Smoothed Quantile Regression with Post-Processing — post_processed_grad_descent • quantspace

Smoothed Quantile Regression with Post-Processing

post_processed_grad_descent(
  X,
  y,
  tau,
  lambda = 0,
  nwarmup_samples = 0.1 * nrow(X),
  lp_size = 10000,
  intercept = NULL
)

Arguments

X: design matrix
y: outcome variable
tau: target quantile
lambda: optional weight on penalty function
nwarmup_samples: number of samples to use for warmup in approximat quantile regression
lp_size: size of linear programming problem passed to the simplex algorithm
intercept: integer for location of intercept column

Details

This function performs smoothed quantile regression w/ post-processing to ensure accuracy of the approximate first-order method.