# How can i make comparison faster

1 view (last 30 days)
Mats on 26 Oct 2017
Commented: Andrei Bobrov on 26 Oct 2017
My aim with the loop is to pick out all elements of C that are within range r of rd and range t of td (rd and td are range matrices). Then multiply the elements together and add them to a sum.
This loop is called about a million times and is relatively slow.
Does anyone have an idea of how I can increase the speed of the loop? (I have tried code generation which turned out slower, and parallelism)
for i = 1:indN
D = C(rd(ind(i),:)<r & td(ind(i),:)<t); % Check for <r and <t
if (isempty(D)) D=0; end
gSum = gSum + prod(D);
end
Profiling gives that D=C(.. line takes 84,7% of the entire run time (the other lines in the loop next to nothing)
Jan on 26 Oct 2017
Please provide some test data. Optimizing code depends on the inputs also. E.g. it matters, if there are repetitions in ind, or if C is a [1e7 x 1e2] or [1e2 x 1e7] matrix.

Jan on 26 Oct 2017
index = (rd(ind, :) >= r | td(ind, :) >= t);
D = C;
D(index) = 1;
gSum = sum(prod(D, 2));
Andrei Bobrov on 26 Oct 2017
+1! Nice!
Small fixed:
D = ones(numel(ind),1)*C(:)';

Andrei Bobrov on 26 Oct 2017
Edited: Andrei Bobrov on 26 Oct 2017
[ii,jj] = find(rd(ind,:) < r & td(ind,:) < t);
gSum = sum(accumarray(ii,C(jj),[],@(x)prod(x)));