Cross-validated Quantile Boosting for high-dimensional data

Applying a greedy algorithm to minimize the expected smoothed check loss. Employs the Weak Restricted Greedy Algorithm. For the orthogonal variant, set stepsize to NULL.

cv_qboost(
  X,
  Y,
  tau = 0.5,
  m_stop = 1,
  h = 0.1,
  kernel = "Gaussian",
  stepsize = NULL,
  n_folds = 5
)

Arguments

X

(matrix)
A matrix of covariates.

Y

(numeric())
A vector of responses.

tau

(numeric(1L))
A quantile.

m_stop

(integer(1L))
The number of boosting steps.

h

(numeric(1L))
The bandwith for smoothing.

kernel

(character(1L))
The kernel for smoothing.

stepsize

(numeric(1L))
The stepsize of the boosting procedure. Set to NULL for WCGA.

n_folds

(integer(1L))
Number folds used.

Value

An object with S3 class qboost

• coeff_path The cofficients along the boosting steps as a matrix (starts with a zero vector).

• params Additional parameters.

See also

predict.qboost