intnet: Cross validation interaction screening

In jaredhuling/intscreen:

Description

This function implements CV interaction screening

Usage

intnet(x, y, which.cols = 1:ncol(x), nints = 100,
  heredity = c("none", "weak", "strong"),
  resample.type = c("bootstrap", "cv"), nsplits = 10,
  train.frac = 0.75, fraction.in.thresh = 1, verbose = FALSE,
  modifier = NULL, ...)

Arguments

x	matrix of predictors
y	vector of observations
which.cols	integer vector indicating which columns of x to check for interactions
nints	integer number of top interactions to screen
heredity	either "weak", "strong", or "none"
resample.type	either "cv" for cross validation or "bootstrap" for bootstrap approach
nsplits	integer number of cross validation splits to run. defaults to 10
train.frac	fraction of data used for each split. defaults to 0.75. Only used for resample.type = "cv"
fraction.in.thresh	fraction of times across the nsplits CV splits each interaction is required in the top nints interactions in order to be selected
modifier	effect modifier
...	arguments to be passed to glmnet

Examples

library(intnet)

set.seed(1)
x <- matrix(rnorm(100 * 350), ncol = 350)
y <- rnorm(100) + x[,1] * x[,2] * 0.5 - x[,3] * x[,4] * 0.5


## require that each interaction be in the top 50 ints 100% of the 15 splits
intmod <- intnet(x, y, nints = 50, nsplits = 15, fraction.in.thresh = 1)

plot(intmod)

jaredhuling/intscreen documentation built on May 19, 2019, 9:38 p.m.