HEROS Beta 0.2.5

PyPI build fixes

HEROS Beta 0.2.4

Fixes for PyPI automated package build.

HEROS Beta 0.2.3

First PyPI release of HEROS. Added github workflow, reorganized tests, and added files for building PyPI package.

HEROS Beta 0.2.2

Adds a number of minor fixes and updates to the algorithm codebase ensuring the functionality of the data on multiclass and binary class outcomes, data with missing values, and data with both quantitative and categorical feature values. Also identified and fixed some minor bugs in rule discovery that were slowing down successful rule discovery. Lastly added a new HEROS logo.

HEROS Beta 0.2.1

This release cleans up the code and improves the demonstration notebook but maintains the same algorithm functionality used in the original HEROS paper. Includes updates to the README, Requirements, code organization, and the demonstration notebook. Within the demonstration notebook and code-base we've added functions to generate model prediction explanations for individual predictions, new visualizations, and examples of generating model feature importance estimates and generating an ROC curve with HEROS predictions. This release was originally completed on April 14th 2025.

HEROS Beta 0.2

This release pairs with our accepted/revised HEROS publication at GECCO 2025. The algorithm and paper was revised after the initial submission to fix some issues identified in the original paper submission. This current release represents the code used to evaluate the algorithm used in the final published paper evaluations. It was originally completed on April 13, 2025.

HEROS Beta 0.1

Initial release of HEROS (Heuristic Evolutionary Rule Optimization System). This release pairs with the initial manuscript submission to GECCO 2025 including early analysis and comparison of HEROS to the scikit-ExSTraCS algorithm. This version of HEROS includes a two phase algorithm (run without alternation between phases), where Phase I focuses on the discovery and optimization of candidate rules and Phase II focuses on using a population of final rules discovered by Phase I to learn optimal accurate and compact rule-sets. Both phases of the algorithm are uniquely driven by respective Pareto-inspired multi-objective fitness functions. While the code in this release has been designed to accommodate discrete or quantitative features, as well as binary or multiclass outcomes, thus far we have only thoroughly debugged and evaluated it for discrete features and binary outcomes. Many improvements and expansions of this algorithm are planned for the future. This version of the code was originally completed on January 29th 2025.

Alpha 0.1.0

An initial working implementation of HEROS with both phase 1 and phase 2 operational. Not yet fully tested or optimized, with potential remaining bugs under untested scenarios. Originally completed on Nov. 6, 2024.