Release Notes#

This is the list of changes to MLCompare between each release. For full details, see the commit logs.

Version 1.2.0#

### Pipelines - Created a data_pipeline function for performing only data retrieval and processing - Expanded the generated model performance metrics and added a required argument to full_pipeline for specifying whether the pipeline is being used for regression or classification tasks

### DatasetProcessor - Refactored the class to store the train-test split data for easier processing - Added a handle_nan method which can drop, forward-fill, and backward-fill missing values - Added label encoding ordinal encoding, and target encoding methods - Added several scaling and transformation methods from sklearn: StandardScaler, MinMaxScaler, MaxAbsScaler, RobustScaler, PowerTransformer, QuantileTransformer, and Normalize

### Documentation - Created a new homepage - Updated the layout of the API Reference page - Added content to the Release Notes page - Improved various docstrings - Made multiple updates to the README including adding a “Planned Additions” section

### Other - Added a ResultsWriter class, responsible for directory and file naming and creation throughout pipelines

  • Implemented directory and file name incrementing to prevent overwrites

  • Changed the default directory name to use the current timestamp to ensure uniqueness

  • Improved how saving model results is handled

  • Removed the DataProcessor class in favor of pipelines

  • Migrated several high-level functions being used within pipelines to a new module: processing.py

  • Improved unit test coverage

Version 1.1.0#

  • Refactored DatasetProcessor, moving save_directory from a class attribute to a method argument

  • Added type validation to several methods within DatasetProcessor

  • Updated docstrings for the dataset_processor module

  • Updated unit tests for DatasetProcessor

  • Added optimal device selection for PyTorch models as default behavior

  • Corrected a logging issue with model processing

Version 1.0.1#

  • Updated the project versioning to dynamically use the version in mlcompare/__init__.py

  • Modified the package attributes displayed on PyPi including adding links to documentation

  • Added a link to the documentation to the library __init__

  • Created a GitHub action for publishing newly tagged versions to PyPi

Version 1.0.0#

Initial Release

On this page
Show Source