Release v2.0.5.post1

Changes

• See CHANGELOG.md for detailed changes

Installation

pip install fisher_scoring==v2.0.5.post1

v2.0.5

✨ New Features

• RobustLogisticRegression: Outlier-resistant classification with epsilon-contamination
• Empirical Fisher Information: Support for Poisson & Negative Binomial models
• Enhanced Fisher Scoring: Consistent implementation across all GLMs

🔧 Improvements

• Converted NegativeBinomialRegression to proper Fisher scoring
• Added comprehensive statistical inference methods
• Enhanced numerical stability

📦 Installation

pip install fisher-scoring==2.0.5

v2.0.4

• v2.0.4
  • Added a beta version of Poisson and Negative Binomial regression using Fisher Scoring.
  • Changed naming conventions for simplicity and consistency.
  • Changed poetry to uv for packaging.

Fisher Scoring v2.0.2

Fisher Scoring v2.0.2

Changelog

Overview

The Fisher Scoring package has undergone significant performance improvements and feature updates from v2.0 onward, enhancing matrix calculation efficiency, model speed, and usability across all regression types.

New in v2.0.2

• FisherScoringMultinomialRegression: Enhanced flexibility by removing fixed data types for numpy arrays, improving compatibility with diverse datasets.
• Code Optimization: Refined matrix operations to maintain peak performance across Fisher Scoring modules.
• Documentation & Image Display: Adjusted README formatting and hosted images to ensure accurate display on PyPI, enhancing readability and visual guidance.

Summary of Major Changes Since v2.0

Performance Improvements

With optimized matrix calculations, all models now exhibit faster training times and reduced memory usage. The streamlined operations provide up to 290x speed improvements:

• Multinomial Logistic Regression: Reduced training time from 125.10s to 0.43s (~290x speedup).
• Logistic Regression: Reduced training time from 0.24s to 0.05s (~5x speedup).
• Focal Loss Logistic Regression: Reduced training time from 0.26s to 0.01s (~26x speedup).

Fixes and Usability Enhancements

• Verbose Parameter for Focal Loss: Fixed the verbose logging for FisherScoringFocalRegression, ensuring informative output during model training.
• Improved Documentation: Expanded README to provide clear installation instructions, feature descriptions, and example usage for all models, enhancing user experience.
• External Image Display on PyPI: Ensured hosted images display properly in the README for a polished PyPI appearance.

Fisher Scoring v2.0.3

Fisher Scoring v2.0.3

Changelog

New in v2.0.3

• predict_ci method with the ability to compute confidence intervals for predicted probabilities, supporting both "logit" and "proba" methods. Both methods are based on the delta method with a slight difference in computation. The logit method is adopted from the book "Applied Logistic Regression" and the delta method from this stackoverflow thread.
• Introduction of model parameters (standard errors, Wald scores, confidence intervals) for FisherScoringFocalRegression allowing to use inference about maximum likelihood estimates obtained with this model.

Fisher Scoring v0.1.4

Fisher Scoring v0.1.4

This release introduces some changes to the functionality and outputs of logistic regression and correct log likelihood display for multinomial logistic regression model.

Changelog

• Added minor changes to the model code for logistic regression and multinomial logistic regression.

Fisher Scoring v0.1.3

Fisher Scoring v0.1.3

This release introduces enhanced statistical outputs for Multinomial Regression, including the calculation of coefficients, standard errors, p-values, and confidence intervals for each class.

Changelog

• Added the calculation of coefficients, standard errors, p-values, and confidence intervals for each class in Multinomial Regression.

Fisher Scoring v0.1.2

Fisher Scoring v0.1.2

This release includes downgrading dependencies of scikit-learn and numpy for better handling with other packages.

Changelog

• Updated dependency versions

Fisher Scoring v0.1.1

Fisher Scoring v0.1.1

This release includes updates and improvements for the Fisher Scoring algorithm implementations. The following features are included:

• Fisher Scoring for logistic regression
• Fisher Scoring for multinomial regression
• Fisher Scoring with focal loss

Changelog

• Added support for Python 3.9+
• Updated dependency versions
• Improved documentation

v.0.1.0

Initial release