Architecture

Overview of the boosters codebase architecture.

Crate Structure

crates/
└── boosters/               # Core library
    ├── src/
    │   ├── data/           # Dataset, binning, histograms
    │   ├── model/          # Tree/linear model storage
    │   ├── training/       # Training algorithms
    │   ├── inference/      # Prediction
    │   ├── objective/      # Loss functions
    │   └── explainability/ # SHAP, feature importance

packages/
├── boosters-python/        # PyO3 bindings
│   ├── src/                # Rust → Python wrappers
│   └── python/boosters/    # Pure Python code (sklearn)
└── boosters-eval/          # Benchmarking
    └── src/boosters_eval/  # CLI and evaluation framework

Key Design Decisions

See the RFCs section for detailed RFCs on:

• Data representation and binning

• Tree storage (Structure-of-Arrays)

• Histogram-based training

• Parallel processing strategy

Contributing Guidelines

For detailed contribution guidelines, see the Development Setup section.