Extracting values from last dimension of 3D numpy array

Question

I'm trying to extract from a 3D matrix of values a 2d matrix were the last dimension has values from the last dimension of the 3d matrix. For example if P of dimensions [2,2,3] =

[ [[5, 1, 5], [9, 9, 4]], [[0, 9, 8], [8, 6, 8]] ]

what is the index matrix in order to get the out matrix [[1, 9],[0, 8]] where 1 is the 2nd element of the first column of the first row, 9 is the 1st element of the first row second column, 0 is the 1st element of the second row first column and 8 is the 3rd element of the second row second column? The idea is that, for each column I have k different scores. I want to retrieve for each column a different score for which I know the index.

I'm a bit confused with advanced indexing in Numpy and I'm not figuring it out by my self. Thanks!

How are you getting that pattern of 2nd element, 1st element, 1st element & 3rd element? Or is it given as a separate indexing variable? — Divakar
– Divakar, Commented Jun 10, 2018 at 15:36
It can be another matrix that I use to index. The problem is given the matrix P and the matrix out what is the matrix I such that P[I] = out? I understand that I could be actually an expression for indexing or advanced indexing in numpy. What I mean is that I know which index I have to pick from P, what I don't know is how to index P in the correct way to have the right output. — Tommaso Pasini
– Tommaso Pasini, Commented Jun 10, 2018 at 15:39
@TommasoPasini that may not be easy. There a duplicate elements in P. — J...S
– J...S, Commented Jun 10, 2018 at 15:42

Divakar · Accepted Answer · 2018-06-10 15:45:18Z

3

I am assuming there's an indexing array to index into the last axis. Let's call it idx. For the given sample with the given text in the question, it would be -

idx = np.array([[1,0],[0,2]])

Specifically, this was extracted from the quoted text :

1 is the 2nd element of the first column of the first row, 9 is the 1st element of the first row second column, 0 is the 1st element of the second row first column and 8 is the 3rd element of the second row second column

To solve, we will use open grid with np.ogrid to index into the first two axes of the input array -

m,n = idx.shape
I,J = np.ogrid[:m,:n]
out = A[I,J,idx]

Sample run -

In [57]: A
Out[57]: 
array([[[5, 1, 5],
        [9, 9, 4]],

       [[0, 9, 8],
        [8, 6, 8]]])

In [59]: idx = np.array([[1,0],[0,2]])

In [60]: m,n = idx.shape

In [61]: I,J = np.ogrid[:m,:n]

In [62]: A[I,J,idx]
Out[62]: 
array([[1, 9],
       [0, 8]])

answered Jun 10, 2018 at 15:45

Divakar

222k19 gold badges273 silver badges374 bronze badges

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

1 Comment

Tommaso Pasini Over a year ago

perfect! I already tried this way before and I don't know why it didn't work probably I missed something. Thank you very much! Could you explain why as idx I need a 2 dimensional matrix? Thanks

Collectives™ on Stack Overflow

Extracting values from last dimension of 3D numpy array

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related