Quantile Regression approximated w/ huber loss

Quantile Regression approximated w/ huber loss

rq.fit.agd(
  X,
  y,
  tau = 0.5,
  weights = NULL,
  control,
  lambda,
  smoothing_window = 1e-10,
  beta_tol = 1e-05,
  check_tol = 1e-05,
  maxiter = 1000,
  n_samples = min(c(ceiling(nrow(X)/10), 10000)),
  init_beta = NULL,
  scale = 1,
  intercept = NULL,
  ...
)

Arguments

X

design matrix

y

outcome vector

tau

target quantile

weights

optional weight vector

control

ignored for now

lambda

ignored for now

smoothing_window

neighborhood around 0 which is smoothed by either typical least squares or appropriately tilted least squares loss function

beta_tol

stopping rule based on max value of gradient

check_tol

stopping rule based on change in the loss function

maxiter

largest number of iterations allowed

n_samples

number of observations to use in "warmup" regression

init_beta

initial guess at betas

scale

whether to scale x and y variables in regression

intercept

optional integer indicating intercept column that identifies initial values

...

other arguments, ignored for now