RFCs

Boosters is developed using a Request for Comments (RFC) process. Each major feature or design decision is documented before implementation.

RFCs describe the “why” and “how” of implementation decisions.

Core

• RFC-0001: Dataset
• RFC-0002: Tree and Forest Representation
• RFC-0016: Model Serialization

Preprocessing

• RFC-0003: Binning and Quantized Data Storage
• RFC-0004: Exclusive Feature Bundling (EFB)
• RFC-0012: Categorical Features

Training

• RFC-0005: Objectives, Metrics, and Output Transforms (vNext)
• RFC-0006: Sampling Strategies
• RFC-0007: Histogram Building and Optimization
• RFC-0008: GBDT Training

Inference & Analysis

• RFC-0009: GBDT Inference
• RFC-0013: Explainability

Linear Models

• RFC-0010: GBLinear
• RFC-0011: Linear Leaves

Python & Tooling

• RFC-0014: Python Bindings
• RFC-0015: Evaluation Framework
• RFC-0017: Documentation and READMEs

RFC Status Overview

RFC	Title	Status
0001	Dataset	✅ Complete
0002	Trees	✅ Complete
0003	Binning	✅ Complete
0004	Exclusive Feature Bundling	✅ Complete
0005	Objectives & Metrics	✅ Complete
0006	Sampling	✅ Complete
0007	Histograms	✅ Complete
0008	GBDT Training	✅ Complete
0009	GBDT Inference	✅ Complete
0010	GBLinear	✅ Complete
0011	Linear Leaves	✅ Complete
0012	Categorical Features	✅ Complete
0013	Explainability	✅ Complete
0014	Python Bindings	✅ Complete
0015	Evaluation Framework	✅ Complete
0016	Model Serialization	✅ Complete
0017	Documentation	🔄 In Progress

Creating an RFC

1. Copy docs/rfcs/TEMPLATE.md

2. Fill in the template sections

3. Submit as a PR for review

4. Once approved, begin implementation

See the RFC template on GitHub for the full template.