Changelog

This document provides a high-level view of the changes made to PG-SUI for each release.

Unreleased

Fixed bug where supervised imputers would fail due to duplicated self argument.
Fixed bug where ImputeNLPCA would run ImputeUBP instead.
Fixed gt_probability heatmap plot. It works correctly now.
Fixed issues where plot directories were not being created.
Fixed bugs where non-ML imputers would not decode the integer genotypes.
Renamed gt probability plot to simulated_genotypes.
Fixed default prefix argument for supervised imputers. It now aligns with the unsupervised imputers as imputer.
Fixed bugs where ImputeKNN and ImputeRandomForest would not run.
Set max pandas version to prevent future warnings.
Added warnings.simplefilter to each module to catch FutureWarning

Unsupervised models now impute on nucleotide-encoded multilabel data instead of 012 data.
Much higher metric scores due to the new multilabel encodings (less class imbalance)
Unsupervised grid searches are faster due to less redundant code and removig unnecessary calculations with the scoring functions.

Documentation updates to make it easier to understand all the available arguments and what they are for.
Refactored estimators.py and scorers.py for better modularity and easier code maintenance.

012-encoded inputs for unsupervised imputers. Now uses multilabel nucleotide (4 class) inputs.

Initial release with four unsupervised neural network models, three supervised IterativeImputer models, and four non-machine-learning imputers.