Correctly indexing a multidimensional Numpy array with another array of indices

Question

I'm trying to index a multidimensional array P with another array indices. which specifies which element along the last axis I want, as follows:

import numpy as np

M, N = 20, 10

P = np.random.rand(M,N,2,9)

# index into the last dimension of P
indices = np.random.randint(0,9,size=(M,N))

# I'm after an array of shape (20,10,2)
# but this has shape (20, 10, 2, 20, 10)
P[...,indices].shape

How can I correctly index P with indices to get an array of shape (20,10,2)?

If that's not too clear: For any i and j (in bounds) I want my_output[i,j,:] to be equal to P[i,j,:,indices[i,j]]

I am probably being slow, but how can you use a 2D array to index along one axis ("the last axis" in your question)? Surely, indexing along a single axis requires one coordinate and not two? — NPE
– NPE, Commented Mar 27, 2013 at 14:09
I didn't explain it very well, which may well be related to why I am stuck here. For any i and j I want my_output[i,j,:] to be equal to P[i,j,:,indices[i,j]] — YXD
– YXD, Commented Mar 27, 2013 at 14:13

Jaime · Accepted Answer · 2013-03-27 15:59:22Z

2

I think this will work:

P[np.arange(M)[:, None, None], np.arange(N)[:, None], np.arange(2),
  indices[..., None]]

Not pretty, I know...

This may look nicer, but it may also be less legible:

P[np.ogrid[0:M, 0:N, 0:2]+[indices[..., None]]]

or perhaps better:

idx_tuple = tuple(np.ogrid[:M, :N, :2]) + (indices[..., None],)
P[idx_tuple]

edited Mar 27, 2013 at 15:59

answered Mar 27, 2013 at 14:19

Jaime

67.7k19 gold badges128 silver badges164 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Correctly indexing a multidimensional Numpy array with another array of indices

1 Answer 1

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related