Add partial support for L-Moments

[huard]

Pull Request Checklist:

• This PR addresses an already opened issue (for bug fixes / features)
  • This PR partially fixes Fit distributions using L-moments in addition to ML #33
• Tests for the changes have been added (for bug fixes / features)
• Documentation has been added / updated (for bug fixes / features)
• HISTORY.rst has been updated (with summary of main changes)
• bumpversion (minor / major / patch) has been called on this branch
• Tags have been pushed (git push --tags)

• What kind of change does this PR introduce?
  Adds a pwm_fit method, estimating statistical distribution parameters from a sample using probability weighted moments.
  Only a few distributions are supported.

• Does this PR introduce a breaking change?
  No.

• Other information:
  Requires the lmoments3 library, but it is unmaintained anymore. It's not clear if we should add it as a dependency or work on a more robust solution. Needs to be installed from master because the last release is not compatible with recent scipy releases.

[coveralls]

Pull Request Test Coverage Report for Build 2191

• 17 of 18 (94.44%) changed or added relevant lines in 1 file are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage increased (+0.005%) to 91.77%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
xclim/indices/generic.py	17	18	94.44%

Files with Coverage Reduction	New Missed Lines	%
xclim/indices/generic.py	1	90.59%

Totals
Change from base Build 2182:	0.005%
Covered Lines:	2899
Relevant Lines:	3159

---

💛 - Coveralls

[huard]

@Zeitsperre How do you suggest we deal with the lmoments3 dependency? Since it is unmaintained, I'm not sure it's a good idea to include this "officially" in the requirements, but we probably still want to run it in the travis test suite.

[Zeitsperre]

Fully agree that adding unmaintained libraries to the requirements is bad form. Plenty of ways forward though.

There's a couple of ways we could deal with it from the CI side:

• Add a new CI build for it (expensive on resources)
• Retrofit an existing CI build for it (like the py36-nosubset build)

In terms of how to proceed requirements-wise, we can treat functions reliant on it the same way we treat plotting functions, (IE creating a global variable like LM3_INSTALLED in the module that uses it, aborting functions that need it when it isn't there).

If we want users to be able to run tests locally with lmoments3 installed, we would put the dependency in the tox.ini, and if we only want to run it on Travis CI, we can add it to the .travis.yml.

[Zeitsperre]

Whichever way forward you want to try, let me know and I can push some changes here.

[huard]

@Zeitsperre I suggest we go with tox.ini to allow local testing. Ok to bundle the dependency with another suite. We could have a test environment dedicated to all optional external dependencies like roocs. Note that the lmoments3 library has to be installed from master.
pip install git+https://github.com/OpenHydrology/lmoments3.git

[huard]

@aulemahal Combined fit and fit_pwm into one single function.

[aulemahal]

Woupelaye. The build failures are due to OpenHydrology/lmoments3#6, which is open and inactive since june 2019...

[huard]

Ah right, it's the develop branch.

[huard]

change comment to use develop branch.

[Zeitsperre]

Added in the formatting exceptions. Having the import in a function with no space afterwards breaks two hooks, haha.