Further enhancements in ADMMOptimizer

[adekusar-drl]

Summary

Removed soft depenedency on CplexOptimizer, now MinEigenOptimizer and SlsqpOptimizer are used by default.
Added repr support to ADMMParameters.
Added getter/setter for the parameters to ADMMOptimizer.

[woodsp-ibm]

After these changes there is no test then of ADMMOptimizer with the CPLEXOptimizer even specifying it explicitly. Maybe the latter does not need to be included in the testing here in what is really more of an integration test since the same function of ADMM is tested using other optimizers.

[woodsp-ibm]

One other comment - maybe the ADMMOptimizer docstring should state the default qubo and continuous optimizers that are selected internally - or at least say if left as default then a suitable one will chosen internally even if we don't care to explicitly name which one.

[adekusar-drl]

@woodsp-ibm, since we made dependency on Cplex optional and tests with Cplex are not run by the build system (well, they are skipped, correct me if I'm wrong here) I think it is better to use other optimizers internally, especially once we have an SLSQP wrapper. That's why I removed Cplex in the implementation and in the tests as well.

I will update the docstring tomorrow, it is a good point to mention how internal optimizers are chosen if they are not specified.

[woodsp-ibm]

I think tests with Cplex are run under Python 3.7 and the reason for the skip is that Cplex is not yet supported for 3.8 under which we also test in the build.

[adekusar-drl]

Yes, you're right, Cplex does not support python 3.8. Anyway, there's a test where a non-default (Cobyla) continuous optimizer is used, just in case.
Updated the docstring, please review/approve.

[woodsp-ibm]

I think for this ":class:NumPyMinimumEigensolver" to resolve as a clickable link it would have to be fully qualified ":class:~qiskit.aqua.algorithms.NumPyMinimumEigensolver" The ~ means it will show just the last part so it would look like you intended. (The other ref there should work as is since its in the currentmodule.) BTW you can check check the links in html output by cd to docs folder and there 'make html' which will build out the html and you find the index.html in a subfolder and you can navigate to this class from there

[adekusar-drl]

At least in my setup a link ":class:NumPyMinimumEigensolver" is clickable. I have verified this before committing by building html documentation. I guess if you have an import, then you don't need a fully qualified class name.
Or you think it is better to have a full name?
P.S. build fails due other reasons, a fews PR in the same situation.

[woodsp-ibm]

Ok, with the import - I guess I should have thought. I have had to fix up a few of these where another class in another module has been referenced and its not been working. Its fine as-is then, no need to make it fully qualified.