Skip to content

Trainer

Overview

Key features

  • Highly configurable: Trainer provides unified API and configs to train and evaluate predictive models using various ML libraries (XGBoost, PyTorch, RLLib)
  • Integrated with Featurizer: Use distributed in-memory FeatureLabelSet data structure to train your predictive models without extra data pipelines
  • Scalability: Data-parallel distributed training for all supported frameworks
  • Hyperparameter optimization: Use Ray Tune to optimize hyperparameters and pick the best model
  • Model and metadata storage: Trainer provides easy API for model access and metadata discovery by integrating with MLFlow