Overview

• This chapter focuses on partial dependence plots (PDP), also referred to as partial- dependence (PD) profiles.
  
• Extremely popular technique as global model explainer
  
• Available in multiple packages such as DALEX, iml, and pdp, and PDPbox
  
• Core idea: PDP plots show how model predictions change as a function of an explanatory variable
  
• Can be produced used on all observations or focusing on key subsets
  
• Useful for comparing multiple models:
  • Agreement between profiles establishes confidence
  • Disagreement may suggest a way to improve a model
    
  • Evaluation of model performance at boundaries