Loopin Matrix Multiplication Performance Improvement
1 view (last 30 days)
Show older comments
Tommaso Belluzzo on 3 May 2020
Hi all! I'm struggling to reduce computation time on a function I created:
function [beta,covariance,residuals] = hac_regression(y,x,ratio)
t = length(y);
[beta,~,residuals] = regress(y,x);
h = diag(residuals) * x;
q_hat = (x.' * x) / t;
o_hat = (h.' * h) / t;
l = round(ratio * t,0);
for i = 1:(l - 1)
o_tmp = (h(1:(t-i),:).' * h((1+i):t,:)) / (t - i);
o_hat = o_hat + (((l - i) / l) * (o_tmp + o_tmp.'));
covariance = (q_hat \ o_hat) / q_hat;
Below, a result of a profiling run:
I think nothing can be done with respect to built-in "regress" call.
But I'm wondering if the loop can somehow be optimized in order to reduce the overhead. On computations performed on very large datasets, even a small 5% improvement may dramatically reduce the overall computation time.
Below a reproducible example:
y = rand(100,1);
x = rand(100,3)
[beta,covariance,residuals] = hac_regression(y,x,0.1);
Thanks for your help!
Find more on Mathematics 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!