Machine learning system design patterns

This repository contains the material for the book: Machine learning system design patterns.

Topics

• Common training workflow structure
• Decoupling feature creation from training & serving workflow
• Embed pre & post processing logic in the model
  • Embedded in the model object
  • Separate task of the pipeline (batch-inference) (TODO)
• Model registry
• Serving patterns
  • Batch prediction (TODO)
  • Online serving (TODO)
    • Online serving structure
  • Event-based serving (TODO)
• Data validation for inference
  • Batch inference
  • Online inference
  • Distribution skew versus drift
  • Metrics to determine distribution skew / drift
• Infrastructure setup
  • Sandbox, staging & production
  • Promoting between staging & production

Render the Markdown

In order to render the markdown to HTML first install mdbook & mdbook-mermaid package:

cargo install mdbook
cargo install mdbook-mermaid

To build the book, run:

mdbook-mermaid install
mdbook build

To continually serve the book, run:

mdbook serve