87d57a3d34a1eef2c6ad5519741710e3321f136c,DialogueRNN/model.py,BiE2EModel,forward,#BiE2EModel#,304

Before Change

            for t in emotions:
                att_em, alpha_ = self.matchatt(emotions,t,mask=umask)
                att_emotions.append(att_em.unsqueeze(0))
                alpha.append(alpha_[:,0,:])
            att_emotions = torch.cat(att_emotions,dim=0)
            hidden = F.relu(self.linear(att_emotions))
        else:

After Change

        qmask = qmask.unsqueeze(1).expand(-1, T1.size(1), -1)

        umask = torch.FloatTensor([1,1,1]).type(T1.type())
        umask = umask.expand(-1, T1.size(1))

        emotions_f, alpha_f = self.dialog_rnn_f(U, qmask) // seq_len, batch, D_e
        emotions_f = self.dropout_rec(emotions_f)
        rev_U = self._reverse_seq(U, umask)
        emotions_b, alpha_b = self.dialog_rnn_r(rev_U, qmask)
        emotions_b = self._reverse_seq(emotions_b, umask)
        emotions_b = self.dropout_rec(emotions_b)
        emotions = torch.cat([emotions_f,emotions_b],dim=-1)
        //print(emotions)
        emotions = self.dropout_rec(emotions)

        //emotions = emotions.unsqueeze(1)
        if att2:
            att_emotion, _ = self.matchatt(emotions,emotions[-1])
            hidden = F.relu(self.linear1(att_emotion))
        else:
            hidden = F.relu(self.linear1(emotions[-1]))
        //hidden = F.relu(self.linear2(hidden))

Instances