Hi, I am trying to vectorize following for loop. Need some help??

    accumulator=zeros(numThetas,numRhos);

    accx_n=zeros(numEdgePixels,numThetas);

    for j=1:numThetas

    accumulator(j,:)=[0 histcounts(accx_n(:,j),rho)];

    end

[Walter Roberson]

We do not know whether rho is a scalar (acting as a bin count) or a vector (acting as edge information)

[Parag Patil]

Its a row vector. Say, rho=[-1.3415:1:1.3415] Thank you.

Answer 1

[Guillaume]

This is a case where the 'old' histogram functions work better than the newer one. If you pass a matrix to histc, it returns the histogram of each column, exactly what you're doing right now with your loop.

You have to watch out that histc and histcounts do not behave exactly the same with regards to the edges (assuming rho is an edge vector), so you may have to modify your rho slightly.

accumulator = [zeros(size(accx_n, 1), 1), histc(accx_n, rho)]

[Parag Patil]

Thanks Guillaume for quick reply.. histc worked. Can u suggest any solution using histcounts. As I am using GPU for calculations and histcounts is overloaded to work on GPU. Thank you.

[Guillaume]

Unfortunately, there's no workaround for histcounts. The best you could do is parallelise the loop with parfor.

edit: saying that you can reproduce your usage of histcounts with discretize and accumarray. As discretize uses the same binning method as histcounts you 'll get exactly the same result:

bins = discretize(accx_n, rho);
rows = repmat(1:size(accx_n, 2), size(accx_n, 1), 1);
accumulator = [zeros(size(accx_n, 2), 1) accumarray([rows(:), bins(:)], 1)]

No guarantee that it is faster than the loop, due to the matrix resizing of rows and bins.

[Parag Patil]

Thanks..histc worked but I will try this one too.