Changelog

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

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.0.0] - 2026-01-20

Added

Core Models

• GradientBoosting - Standard gradient boosting for regression/classification
• MultiClassGradientBoosting - Multi-class classification with softmax
• DART - Dropout regularized trees
• OpenBoostGAM - GPU-accelerated interpretable GAM

Distributional Models (NaturalBoost)

• NaturalBoostNormal - Gaussian distribution
• NaturalBoostLogNormal - Log-normal for positive values
• NaturalBoostGamma - Gamma distribution
• NaturalBoostPoisson - Count data
• NaturalBoostStudentT - Heavy tails
• NaturalBoostTweedie - Insurance claims (Kaggle favorite)
• NaturalBoostNegBin - Sales forecasting (Kaggle favorite)

Advanced Features

• LinearLeafGBDT - Linear models in tree leaves
• GPU acceleration via Numba CUDA
• Multi-GPU support via Ray
• GOSS sampling (LightGBM-style)
• Mini-batch training for out-of-core datasets
• Memory-mapped array support

sklearn Integration

• OpenBoostRegressor - sklearn-compatible regressor
• OpenBoostClassifier - sklearn-compatible classifier
• OpenBoostDistributionalRegressor - Distributional regressor
• OpenBoostLinearLeafRegressor - Linear leaf regressor

Callbacks

• EarlyStopping - Stop training when validation metric stops improving
• Logger - Print training progress
• ModelCheckpoint - Save best models during training
• LearningRateScheduler - Dynamic learning rate

Loss Functions

• MSE, MAE, Huber, Quantile (regression)
• LogLoss, Softmax (classification)
• Poisson, Gamma, Tweedie (count/positive data)
• Custom loss function support

Growth Strategies

• Level-wise (XGBoost-style)
• Leaf-wise (LightGBM-style)
• Symmetric/Oblivious (CatBoost-style)

Utilities

• compute_feature_importances() - Gain-based importance
• suggest_params() - Automatic parameter suggestions
• cross_val_predict() - Out-of-fold predictions
• evaluate_coverage() - Prediction interval validation

Performance

• NaturalBoost: 1.3-11x faster than NGBoost
• OpenBoostGAM: 25-43x faster than InterpretML EBM on GPU
• Standard GBDT: Comparable to XGBoost (within 5% RMSE)

Documentation

• Comprehensive README with examples
• Quickstart guide
• Uncertainty quantification tutorial
• Custom loss function tutorial
• XGBoost migration guide
• 13 runnable examples

Development Phases

This release represents 22 phases of development:

• Phase 1-7: Core implementation
• Phase 8-9: Growth strategies and loss functions
• Phase 10-11: Feature importance and custom loss
• Phase 12-14: Callbacks, sklearn integration, regularization
• Phase 15-16: Distributional GBDT (NaturalBoost)
• Phase 17-18: Large-scale training, multi-GPU
• Phase 19-21: Integration testing, CUDA verification
• Phase 22: Pre-launch QA