9. Ray Machine Learning

This chapter will focus on Ray machine learning, mainly introducing libraries such as Ray Train and Ray Tune. These libraries are based on Ray’s distributed computing capabilities, are targeted at machine learning and deep learning applications, and deeply integrate machine learning libraries such as PyTorch, TensorFlow, Hugging Face Transformers, XGBoost scikit-learn, etc.

• 9.1. Ray Train
  • Key steps
  • Example: Image Classification
  • Difference from native PyTorch
    • Difference from stand-alone program
    • Difference from DistributedDataParallel
  • Data reading
  • ScalingConfig
  • Monitoring
  • Checkpoint
• 9.2. Ray Tune
  • Key components
  • Trainable
    • Example: Image classification
  • Search Space
  • Search algorithm and scheduler
  • Example: Airplane Delay Prediction
  • Example: Image classification based on PBT
• 9.3. Ray Serve
  • Key concepts
  • Example: Large language model reasoning
    • Server
    • Client
    • Deployment parameters
    • Configuration file