refactor: MLE API

[lukasheinrich]

Description

changes the optimizer API to only be about minimization. Actual fitting (in the sense of maximum likelihood estimation) is moved to pyhf.infer.mle

#531
#687

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

* Simplify optimizer API to only have 'minimize()'
* Add option to return minimized function in 'minimize()'
* Add optin to return uncertainties in the parameters for 'minimize()'
* Move concrete fits that are needed for maximum likelihood estimation / inference to pyhf.infer.mle

[lukasheinrich]

This works then

[codecov]

Codecov Report

Merging #690 into master will increase coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #690      +/-   ##
==========================================
+ Coverage   95.15%   95.17%   +0.01%     
==========================================
  Files          50       51       +1     
  Lines        2746     2754       +8     
  Branches      385      390       +5     
==========================================
+ Hits         2613     2621       +8     
  Misses         88       88              
  Partials       45       45

Flag	Coverage Δ
#unittests	95.17% <100%> (+0.01%)	⬆️

Impacted Files	Coverage Δ
src/pyhf/optimize/__init__.py	100% <ø> (ø)	⬆️
src/pyhf/optimize/opt_pytorch.py	100% <ø> (ø)	⬆️
src/pyhf/infer/mle.py	100% <100%> (ø)
src/pyhf/infer/test_statistics.py	100% <100%> (ø)	⬆️
src/pyhf/infer/utils.py	100% <100%> (ø)	⬆️
src/pyhf/optimize/autodiff.py	100% <100%> (ø)	⬆️
src/pyhf/optimize/opt_minuit.py	100% <100%> (ø)	⬆️
src/pyhf/optimize/opt_scipy.py	100% <100%> (ø)	⬆️
src/pyhf/optimize/opt_tflow.py	100% <100%> (ø)	⬆️

---

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 3a2a1d0...26381ae. Read the comment docs.

[lukasheinrich]

the only reason CI is failing is due to an addition of a new file as part of the refactoring

[matthewfeickert]

the only reason CI is failing is due to an addition of a new file as part of the refactoring

Later today I can work on having Codecov only fail if there is a greater than 0.2% change or something like that.

[matthewfeickert]

This is all looking good! If you can take a look at the Codefactor issue and then make a call either way on it that would be nice. I think I'm going to need a bit more time to think about the discussion around fval, but perhaps @kratsg has had time for additional thoughts here.

[kratsg]

I still think return_fval, and fixed_values need to be changed.