Ray Train User Guides

• Data Loading and Preprocessing
  • Quickstart
  • Starting with PyTorch data
  • Splitting datasets
  • Random shuffling
  • Enabling reproducibility
  • Preprocessing structured data
  • Performance tips
• Configuring Scale and GPUs
  • Increasing the number of workers
  • Using GPUs
  • Setting the resources per worker
  • Setting the communication backend (PyTorch)
  • Trainer resources
• Configuring Persistent Storage
  • Cloud storage (AWS S3, Google Cloud Storage)
  • Shared filesystem (NFS, HDFS)
  • Local storage
  • Custom storage
  • Overview of Ray Train outputs
  • Advanced configuration
• Monitoring and Logging Metrics
  • How to obtain and aggregate results from different workers?
• Saving and Loading Checkpoints
  • Saving checkpoints during training
  • Configure checkpointing
  • Using checkpoints after training
  • Restore training state from a checkpoint
• Experiment Tracking
  • Getting Started
  • Examples
  • Common Errors
• Inspecting Training Results
  • Viewing metrics
  • Retrieving checkpoints
  • Accessing storage location
  • Viewing Errors
  • Finding results on persistent storage
• Handling Failures and Node Preemption
  • Automatically Recover from Train Worker Failures
  • Restore a Ray Train Experiment
• Reproducibility
• Hyperparameter Optimization
  • Key Concepts
  • Basic usage
  • How to configure a Tuner?
  • Search Space configuration
  • Train - Tune gotchas
  • Advanced Tuning