sgsl: Fit the sparse group-subgroup lasso (SGSL)

In rakheon/sgslasso: Sparse group-subgroup lasso for identification of important regressor groups, subgroups and individuals

Description

Fit the sparse group-subgroup lasso (SGSL)

Usage

sgsl(
  x,
  y,
  type = c("lasso", "group", "ggroup", "ggroupind", "sgsl", "groupsgl")[1],
  index.subgroup,
  tau = NULL,
  delta = 2,
  delta.group = 2,
  delta.subgroup = 2,
  delta.ind = 2,
  standardize = TRUE
)

Arguments

x	p by N matrix of predictors (N: sample size, p: number of predictors)
y	1 by N matrix of response variable
type	One of "lasso" (for the standard lasso), "group" (for the group lasso), "ggroup" (for the group lasso among subgroups), "ggroupind" (for the lasso with individual features), "sgsl" (for the sparse-group-subgroup lasso) or "groupsgl (for the sparse group lasso at subgroup level).
index.subgroup	index for subgroups
tau	multiplier for using a multiplicative grid for penalty parameter lambda, starting at maximal lambda value
delta	Among the lasso solution path, the best descriptive model is the one which minimizes the loss function: (residual sum of squares)/(estimator of the model error variance) - (sample size) + delta*(number of predictors in the selected model). If delta = 2, this loss function is Mallows' Cp.
delta.group	delta applied to C_p criterion for group lasso
delta.subgroup	delta applied to C_p critierian for group lasso among subgroups
delta.ind	delta applied to C_p criterion for lasso with individual features
standardize	logical. TRUE for standardizing the data.

Value

out: indicators of the selected predictors. 1 for selected predictors and 0 for not selected predictors

Examples

set.seed(1)
x <- matrix(rnorm(360), nrow=12)
y <- 0.5*x[1,] + 0.5*x[2,] + 1.0*x[4,] + matrix(rnorm(30), nrow=1)
index.subgroup <- matrix(NA,nrow=3,ncol=12)
index.subgroup[1,1:2]=1; index.subgroup[1,3:4]=2
index.subgroup[2,5:6]=3; index.subgroup[2,7:8]=4
index.subgroup[3,9:10]=5; index.subgroup[3,11:12]=6
out_lasso <- sgsl(x,y,type="lasso",index.subgroup = index.subgroup)
out_group <- sgsl(x,y,type="group",index.subgroup = index.subgroup,tau=0.94)
out_ggroup <- sgsl(x,y,type="ggroup",index.subgroup = index.subgroup,tau=0.94)
out_ggroupind <- sgsl(x,y,type="ggroupind",index.subgroup = index.subgroup,tau=0.94)

rakheon/sgslasso documentation built on March 1, 2020, 11:46 a.m.