Cvxprob/newcnstr error

1

Hi all, i ran the code as below and i got some errors as stated, can someone tell me what is the error or anything that i should do to make this coding run?

 %% Find sparse coefficients using BP

 coeffMat = zeros(numTrain, numTest);
 % there are numTest numTrain-dimensional coefficients

 tic;
 cvx_clear;

for testIdx = 1:numTest

if rank(dictionary) == size(dictionary,1)
cvx_begin
    cvx_quiet(true);
    variable coeff(numTrain)
    minimize(norm(coeff, 1));
    subject to
        dictionary'*coeff==testStruct.testMat(testIdx,:)';
cvx_end
else
	coeff = zeros(numTrain);
end
disp(testIdx/numTest);

% Uncomment for a faster but less accurate implementation of BP
%         [coeff, iterationCount] = ...
%             SolveDALM_fast(dictionary', testStruct.testMat(testIdx,:)');

coeffMat(:, testIdx) = coeff; % save coefficients
end
toc;

Error:
Error using cvxprob/newcnstr (line 87)
Matrix dimensions must agree.

Error in == (line 12)
b = newcnstr( evalin( ‘caller’, ‘cvx_problem’, ‘[]’ ), x, y, ‘==’ );

Error in testsparseclassify (line 58)
_ dictionary’*coeff==testStruct.testMat(testIdx,:)’;_

2

Two possibilities: Either the dimensions really don’t agree, or you have encountered a bug in CVX.

For the former possibility, what are the dimensions of dictionary, coeff, and testStruct.testMat(testIdx,:) ?

For the latter possibility, what version of CVX are you using? If CVX 3.0 beta (which is buggy), try CVX2.1