Curve Analysis class

[nkanazawa1989]

Summary

This PR adds CurveAnalysis class that provides basic functionalities to extract curve data and perform proper error handling for subclasses.

Details and comments

Based on this base class, we can easily implement several fitting subclasses. Data extraction and error handling is nicely performed behind the scene. This class overrides BaseAnalysis._run_analysis and provides several function-wise private methods that subclasses can override. For example _run_fitting, _create_figure, _data_processing, and _post_processing. Internal data representation CurveEntry (named tuple) is introduced to manage curve data.

To define new analysis, basically you can override base class attributes.

For example, T1 experiment fit may be written as

class T1Analysis(CurveAnalysis):
    __x_key__ = "delay"
    __fit_funcs__ = [library.exponential]
    __param_names__ = ["a", "tau", "b"]

and IRB fit may be written as

class IRBAnalysis(CurveAnalysis):
    __x_key__ = "ncliffs"
    __series__ = [
        SeriesDef(
            name="standard_rb",
            param_names=["a", "alpha_std", "b"],
            fit_func_index=0,
            filter_kwargs={"interleaved": False}
        ),
        SeriesDef(
            name="interleaved_rb",
            param_names=["a", "alpha_int", "b"],
            fit_func_index=0,
            filter_kwargs={"interleaved": True}
        )
    ]
    __fit_funcs__ = [library.exponential]
    __param_names__ = ["a", "alpha_std", "alpha_int", "b"]

As you can see subclasses don't need to write their own code. This will give us two benefits:

• This drastically simplifies unittests and maintenance overhead (you only need to write unittest for method that you override).
• We can easily serialize the analysis class (if method is not overridden).

Hint for review

• Details are written in the CurveAnalysis class docstring.

• Actual usages are shown in unittest.

• Flow of fit operation

  1. _run_analysis is called
  2. _run_analysis calls _extract_curves.
  3. _extract_curves calls _data_processing (so that subclass can override the logic).
  4. _extract_curves generates CurveEntrys and return them.
  5. _run_analysis calls _run_fitting (so that subclass can override the logic, i.e. initial guess/weights/bounds...)
  6. _run_fitting calls _series_curve_fit
  7. _series_curve_fit performs a linearized multi-objective non-linear least squares fit and returns AnalysisResults.
  8. _run_analysis calls _post_processing followed by _create_figure.
  9. _run_analysis returns analysis results and figures

[yaelbh]

Overall it looks very good. My comments are mostly questions/hesitations and typos.

[yaelbh]

Also here, not sure whether to return a dictionary or an AnalysisResultV1 object

[eggerdj]

I left some ideas on how to better deal with the data processing.

[gadial]

Most of the PR is really excellent and will save some of the current complexity in RB; however, I'm not sure if it'll be possible to use RB with it without some further additions (in particular the handling of p0).

[nkanazawa1989]

Thanks for reviewing! @eggerdj @gadial @yaelbh
I updated the PR and I think it's ready for the second review.

[eggerdj]

Thanks for adding the training logic. I still have a few questions and comments related to it.

[eggerdj]

level2_probability can be removed, see my comment above.

[eggerdj]

Looks mostly good. Still a few comments and questions.

[eggerdj]

LGTM