Classify qubit states

[Edoardo-Pedicillo]

This PR introduces the folder qibocal/extras, where there are the classifiers benchmarks.
Checklist:

• Reviewers confirm new code works as expected.
• Tests are passing.
• Coverage does not decrease.
• Documentation is updated.

[codecov]

Codecov Report

Merging #261 (69f2f68) into main (4cfb560) will increase coverage by 2.72%.
The diff coverage is 51.65%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #261      +/-   ##
==========================================
+ Coverage   35.11%   37.84%   +2.72%     
==========================================
  Files          30       42      +12     
  Lines        2355     2801     +446     
==========================================
+ Hits          827     1060     +233     
- Misses       1528     1741     +213     

Flag	Coverage Δ
unittests	37.84% <51.65%> (+2.72%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/qibocal/fitting/classifier/ada_boost.py	0.00% <0.00%> (ø)
src/qibocal/fitting/classifier/gaussian_process.py	0.00% <0.00%> (ø)
src/qibocal/fitting/classifier/naive_bayes.py	0.00% <0.00%> (ø)
src/qibocal/fitting/classifier/nn.py	0.00% <0.00%> (ø)
src/qibocal/fitting/classifier/qubit_fit.py	0.00% <0.00%> (ø)
src/qibocal/fitting/classifier/random_forest.py	0.00% <0.00%> (ø)
src/qibocal/fitting/classifier/rbf_svm.py	0.00% <0.00%> (ø)
src/qibocal/fitting/methods.py	71.92% <20.00%> (+0.50%)	⬆️
src/qibocal/fitting/classifier/run.py	81.61% <81.61%> (ø)
src/qibocal/fitting/classifier/utils.py	83.33% <83.33%> (ø)
... and 3 more

[alecandido]

@Edoardo-Pedicillo @scarrazza from what I understood "Ramiro"'s method is a linear SVM, simply manually fitted.

Is there any value in using that, instead of an actual SVM? (or just the Naive Bayes, that is orders of magnitude faster)

Of course, the current implementation is definitely useful for benchmarking, and to prove the effectiveness and performances of the other methods.

[alecandido]

Why do you need extra IO support?

[Edoardo-Pedicillo]

Because the workflow has failed, I think that the problem with tensorflow-io-gcs-filesystem is that there was a new release on Mar 29,2023 that has not the Windows wheels. Maybe I should put tensorflow-io-gcs-filesystem = "^0.31".

[alecandido]

No, the problem is in a dependency of a dependency, do not add it to your own to make it passing on Windows.

We can think about more refined solutions, in particular dropping who depends on that, or lowering the related version, but if you don't use it directly, don't declare it among your deps :)

[Edoardo-Pedicillo]

Yes, I know, obviously tensorflow has this dependecy, I checked but downgrading it doesn't solve the problem. I can try with the other dependencies, but this could be a lot time-consuming. Do you have some hints to solve it as soon as possible?

[alecandido]

It is definitely a non-trivial problem. Proper solutions are:

• TensorFlow IO releasing wheels for Windows for v0.32.0 tensorflow-io 0.32.0 release tensorflow/io#1789 (comment)
• Poetry support to resolve indirect deps for multiple platforms Platform dependent indirect dependencies python-poetry/poetry#7746 (or at least invalidate the lock file for unsupported platforms)

I'm trying to consider if there are other possible workarounds, in order to avoid a direct dependency on tensorflow-io-gcs-filesystem, that we don't use.

[alecandido]

Last few comments, but there is nothing blocking :)

After these we can merge (if needed, even before).

[alecandido]

Let's try this way: we won't have tf on Windows, but we're mostly not even using it.

[Edoardo-Pedicillo]

@scarrazza this PR is ready to be merged.

[scarrazza]

Very good, thanks.