coxstream: Fit a Cox proportional hazards model via streaming...

In coxstream: Memory-Efficient Cox Proportional Hazards via Streaming Newton-Raphson

Fit a Cox proportional hazards model via streaming Newton-Raphson

Description

Fits the Cox PH model using a single descending-time-order pass per Newton-Raphson iteration. Peak RAM is O(p^2) regardless of n, making it suitable for large datasets. Produces identical coefficients to survival::coxph() with Efron tie correction.

Usage

coxstream(
  formula,
  data,
  init = NULL,
  max_iter = 25L,
  tol = 1e-09,
  verbose = FALSE
)

Arguments

formula	A formula with a survival::Surv() response, e.g. Surv(time, event) ~ x1 + x2.
data	A data frame containing the variables in formula.
init	Optional numeric vector of starting values for beta (length p). Defaults to zero.
max_iter	Maximum Newton-Raphson iterations. Default 25.
tol	Convergence tolerance on the max absolute score element. Default 1e-9.
verbose	Currently unused; reserved for future per-iteration output. Default FALSE.

Value

An object of class "coxstream" with components:

coefficients	Named numeric vector of fitted coefficients.
var	Variance-covariance matrix (inverse of observed information).
loglik	Log-likelihood at convergence.
n_iter	Number of NR iterations taken.
n	Number of rows.
formula	The formula used.
call	The matched call.

Examples

library(survival)
fit <- coxstream(Surv(time, status) ~ age + sex, data = lung)
coef(fit)

coxstream documentation built on June 20, 2026, 5:07 p.m.