QA_CNN_point_wise.py 文件源码

def attentive_pooling(self,input_left,input_right):
        Q = tf.reshape(input_left,[self.batch_size,self.max_input_left,len(self.filter_sizes) * self.num_filters],name = 'Q')
        A = tf.reshape(input_right,[self.batch_size,self.max_input_right,len(self.filter_sizes) * self.num_filters],name = 'A')

        # G = tf.tanh(tf.matmul(tf.matmul(Q,self.U),\
        # A,transpose_b = True),name = 'G')
        first = tf.matmul(tf.reshape(Q,[-1,len(self.filter_sizes) * self.num_filters]),self.U)
        second_step = tf.reshape(first,[self.batch_size,-1,len(self.filter_sizes) * self.num_filters])
        result = tf.matmul(second_step,tf.transpose(A,perm = [0,2,1]))
        G = tf.tanh(result)
        # column-wise pooling ,row-wise pooling
        row_pooling = tf.reduce_max(G,1,True,name = 'row_pooling')
        col_pooling = tf.reduce_max(G,2,True,name = 'col_pooling')

        attention_q = tf.nn.softmax(col_pooling,1,name = 'attention_q')
        attention_a = tf.nn.softmax(row_pooling,name = 'attention_a')

        R_q = tf.reshape(tf.matmul(Q,attention_q,transpose_a = 1),[self.batch_size,self.num_filters * len(self.filter_sizes),-1],name = 'R_q')
        R_a = tf.reshape(tf.matmul(attention_a,A),[self.batch_size,self.num_filters * len(self.filter_sizes),-1],name = 'R_a')

        return R_q,R_a