A Partial Dependence Plot (PDP) is an instrument used to help understanding the relationship between a feature F and the output y of a model. Given a sample S, all features except F are frozen, and F is varied through the range seen for all samples, while capturing the outputs y. This builds an Individual Conditional Expectation (ICE) plot for the sample S. Averaging such plots for all samples yields the Partial Dependence Plot for feature F.