How do I speed up loops in the code?

2 次查看（过去 30 天）

RP 2023-2-2

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1904971-how-do-i-speed-up-loops-in-the-code

编辑： DGM 2023-2-2

I have the following code which consists of three loops whicjh is taking a long time to run. Any ideas how to speed it up?

C=ones(NumSims, NumScen);

for i=1:NumScen

B=copularnd('Gaussian',PSDMatrix,NumSims);

C(:,i)=icdf('poisson',B(:,i),A(i,1));

M=cumsum(C);

Maxsim=M(end,:);

for i=1:NumScen

Col(1,i)=max(C(:,i));

Col2=max(Col);

LossFlag=zeros(NumSims,Col2);

end

for i=1:NumScen

for j=1:NumSims

for k=1:Col2

if k<=C(j,i)

LossFlag(j,k,i)=1;

else

3 个评论
显示 1更早的评论隐藏 1更早的评论

RP 2023-2-2

NumSims=10000

NumScen = 25

Col2=20

DGM 2023-2-2

编辑：DGM 2023-2-2

在 MATLAB Online 中打开

Since the pasted code is incomplete, I can only guess what's missing. I see a bunch of nested and/or improperly terminated loops. There appear to be two 1:NumScen loops that run inside of another for no apparent reason. I suspect that this can be simplified, but not without knowing what's missing.

C=ones(NumSims, NumScen);
for i=1:NumScen % everything is inside this loop
    B=copularnd('Gaussian',PSDMatrix,NumSims);
    C(:,i)=icdf('poisson',B(:,i),A(i,1));
    
    % these get overwritten each time, but aren't apparently used
    % if they aren't used, get rid of them
    % if they can be calculated outside the loop, move them outside the loop
    % if only Maxsim is needed, get rid of M
    % Maxsim = sum(C,1);
    M=cumsum(C);
    Maxsim=M(end,:);
    
    for i=1:NumScen % ... so this loop appears redundant or misplaced
        % ... but idk if Col needs to be fully written
        % if so, this could just be vectorized -- but idk what size Col is
        % if Col is only used to calculate Col2, then this simplifies
        Col(1,i)=max(C(:,i)); % i'm assuming Col is a vector?
        Col2=max(Col); % ambiguous dimension; again, suggests Col is a vector
        LossFlag=zeros(NumSims,Col2); % not preallocated to the size used later
    end
    for i=1:NumScen % ... and this loop appears redundant or misplaced
        for j=1:NumSims
            for k=1:Col2
                % this should be able to be reduced to a single relational
                % test for the entire array, without any loops needed
                if k<=C(j,i)
                    LossFlag(j,k,i)=1;
                else
                    0; % this does nothing
                end % missing??
            end % missing??
        end % missing??
    end % missing??
end % missing??

Perhaps something like this? (not tested)

% populate C
C=ones(NumSims, NumScen);
for i = 1:NumScen % everything is inside this loop
    B = copularnd('Gaussian',PSDMatrix,NumSims);
    C(:,i) = icdf('poisson',B(:,i),A(i,1));
end
% assuming this can be moved and that M is not needed itself
Maxsim = sum(C,1);
% assuming Col is only needed to calculate Col2
Col2 = max(C,[],'all');
k = 1:Col2; % a row vector
LossFlag = k <= permute(C,[1 3 2]); % compare to generate a 3D logical array

These are a lot of gross assumptions on my part, but that's all I can do.

Consider the example:

outvector = zeros(1,10); % a test vector
n = 0;
for k = 1:10
    for k = 1:10 % trying to alter loop index within a loop is bad
        % this runs 100 times, not 10
        n = n+1;
        outvector(k) = n;
    end
end
% 90% of all the results are just discarded
outvector
outvector = 1×10
    91    92    93    94    95    96    97    98    99   100