interaction_search: interaction_search
In gm209/gmtools: A collection of generic ML functions for Insurance Risk Modelling

View source: R/find_interactions_untested.R

This function runs the XGBfi utility on a trained xgboost model to identify interactions to be included in future model interations.

1
2
3

interaction_search(mdl, features, XGBfiBin = "~/Kaggle/xgbfi/bin/",
  filterExp = NULL, XGBfiParams = list(d = 3, g = -1, t = 100, k = 100, h =
  10), intern = FALSE)

`mdl`	A trained xgboost model of class xgb.Booster
`features`	A list of the feature names used in the model (ordering must be consistent with the dmatrix used to train the model)
`XGBfiBin`	The file path to the bin of the XGBfi utility
`filterExp`	A character expression that tells the process how to filter the XGBfi results. e.g. Gain > 2000
`XGBfiParams`	A named list that contains the parameters to be passed to the cli of the XGBfi utility.
`intern`	Passed to the system command (internalise console output or not)