4. Inspection

Predictive performance is often the main goal of developing machine learning models. Yet summarizing performance with an evaluation metric is often insufficient: it assumes that the evaluation metric and test dataset perfectly 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 to understand the model’s underlying issue. The sklearn.inspection module provides tools to help understand the predictions from a model and what affects them. This can be used to evaluate assumptions and biases of a model, design a better model, or to diagnose issues with model performance.

Examples

• Common pitfalls in the interpretation of coefficients of linear models

• 4.1. Partial Dependence and Individual Conditional Expectation plots
  • 4.1.1. Partial dependence plots
  • 4.1.2. Individual conditional expectation (ICE) plot
  • 4.1.3. Mathematical Definition
  • 4.1.4. Computation methods
• 4.2. Permutation feature importance
  • 4.2.1. Outline of the permutation importance algorithm
  • 4.2.2. Relation to impurity-based importance in trees
  • 4.2.3. Misleading values on strongly correlated features