Release v0.1.6a3

Changes

See CHANGELOG.md for detailed changes.

Installation

pip install fastwoe==0.1.6a3

What's New in v0.1.6a3

Check the CHANGELOG for full details.

Release v0.1.6a0-test

Changes

See CHANGELOG.md for detailed changes.

Installation

pip install fastwoe==0.1.6a0-test

What's New in v0.1.6a0-test

Check the CHANGELOG for full details.

Release v0.1.5

Changes

Performance Fix & Code Cleanup: Eliminated DataFrame fragmentation warning and removed debug statements

Bug Fixes

• DataFrame Fragmentation Warning: Fixed PerformanceWarning: DataFrame is highly fragmented in transform() method
  • Root cause: Iteratively adding columns to DataFrame with woe_df[col] = woe_values caused memory fragmentation
  • Solution: Collect all WOE columns in a dictionary first, then create DataFrame in one operation
  • Performance improvement: Eliminates repeated memory reallocation during transform
  • User impact: No more annoying performance warnings when transforming data
• Debug Print Statement: Removed leftover debug print("FAISS is available:", faiss) statement in FAISS KMeans binning
  • Cleaned up console output when using binning_method='faiss_kmeans'
• Code Quality: Improved transform method efficiency following pandas best practices

Technical Details

• Changed from: for col in columns: woe_df[col] = values (causes fragmentation)
• Changed to: woe_columns = {col: values for col in columns}; woe_df = pd.DataFrame(woe_columns) (single allocation)
• Follows pandas recommendation to use pd.concat(axis=1) or dict-based DataFrame construction

Testing

• All 102 tests passing successfully ✅
• Verified no fragmentation warnings with multi-feature datasets
• Backward compatible: transform output unchanged

Installation

pip install fastwoe==0.1.5

See CHANGELOG.md for full details.

v0.1.5rc1

Clean API refactoring and Pythonic input handling.

• Smart feature naming from monotonic constraints
• 1D array support with auto-reshaping
• Eliminated conversion warnings
• All 161 tests passing

v0.1.5a1: Multiclass WOE Refactoring & Bug Fixes

🎉 Release v0.1.5a1: Multiclass WOE Refactoring & Bug Fixes

🏗️ Code Organization

• Multiclass Separation: Extracted multiclass functionality into fastwoe_multiclass.py
  • Created MulticlassWoeMixin for clean separation of concerns
  • Reduced main file complexity by ~300 lines
• Type Stubs: Added typings/ folder for better IDE support

🐛 Multiclass WOE Bug Fixes

• Fixed missing X parameter causing zero counts and divide-by-zero warnings
• Fixed array flattening issue causing NaN values
• Fixed WOE calculation formula (removed incorrect epsilon)
• Fixed probability prediction logic with proper simple-vs-composite hypothesis (Good, 1950)
• Fixed double sigmoid application and loop storage issues

🔧 Monotonic Constraints Fix

• Fixed infinity symbol parsing in bin labels (-∞, ∞)
• Proper handling of infinite bounds in isotonic regression

✅ Testing & Quality

• All 156 tests passing
• 73% test coverage
• Type checking clean (ty check)
• Linting clean (ruff check & ruff format)

📚 Documentation

• Updated README with multiclass monotonic constraints support
• Updated examples: fastwoe_multiclass.ipynb, fastwoe_monotonic.ipynb
• Added comprehensive multiclass WOE guide

See CHANGELOG.md for full details.

Release v0.1.4.post2: Monotonic Constraints Support

Monotonic Constraints Support

This release adds comprehensive monotonic constraints support across all binning methods, essential for credit scoring compliance and regulatory requirements.

New Features:

• Monotonic Constraints: Complete implementation across all binning methods
• Tree Method: Native scikit-learn monotonic constraints
• KBins Method: Isotonic regression post-processing
• FAISS KMeans Method: Isotonic regression post-processing
• Constraint Values: 1 (increasing), -1 (decreasing), 0 (no constraint)
• Comprehensive Testing: Extensive test coverage for all scenarios
• Enhanced Documentation: Updated README and examples

Technical Details:

• API: Added monotonic_cst parameter to FastWoe constructor
• Validation: Comprehensive input validation with clear error messages
• Binning Info: Constraints stored in binning_info_ and displayed in summaries
• Isotonic Regression: Uses scikit-learn's IsotonicRegression for KBins/FAISS
• Backward Compatibility: Fully compatible with existing code

Examples:

• New: examples/fastwoe_monotonic.py - Comprehensive demonstration
• Shows all binning methods with constraints
• Compares KBins strategies (uniform, quantile, kmeans)
• Analyzes monotonic patterns and performance

Bug Fixes:

• Fixed FAISS API compatibility issues
• Fixed type checking configuration
• Improved exception handling in tests

Installation: pip install fastwoe==0.1.4.post2

Release v0.1.4: Multiclass Support & Enhanced Tree Binning

Major Features

🎯 Multiclass WOE Encoding

• One-vs-rest approach for targets with 3+ classes
• Automatic detection of multiclass vs binary targets
• Multiple output columns per feature for multiclass scenarios
• Class-specific methods: predict_proba_class(), predict_ci_class()
• Support for string/integer labels in multiclass scenarios

🌳 Enhanced Tree Binning

• Decision tree as default binner for numerical features
• Fixed NaN values in the last bin for numerical features
• Optimized default parameters: max_depth=3, random_state=42
• Unified binner_kwargs API for consistent parameter passing

Improvements

🔧 Bug Fixes

• Fixed NaN values appearing in the last bin for numerical features
• Improved error handling and validation
• Enhanced type checking with comprehensive configuration

📚 Documentation & Examples

• Updated README with multiclass examples and API documentation
• Added comprehensive multiclass example (fastwoe_multiclass.py)
• Enhanced type checking configuration for examples
• Updated CHANGELOG with detailed release notes

🧪 Testing & Quality

• Fixed all type checking issues across core library, tests, and examples
• Added comprehensive test coverage for multiclass functionality
• Enhanced CI/CD pipeline with proper type checking

Breaking Changes

• Decision tree is now the default binner for numerical features (was KMeans)
• Multiclass targets automatically detected and handled with one-vs-rest encoding

Migration Guide

If you were using KMeans binning explicitly, update your code:

# Old (explicit KMeans)
encoder = FastWoe(numerical_binner='kmeans')

# New (default decision tree, or explicit KMeans)
encoder = FastWoe()  # Uses decision tree by default
encoder = FastWoe(numerical_binner='kmeans')  # Still works

Performance Improvements

• Faster numerical binning with optimized decision tree parameters
• Improved memory efficiency for multiclass scenarios
• Enhanced parallel processing for large datasets

---

Full Changelog: v0.1.3...v0.1.4

Release v0.1.3.post1: Enhanced Statistical Analysis & Developer Experience

Enhanced Statistical Analysis & Developer Experience

• IV Standard Errors: Add confidence intervals and significance testing for Information Value calculations
• Series Input Support: Fix 'Series' object has no attribute 'columns' error
• Type Checking Integration: Integrate pyrefly type checking with CI workflows
• FAISS GPU Support: Add fastwoe[faiss-gpu] optional dependency and improve dependency management
• Improved UX: Change default warn_on_numerical=False (numerical binning is now core feature)
• Local Testing: Add act integration for local GitHub Actions testing
• Enhanced Documentation: Comprehensive guides and examples with improved mathematical framework
• Development Automation: Add Make commands for streamlined workflows

Release v0.1.2.post4

Changes

• See CHANGELOG.md for detailed changes

Installation

pip install fastwoe==v0.1.2.post4

Release v0.1.2.post3

Changes

• See CHANGELOG.md for detailed changes

Installation

pip install fastwoe==v0.1.2.post3