feat: Add percentile function to tensorlib

[matthewfeickert]

Description

• Partially addresses Issue Add a percentile function to tensorlib #815
• Resolves Add a percentile function to tensorlib #815
• Resolves Tracking of TensorFlow and Python 3.9 harmonization #1573

Add a percentile function to the tensor backends that adheres to NumPy's numpy.percentile API.

To get quotation marks to look okay for the interpolation method section of the docstrings I used \\ to escape them given this Stack Overflow question. There may be a better way.

Docs build at: https://pyhf--817.org.readthedocs.build/en/817/api.html#backends

Checklist Before Requesting Reviewer

• Tests are passing
• "WIP" removed from the title of the pull request
• Selected an Assignee for the PR to be responsible for the log summary

Before Merging

For the PR Assignees:

• Summarize commit messages into a comprehensive review of the PR

* Add percentile function to the tensor backends
* Add tests for percentile and its interpolation methods
   - JAX requires additional dtype support with the 'linear' interpolation method
     c.f. https://github.com/google/jax/issues/8513
   - PyTorch has yet to implement interpolation method options
   - c.f. https://github.com/scikit-hep/pyhf/issues/1693

[codecov]

Codecov Report

Merging #817 (eee1d30) into master (902052f) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #817   +/-   ##
=======================================
  Coverage   98.05%   98.05%           
=======================================
  Files          64       64           
  Lines        4212     4220    +8     
  Branches      587      587           
=======================================
+ Hits         4130     4138    +8     
  Misses         49       49           
  Partials       33       33           

Flag	Coverage Δ
contrib	25.35% <50.00%> (+0.04%)	⬆️
doctest	61.13% <100.00%> (+0.07%)	⬆️
unittests	96.37% <100.00%> (+<0.01%)	⬆️

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

Impacted Files	Coverage Δ
src/pyhf/tensor/jax_backend.py	97.12% <100.00%> (+0.04%)	⬆️
src/pyhf/tensor/numpy_backend.py	98.49% <100.00%> (+0.02%)	⬆️
src/pyhf/tensor/pytorch_backend.py	98.43% <100.00%> (+0.02%)	⬆️
src/pyhf/tensor/tensorflow_backend.py	97.26% <100.00%> (+0.03%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 902052f...eee1d30. Read the comment docs.

[matthewfeickert]

Note to self: Should give more information in docstrings of interpolation methods and note what is the default interpolation method ("linear").

[kratsg]

should be out with pytorch/pytorch#48523 being done?

[kratsg]

Nope -- need 1.7.2 (https://github.com/pytorch/pytorch/releases/tag/v1.7.1)

[matthewfeickert]

Yeah, sadly. I just keep rebasing these PRs so that they don't fall too far behind and make it harder later on.

[matthewfeickert]

Skipping PyTorch as interpolation kwargs aren't implemented yet as of torch v1.10.0. c.f. #815 (comment)

Skipping JAX so as to have separate test below given floating point issues with "linear" option.

[matthewfeickert]

Hopefully these tests are tolerable at the moment, though obviously not ideal.

[matthewfeickert]

RE: the deviations seen for JAX compared to all other backends, I opened up jax-ml/jax#8513.

[matthewfeickert]

Issue #1693 exists to monitor the PyTorch and JAX issues, so this can PR can move forward.