How to vectorize a complex function?

Hello,

I have a rather complex function but I need to vectorize it in order to be able to use a matlab optimization package. My function is

g=@(par) nonbcon(par,J,N,X0,X,EZ,H,H_coeff,Coeff,knots,mesh_fine,M,SplineType,w);

You do not need to know the details of this function. g, accepts a row vector "par" and then as output gives either 0 or 1. However, I need to vectorize g so that it accepts a matrix A and performs the same operation on each row of A. I tried to use arrayfun but it is a bit complicated for me. I need something like the following (of course, this is wrong):

G=@(y)arrayfun(@(x) nonbcon(x),y(i,:));

but, I do not know how to modify G to do the job for me.

Any idea?

Thanks in advance!

Babak

[Torsten]

Unclear.

[Jiri Hajek]

Hi, why don't you use a for loop? Especially if your function is a complex one, as you suggest, then the overhead of the loop will be negligible...

for i=1:N
    g(i)=@(parMatrix(i,:)) nonbcon(par,J,N,X0,X,EZ,H,H_coeff,Coeff,knots,mesh_fine,M,SplineType,w);
end

[Torsten]

This will error - a vector of function handles can only be defined as a cell array.

[Mohammad Shojaei Arani]

OK. Thanks!

It seems that my only option is to use for-loop as this is not aparently paralaleisable

Answer 1

[Paul]

Hi Babak,

Here is a one line solution. Whether or not it's more efficient/effective/clearer than the loop is an open question.

Test data for input

rng(101);
test = rand(4,3);
sum(test,2)

ans = 4×1

1.9232 1.5945 0.8897 1.4173

Function that accepts a row vector and returns 0 or 1

g = @(par) double(sum(par,2)>1);

Loop solution

r1 = zeros(4,1);
for ii = 1:4,
    r1(ii) = g(test(ii,:));
end
r1

r1 = 4×1

1 1 0 1

One line solution

r2 = cellfun(g,mat2cell(test,ones(size(test,1),1)))

r2 = 4×1

1 1 0 1