Consider the numpy 2d array shown:
a = [[8, 16, 4, 1, 0, 5],
[3, 0, 0, 11, 9, 7],
[5, 5, 8, 5, 15, 5],
[2, 0, 2, 14, 2, 0],
[0, 1, 2, 3, 4, 15]]
I want to find all row pairs from the array, placing them into a 3d array. When selecting row pairs, repeat rows are allowed, and the order of the 2 rows is not important.
In the example, there are 15 such 2-row permutations, so I’m hoping to get a 3d array that is 15 layers deep:
[[8, 16, 4, 1, 0, 5],
[8, 16, 4, 1, 0, 5]],
[[8, 16, 4, 1, 0, 5],
[3, 0, 0, 11, 9, 7]],
[[8, 16, 4, 1, 0, 5],
[5, 5, 8, 5, 15, 5]],
...etc
[[2, 0, 2, 14, 2, 0],
[2, 0, 2, 14, 2, 0]],
[[2, 0, 2, 14, 2, 0],
[0, 1, 2, 3, 4, 15]],
[[0, 1, 2, 3, 4, 15],
[0, 1, 2, 3, 4, 15]]]
The actual starting arrays can be quite large, so I’m hoping an efficient solution could be suggested.
1
I think you’ll have to go outside of numpy to get all the pairs, for instance using itertools.combinations_with_replacement from the standard library. Example:
import itertools
import numpy as np
a = [[8, 16, 4, 1, 0, 5],
[3, 0, 0, 11, 9, 7],
[5, 5, 8, 5, 15, 5],
[2, 0, 2, 14, 2, 0],
[0, 1, 2, 3, 4, 15]]
out = np.array(list(itertools.combinations_with_replacement(a, 2)))
You can use itertools.combinations_with_replacement to generate the pairs, then slice:
from itertools import combinations_with_replacement
N = 2
idx = np.array(list(combinations_with_replacement(range(a.shape[0]), r=N)))
out = a[idx]
Output:
array([[[ 8, 16, 4, 1, 0, 5],
[ 8, 16, 4, 1, 0, 5]],
[[ 8, 16, 4, 1, 0, 5],
[ 3, 0, 0, 11, 9, 7]],
[[ 8, 16, 4, 1, 0, 5],
[ 5, 5, 8, 5, 15, 5]],
[[ 8, 16, 4, 1, 0, 5],
[ 2, 0, 2, 14, 2, 0]],
[[ 8, 16, 4, 1, 0, 5],
[ 0, 1, 2, 3, 4, 15]],
[[ 3, 0, 0, 11, 9, 7],
[ 3, 0, 0, 11, 9, 7]],
[[ 3, 0, 0, 11, 9, 7],
[ 5, 5, 8, 5, 15, 5]],
[[ 3, 0, 0, 11, 9, 7],
[ 2, 0, 2, 14, 2, 0]],
[[ 3, 0, 0, 11, 9, 7],
[ 0, 1, 2, 3, 4, 15]],
[[ 5, 5, 8, 5, 15, 5],
[ 5, 5, 8, 5, 15, 5]],
[[ 5, 5, 8, 5, 15, 5],
[ 2, 0, 2, 14, 2, 0]],
[[ 5, 5, 8, 5, 15, 5],
[ 0, 1, 2, 3, 4, 15]],
[[ 2, 0, 2, 14, 2, 0],
[ 2, 0, 2, 14, 2, 0]],
[[ 2, 0, 2, 14, 2, 0],
[ 0, 1, 2, 3, 4, 15]],
[[ 0, 1, 2, 3, 4, 15],
[ 0, 1, 2, 3, 4, 15]]])