Set qcp to True for cvxpy Problem.solve in qp_solver

[mtreinish]

Summary

The legacy tutorial in the Qiskit/qiskit-tutorials repo for running
QSVM classification is failing with newer versions of aqua that use
cvxpy. The error message returned suggests setting the qcp flag to true
on solve(). This treats the problem as disciplined quasiconvex program
instead of a disciplined convex program and seems to avoid the hard
failure in the tutorial.

Details and comments

Fixes Qiskit/qiskit-tutorials#928

[rraymondhp]

Does this really solve the issue? I tried to set the qcp=True but it does not work. And gave another error as below:

~/miniconda3/envs/qiskit_keio/lib/python3.6/site-packages/qiskit/aqua/utils/qp_solver.py in optimize_svm(kernel_matrix, y, scaling, max_iters, show_progress)
     79         [G@x <= h,
     80          A@x == b])
---> 81     prob.solve(verbose=show_progress,qcp=True)
     82     result = np.asarray(x.value).reshape((n, 1))
     83     alpha = result * scaling

~/miniconda3/envs/qiskit_keio/lib/python3.6/site-packages/cvxpy/problems/problem.py in solve(self, *args, **kwargs)
    393         else:
    394             solve_func = Problem._solve
--> 395         return solve_func(self, *args, **kwargs)
    396 
    397     @classmethod

~/miniconda3/envs/qiskit_keio/lib/python3.6/site-packages/cvxpy/problems/problem.py in _solve(self, solver, warm_start, verbose, gp, qcp, requires_grad, enforce_dpp, **kwargs)
    731             if qcp and not self.is_dcp():
    732                 if not self.is_dqcp():
--> 733                     raise error.DQCPError("The problem is not DQCP.")
    734                 reductions = [dqcp2dcp.Dqcp2Dcp()]
    735                 if type(self.objective) == Maximize:

DQCPError: The problem is not DQCP.

[t-imamichi]

An issue is posted again. Qiskit/qiskit-tutorials#932

[woodsp-ibm]

@rraymondhp As you have seen there is still a problem that this did not take care of. Someone else contacted me directly noting the above. We briefly looked at it and in the meanwhile, since they wanted to move forwards, I suggested they might go back to 0.7.1 where cvxopt was still being used.