4. Inspection

Predictive performance is often the main goal of developing machine learningmodels. Yet summarising performance with an evaluation metric is ofteninsufficient: it assumes that the evaluation metric and test datasetperfectly reflect the target domain, which is rarely true. In certain domains,a model needs a certain level of interpretability before it can be deployed.A model that is exhibiting performance issues needs to be debugged for one tounderstand the model’s underlying issue. Thesklearn.inspection module provides tools to help understand thepredictions from a model and what affects them. This can be used toevaluate assumptions and biases of a model, design a better model, orto diagnose issues with model performance.

• 4.1. Partial dependence plots
• 4.2. Permutation feature importance
  • 4.2.1. Relation to impurity-based importance in trees
  • 4.2.2. Strongly correlated features