Assigning identical array indices at once in Python/Numpy

Question

I want to find a fast way (without for loop) in Python to assign reoccuring indices of an array. This is the desired result using a for loop:

import numpy as np
a=np.arange(9, dtype=np.float64).reshape((3,3))
# The array indices: [2,3,4] are identical.
Px = np.uint64(np.array([0,1,1,1,2]))
Py = np.uint64(np.array([0,0,0,0,0]))
# The array to be added at the array indices (may also contain random numbers).
x = np.array([.1,.1,.1,.1,.1])

for m in np.arange(len(x)):
    a[Px[m]][Py[m]] += x

print a
%[[ 0.1  1.  2.]
%[ 3.3  4.  5.]
%[ 6.1  7.  8.]]

When I try to add x to a at the indices Px,Py I obviously do not get the same result (3.3 vs. 3.1):

a[Px,Py] += x
print a
%[[ 0.1  1.  2.]
%[ 3.1  4.  5.]
%[ 6.1  7.  8.]]

Is there a way to do this with numpy? Thanks.

First of all, I would group the values together, so that you have a list of tuples (Px,Py). Than sort this list, count the occurences, multiply x with that number and add to the array. But somehow numpy seems to skip double entrys....strange. — Dschoni
– Dschoni, Commented Aug 16, 2013 at 13:20

Jaime · Accepted Answer · 2013-08-16 14:59:19Z

7

Yes, it can be done, but it is a little tricky:

# convert yourmulti-dim indices to flat indices
flat_idx = np.ravel_multi_index((Px, Py), dims=a.shape)
# extract the unique indices and their position
unique_idx, idx_idx = np.unique(flat_idx, return_inverse=True)
# Aggregate the repeated indices 
deltas = np.bincount(idx_idx, weights=x)
# Sum them to your array
a.flat[unique_idx] += deltas

edited Aug 16, 2013 at 14:59

answered Aug 16, 2013 at 13:33

Jaime

67.7k19 gold badges128 silver badges164 bronze badges

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

2 Comments

Daniel Over a year ago

Bah, beat me with the exact same answer by 4 seconds. Although, I was sincerely hoping that there was a better way to do this.

Paul Over a year ago

Thanks! I like the way you used the np.ravel_multi_index function to get the flat indices and then use the np.bincount with the optional weights parameter.

Collectives™ on Stack Overflow

Assigning identical array indices at once in Python/Numpy

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related