2.2.1 - 2020-06-30

Fixed

• #65 Remove six dependency entirely.

2.2.0 - 2020-06-29

Changed

• #57 Mark the Fortran code as linguist-vendored so that GitHub classifies this project as Python.
• #62 Update the cross-validation for users to be able to define groups of observations, which is equivalent with foldid of cvglmnet in R.
• #64
  • Python version support: Add v3.8, and drop v3.4 + v3.5.
  • Maintenance: Drop versioneer; update and pin dependencies for development.

2.1.1 - 2019-03-11

Fixed

• #55 Include all Fortran source code in source tarball; exclude autogenerated C.

2.1.0 - 2019-03-11

Added

• #29 Provide understandable error messages for more glmnet solver errors.
• #31 Expose max_features parameter in ElasticNet and LogitNet.
• #34 Use sample weights in LogitNet.
• #41 Add lower_limits and upper_limits parameters to ElasticNet and LogitNet, allowing users to restrict the range of fitted coefficients.

Changed

• #44 Change CircleCI configuration file from v1 to v2, switch to pytest and test in Python versions 3.4 - 3.7.
• #36 Convert README to .rst format for better display on PyPI (#35).
• #54 Use setuptools in setup.py and update author in metadata.

Fixed

• #24 Use shuffled splits (controlled by input seed) for cross validation (#23).
• #47 Remove inappropriate __init__.py from the root path (#46).
• #51 Satisfy scikit-learn estimator checks. Includes: Allow one-sample predictions; allow list inputs for sample weights; Ensure scikit-learn Estimator compatibility.
• #53 Return correct dimensions for 1-row predictions, with or without lambda path, in both LogitNet and ElasticNet (#52, #30, #25).