User Guide

This section provides practical guidance on how to get started with cuML in your own projects to run classic ML algorithms blazingly fast on NVIDIA GPUs. You can simply read through the examples or try them out yourself. If you want to run code directly, have a look at the notebooks within the notebooks as part of the cuML GitHub repository instead.

Note

This guide describes how to use cuML’s GPU-accelerated estimators and functions directly within your own code. See the the next section if you want to accelerate existing scikit-learn, UMAP, and HDBSCAN code with zero code changes.

• Training and Evaluating Machine Learning Models
  • Classification
  • Clustering
  • Regression
• Model Serialization and Persistence
  • Single GPU Model Serialization
  • Using joblib for Model Serialization
  • Distributed Model Serialization
  • Converting Between cuML and scikit-learn Models
  • Exporting Random Forest Models for CPU-Only Deployment
• Supported Versions
  • Required Runtime Dependencies
  • Dask Runtime Dependencies
  • Optional Runtime Dependencies
  • RAPIDS Dependencies