Closest Points between two datasets without using pdist2
1 vue (au cours des 30 derniers jours)
Afficher commentaires plus anciens
Alberto Belvedere
le 24 Oct 2020
Commenté : Alberto Belvedere
le 25 Oct 2020
Hi, i have two matrices A, of size mx2, and B, of size nx2.
Each row of both matrices identifies a point in 2D space. What i want to do is to write a code, that does not involve loops and pdist2, as fast as possible, that tells me the indices of the row of both A and B such that the distance squared of the two points is the minimum one.
Example:
A=[5 6;
1 2;
3 4
1 8];
B=[3 0;
2 1;
4 1;
3 5;
1 2];
My function must be like [indA,indB]=function(A_matrix,B_matrix)
I want as output [2,5]=function(A,B)
I found a solution using for-loops but i really would like to find a solution using repmat that involves vectorization.
Thanks
4 commentaires
Walter Roberson
le 25 Oct 2020
repmat is slower than implicit expansion in many cases.
There are vectorized ways to get indices of the minimum, but they are not necessarily faster than using find() (would have to be tested) and would have problems with ties.
Réponse acceptée
Walter Roberson
le 25 Oct 2020
A=[5 6;
1 2;
3 4
1 8];
B=[3 0;
2 1;
4 1;
3 5;
1 2];
P = permute(sum((A-permute(B,[3 2 1])).^2,2),[1 3 2]) %will be size(A,1) by size(B,1)
A_idx_in_B = sum(cumprod(P ~= min(P,[],1),1),1)+1
B_idx_in_A = sum(cumprod(P ~= min(P,[],2),2),2)+1
This code resolves ties in favor of the first match.
3 commentaires
Walter Roberson
le 25 Oct 2020
The idx_in variables are the indices of the minimum distance. For example the closest entry in B to A(4,:) is the 4th entry of A_idx_in_B which is 3, so A(4,:) is closest to B(3,:)
Plus de réponses (1)
Mitchell Thurston
le 24 Oct 2020
Modifié(e) : Mitchell Thurston
le 24 Oct 2020
Came up with a solution:
[m,~] = size(A);
[n,~] = size(B);
A_rep = repmat(A,n,1);
B_rep = B';
B_rep = repmat(B_rep(:)',m,1);
dist = hypot( A_rep(:,1)-B_rep(:,1:2:end), A_rep(:,2)-B_rep(:,2:2:end) );
ind = find(dist == min(dist));
indB = floor((ind-1)./m)+1
indA = mod(ind-(indB-1)*m,n)
3 commentaires
Mitchell Thurston
le 24 Oct 2020
Not as far as I know, this is the method I've always used for cases like this. What is nice about this though is if there's a tie for the closest it will return all of those indicies of the tie.
Voir également
Catégories
En savoir plus sur Matrix Indexing dans Help Center et File Exchange
Produits
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!