Preparing for SKLL v2.0 Release

[desilinguist]

(Note: Due to the issues discovered on Windows which necessitated the integration of Azure Pipelines for regular testing, this release PR will be longer than usual. Thanks for your patience!)

• Code changes to get tests to run and pass on Windows

  • Explicitly tell SKLL readers and writers to read and write files in UTF-8.
  • Add newline='' whenever writing out plain text files (summary files, folds file, etc.) to prevent Windows from adding an extra carriage return in CSV/TSV files.
  • Modify test_input.py and test_logutils.py to deal with "file is already open" errors on Windows.
  • Explicitly close matplotlib figure instances to prevent increased memory usage.
  • Modify the custom NumpyTypeEncoder JSON encoder to also include np.int32 since those can be produced on Windows.
  • We cannot figure out why the test test_send_warnings_to_log() does not pass on Windows even though capturing sklearn warnings on Windows, so we are just skipping it for now. I have filed an issue to investigate it in the future.

• Add Windows testing

  • Add azure-pipelines.yml to configure the Azure Pipelines build.
  • Add DistributeTests.ps1 powershell script to split up tests to run in parallel the same way they are done for Travis CI.
  • Remove Appveyor integration.

• Update conda recipe

  • Update version number.
  • Use a more efficient command for the build script that does not involve installing depenencies again as part of the build process.
  • Sort build and run requirements
  • Add license file and update license short name.
  • Update README for recipe.

• Update documentation

  • Add installation section to README.rst and point to the installation section on ReadTheDocs.
  • Move requirements section up in README.rst.
  • Update conda badge in README.rst.
  • Update installation command in getting_started.rst.
  • Update release process documentation to add package building and testing commands.

• Sort entries in requirements.txt

• Update version number in version.py.

• Update setup.py and MANIFEST.in to exclude tests, examples, and documentation from the PyPI package.

Please run the following tests before approving this PR:

• Go through the ReadTheDocs documentation that I built specifically for this branch.
• Test the conda package by either installing it in your environment (conda install -c conda-forge -c ets skll), or creating an environment with it (conda create -n skllenv -c conda-forge -c ets skll). These commands will use python 3.7. You can also add python=3.6 before skll to test with python 3.6.
• To test the TestPyPI package using pip, use pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple skll.
• Please test on your own Macs, Linux servers, as well as Windows, if possible.

[codecov[bot]]

Codecov Report

Merging #580 into master will increase coverage by <.01%. The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #580      +/-   ##
==========================================
+ Coverage   95.02%   95.02%   +<.01%     
==========================================
  Files          20       20              
  Lines        2973     2974       +1     
==========================================
+ Hits         2825     2826       +1     
  Misses        148      148

Impacted Files	Coverage Δ
skll/data/writers.py	94.38% <100%> (ø)	:arrow_up:
skll/data/readers.py	90.51% <100%> (ø)	:arrow_up:
skll/experiments.py	95.52% <100%> (ø)	:arrow_up:
skll/logutils.py	100% <100%> (ø)	:arrow_up:
skll/version.py	100% <100%> (ø)	:arrow_up:

---

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 e617e71...6b1e5b5. Read the comment docs.

[ghost]

Looks good to me thanks! I tested the conda/pip packages and ran the tests on MacOS and didn't find any issues.

I also tested them via pip/conda and ran examples, and worked fine.