Fast Weibull AFT Regression (R Wrapper: Rcpp Backend or survival)

Fits the Weibull accelerated failure time model documented in full at fast_weibull_regression_general_cpp (\(\log T_i = \eta_i + \sigma W_i\), \(\eta_i = x_i^\top\beta\), right-censoring only), via either that C++ backend (use_rcpp = TRUE, the default) or survreg with dist = "weibull" (use_rcpp = FALSE) as a fallback/cross-check implementation.

Usage

fast_weibull_regression(
  y,
  dead,
  X,
  use_rcpp = TRUE,
  estimate_only = FALSE,
  optimization_alg = "lbfgs",
  warm_start_params = NULL,
  warm_start_fisher_info = NULL
)

Arguments

y: Observed survival/censoring times (must be positive).
dead: Event indicator: 1 for an exactly observed event, 0 for right-censored (survival known only to exceed y[i]).
X: A numeric matrix of predictor variables. It is assumed that an intercept column (e.g., a column of ones) is already included in X if desired.
use_rcpp: Logical. If TRUE (default), use the optimized Rcpp implementation. If FALSE, use survreg.
estimate_only: Logical. If TRUE, skip variance-covariance matrix calculation for speed. Only has an effect when use_rcpp = TRUE.
optimization_alg: Optimization algorithm: "lbfgs" (default) or "newton_raphson". Only has an effect when use_rcpp = TRUE.
warm_start_params: Optional starting values for \([\beta, \log\sigma]\). Only has an effect when use_rcpp = TRUE.
warm_start_fisher_info: Optional initial curvature (Fisher/observed information) matrix. Only has an effect when use_rcpp = TRUE.

Value

A list containing the following components:

coefficients: A numeric vector of the estimated Weibull regression coefficients \(\hat\beta\), including the intercept.
log_sigma: The logarithm of the fitted scale parameter \(\hat\sigma\) of the Weibull AFT distribution.
vcov: The variance-covariance matrix of the estimated coefficients, or NULL if estimate_only = TRUE (Rcpp path only).
neg_log_lik: (Rcpp path only) The negative log-likelihood at the fitted parameters.
fisher_information: (Rcpp path only) The observed information matrix, or NULL if estimate_only = TRUE.
std_errs: (survival path only) The coefficient standard errors, sqrt(diag(vcov)).

Details

When use_rcpp = TRUE, an intercept column is prepended to X automatically if not already present (detected as a first column of all 1s), fitting always starts from a zero cold start (smart_cold_start = FALSE is hardcoded, regardless of whether warm_start_params is supplied), and a non-converged C++ fit is escalated to an R-level stop() rather than returned silently.

When use_rcpp = FALSE, any existing intercept column is stripped and survreg is left to add its own; remaining covariate columns are first passed through drop_linearly_dependent_cols to remove collinear columns before fitting (silently — no error or warning is raised for dropped columns). estimate_only, optimization_alg, warm_start_params, and warm_start_fisher_info have no effect on this path — survreg always computes the full variance-covariance matrix, and std_errs (from sqrt(diag(vcov))) is included only in this path's return value, not the Rcpp path's. Both non-finite coefficients and (unless estimate_only = TRUE, which is ignored on this path regardless) non-finite variance-covariance entries from survreg are escalated to an R-level stop().

Examples