glmPenaltyFit: Fit the generalized linear model with a vector of penalty...

In PingYangChen/elnglm: Generalized Linear Model with Elastic Net Penalty

Description

Fit the generalized linear model with a vector of penalty parameters.

Usage

glmPenaltyFit(
  y,
  x,
  family = c("gaussian", "binomial", "multinomial"),
  lambdaLength = 100L,
  minLambdaRatio = 0.001,
  lambdaVec = NULL,
  alpha = 0.5,
  standardize = TRUE,
  maxit = 100L,
  tol = 1e-04,
  ver = c("r", "arma")
)

Arguments

y	the vector of the response variable.
x	the matrix of the predictors.
family	string. One of the response families, "gaussian", "binomial" or "multinomial".
lambdaLength	integer. The number of tuning penalty parameters. The default is 100.
minLambdaRatio	double. The ratio of the minimal value to the maximal value of the penalty parameter. The default is 1e-3.
lambdaVec	vector. The optional input of the tuning penalty parameters. The default is NULL that the function automatically computes the maximal value of the penalty parameter and generates a sequence of penalty parameter values of length lambdaLength.
alpha	double. The elastic net parameter between 0 and 1. The default value is 0.5.
standardize	boolean. If TRUE, the function first standardizes the predictor matrix.
maxit	integer. The number of maximal iterations of the coordinate descent algorithm. The default is 100.
tol	double. The value of the convergence tolerance of the coordinate descent algorithm. The default is 1e-4.
ver	string. The version of the coordinate descent engine, "r": R codes or "arma": C++ codes with armadillo library.

Examples

# Generate data of continuous response
trueb0 <- 1
trueact <- c(1, 1, 1, 0, 0, 0, 0, 0, 0, 0)
trueb <- runif(10, -1, 1)*10
trueb[which(trueact == 0)] <- 0 
df <- glmDataGen(n = 500, d = 10, family = "gaussian", trueb0, trueb, s = 0.5, seed = 1)

# Run cross-validation
mdl <- glmPenaltyFit(y = df$y, x = df$x, family = "gaussian", lambdaLength = 100,
                     minLambdaRatio = 1e-3, maxit = 1e5, tol = 1e-7, alpha = 0.5, ver = "arma")
# Estimated intercept of the best model
print(mdl$b0)
# Estimated coefficients of the best model
print(mdl$b)

PingYangChen/elnglm documentation built on Jan. 6, 2022, 5:50 a.m.