Sparse for element by element operation

If we want to perform the product of two sparse matrices or sparse and a full matrix , sparse function is very useful as it avoids multiplications with zero. However when we do element by element multiplication of two sparse matrices or a sparse matrix and a full matrix it is observed that it takes a considerable amount of time as compared to the multiplication of full matrices. Here is a concrete example

    A = sprand(1000,1000,0.005);
    B = rand(1000,1000);
    Af = full(A);
    timeit(@() Af.*B)

ans =

     5.1227e-04

    timeit(@() A.*B)

ans =

      0.0011

How do I tackle this issue when I am looking for a considerable lower time than the multiplication of two full matrices?

Answer 1

[Bruno Luong]

A = sprand(1000,1000,0.005);
B = rand(1000,1000);

tic
Af = full(A);
AB=Af.*B;
toc % Elapsed time is 0.011628 seconds.

tic
AB=A.*B;
toc % Elapsed time is 0.004143 seconds.

Recommended solution

    tic
    [i,j,a]=find(A);
    b=B(sub2ind(size(B),i,j));
    AB=sparse(i,j,a.*b);
    toc % Elapsed time is 0.001570 seconds.

[D_coder]

I dont think this will be very useful for me as I am running the function inside the loop

[Bruno Luong]

Why you think my code assumes some loop or not ? It just carries out the element-wise product A.*B like other methods.

[D_coder]

It's not your code that assumes a loop , my code has a loop that runs for 9000 times and I have already tried doing sparse inside it , it gives me a worse performance than the original thing without sparse

Answer 2

[Bruno Luong]

You might also interested in this FEX

[i,j]=find(A);
b=B(sub2ind(size(B),i,j));
AB=setsparse(A,i,j,b,@times);