Changelog¶

All notable changes to this project will be documented in this file.

[v0.1.1] - Metadata Hotfix¶

Core evaluate() Engine: Automatically slice datasets up to depth $N$ and test ML model performance across subgroups.
Statistical Rigor: Auto-switching statistical backends (Z-Test, Fisher’s Exact, Bootstrapping) to guarantee robust comparisons against global baseline metrics.
Visual Output System: Integrated matplotlib renderers generating automated, color-coded heatmaps and “Worst Segments” bar charts.
Universal Metric Support: Built-in support for classification metrics (accuracy, f1, precision, recall) and regression metrics (mae, rmse, r2, mse).
Exporters: Added pyslicekit.to_csv() and pyslicekit.to_json() for seamless external auditing and dashboard integration.
Documentation Suite: Sphinx-based documentation containing Getting Started guide, User Guide, FAQ, and complete API reference.

Visualizations currently rely entirely on matplotlib (interactive charts like Plotly are not yet supported).
Very deep slicing (depth > 3) on wide datasets may cause significant performance degradation and memory usage due to combinatorial explosion.

Overlapping text on the heatmap Y-axis labels when column names or categorical string values are exceptionally long.
High variance in Bootstrap Confidence Intervals when dealing with extremely small, skewed regression segments near the min_samples boundary.

Performance Enhancements: Integration of multiprocessing/parallelization to handle massive datasets and deeper slice combinations faster.
Interactive Dashboard: Export capabilities that automatically generate an interactive HTML dashboard.
Custom Metrics: Allowing users to pass their own callable metric functions instead of just pre-defined string names.