Is it possible to vectorize this ?

A matrix A(100x1000) a selective column defined in B(1x1000). The required processing is the mean of non zero elements of each column of A referenced by B. Without the use of loops.

Regards

Answer 1

[Matt J]

A(~A)=nan;
result=mean(A(:,B),1,'omitnan')

[Omar Ali Muhammed]

Dear, it is wonderful.

How can we 'omitnan' if we replace mean by trimmean?

[Matt J]

Dear, it is wonderful.

I'm very glad. Please Accept-click the answer to indicate so.

How can we 'omitnan' if we replace mean by trimmean?

What would you be excluding?Zeros again? If so,

A(~A)=inf;

Answer 2

[Jan]

AB     = A(:, B);
result = sum(AB, 1) ./ sum(AB ~= 0, 1);

mean('omitnan') replaces the NaNs by zeros for the summation and calculates the number of non-NaNs by sum(~isnan(AB)). Therefore this code should be faster, because it avoids replacing zeros by NaNs and back to zeros again.

[Matt J]

because it avoids replacing zeros by NaNs and back to zeros again.

+1. Although, I would argue, they probably should have been NaNs instead of zeros from the very beginning.