tensorflow 实现 前一个tensor 的最后一维 矩阵相乘 后一个tensor 的第一维

import tensorflow as tf

def multiply_tensors(tensor1, tensor2):
"""Multiplies two tensors in a matrix-like multiplication based on the
last dimension of the first tensor and first dimension of the second
tensor.
Inputs:
tensor1: A tensor of shape [a, b, c, .., x]
tensor2: A tensor of shape [x, d, e, f, ...]
Outputs:
A tensor of shape [a, b, c, ..., d, e, f, ...]
"""
sh1 = tf.shape(tensor1)
sh2 = tf.shape(tensor2)
len_sh1 = len(tensor1.get_shape())
len_sh2 = len(tensor2.get_shape())
prod1 = tf.constant(1, dtype=tf.int32)
sh1_list = []
for z in range(len_sh1 - 1):
sh1_z = sh1[z]
prod1 *= sh1_z
sh1_list.append(sh1_z)
prod2 = tf.constant(1, dtype=tf.int32)
sh2_list = []
for z in range(len_sh2 - 1):
sh2_z = sh2[len_sh2 - 1 - z]
prod2 *= sh2_z
sh2_list.append(sh2_z)
reshape_1 = tf.reshape(tensor1, [prod1, sh1[len_sh1 - 1]])
reshape_2 = tf.reshape(tensor2, [sh2[0], prod2])
result = tf.reshape(tf.matmul(reshape_1, reshape_2), sh1_list + sh2_list)
assert len(result.get_shape()) == len_sh1 + len_sh2 - 2
return result