📦 Changelog - RegressionMadeSimple

View the project on GitHub

• Model Registry Pattern: Access models via rms.models.Linear, rms.models.Quadratic, rms.models.Cubic, and rms.models.CustomCurve
• Enhanced Base Class: Completely refactored base class with common functionality for all models
• Model Serialization: Save and load trained models with save_model() and load_model() methods using joblib
• Additional Scoring Metrics: New methods including r2_score(), mae(), and rmse() for comprehensive model evaluation
• Class-based Model Specification: New recommended API using model classes instead of strings (e.g., model=rms.models.Linear)

• Code Reduction: Eliminated ~60% of duplicate code across model classes through shared base functionality
• Better Architecture: Centralized common train/test split logic, plotting, and metric computation in BaseModel
• Type Hints: Added comprehensive type hints throughout the codebase for better IDE support
• Improved Documentation: Comprehensive README with migration guide and advanced examples
• Better Organization: Models now live in dedicated models/ submodule for cleaner structure

• New Recommended API: Use class-based model specification: model=rms.models.Linear instead of model='linear'
• Backward Compatible: Old string-based API still works but shows deprecation warnings
• Migration Required: String-based model specification will be removed in v4.0.0. See Migration Guide to v3.0.0

• Rewrote the whole package. More info please consult the Migration Guide and more notes.

• Added plotting compatibility for Matplotlib
• Added a early Logging class (but did not make any log infos in the actual code (TODO))

• Small code improvements to ComplexCurves. (Still developing, use at your own risk)

• Added support for ComplexCurves (early dev version)

• Added Quadratic and Cubic regression models with clean, user-friendly APIs
• Introduced wrapper.py with LinearRegression.fit(...) interface
• Models support both auto and pre-split data with RMS-style constructor args

• Refined package structure for better modularity and clarity
• Updated __init__.py to expose all new models and wrapper interfaces
• Reinforced options system to support future model configurations

• 🔧 Introduced rms.options for global configuration using SimpleNamespace
• 🧠 Added support for .save(), .load(), and .reset() for RMS options
• 🧪 Configurable Linear constructor: honors rms.options.training + accepts pre-split data

• Internal logic streamlined to use config fallbacks
• Better defaults, simplified structure for future models
• Updated __init__.py to expose config functions

• Added .summary() method to Linear model
• Support for pre-split data in Linear constructor via X_train, y_train, X_test, y_test
• Cleaner, more flexible preworks utilities

• Streamlined logic in Linear and preworks using real-world usage experience
• Renamed/reorganized function names for clarity
• Improved error handling for .plot() and .mse() when test data is unavailable