GBNet: XGBoost and LightGBM Modules for PyTorch (github.com)

A new framework named GBNet has been introduced, integrating XGBoost and LightGBM with PyTorch for advanced gradient boosting model development. This innovation allows users to leverage PyTorch's automatic differentiation capabilities, enabling them to easily define custom loss functions and complex model architectures without the need for manual gradient and Hessian calculations. This significantly lowers the barrier for researchers and practitioners looking to employ gradient boosting methods in innovative neural network architectures. GBNet comprises three main components—XGBModule, LGBModule, and GBLinear—which facilitate the integration of traditional gradient boosting techniques with PyTorch's deep learning framework. Notable advancements include specialized models for tasks such as forecasting, ordinal regression, and survival analysis, outperforming existing benchmarks like Meta's Prophet algorithm in certain cases. Moreover, GBNet supports joint training of multiple models, enhancing its utility for multi-output scenarios. This framework not only streamlines the model development process but also opens up new avenues for applications in machine learning, particularly in fields requiring advanced statistical modeling and analysis.