Welcome to PySliceKit’s documentation!

GitHub Repository: github.com/AnshumanTiwari2006/PySliceKit

The Problem: Relying on global metrics like “95% accuracy” masks critical algorithmic bias, data drift, and localized underfitting where your model is secretly failing.

PySliceKit is an automated detective for Machine Learning models that solves this by doing five things automatically:

1. Bins continuous columns into quartiles.

2. Cross-Products features to find intersectional failures.

3. Applies Statistical Rigor (Z-Tests, Fisher’s Exact, Bootstrapping) to ensure failures are real.

4. Flags low-sample segments.

5. Enforces a Visual Contract where “Red always means bad”, regardless of metric direction.

Contents:

• Getting Started
  • Installation and Dependencies
  • Supported Metrics
  • What it Returns
  • Complete Minimal Example
• User Guide
  • The Problem with Global Metrics
  • The PySliceKit Solution
  • How PySliceKit Processes Your Data
  • Understanding the Gap Sign
  • How PySliceKit Decides if a Gap is Real
  • Reading the Charts
  • Walkthrough 1: Regression (California Housing)
  • Walkthrough 2: Classification (Breast Cancer)
  • Working with SliceResult Objects
  • Choosing the Right Parameters
  • Common Mistakes and How to Fix Them
  • Exporting Results
  • Next Steps
• API Reference
  • Core Functions
  • Data Types
  • Exporters
  • Exceptions
• Frequently Asked Questions (FAQ)
  • 1. What is PySliceKit?
  • 2. Why shouldn’t I just use SHAP or LIME?
  • 3. Do I need to re-train my model?
  • 4. What metrics are supported?
  • 5. How does PySliceKit know if a drop in performance is “real”?
  • 6. What does the depth parameter do?
  • 7. My dataset is massive, will depth=3 be too slow?
  • 8. Can I pass PyTorch or TensorFlow models?
  • 9. Why are some bars in the bar chart faded?
  • 10. What does the warning icon mean in the heatmap?
  • 11. Why does the renderer enforce “Red means bad”?
  • 12. Can I export the raw data instead of visualizations?
  • 13. Does PySliceKit automatically bin continuous columns?
  • 14. How can I change the chart colors?
  • 15. Does it support multi-class classification?
  • 16. What happens if I pass highly correlated columns?
  • 17. Can I integrate PySliceKit into my CI/CD pipeline?
  • 18. What is the difference between metric_value and overall_metric?
  • 19. How do I handle missing data (NaNs) in my slice columns?
  • 20. Is PySliceKit free for commercial use?
• Changelog
  • [v0.1.1] - Metadata Hotfix
  • [v0.1.0] - Initial Release

Indices and tables

• Index

• Module Index

• Search Page