coxnet: Cox regression via penalized maximum likelihood using C++...

In glmnet: Lasso and Elastic-Net Regularized Generalized Linear Models

Cox regression via penalized maximum likelihood using C++ engine

Description

This function calls the C++ implementation of Cox regression with elastic net regularization. It handles both right-censored and left-truncated (start, stop) survival data using the Breslow or Efron method for ties. For stratified Cox models, it uses an IRLS approach with integrated C++ gradient/Hessian computation.

Usage

coxnet(
  x,
  is.sparse,
  y,
  weights,
  offset,
  alpha,
  nobs,
  nvars,
  jd,
  vp,
  cl,
  ne,
  nx,
  nlam,
  flmin,
  ulam,
  thresh,
  isd,
  vnames,
  maxit,
  pb,
  efron = FALSE
)

Arguments

x	Design matrix, of dimension nobs x nvars.
is.sparse	Logical, is x a sparse matrix?
y	Survival response variable, must be a Surv or stratifySurv object.
weights	Observation weights.
offset	Offset for the linear predictor.
alpha	The elastic net mixing parameter.
nobs	Number of observations.
nvars	Number of variables.
jd	Excluded variable indices (1-indexed, first element is count).
vp	Penalty factors for each coefficient.
cl	Coefficient limits matrix (2 x nvars).
ne	Maximum number of variables in the model.
nx	Maximum number of variables ever to be nonzero.
nlam	Number of lambda values.
flmin	Minimum lambda ratio.
ulam	User-supplied lambda sequence.
thresh	Convergence threshold.
isd	Standardize flag.
vnames	Variable names.
maxit	Maximum number of iterations.
pb	Progress bar object.
efron	Logical; if TRUE use Efron method for ties, otherwise Breslow.

Value

An object of class "coxnet" with components:

a0	NULL (Cox model has no intercept)
beta	Sparse coefficient matrix
df	Number of nonzero coefficients per lambda
dim	Dimension of coefficient matrix
lambda	Lambda sequence used
dev.ratio	Fraction of null deviance explained
nulldev	Null deviance
npasses	Number of coordinate descent passes
jerr	Error code
offset	Logical indicating if offset was used

glmnet documentation built on May 4, 2026, 9:09 a.m.