Most efficient way to calculate every L2 distance between vectors of vector array A and vectors of vector array B?

Question

I need to implement an algorithm. But it takes a lot of time to compute and I need to make it as fast as possible.

Right now I have two numpy arrays:

Array A -> 2000 vectors of 512 elements,

Array B -> 1000 vectors of 512 elements.

I need to calculate every single distance between the vectors from Array A and Array B. Right now, I take 1 vector from array A, and calculate it's distances to all vectors in Array B as follows:

np.sum(np.abs(B-A[0])**2,axis=-1)**(0.5)

But using this I have to loop for 2000 cycles and it takes a lot of time.

Any alternatives?

@Mad Physicist, this question is different because it's about all pairwise distances, not just the distance between two vectors. — orlp, Jun 30 '21 at 19:38
@orlp. Does `cdist` return what you are looking for? Are you not asking for a 1000x2000 matrix of distances? — Mad Physicist, Jun 30 '21 at 19:38
I only used what @orlp proposed and it provided what I needed. — rocketsfallonrocketfalls, Jun 30 '21 at 20:54

score 2 · Accepted Answer · answered Jun 30 '21 at 19:02

2

sklearn.metrics.pairwise_distances solves exactly this problem.

answered Jun 30 '21 at 19:02

orlp

112,504
36
218
315

Most efficient way to calculate every L2 distance between vectors of vector array A and vectors of vector array B?

1 Answers1