Question

考虑我有一组向量A，我想将A中的每个向量与张量T相乘，最终得到一个张量y，其中每个切片矩阵是v中A与T之间的向量X相乘的结果：

如果tf.reduce_sum(tf.expand_dims(v,2)*T,1)只包含一个向量，则以下代码有效（感谢matmul function for vector with tensor multiplication in tensorflow中的答案）：

但如果A由几个向量组成，乘法代码怎么样？

例如，我对2（T向量）和A = tf.constant([1,2,3,4], shape=[2,2], dtype=tf.float32) #v1 =[1,2], v2=[3,4] T = tf.constant([1,2,3,4,5,6,7,8], shape=[2,2,2], dtype=tf.float32)具有以下值：

我想通过将T乘以[[[ 7. 10.] [19. 22.]] [[15. 22.] [43. 50.]]]得到以下输出：

y = AX +b

这个问题的应用是线性回归X中的批量梯度下降，其中我有一组向量但不是训练权重矩阵T，我想训练一个张量y，因此输出T将是一个张量，其中张量中的每个矩阵都是输入向量乘以v的输出。

请注意，通常，当我们将维度1*n的向量T与维度m*n*k的张量m*k/m*1*k相乘时，我们希望得到矩阵/张量维度m。这意味着我们的张量具有n*k个具有维度v的矩阵，并且en乘以每个矩阵，并且所得到的向量堆叠在一起。

Answer 1

尝试tf.tensordot：

dpkg: rvm: dependency problems, but removing anyway as you requested:
 ruby-full depends on ruby; however:
  Package ruby is not configured yet.
  Package rvm which provides ruby is to be removed.
 rake depends on ruby | ruby-interpreter; however:
  Package ruby is not configured yet.
  Package rvm which provides ruby is to be removed.
  Package ruby-interpreter is not installed.
 ruby-test-unit depends on ruby | ruby-interpreter; however:
  Package ruby is not configured yet.
  Package rvm which provides ruby is to be removed.
  Package ruby-interpreter is not installed.

(Reading database ... 344123 files and directories currently installed.)
Removing rvm (1.29.3-1) ...
groupdel: cannot remove the primary group of user 'josh'
dpkg: error processing package rvm (--remove):
 subprocess installed post-removal script returned error exit status 8
Errors were encountered while processing:
 rvm
E: Sub-process /usr/bin/dpkg returned an error code (1)

或者，

tf.tensordot(A, T, axes=[-1, -2])

tf.squeeze(tf.tensordot(A[:, None], T, axes=[2, 1]))这会给出

eval

N-D张量矩阵乘以张量流

1 个答案: