Multiply matrix with other matrix of different shapes in keras backend

Question

I'm trying to implement an attention model based this model but I want my model to not just look one frame to decide the attention for that frame, I want a model that will try to look at the frame in respect to the whole sequence. So what I'm doing is multiplying each frame by a sequence vector, which is the output of a lstm (return_sequence=False)

These are the modified functions:

def build(self, input_shape):

    self.W = self.add_weight((input_shape[-1],),
                             initializer=self.init,
                             name='{}_W'.format(self.name))
    if self.lstm_size is None:
        self.lstm_size = input_shape[-1]
    self.vec_lstm = LSTM(self.lstm_size, return_sequences=False)
    self.vec_lstm.build(input_shape)
    self.seq_lstm = LSTM(self.lstm_size, return_sequences=True)
    self.seq_lstm.build(input_shape)
    self.trainable_weights = [self.W]+self.vec_lstm.trainable_weights + self.seq_lstm.trainable_weights
    super(Attention2, self).build(input_shape)  # Be sure to call this somewhere!

def call(self, x, mask=None):
    vec = self.vec_lstm(x)
    seq = self.seq_lstm(x)

#
    eij = # combine seq and vec somehow?
#
    eij = K.dot(eij,self.W)
    eij = K.tanh(eij)
    a = K.exp(eij)
    a /= K.cast(K.sum(a, axis=1, keepdims=True) + K.epsilon(), K.floatx())
    a = K.expand_dims(a)
    weighted_input = x * a
    attention = K.sum(weighted_input, axis=1)
    return attention

The naive code for combining the 2 matrices would be:

eij = np.zeros((batch_size,sequence_length,frame_size))
for i,one_seq in enumerate(seq):    
    for j,timestep in enumerate(one_seq):
        eij[i,j] = timestep*vec[i]  

I would appreciate help in implementing this with keras backend.

Thank you!

Answer 1

This seems to provide the result I want:

vec = vec_lstm(x)        
seq = seq_lstm(x)
repeat_vec = K.repeat(vec,seq.shape[1])
eij = seq * repeat_vec