Summing of non-zero elements of columns using indexing

I have idx, is there a way of using idx to make the sum(matrix,1) in the below code cheaper. In another words, is it possible to use idx for calculating sum of 'matrix'(see below code) along the columns to get a row vector.

 matrix(matrix<thresh) = 0 ;
 idx = find(matrix~=0);
    while condition is true
        matrix(idx) = matrix(idx).*another matrix(idx) %matrix is sparse and has lot of zeros ;numel(matrix) = 1000000
        new_matrix(a,:) = b*sum(matrix,1) %here b is an integer, make this line efficient and faster so it takes only non zero elements for summing
    end

Answer 1

[Walter Roberson]

Yes, it is possible to use idx to reduce the number of additions.

matrix(matrix<thresh) = 0 ;
idx = find(matrix~=0);
[idxr, idxc] = ind2sub(size(matrix), idx);
numcol = size(matrix,2);
while condition is true
   matrix(idx) = matrix(idx).*another matrix(idx) %matrix is sparse and has lot of zeros ;numel(matrix) = 1000000
   for col = 1 : num2col
     matched_rows = idxr(idxc==col);
     new_matrix(a,col) = b * sum(matrix(matched_rows, col));
   end
end

I would be surprised if this is more efficient, but it is possible . It would probably be more efficient to loop calculating the column-wise indices into a cell array and indexing into that.

[D_coder]

do you have any alternative code without use of for loop coz the while loop runs 9000 times so thats gonna consume a lot of time.accumarray()?

Answer 2

[Akira Agata]

If your matrix contains many zeros, I would recommend converting it to sparse matrix and applying sum function, like the following. It runs faster.

matrix = sparse(matrix); % Convert to sparse matrix
summedValue = sum(matrix,1);

[D_coder]

no it doesnt , inside a loop , conversion to sparse takes a huge amount of time

[Walter Roberson]

But you would convert to sparse outside the loop.

[D_coder]

No then my element by element multiplication would suffer