stanfordmlgroup/ngboost

What it solves

NGBoost addresses the limitation of traditional boosting algorithms that typically provide point predictions. Instead, it enables probabilistic prediction, allowing models to output a full probability distribution rather than a single value.

How it works

It implements Natural Gradient Boosting, a method that uses the natural gradient to optimize the parameters of a probability distribution. The library is built on top of Scikit-Learn and is designed to be modular, allowing users to choose their own scoring rules, probability distributions, and base learners.

Who it’s for

Data scientists and machine learning practitioners who need to quantify uncertainty in their predictions through probabilistic forecasting.

Highlights

• Built on top of Scikit-Learn for easy integration.
• Modular design supporting various distributions and base learners.
• Scalable implementation of Natural Gradient Boosting.
• Supports both point predictions and full probability distribution outputs.