Vectorize product cell vectors

2 views (last 30 days)
I need to efficiently compute the quantity where and , , are vectors and are sparse matrices.
If the matrices were not sparse, then I could use a 3D array and probably vectorize some of the operations.
Right now, I am using a cell structure to store the matrices and I store the vectors in matrices.
Here's the relevant part of the code that I am using:
res = 1;
for k = 1:N
res = res * (u(:, k).' * A{k} * v(:, k));
end
For reference, in my case and 0, and the loop takes about 2 seconds. In my code, this loop is called several times, so it would make a great impact to reduce its runtime.
How can I improve the performance of the loop? Can it be vectorized? Should I be computing it in another way?
Thank you.

Accepted Answer

Srivardhan Gadila
Srivardhan Gadila on 17 Nov 2021
I think that, you are already representing the matrices of the cell array A in the sparse form and in that case you are already having the best performance. You can refer to the documentation page of Sparse Matrices and the following MATLAB Answer: Is there a faster way to multiply a sparse and full matrix than standard multiplication in Matlab? for any additional information.
Although, I have tried experimenting the code by making use of gpuArray, and it seems that for m = 100,000 and n = 100 it was consuming less time of all, on my machine. At the same time for lower values of m it is slower. Once try executing the below code:
clc; clear all
%%
m = 100000; N = 100;
u = randn(m,N); v = randn(m,N);
ug = gpuArray(u); vg= gpuArray(v);
A = {}; As = {}; Asg = {};
for i = 1:N
As{i} = sprand(m,m,1/m);
A{i} = full(As{i});
Asg{i} = gpuArray(As{i});
end
%% Not using sparse representation
tic
res = 1;
for k = 1:N
res = res * (u(:, k).' * A{k} * v(:, k));
end
toc
%% Using sparse representation
tic
res = 1;
for k = 1:N
res = res * (u(:, k).' * As{k} * v(:, k));
end
toc
%% Using arrayfun
tic
res = arrayfun(@(k)fun(As,u,v,k),1:N);
res = prod(res);
toc
%% Using gpuArray
tic
res = 1;
for k = 1:N
res = res * (ug(:, k).' * Asg{k} * vg(:, k));
end
toc
function out = fun(A,u,v,k)
out = (u(:, k).' * A{k} * v(:, k));
end

More Answers (0)

Categories

Find more on Loops and Conditional Statements in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!