If I have a set of points:
points = np.random.randint(0, 11, size=10)
print(points)
[ 5 4 9 7 4 1 2 10 4 2]
And if I want to get a matrix representing the distance from each point to each other point, I can do so like this:
def get_diff_array(values):
dX = values - values[0]
tmp = [dX]
for i in range(1, len(dX)):
tmp.append((dX - dX[i]))
return np.array(tmp)
print(get_diff_array(points))
[[ 0 -1 4 2 -1 -4 -3 5 -1 -3]
[ 1 0 5 3 0 -3 -2 6 0 -2]
[-4 -5 0 -2 -5 -8 -7 1 -5 -7]
[-2 -3 2 0 -3 -6 -5 3 -3 -5]
[ 1 0 5 3 0 -3 -2 6 0 -2]
[ 4 3 8 6 3 0 1 9 3 1]
[ 3 2 7 5 2 -1 0 8 2 0]
[-5 -6 -1 -3 -6 -9 -8 0 -6 -8]
[ 1 0 5 3 0 -3 -2 6 0 -2]
[ 3 2 7 5 2 -1 0 8 2 0]]
My question is: is there a faster numpy-specific way to calculate this? Currently it takes ~0.44 sec for 10k points, which seems slow. This is really just a learning question to try to understand numpy better.
>Solution :
You can use values[:, np.newaxis] - values[np.newaxis, :]:
import numpy as np
def diff(A):
return A[:, np.newaxis] - A[np.newaxis, :]
print(diff(np.random.randint(0, 11, size=10)))
Prints
[[ 0 -7 -8 -3 -6 -6 -4 -5 -10 -3]
[ 7 0 -1 4 1 1 3 2 -3 4]
[ 8 1 0 5 2 2 4 3 -2 5]
[ 3 -4 -5 0 -3 -3 -1 -2 -7 0]
[ 6 -1 -2 3 0 0 2 1 -4 3]
[ 6 -1 -2 3 0 0 2 1 -4 3]
[ 4 -3 -4 1 -2 -2 0 -1 -6 1]
[ 5 -2 -3 2 -1 -1 1 0 -5 2]
[ 10 3 2 7 4 4 6 5 0 7]
[ 3 -4 -5 0 -3 -3 -1 -2 -7 0]]