Shortening/reducing computing time of for with while loop

Hello, I have some code set up but it takes a bit too long for my liking to run. May I please have some ideas or suggestions on how to optimize it? "some_function" takes up the most time in this code. Thank you.

    subjectlist = importdata('subjectlist.txt')  
    files = dir('*.txt');
    for i = 1:length(subjectlist);
      temp = load(files(i).name);
      corrtemp = corrcoef(temp);
      corrtemp(logical(eye(size(corrtemp)))) = 0;
      corrtemp(corrtemp < 0) = 0;
      counter = 1;
      uniqueedges = (nodes^2)-nodes/2;
          for j = .1:.01:.4
          loop_limit = 200;
          a = corrtemp > 0;
          k = 1;
              while (nnz(triu(a))/uniqueedges > j && loop_limit > 0)
                  a = corrtemp > prctile(corrtemp(:),k+5);
                  k = k + .5;
                  loop_limit = loop_limit - 1:
              end
          sparse{i,counter} = corrtemp.*a;
          kanye{i,counter} = mean(mean(sparse{i,counter},2));
          usher{i,counter} = some_function(sparse{i,counter});
          eminem{i,counter} = some_function(sparse{i,counter});
          counter = counter + 1;
      end
    end

[John BG]

would you please attach subjectlist.txt, part of it, or a text file with data that you agree it's reasonably close to the data you want to process?

thanks for time and attention, awaiting answer

John BG

<mailto:[e-mail] [e-mail]>

[Walter Roberson]

What is the code intended to do?

Answer 1

[Jan]

Start with a pre-allocation of the output:

% Before the loop:
nSubject = length(subjectlist);
nJ       = 31;                   % length(0.1:0.01:0.4)
sparse   = cell(nSubject, nJ)
kanye    = cell(nSubject, nJ);
usher    = cell(nSubject, nJ);
eminem   = cell(nSubject, nJ);

Then use the profiler to find the bottleneck of the code. If this e.g. the load command dur to the slow disk (or network drive?) access, it is not worth to improve the calculations.

If it is prctile, the while loop can be replaced by a smarter binary search, which reduces the number of tests until the limit is found.