Qiskit / qiskit-ibm-runtime / 28514905957

87%

push

github

Add ZNE extrapolation module (#2889)

<!--
⚠️ The pull request title should be short detailed and understandable
for all.
⚠️ Also, please add a release note file using reno if the change needs
to be
  documented in the release notes.
⚠️ If your pull request fixes an open issue, please link to the issue.

✅ I have updated the documentation accordingly.
✅ I have read the CONTRIBUTING document.
-->

### Summary

This PR introduces an extrapolation module used for ZNE. It introduces
one public function, `process_extrapolated_expectation_values`, which is
identical in name and signature to the legacy implementation. We may
need to expose more functions from this module when we add PEA -- TBD.

To reduce software complexity, all non-fallback models will now be
implemented with NumPy and piped through the same `scipy.curve_fit` call
to fit to the data. Additionally, the variance for each extrapolated
point will be calculated in the same way, regardless of model, using the
method described
[here](https://www.astro.rug.nl/software/kapteyn/kmpfittutorial.html#confidence-and-prediction-intervals)
(same method used for non-linear variance estimation by legacy fitting
software internally). Piping all fits/extraps through the same workflow
allowed for a significant reduction in branches for different models,
and thus a reduction in code complexity.

This function is backward compatible with the legacy function of the
same name; however, different behaviors may be noticed in degenerate
cases. For example, using `exponential` on a curve with a few (~5)
linear points may now lead to the fit being (correctly) rejected;
whereas before, the extrapolated value may have collapsed to zero with
an erroneously small variance (will show this in unit tests).

### Details and comments
Fixes #2848

This implementation uses `numpy/scipy` to perform curve fitting and
extrapolation.

- [x] Add test suite for new module
- [x] Beef up public docstring. Include references to the heuris... (continued)

185 of 185 new or added lines in 1 file covered. (100.0%)

8476 of 9712 relevant lines covered (87.27%)

6.92 hits per line