vectorize a loop

Hi there I have the following loop, is there a way to vectorize it?

   for i=1:10000
for j=i:10000
s = zeros(size(P));
[lead,lag] = movavg(P,i,j,'e');
s(lead>lag) = 1;
s(lag>lead) = -1;
r  = [0; s(1:end-1).*diff(P)-abs(diff(s))*cost];
sh(i,j) = scaling*sharpe(r,0);
end
end

[Walter Roberson]

Is there a difference between this function and the one you were previously asking about vectorizing?

Answer 1

[Jan]

When movavg() is the bottleneck, a vectorization will not be remarkably faster. So please use either the profiler and some tic/toc measurements to find out, where the most time is spent.

Of course the repeated calculation of "diff(P)" should be avoided by using a temporary variable created before the loops. So at first I'd start with a cleaned loop:

s = zeros(size(P));
sh = zeros(10000, 10000);  % pre-allocate!!!
diffP = diff(P);
for i=1:10000
  for j=i:10000
    s(:) = 0;   % Faster than zeros()
    [lead,lag] = movavg(P,i,j,'e');
    s(lead>lag) = 1;
    s(lag>lead) = -1;
    r  = [0; s(1:end-1) .* diffP - abs(diff(s))*cost];
    sh(i,j) = scaling*sharpe(r,0);
  end
end

[Andrei Bobrov]

s = sign(lead - lag);