Cross-validated tuning for coxtrans

Selects penalty parameters by minimising the held-out partial likelihood deviance over a full lambda1 x lambda2 x lambda3 grid. Supports both the lambda.min rule (minimum CV deviance) and the lambda.1se rule (most sparse model within one standard error of the minimum).

Usage

cv.coxtrans(
  formula,
  data,
  group,
  target,
  prior_matrix = NULL,
  nlambda = 10,
  lambda.min.ratio = 0.001,
  nfolds = 10,
  penalty = c("lasso", "MCP", "SCAD"),
  ncores = 1,
  nthreads = 1L,
  seed = 42,
  verbose = FALSE,
  ...
)

# S3 method for class 'cv.coxtrans'
coef(object, s = c("lambda.min", "lambda.1se"), ...)

# S3 method for class 'cv.coxtrans'
predict(object, s = c("lambda.min", "lambda.1se"), ...)

Arguments

formula

A formula with a Surv response.

data

A data frame.

group

A factor indicating group membership.

target

The target group level.

prior_matrix

Optional G x (K-1) transfer constraint matrix.

nlambda

Number of lambda values per dimension. Default 10.

lambda.min.ratio

Smallest/largest lambda ratio. Default 1e-3.

nfolds

Number of CV folds. Default 10.

penalty

"lasso", "MCP", or "SCAD".

ncores

Number of R worker processes for parallel grid evaluation. Default 1.

nthreads

Number of OpenMP threads per coxtrans() call. Default 1. Total CPU usage is ncores x nthreads.

seed

Random seed.

verbose

Print progress.

...

Passed to coxtrans.

object

A cv.coxtrans object.

s

Which model to predict from: "lambda.min" (default) or "lambda.1se".

Value

An object of class cv.coxtrans with fields: grid (full lambda grid), cvm (mean deviance per grid point), cvsd, cvup, cvlo, nzero (mean non-zero target coefficients), lambda.min (optimal lambda list), lambda.1se (1SE-rule lambda list), index.min, index.1se, coxtrans.fit (final model at lambda.min), coxtrans.fit.1se (final model at lambda.1se), call, name.

Examples

result <- cv.coxtrans(Surv(time, status) ~ . - group - id,
  sim2, sim2$group,
  target = 1, nlambda = 5, penalty = "SCAD"
)
result
#> cv.coxtrans
#> 
#> Call: cv.coxtrans(formula = Surv(time, status) ~ . - group - id, data = sim2, 
#>     group = sim2$group, target = 1, nlambda = 5, penalty = "SCAD")
#> 
#> Measure: Partial Likelihood Deviance
#> 
#> lambda.min:  l1=0.0588  l2=0.0177  l3=0.2154
#>   Deviance:  3.2739 (+/- 0.1211)   Non-zero: 5
#> 
#> lambda.1se:  l1=0.0588  l2=0.0006  l3=2e-04
#>   Deviance:  3.3161 (+/- 0.1089)   Non-zero: 4