ccrc_model.py 文件源码

python
阅读 26 收藏 0 点赞 0 评论 0

项目:Constituent-Centric-Neural-Architecture-for-Reading-Comprehension 作者: shrshore 项目源码 文件源码
def get_candidate_answer_final_representations(self, candidate_answer_hidden_list):
        inputs=tf.expand_dims(candidate_answer_hidden_list,axis=0)
        sequence_length=tf.gather(tf.shape(inputs),1)
        sequence_length=tf.expand_dims(sequence_length, 0)
        #with tf.variable_scope('candidate_answer_generation_forward',reuse=True):
        #    fwcell=rnn.BasicLSTMCell(self.config.hidden_dim, activation=tf.nn.tanh) 
        #with tf.variable_scope('candidate_answer_generation_backward',reuse=True):
        #    bwcell=rnn.BasicLSTMCell(self.config.hidden_dim, activation=tf.nn.tanh)
        chain_outputs, chain_state=tf.nn.bidirectional_dynamic_rnn(self.fwcell, self.bwcell, inputs, 
            sequence_length, initial_state_fw=self._fw_initial_state, initial_state_bw=self._bw_initial_state,scope='candidate_answer_{}'.format(self.scope_index))

        self.scope_index+=1
        chain_outputs=tf.concat(chain_outputs, 2)
        chain_outputs=tf.gather(chain_outputs, 0)
        output=tf.gather(chain_outputs, tf.subtract(tf.gather(tf.shape(chain_outputs),0),1))
        return output #[2*hidden_dim]
评论列表
文章目录


问题


面经


文章

微信
公众号

扫码关注公众号