How-To Guides

Task-oriented guides with code snippets for common operations.

• Integrating with sklearn Transformers
  • Accept JSON input
  • Accept DataFrame input
  • Scale numeric features
  • Preprocess enriched features in a shared-learner bandit
  • Access pipeline internals
• Choosing and Tuning a Decay Rate
  • Start with no decay
  • Decouple decay from updates
  • Choose a decay rate
  • Avoid over-decay
• Writing Custom Reward Functions
  • Map binary outcomes to profit
  • Use context in the reward function
  • Express non-linear utility
  • Batch reward functions for shared-learner bandits
  • If something goes wrong
• Handling Delayed Rewards
  • The basic pattern
  • Batch updates for the same arm
  • Multiple pulls before any update
  • Interaction with decay
  • Non-contextual agents
  • If something goes wrong
• Deploying to Production
  • Serialize with joblib
  • Reseed the RNG after loading
  • Add and remove arms at runtime
• Working with Sparse Features
  • Enable sparse mode
  • Install CHOLMOD for production workloads