Search for optimal lambda via cross-validation

lasso_cv_search(
  x,
  y,
  tau = 0.5,
  weights = NULL,
  method = "two_pass",
  intercept = TRUE,
  nfolds = 10,
  foldid = NULL,
  nlambda = 100,
  eps = 1e-04,
  init.lambda = 2,
  parallel = T,
  coef.cutoff = 1e-05,
  thresh = 0.01,
  ...
)

Arguments

x: design matrix for regression
y: outcome variable for regression
tau: target quantile
weights: optional weights for regression
method: method to be used for penalized quantile regression (usually one of "sfn" or "br")
intercept: Whether to model the intercept or not
nfolds: number of folds to use for crossvalidation
foldid: optional pre-specified fold identifier (for example, if you want the folds to satisfy underlying data groupings)
nlambda: number of lambdas to search over
eps: smallest lambda used in search
init.lambda: initial lambda for search
parallel: whether to run cv scoring in parallel or not
coef.cutoff: what cutoff to use for "0" coefficients
thresh: threshhold for what counts as a "sparse enough" solution for the top of the grid
...: other parameters to pass on to fitting method

Details

Searches for a range of lambdas by first expanding the max lambda if necessary, then finding the smallest lambda that sets all coefficients to zero. Then it computes kfold CV scores along a grid of lambdas, returning the scores and the smallest lambda.