attention_tf.py

#! -*- coding: utf-8 -*-

import tensorflow as tf

tf.enable_eager_execution()

'''
inputs是一个形如(batch_size, seq_len, word_size)的张量；
函数返回一个形如(batch_size, seq_len, position_size)的位置张量。
'''
def Position_Embedding(inputs, position_size):
    batch_size, seq_len = tf.shape(inputs)[0], tf.shape(inputs)[1]
    # 分母10000^(2i/postion_size)
    position_j = 1. / tf.pow(10000.,
                             2 * tf.range(position_size / 2, dtype=tf.float32) / position_size)  # shape(p_size/2,)
    position_j = tf.expand_dims(position_j, 0)  # shape(1,p_size/2)
    # 位置编码--分子p
    position_i = tf.range(tf.cast(seq_len, tf.float32), dtype=tf.float32)  # shape(seq_len,)
    position_i = tf.expand_dims(position_i, 1)  # shape(seq_len,1)
    # 合起来
    position_ij = tf.matmul(position_i, position_j)  # shape(seq_len,p_size/2)
    position_ij = tf.concat([tf.cos(position_ij), tf.sin(position_ij)], 1)  # shape(seq_len,p_size)

    position_embedding = tf.expand_dims(position_ij, 0) #shape(1,seq_len,p_size)
    position_embedding += tf.zeros((batch_size, seq_len, position_size))    #shape(batch_size,seq_len,p_size)

    return position_embedding

'''
inputs是一个二阶以上的张量，代表输入序列，比如形如(batch_size, seq_len, input_size)的张量；
seq_len是一个形如(batch_size,)的张量，代表每个序列的实际长度，多出部分都被忽略；
mode分为mul和add，mul是指把多出部分全部置零，一般用于全连接层之前；
add是指把多出部分全部减去一个大的常数，一般用于softmax之前。
'''
def Mask(inputs, seq_len, mode='mul'):
    if seq_len == None: #double check
        return inputs
    else:
        mask = tf.cast(tf.sequence_mask(seq_len), tf.float32)
        for _ in range(len(inputs.shape) - 2):
            mask = tf.expand_dims(mask, 2)
        if mode == 'mul':
            return inputs * mask
        if mode == 'add':
            return inputs - (1 - mask) * 1e12


'''
普通的全连接
inputs是一个二阶或二阶以上的张量，即形如(batch_size,...,input_size)。
只对最后一个维度做矩阵乘法，即输出一个形如(batch_size,...,ouput_size)的张量。
'''
def Dense(inputs, ouput_size, bias=True, seq_len=None):
    input_size = int(inputs.shape[-1])  #只操作最后一个
    W = tf.Variable(tf.random_uniform([input_size, ouput_size], -0.05, 0.05))   #shape(in_size,out_size),range(-0.05,0.05)
    if bias:
        b = tf.Variable(tf.random_uniform([ouput_size], -0.05, 0.05))   #shape(out_size),range(-0.05,0.05)
    else:
        b = 0
    outputs = tf.matmul(tf.reshape(inputs, (-1, input_size)), W) + b    #inputs.shape(*,in_size) * W + b
    outputs = tf.reshape(outputs, \
                         tf.concat([tf.shape(inputs)[:-1], [ouput_size]], 0)
                         )  #ensure that last dimension is equal to out_size
    if seq_len != None:
        outputs = Mask(outputs, seq_len, 'mul')
    return outputs


'''
Multi-Head Attention的实现
'''
def Attention(Q, K, V, nb_head, size_per_head, Q_len=None, V_len=None):
    # 对Q、K、V分别作线性映射
    Q = Dense(Q, nb_head * size_per_head, False)
    Q = tf.reshape(Q, (-1, tf.shape(Q)[1], nb_head, size_per_head))
    Q = tf.transpose(Q, [0, 2, 1, 3])
    K = Dense(K, nb_head * size_per_head, False)
    K = tf.reshape(K, (-1, tf.shape(K)[1], nb_head, size_per_head))
    K = tf.transpose(K, [0, 2, 1, 3])
    V = Dense(V, nb_head * size_per_head, False)
    V = tf.reshape(V, (-1, tf.shape(V)[1], nb_head, size_per_head))
    V = tf.transpose(V, [0, 2, 1, 3])   #shape(-1,nb_head,tf.shape(V)[1],size_per_head)

    # 计算内积，然后mask，然后softmax
    A = tf.matmul(Q, K, transpose_b=True) / tf.sqrt(float(size_per_head))
    A = tf.transpose(A, [0, 3, 2, 1])
    A = Mask(A, V_len, mode='add')
    A = tf.transpose(A, [0, 3, 2, 1])
    A = tf.nn.softmax(A)
    # 输出并mask
    O = tf.matmul(A, V)
    O = tf.transpose(O, [0, 2, 1, 3])
    O = tf.reshape(O, (-1, tf.shape(O)[1], nb_head * size_per_head))
    O = Mask(O, Q_len, 'mul')
    return O