Mapping locations of similar rows in two matrices

Dear all,

I have two matrices A and B of respective sizes [na,m] and [nb,m], with na>=nb. Both matrices have the same number of columns. All the rows in B are unique while the rows in A can be duplicated, but are all members of B. I would like to create a mapping between A and B and one very inefficient way to do that is as follows:

out=nan(na,1);
for ii=1:na
    % for each row of A, I search its location in B
    out(ii)=find(all(bsxfun(@minus,A(ii,:),B),2)==0);
end

This works well but becomes prohibitively expensive for large na and nb. And so my question is whether there is a way of getting rid of the for loop and/or creating this mapping differently.

Thanks,

P.

Answer 1

[Guillaume]

This is exactly what ismember is for. The 2nd output argument will give you the locations:

[~, out] = ismember(A, B, 'rows');

[Harry]

Good answer. (Showing usage without a loop).

[Patrick Mboma]

It works well but still a bit slow for my purpose, maybe it is possible to further reduce the computation time by writing a specialized function for exactly the purpose of comparing the rows of two matrices?

[Guillaume]

For that specialised function to be any faster than ismember you would have to write it as a mex file.

And if ismember(..., 'rows') is implemented efficiently, it's possible the only saving you could get would be to skip the parsing of the 'rows' argument

[Patrick Mboma]

A million thanks Guillaume

Answer 2

[the cyclist]

Use the ismember function, with the 'rows' option.