This package estimates multi-reponse quantile regression in the context that the resulting predictions should be “smooth”. The typical use case would be multi-horizon time-series forecasting wherein we believe that the responses should vary smoothly.
See the related pre-print: https://arxiv.org/abs/2202.09723
Installation
You can install the development version of smoothqr from GitHub with:
# install.packages("remotes")
remotes::install_github("dajmcdon/smoothqr")Example
This is a basic example.
library(smoothqr)
library(ggplot2)
x <- -99:99 / 100 * 2 * pi
y <- sin(x) + rnorm(length(x), sd = .1)
XY <- lagmat(y[1:(length(y) - 20)], c(-20:20))
Y <- XY[, 1:20]
X <- XY[, 21:ncol(XY)]
tt <- smooth_qr(X, Y, c(.1, .5, .9), aheads = 20:1)
pl <- predict(tt, newdata = X[max(which(complete.cases(X))), , drop = FALSE])
pll <- dplyr::bind_rows(lapply(pl, tibble::as_tibble), .id = "ahead")
pll$ahead <- rev(x[length(y) - 19:0])
ggplot(pll) +
geom_ribbon(
aes(x = ahead, ymin = `tau = 0.1`, ymax = `tau = 0.9`),
fill = "steelblue"
) +
geom_point(data = data.frame(x, y), aes(x, y), color = "lightgrey") +
geom_vline(xintercept = x[length(y) - 20], linetype = 2) +
stat_function(fun = sin, color = "black") +
geom_line(aes(x = ahead, y = `tau = 0.5`), color = "darkorange") +
theme_bw()