张量流中 tf.transpose 的时间复杂度是多少？

Question

For example, assume we have an tensor a with input shape (10,32,32,3) Which means 10 32*32 RGB pictures with NHWC format

and I want to use tf.transpose(b, perm=[0, 3, 1, 2]) to convert it into NCHW format

What is the time complexity of this operation?

---------FORWARD---------

I found that the time complexity of numpy transpose function is O(1), according to the following link

https://www.thetopsites.net/article/58279082.shtml

Is that the same in tf?

Answer 1

Matrix transposition for a 2D matrix is O(dim1*dim2) .

For a 3D matrix it would be O(dim1*dim2*dim3) and so on.

Answer 2

According to the source :

@compatibility(numpy)
  In `numpy` transposes are memory-efficient constant time operations as they
  simply return a new view of the same data with adjusted `strides`.
  TensorFlow does not support strides, so `transpose` returns a new tensor with
  the items permuted.
@end_compatibility

That means that the complexity depends on the dimensions of your tensor. Assuming that you have k dimensions of n -size each, it would mean O(n^k) .