Enable Parfor or Vectorization

3 次查看（过去 30 天）

Brian 2012-6-27

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/42201-enable-parfor-or-vectorization

I'm looking for a faster way to calculate these loops. I've read some about vectorization on Loren's page but I'm not sure that would apply here. I also have the parallel computing toolbox, but because of how I'm writing to my preallocated data arrays, I get the message "Valid Indices are restricted in parfor loops." Can someone take a look at my code and make a suggestion how to get parfor to work and/or use vectorization? Essentially I'm calculating a 21 day return for every security, for every day (past 21 days) in my dataset.

Thanks a lot, Brian

parfor i = 1:SecCnt
    tickidx = strcmp(UnqSec(i,1),Cusips);
    for d = 21:DatesCnt
        RollRtns(i*(d-21)+1,1) = prod(Rtns(Rtns(:,2)>d-21 & Rtns(:,2)<=d & tickidx));
        RollCusips(i*(d-21)+1,1) = UnqSec(i,1);
        RollDates(i*(d-21)+1,1) = UnqDates(d,1);
     end
end

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Brian 2012-7-2

编辑：Brian 2012-7-2

在 MATLAB Online 中打开

I have edited this code so that it now works in parallel but is still rather slow. I've read about vectorization and I'm wondering if it's possible for this code. This takes about 30 seconds to in serial and 10 seconds to run in parallel because of the large data overhead. I'm just trying to speed this up a bit.

parfor i = 1:SecCnt
    NestLoop = NaN(1,DatesCnt-21);
    tickidx = strcmp(UnqSec(i,1),Cusips);
    for d = 21:DatesCnt
        NestLoop(d-20) = prod(Rtns(Rtns(:,2)>d-21 & Rtns(:,2)<=d & tickidx));
    end
    RollRtns(i,:) = NestLoop;
end