Decreasing the computing time of the for loop

Hello,

I am looking for some help to deal with this loop, and speed up the computing time. Would anyone have some advice as to how I can make it more efficient as I have already tried preallocation so I think it is the if statement that is slowing down the computing time.

A = zeros(1069,30659,16);
idx = reshape(1:16,4,4);
for j = 1:4 
  for k = 1:4 
    for n = 1:30658
      for i = 1+j:(1069-k)
        if ~isnan(logRet(i-j:i-1,n)) 
          if nansum(logRet(i-j:i-1,n)) < 0 
            A(i+k,n,idx(j,k)) = -1*(logRet(i+k,n));
          else
            A(i+k,n,idx(j,k)) = (logRet(i+k,n)); 
          end
        end
      end
    end
  end
end

Another question that I have is say if I wanted to loop from for j =1:12, and for k = 1:12, the end result would provide me with too large of a matrix so how would I be able to achieve this? Would it be to break up the loop?

Any help is appreciated and thanks in advance.

Answer 1

[Walter Roberson]

In your

   if ~isnan(logRet(i-j:i-1,n))

that is the same as

if all(~isnan(logRet(i-j:i-1,n)))

which will only succeed if none of the entries are nan. But if none of the entries are nan, then there is no point in using nansum() on the next line when it would be more efficient to use sum().

Also, look a bit more closely and notice that idx(j,k) is not varying in the "for i" loop, or in the "for n" loop that surrounds that. It would therefore be more efficient to assign idx(j,k) to a variable each time that k changes, and use that variable instead of recalculating idx(j,k) each time.

[Putsandcalls]

Unfortunately, I have a bunch of NaN's segregated everywhere. I guess I will just have to wait it out or break down the tasks.

[Walter Roberson]

When "if ~isnan(logRet(i-j:i-1,n))" is true, none of the entries in logRet(i-j:i-1,n) can possibly be NaN, because the "if" would fail if any of the values in "~isnan(logRet(i-j:i-1,n))" were false, which would occur if any of logRet(i-j:i-1,n) are nan.

So, under the condition that none of logRet(i-j:i-1,n), which is protected against by your "if", then you can use sum() instead of nansum()