Add SklearnIQDiscriminator.

[hodgestar]

Summary

This PR adds a new measurement IQ discriminator that accepts any sklearn classifier as an input and uses that classifier to categorize measurements.

Details and comments

This is my first qiskit PR, so feedback welcome!

I couldn't find where the measurement docs should live. If someone lets me know where to start some, I can add them as part of this PR.

[claassistantio]

All committers have signed the CLA.

[claassistantio]

Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[eggerdj]

Thank you very much for your contribution. This is looking very good.
On top of the requested changes, could you also add the following to the example notebook (it is in qiskit-ignis/examples/DiscriminatorTutorial.ipynb) to highlight your work?

from sklearn.svm import SVC

svcs = {
    0: SVC(C=1., kernel="rbf", gamma="scale"),
    1: SVC(C=1., kernel="rbf", gamma="scale")
}

svc_discriminators = {}

for q in qubits:
    svc_discriminators[q] = SklearnIQDiscriminator(svcs[q], result, [q], ['0', '1'])
    
fig, ax = plt.subplots(1, 2, figsize=(14,5))

svc_discriminators[0].plot(ax[0], flag_misclassified=True, show_boundary=True)
svc_discriminators[1].plot(ax[1], flag_misclassified=False, show_boundary=True);

I suggest adding this at the end of the section entitled Single-qubit discriminator

[eggerdj]

Made a suggestion to make is more transparent to the user what we require from the classifier instance.

[hodgestar]

@eggerdj Thanks for the review! I've added the section to the tutorial notebook, but I don't think I have pulse-level access to any multi-qubit machine, so I'm unsure how to run the notebook cleanly and generate new output. Any suggestions?

[hodgestar]

I think all comments have been addressed and that this is ready for further review.

[eggerdj]

@hodgestar This looks good. A few minor details: lint checks are failing.
Regarding the IQ data: there is a pickled data set test_result.pickle in the example folder which contains data from one of the IBM Q backends. You can directly load this without having to interact with the backend by running the cell with

if use_prerun_data:
    #Use pickle to load existing data
    import pickle
    from qiskit.result import Result

    with open('test_result.pickle', 'rb') as handle:
        res = pickle.load(handle)

    result = Result.from_dict(res)
else:
    result = job.result(timeout=3600)

This will allow you to test all sorts of classifiers.

[eggerdj]

Code looks good but style checks are failing (see https://travis-ci.com/Qiskit/qiskit-ignis/jobs/267623281). Please see suggestions below.

[hodgestar]

Linting issues fixed (passing locally, waiting for Travis to finish). I updated the discriminator tutorial output.

Note: Updating the notebook isn't quite trivial because even with use_prerun_data the it requires a suitable quantum computer backend (it tries to set up operations on qubits 0 & 1, tries to measure qubits 0-19, uses the pulse backend, etc). Rather than heavily edit the notebook in this PR, I fudged things by skipping the code blocks that set up the schedule and manually fixing the execution count afterwards to make the notebook look nice.

[eggerdj]

Looks good to me. Thanks a lot for this, really cool stuff.

[eggerdj]

@hodgestar You actually don't need to run the schedules or define them. If you just want to play with the pickled data you need to run only the imports, this block:

qubits = [0, 1]
schedules = []
meas_buffer = 2
shots = 512
experiment_name = 'X90p'

and the data loading part. Then you can create discriminators.

[eggerdj]

@dcmckayibm This looks good. I am happy to merge this.

[hodgestar]

@eggerdj Re notebook: Yep, that's what I did, but it's nice to be able to just restart & run all so that one has confidence the notebook runs correctly end to end. Anyway, it's good for now I think. Thanks for reviewing!

[hodgestar]

@eggerdj @dcmckayibm Re-pinging this -- and happy new year!