For numerical features and target, how is the partial dependence calculated? From the tip, there is description "The value of the partial dependence is by how much the prediction is higher or lower than average" what is the average here? I tried to calculated several average values but none of them fit to the output.
For any numerical feature F and any given point (X, Y) of the plot, Y is equal to the difference between: * the average of the prediction on (a sample of) the test set; * and the average prediction of the same data where, for all rows, the feature F is set to X.
I'm not sure I understand what you mean by "target" in this context. Partial dependence can not be computed for the model target but only for model input features.