Quick vectorization question (for loop)

Hi

This loop takes about 0.085 seconds to execute and needs to be executed several thousand times. Is there a way you think I can vectorize this? I really appreciate any help!

wsize = 20;
vin = rand(1200,1);

for idx = wsize:length(vin)
    mstd(idx-wsize+1, :) = std(vin(idx-wsize+1:idx));
end

PS: I don't have Parallel Processing toolbox.

Answer 1

[Matt Fig]

I hope you are pre-allocating your mstd vector. Other than that, this is about 6 times faster. I didn't compare to stdfilt.

for idx = wsize:length(vin)
  x = vin(idx-wsize+1:idx);
  x = x-sum(x)./ wsize;
  mstd(idx-wsize+1, :) = sqrt(sum(x.^2)/(wsize-1));
end
.

.

.

EDIT add comparison based on IA's recommendation

I find the above to not be faster than using stdfilt when repeatedly run. Here is what I used for comparison:

mstd2 = stdfilt(vin,[0;ones(wsize,1)]);
mstd2 = mstd2(wsize/2:end-wsize/2);

There might be a better way to do this with stdfilt, I have no real experience with the function and had to figure it out based on trial and error and doc reading... Perhaps IA can comment.

[Tanmay]

yes, I had pre-allocated this vector. Nice trick there! Thanks!

Answer 2

[Image Analyst]

Do you have the Image Processing Toolbox? it looks like your code is a sliding window of a standard deviation filter. This is done by the function stdfilt() in the Image Processing Toolbox. Alternatively (if you don't have the Image Processing Toolbox) you could use an anonymous function and blockproc (I have a demo for that if you need it).

[Matt Fig]

I had never heard of stdfilt, IA. Thanks!