(Dense(1))

s2rnn = Sequential()
s2rnn.add(embedding_layer)
s2rnn.add(LSTM(128, input_shape=(100, 1)))
s2rnn.add(Dense(1))

model = Sequential()
model.add(Merge([s1rnn,s2rnn],mode='concat'))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='RMSprop', metrics=['accuracy'])
model.fit([X1,X2], Y,batch_size=32, nb_epoch=100, validation_split=0.05)

Answer 1

The problem is not with the merge layer. You need to create two embedding layers to feed in 2 different inputs.

The following modifications should work:

embedding_layer_1 = Embedding(len(word_index) + 1,
                            EMBEDDING_DIM,
                            weights=[embedding_matrix],
                            input_length=50,
                            trainable=False)

embedding_layer_2 = Embedding(len(word_index) + 1,
                            EMBEDDING_DIM,
                            weights=[embedding_matrix],
                            input_length=50,
                            trainable=False)


s1rnn = Sequential()
s1rnn.add(embedding_layer_1)
s1rnn.add(LSTM(128, input_shape=(100, 1)))
s1rnn.add(Dense(1))

s2rnn = Sequential()
s2rnn.add(embedding_layer_2)
s2rnn.add(LSTM(128, input_shape=(100, 1)))
s2rnn.add(Dense(1))