CODEWITHSUNDEEP
arraysnumpyconvolution
jack
jack

Reputation: 923

How to do a 4-d array transpose defined as such in numpy

suppose we have a 4-d matrix of shape (A,B,C,D), f. And we want to make such a transpose where we have f_transpose(i,j,k,c) = f(A-i+1,B-j+1,c,k). This is related to backpropagation of convnet, applying conv operator to get the gradient of conv layer. Can any one help me out?

Upvotes: 0

Views: 70

Answers (1)

user2357112
user2357112

Reputation: 281207

You want to reverse the first two axes and switch the last two. Reversing an axis can be done with a ::-1 slice, and switching axes is numpy.swapaxes:

g = numpy.swapaxes(f, 2, 3)[::-1, ::-1]

Upvotes: 2

Related Questions