MATLAB Answers

0

Vectorizing evaluation of cell array of functions

Asked by Tristan Potter on 18 Sep 2019
Latest activity Commented on by Bruno Luong
on 19 Sep 2019
I have an S-by-1 cell array of function handles, each taking one argument---call it 'funcell'. I also have an S-by-1 vector of numeric values---call it 'inputs'. My goal is to obtain a numeric vector of size S-by-1---call it 'output', such that each element j of 'output' contains the corresponding function evaluated at the corresponding numeric value.
In a for loop, this would just be:
for j=1:S
output(j)=funcell{j}(inputs(j))
end
The problem is that S is very large, so this is slow. Is there a way to vectorize this process for speed?

  1 Comment

"The problem is that S is very large, so this is slow. Is there a way to vectorize this process for speed?"
This is again a wrong preconceived slowness of FOR-LOOP. It is slow because you evaluate many function handles not because it is embeded in a FOR-LOOP.

Sign in to comment.

1 Answer

Answer by James Tursa
on 18 Sep 2019
Edited by James Tursa
on 18 Sep 2019

Not sure this will be any faster since the loop is just hidden, but you can try this:
output = arrayfun(@(x,y)x{1}(y),funcell,inputs);
Did you pre-allocate output before your loop? I'm assuming your example was a typo and was supposed to read:
output(j)=funcell{j}(inputs(j))

  2 Comments

Thanks for the suggestion---it yields a modest speed-up, but computation is still pretty slow.
And yes, that was a typo. I've edited the original post accordingly.
It yields a modest speed-up
Expected, because the slowness is NOT due to the for-loop.

Sign in to comment.