ggInteract_boot: ggInteract_boot

In JBjouffray/ggBRT: Explore and Visualise the Results of Boosted Regression Trees

Description

Function to randomize the response n times to generate a distribution under the null hypothesis of no interaction among predictors for one or several pairs of predictors (following Pinsky and Byler 2015). Model parameters should be the same as for the observed interaction strength. Results allow to compare with observed interaction strength.

Usage

ggInteract_boot(..., nboots = 100, data = NULL, response = NULL,
  predictors = NULL, tc = 3, lr = 0.005, bf = 0.75,
  family = "bernoulli", global.env = T)

Arguments

...	one or several pair of interacting predictors as vectors (e.g. c(pred1, pred2), c(pred3,pred4))
nboots	the number of randomization (default=100)
data	the dataset for the boosted regression trees (BRT) (default=NULL)
response	the response variable of the BRT (default=NULL)
predictors	the set of predictors for the BRT (default=NULL)
tc	tree complexity for the BRT (default=3)
lr	learning rate for the BRT (default=0.005)
bf	bag fraction for the BRT (default=0.75)
family	the distribution family for the BRT (default=“bernoulli”)
global.env	Logical. If TRUE (default) export the result into the R global environment as an object called Interact_boots

Value

Returns a dataframe of interaction sizes with as many columns as there are pairs of predictors (+1) and as many rows as the number of randomizations.

JBjouffray/ggBRT documentation built on Jan. 30, 2021, 8:13 a.m.