MATLAB Answers

Can I make this function faster?

조회 수: 8(최근 30일)
lianne
lianne 7 Feb 2014
답변: J 8 Feb 2014
I need the following function in a program, but it takes a lot of time to run. Does anyone know how I can make it faster? "meting" can be as large as 25000x1 cell, every cell consists of (max, mostly less) 1x2000 char.
if true
% code
function [meetwaarde,power]=databewerking(meting,multiplier,offset)
string=cell2mat(meting);
hexa=regexp(string(1,:),'\w{1,2}','match');
lengtekolom=size(meting,1);
for i=2:lengtekolom
var=regexp(string(i,:),'\w{1,2}','match');
hexa=cat(1,hexa,var);
end
meetaantal=size(meting{1},2)/2;
meetwaarde=zeros(lengtekolom,meetaantal);
deca=hex2dec(hexa);
dec=reshape(deca,lengtekolom,meetaantal);
for i=1:lengtekolom
for j=1:meetaantal
if dec(i,j)==0
meetwaarde(i,j)=0;
else
meetwaarde(i,j)=(multiplier*dec(i,j))-offset;
end
end
end
power=sum(meetwaarde,2);
end
end
Thanks a lot!!
  댓글 수: 2
lianne
lianne 7 Feb 2014
If true isn't in the actual code, it is just to show the code correctly (if you click on {}Code, you get this ;) )

댓글을 달려면 로그인하십시오.

채택된 답변

J
J 8 Feb 2014
Your code is slow because you continuously grow a cell array within this line:
hexa=cat(1,hexa,var);
Since the cell-array is growing every loop, Matlab has to re-alloacate a new memory location for the cell array within each loop. Using vectorized code for this would significantly speed up your code. I think the following code gives the exact same result as your code, but is much faster. For a 5000x1 cell array with 1x2000 char I obtain a speed up of 500x on my laptop.
function [meetwaarde,power]=databewerking(meting,multiplier,offset)
lengtekolom=size(meting,1);
meetaantal=size(meting{1},2)/2;
string=cell2mat(meting);
string = reshape(string.', 2, numel(string)/2).';
deca=hex2dec(string);
dec=reshape(deca,meetaantal,lengtekolom).';
meetwaarde=zeros(lengtekolom,meetaantal);
meetwaarde(dec ~= 0) = (multiplier*dec(dec ~= 0))-offset;
power=sum(meetwaarde,2);

추가 답변(1개)

Dina Irofti
Dina Irofti 7 Feb 2014
Have you tried parallel for loop? See parfor .
  댓글 수: 1
Patrik Ek
Patrik Ek 7 Feb 2014
Parfor seems to be a bit of overkill in this case, since it seems the most of the code could be vectorized. Also, the code takes much time to run and no matlabpools are available, eg if the code is run on the personal laptop, then you may not want to occupy all cores at the same time.

댓글을 달려면 로그인하십시오.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by