# Toy rqss example based on Figure 2 of Polson and Scott (2016, JRSSB) # NB: Solutions are piecewise linear, unlike those of Polson and Scott # whose ADMM procedure does only 30 iterations. dgp <- function(n) { x <- 0:n/n y <- rnorm(n+1, 5 * sin(2 * pi * x), 0.5 + exp(1.5 * sin(4 * pi * x))) data.frame(x = x, y = y) } D <- dgp(1000) plot(D$x, D$y, cex = .5) taus <- 1:9/10 for(i in 1:length(taus)) plot(rqss(y ~ qss(x, lambda = 1/10), tau = taus[i], data = D), rug = FALSE, add = TRUE)