Welcome to GBRL’s documentation!

GBRL is a Python-based Gradient Boosting Trees (GBT) library, similar to popular packages such as XGBoost , CatBoost , but specifically designed and optimized for reinforcement learning (RL). GBRL is implemented in C++/CUDA aimed to seamlessly integrate within popular RL libraries.

Feature Support Matrix

The following table summarizes feature availability by tree type and device:

Feature	Greedy CPU	Greedy GPU	Oblivious CPU	Oblivious GPU
Tree Fitting	✓	✓	✓	✓
Monotonic Constraints	✗	✗	✓ (SGD only)	✓ (SGD only)
Linear LR Scheduler	✓	✗	✓	✓
Constant LR Scheduler	✓	✓	✓	✓
ADAM Optimizer	✓	✗	✓	✗
SGD Optimizer	✓	✓	✓	✓
Control Variates	✓	✗	✓	✗

Note

Monotonic constraints apply to the output dimensions covered by the optimizer (start_idx to stop_idx-1). They require SGD: GBRL raises a ValueError if any optimizer on the model uses Adam, and MultiGBTLearner does not support them at all. See Examples for details.

User Guide:

• Quickstart
  • Prerequisites
  • Installation
• Source Compilation
  • Prerequisites
  • Platform-specific Dependencies
  • Installation
• Examples
  • Basic Usage: Training, Saving, Loading, Copying
  • RL using GBRL
  • Explainability
  • Learning Rate Schedulers
  • Monotonic Constraints