bsxfun to multiply matrix with vector multiple times in for loop --> how faster?

I have a 2D matrix of size NxN, and each row I am multiplying with a 1xN vector with bsxfun in a for loop like this:

for k=1:1000

   b(k)=mean(bsxfun(@times,tb_filter2d,d(k,:)),2);
  end

Is there a way to speed this up? d is of the size 1000xN.

Thank you!!

[Adam]

Please be a little more precise with the exact sizes of each of your variables. You have b, d and tb_filter2d in that expression. What is N and which of these is your NxN matrix? and if N ~= 1000 how exactly does this work?

[Stephen23]

Is b preallocated before the loop?

[James Tursa]

If tb_filter2d is your NxN matrix, how does this even work since the result of the RHS is a vector and you are assigning it to a scalar element location b(k)?

[Rik]

Is there a reason you can't use something like meshgrid to remove the loop?

Answer 1

[Ankita Nargundkar]

1. If you are on MATLAB R2016b, you can use operators directly instead of bsxfun. Refer to this example

2. You can use repmat or reshape to optimize. Refer to this link

Something on similar lines, https://www.mathworks.com/matlabcentral/answers/210352-optimize-speed-up-a-big-and-slow-matrix-operation-with-addition-and-bsxfun

Answer 2

[Matt J]

This shouldn't require more than a single vector-matrix multiplication,

result =(tb_filter2d/N)*d.';