TorchEnsemble-Community/Ensemble-Pytorch

Supports nine ensemble strategies spanning parallel (voting, bagging, adversarial training), sequential (gradient boosting, snapshot ensemble), and mixed architectures for both classification and regression tasks. Provides a unified API with built-in optimizer and scheduler management, allowing users to wrap any PyTorch model as a base estimator and train heterogeneous ensembles with minimal code. Implements both classical ensemble methods and recent deep learning variants like snapshot ensembles and fast geometric ensembling.

1,044 stars. No commits in the last 6 months.