The variable in parfor cant classified

Hello.This is my code:

c_hat = zeros(n,1);
vc = zeros(m,n);
cv = zeros(m,n);

parfor j = 1:n
        checks = find(H(:,j));
        allIncoming = cv(checks,j);
        sumAllIncoming = sum(allIncoming) + LLRs(j);
        c_hat(j) = (1 - sign(sumAllIncoming))/2;
        toCheck = sumAllIncoming -  allIncoming;
        vc(checks,j) = toCheck;
end

It is not runing because "The variable vc in a parfor cannot be classified." Without parfor it is running fine. Is it possible to run it with parfor?

Answer 1

[Matt J]

There are rules about sliced variables and how you can index them in parfor loops. However, you can get around these rules by re-organizing vc and cv as cell arrays,

    c_hat = zeros(n,1);
    vc = num2cell(zeros(m,n),1);
    cv = vc;

    parfor j = 1:n
            checks = H(:,j)~=0;
            allIncoming = cv{j}(checks);
            sumAllIncoming = sum(allIncoming) + LLRs(j);
            c_hat(j) = (1 - sign(sumAllIncoming))/2;
            toCheck = sumAllIncoming -  allIncoming;
            vc{j}(checks) = toCheck;
    end

    vc=[vc{:}]; %convert back to normal array

[freebil]

Really thanks, but i missed to say tha this parfor is inside a bigger loop.Sorry about this.The full function is:

[m n] = size(H);

c_hat = zeros(n,1);

LLRs = 2*y/sigma_2;

vc = zeros(m,n);

cv = zeros(m,n);

for iters = 1:maxiter

      for j = 1:n

          checks = find(H(:,j));

          allIncoming = cv{j}(checks);

          sumAllIncoming = sum(allIncoming) + LLRs(j);

          c_hat(j) = (1 - sign(sumAllIncoming))/2;

          toCheck = sumAllIncoming -  allIncoming;

          vc{j}(checks) = toCheck;
      end

      if( sum(mod(H*c_hat,2)) == 0 )
          return;
      end

      for j = 1:m

          vars = find(H(j,:));

          allIncoming = tanh(vc(j,vars)/2);

          prodAllIncoming = prod(allIncoming);

          toVar = prodAllIncoming./allIncoming;

          cv(j,vars) = 2*atanh(toVar);
      end

end

I cannot parfor the iters because there is return inside this loop.So i have to parfor the insider loops..I would appreciate it if you help me more.After the changes i have the problem"Conversion to cell from double is not possible.

Conversion to cell from double is not possible.

Error in decodeBP (line 57)
        cv(j,vars) = 2*atanh(toVar);

[Matt J]

Conversion to cell from double is not possible.

Perhaps you meant

 cv{j}(vars)=2*atanh(toVar);

[freebil]

Thanks so much,but i have problem again.

[m n] = size(H);

c_hat = zeros(n,1);

LLRs = 2*y/sigma_2;

vc = num2cell(zeros(m,n),1);
cv = vc;

for iters = 1:maxiter

      parfor j = 1:n

          checks = find(H(:,j));

          allIncoming = cv{j}(checks);

          sumAllIncoming = sum(allIncoming) + LLRs(j);

          c_hat(j) = (1 - sign(sumAllIncoming))/2;

          toCheck = sumAllIncoming -  allIncoming;

          vc{j}(checks) = toCheck;
      end

      vc=[vc{:}];

      if( sum(mod(H*c_hat,2)) == 0 )
          return;
      end

      for j = 1:m

          vars = find(H(j,:));

          allIncoming = tanh(vc(j,vars)/2);

          prodAllIncoming = prod(allIncoming);

          toVar = prodAllIncoming./allIncoming;

          cv{j}(vars) = 2*atanh(toVar);
      end

end

I got the error:

Error using decodeBP>(parfor body) (line 35)
Cell contents assignment to a non-cell array object.

Error in decodeBP (line 19)
    parfor j = 1:n

[Jan]

We cannot guess what your "line 19" is. So please explain this clearly.

[Matt J]

@theodor,

Get rid of this

vc=[vc{:}];

and make sure you use vc and cv as a cell array everywhere, e.g.,

   allIncoming = tanh( vc{j}(vars)/2 );

[freebil]

Really thanks for the answers. @Jan Simon line 35 was

vc{j}(checks) = toCheck;

and line 19 was

parfor j = 1:n

@Matt J I changed them but now i have the error

Index exceeds matrix dimensions.

Error in decodeBP (line 52)
        allIncoming = tanh(vc{j}(vars)/2);

[Matt J]

I take it back. Put this back

 vc=[vc{:}];

but add immediately after it

 vc=num2cell(vc,2);

Also, please learn to use DBSTOP to trap and investigate common errors like "Index exceeds matrix dimensions." and "Conversion to cell from double is not possible."