R Graphical Manual

Browse All

Last data update: 2014.03.03

R: Linear regression with variable selection

lmsel

R Documentation

Linear regression with variable selection

Description

lmsel is used to fit linear models with optionally performing variable selection using stepwise regression, lasso or elastic net methods.

Usage

lmsel(formula, data=environment(), varsel=FALSE, criterion="AIC",
direction="backward", indices=NULL, train=0.3, lambda=1000)

Arguments

`formula`	an object of class "formula"; a symbolic description of the model to be fitted.
`data`	an optional data frame, list or environment containg the variables in the model. If not specified, the variables are taken from the current environment.
`varsel`	a method of variable selection to be used. The default is `"FALSE"`. Available methods include: stepwise regression `"step"`, LASSO `"lasso"`, elastic net`"enet"`.
`criterion`	when `varsel="step", criterion` allows to select a method of calculating statistic for model comparison. The default is `"AIC"`. Less liberal, BIC penalty can be used by typing `"BIC"`.
`direction`	the mode of stepwise search, can be one of `"both"`, `"backward"`, or `"forward"`, with a default of `"both"`. If the scope argument is missing the default for `direction` is `"backward"`.
`indices`	vector of `0` and `1` values indicating which observations are to be used as train and test when `varsel="lasso" or "enet"`.
`train`	if `indices=NULL`, the function will randomly assign observations as train and test. `train` specifies what percentage of data will be used as train observations. Can take values from `0.1` to `0.9`.
`lambda`	quadratic penalty parameter for elastic net. The default value is `1000`.

Value

A "lmsel" object is returned, for which print, plot and summary methods can be used.

Author(s)

Michal Knut 1105406k@student.gla.ac.uk.

Examples

data(prostate)
set.seed(10)
lmsel(lpsa~lcavol+lweight+age+lbph+svi+lcp+gleason+pgg45, indices=as.numeric(prostate$train),
data=prostate, varsel="lasso")

data(concrete)
lmsel(CompressiveStrength~., data=concrete, varsel="step", criterion="BIC")