glbin_lcd_c_sparse: Block coordinate gradient descent for logistic group lasso...

In antiphon/glbinc:

Description

Like in Breheny and Huang 2009

Usage

glbin_lcd_c_sparse(X, y, offset, index, eps = 1e-04, lambda,
  lambda.min = 0.001, nlambda = 100, lambda.log = TRUE, dfmax = ncol(X)
  + 1, verb = 1, add_intercept = TRUE, std = TRUE, alpha = 1,
  maxit = 1000, AIC_stop = 0, gc_force = FALSE)

Arguments

X	covariate/design matrix. Global intercept should not be here. sparseMatrix accepted.
y	response 0/1 vector
offset	offset terms
index	grouping. set 0 or NA for unpenalised terms.
eps	convergence criterion
lambda	vector of lambdas
lambda.min	fraction of max lambda to go down to
nlambda	number of penalties
lambda.log	create log-equidistant lambda vec. default: TRUE
dfmax	max df, stop if reached
verb	verbosity
add_intercept	should the global intercept be added. default: TRUE
AIC_stop	default 0. After aic has increased this many steps, halt. 0: go to the end of lambda. If used, should be more than 1.
gc_force	FALSE. force garbage collection before going c-side? Might free memory for large X.

Details

Like in Breheny and Huang 2009 except with an offset term and sparse matrix support (Matrix-class) for large data.

antiphon/glbinc documentation built on July 31, 2019, 11 p.m.