rnn.py 文件源码

def _step(self, m, y, h_, Ur):
        '''Step function for RNN call.

        Args:
            m (T.tensor): masks.
            y (T.tensor): inputs.
            h_ (T.tensor): recurrent state.
            Ur (theano.shared): recurrent connection.

        Returns:
            T.tensor: next recurrent state.

        '''
        preact = T.dot(h_, Ur) + y
        h      = T.tanh(preact)
        h      = m * h + (1 - m) * h_
        return h