vectorization of nested loop

Hello,

I was wondering whether it is possible to make this following nested loop into a vectorization process to make things faster and more efficient.

Y= time series with one column with 10000 observations
[M N]=size(Y);
for INCREMENT=1:100
for WINDOW=1000:INCREMENT:M
for W=WINDOW:INCREMENT:M
BB{INCREMENT,WINDOW-999}(:,W-999)=Y(W-WINDOW+1:W);
end
end
end

Thank you for your time

[Daniel Shub]

While you might be able to make those loops go faster with or without vectorization (try preallocating), you might want to think about how you are using BB. and why you want to hog so much memory by storing Y is a really inaccessible manner.

Answer 1

[Jan]

The code is not valid:

BB{INCREMENT,WINDOW-999}(W-999) = Y(W-WINDOW+1:W)

The lefthand side is a scalar, while the right hand side is a vector when WINDOW differs from 1.

The code will suffer from the missing pre-allocation. The Cell array BB and its contents will grow in each iteration, which will slow down the programm dramatically.

So currently the answer is: No. I cannot vectorize a crashing code.

[Fabricio]

Sorry about the mistake, there's (:,W-999) rather than (W-999) Y= time series with one column with 10000 observations [M N]=size(Y); for INCREMENT=1:100 for WINDOW=1000:INCREMENT:M for W=WINDOW:INCREMENT:M BB{INCREMENT,WINDOW-999}(:,W-999)=Y(W-WINDOW+1:W); end end end

Do you think it's doable?