I want to know how to speed up my for loop,it's a simple for loop.thanks

%% this code cost aroud 6s.

b=[]; %Establish a null matrix

N=length(ampw_1); % N=103;

M=length(amp_m); % M=512;

for j = 1:M-N+1; % j=1：1：410；

index_sm = j; %index_sm = 1:410;

index_em = j+N-1; %index_em = 103:512;

ampw_2 = abs(amp_m(index_sm:index_em))/max(abs(amp_m(index_sm :index_em))); %ampw_2 from 1 to 103; from 2 to 104; from 3 to 105....from 410 to 512

% Every loop，this segment will store 103 * 1 So,the At the end of the loop,the ampw_2 array is 103 * 410 and give b array.In my program, this statement took 4S，and i want to know how to speed it up.

b = [b ampw_2]; %give the result of every loop to b,at the end of the loop , the b array is 103*410;

end

%% then I I built an empty array，the code as follow,But it's only a few seconds shorter,just 0,4s.

N=length(ampw_1); %N=103

M=length(amp_m); %M=512

ampw_2 = zeros(N,M-N+1); %Set up a zero array of 103 * 410;

for j = 1:M-N+1; % j=1：1：410；

index_sm = j; %index_sm = 1:410;

index_em = j+N-1; %index_em = 103:512;

ampw_2(:,j) = abs(amp_m(index_sm:index_em))/max(abs(amp_m(index_sm :index_em)));

end

%% I want to know is there Any method to speed up the program,then I use the matlabpool open 2;this code cost So More time,it's 60s; Just modify:

parfor j=1:M-N+1;

....

end

if you have any method or proposal,please tell me , thank you very much,

[Daniel M]

You don't need to do this in a loop at all.

[ang lee]

Can you be more specific,or for a simple example,I can't run the code you gave me.thank you again

Answer 1

[Daniel M]

You don't need to do this in a loop at all. Here is how to vectorize the creation of the indices from 1:103, ..., 410:512. From there you can probably figure out how to do it without using a loop at all. (Assuming amp_m is a variable and not a function).

n = 103;
m = 410;
indexMatrix = (0:n-1)'+(1:m);

[ang lee]

Can you be more specific,or for a simple example,I can't run the code you gave me.thank you again

[Daniel M]

clear
clc

% in a loop
b=[];                            %Establish a null matrix
N=103;    %  N=103;
M=512;     %  M=512;
amp_m = rand(1,M);

tic
for   j = 1:M-N+1            %   j=1:1:410;
    index_sm = j;                %index_sm =  1:410;
    index_em = j+N-1;        %index_em = 103:512;
    ampw_2 = abs(amp_m(index_sm:index_em))/max(abs(amp_m(index_sm :index_em)));  %ampw_2 from 1 to 103; from 2 to 104; from 3 to 105....from 410 to 512
    %  Every loop,this segment will store 103 * 1  So,the At the end of the loop,the ampw_2 array is 103 * 410 and give b array.In my program, this statement took           4S,and i want to know how to speed it up.
    b =  [b ampw_2];    %give the result of every loop to b,at the end of the loop , the b array is 103*410;
end
toc % Elapsed time is 0.015229 seconds.

%%% vectorized solution
tic
indexMatrix = (0:N-1)'+(1:(M-N+1));
absamp_m = abs(amp_m(indexMatrix));
ampw_2b = absamp_m./max(absamp_m);
b2 = ampw_2b(:)';
toc % Elapsed time is 0.000938 seconds.

isequal(b,b2) % ans = 1

The results are equal. There is a speedup of 16x using the vectorized method. (0.015229/0.000938 = 16.236)

[ang lee]

I am sorry,

indexMatrix = (0:N-1)'+(1:(M-N+1));this code in my computer can't run.It give me the message "Error using + ; MatrixDimensions must agree.". you can run this?

[Daniel M]

You must have an older version of MATLAB that can't do implicit expansion. Here is the fix for that line of code. Tell me if there are other errors.

indexMatrix = bsxfun(@plus,(0:N-1)',(1:(M-N+1)));

[ang lee]

thank you, I can use this code in other version,and your answer is so good.Help me a lot.If I have other question,i hope i can ask you . thank you again,best wishes.