relative.influence: Relative influence of predictors in an xgbm model
In harrysouthworth/mhdm: Medical History Data Mining

Relative influence of predictors in an xgbm model

1	relative.influence(x)

`x`	An object of class 'xgbm'.

Relative influence is referred to as relative importance in the xgboost documentation, but I prefer 'influence' because it keeps the idea distinct from Breiman's variable importance. Also, it should be computed as the sum over all splits on a variable (according to the xgboost documentation), but isn't when there are factors in the model. So here we add the values associated with factors back together. Finally, the documentation for xgb.importance (for which this is a wrapper) states that you need to give it an integer vector of tree numbers indexed at 0. This function takes care of it all by assuming you used cross-validation.

harrysouthworth/mhdm documentation built on Feb. 4, 2022, 12:25 a.m.