Error with parfor loop

I have the following code

ft=zeros(1000,1);
gt=zeros(1000,1);
FT=zeros(1000,1) ;
Gt=zeros(1000,1) ;
CovXY=zeros(length(SP)-999,1) ;
VarX=zeros(length(SP)-999,1) ;
VarY=zeros(length(SP)-999,1) ;
rho=zeros(1,length(SP)-999) ;
x=zeros(1000,1) ;
y=zeros(1000,1);
for j=1:length(SP)-999
x=SP(j :j+999);
y=CAC40(j :j+999) ;
[parameters, ll, Ft, VCV, scores] = heavy(x,1,1) ;
[parameters, ll, Gt, VCV, scores] = heavy(y,1,1) ;  
ft = x ./ sqrt(Ft);
gt = y ./sqrt(Gt);
[CovXY(j, :),VarX(j, :),VarY(j, :)]=DCCA_vec(ft,gt,n);
rho(j)=CovXY(j)/(VarX(j)*VarY(j)) ;
end

I replaced for by a "parfor" then i get a message

Error: The variable CovXY in a parfor cannot be classified.
See Parallel for Loops in MATLAB, "Overview".

If someone can help me.

Thank you

[Kevin Chng]

[CovXY(j, :),VarX(j, :),VarY(j, :)]=DCCA_vec(ft,gt,n);

Your error is here? Do you mind attach your script including all the variables required? So that I can help to try out

[tilfani oussama]

You find in this link https://www.mathworks.com/matlabcentral/answers/423283-reduce-execution-time-in-matlab?s_tid=prof_contriblnk

Answer 1

[Walter Roberson]

On the line

[CovXY(j, :),VarX(j, :),VarY(j, :)]=DCCA_vec(ft,gt,n);

you use CovXY as if it is a 2D array in which you are writing a new row each time.

In the next line

rho(j)=CovXY(j)/(VarX(j)*VarY(j)) ;

you are using CovXY as if it is a vector.

When you use parfor, you need to be consistent in how you access your variables.

Also, the fact that you pre-initialized gt and ft will be taken as evidence that they are to be output from the loop, but because you write over them every time, you would get the last iteration with for, and parfor would refuse to run.

The initialization you are doing of CovXY suggests that you are expecting scalar outputs from DCCA_vec. If so then

[CovXY(j, 1),VarX(j, 1),VarY(j, 1)]=DCCA_vec(ft,gt,n);
rho(j,1)=CovXY(j,1)./(VarX(j,1)*VarY(j,1)) ;

[tilfani oussama]

The most important is to get rho for each iteration, CovXY, VarX and VarY just serves to calculate rho. I think i can write:

[CovXY,VarX,VarY]=DCCA_vec(ft,gt,n);
rho(j)=CovXY./(VarX*VarY) ;

My aim is to speed up the program above, i suggested a "parfor" but if you have other suggestions they are welcome. This program uses two function heavy and DCCA_vect both are optimized, my issue is with the program above.

I'm requesting a help. using parfor doesn't speed up the time for execution

[Walter Roberson]

The general outline for parfor is:

• If you have variables that are being used for temporary storage and are not needed after the loop, then do not initialize them before the loop. If they are completely written over unconditionally, do not initialize them (wastes time and can confuse the parser.) If they are written to in pieces, initialize them to full size inside the parfor
• For output variables needed after the loop, it is still preferred to initialize them before the loop, so that they will definitely have proper size. parfor deliberately runs the last iteration first in case you failed to pre-initialize some variables
• It is permitted to initialize an array before the loop, read from it, and write to it. However, when you do that, all of the subscripts you use in reading should be the same as the subscripts used in writing
• inside parfor, for any one parallel variable, all reads and all writes must use the same subscript expressions
• inside parfor, only have one write to any one parallel variable
• because of the restrictions on reading and writing, you should generally pull out a complete row or column at the beginning of the loop, assigning to a temporary variable, reading and writing to temporary variables, and then at the end, update that one row or column from the temporary variable

For example:

   parfor J = 1 : 10
     output(J,1) = 7;
     output(J,2) = 9;   %NO
   end

and instead

parfor J = 1 : 10
  t = zeros(1,2);
  t(1) = 7;
  t(2) = 9;
  output(J,:) = t;
end

[Walter Roberson]

Your DCCA_vec() appears to be from https://www.mathworks.com/matlabcentral/answers/423283-reduce-execution-time-in-matlab#answer_340928 . I do not know the code for your heavy()