# Speed up fminbnd using vectorization

10 views (last 30 days)
Luca Gagliardone on 12 Aug 2017
Edited: Matt J on 24 May 2018
I am trying to optimize this piece of code. I am using the function fminbnd on a vector, splitting the task on its single entries using a loop.
Would it be possible to speed it up vectorizing the process?
for i = 1:A
for ii= 1:B
for iii = 1:C
fun = @(x) (x * variable(i,ii,iii))^2 ;
[arg_min(i,ii,iii), min_(i,ii,iii)] = fminbnd(fun,0,2);
end
end
end
Thanks for the attention.
Sincerely
Luca

Matt J on 12 Aug 2017
In your example, the solution is always x=0, so a trivial vectorized solution would be
arg_min=zeros(A,B,C);
min_ = arg_min;
More generally, no, vectorization will not help in a situation like this. You could consider parallelizing the loop using PARFOR.

Luca Gagliardone on 16 Aug 2017
Edited: Luca Gagliardone on 16 Aug 2017
What if I create a vector x containing all the possible values for minimization:
x = permute(repmat([0:0.01:2]',[1,A,B,C]),[2,3,4,1]);
variable2 = repmat(variable,[1,1,1,length(x)]);
fun = (x .* variable2).^2;
min_ = min(fun,[],4);
That would do an approximation of the above? Thanks.
Luca

Nick Durkee on 24 May 2018
Edited: Matt J on 24 May 2018
I actually developed a solution to this problem for my research. It's available on the file exchange.