%\VignetteIndexEntry{Getting Started with sssvcqr}
%\VignetteEngine{knitr::knitr}
%\VignetteEncoding{UTF-8}
\documentclass{article}
\usepackage[margin=1in]{geometry}
\usepackage{hyperref}
\title{Getting Started with sssvcqr}
\author{Houjian Hou}
\begin{document}
\maketitle

This vignette shows the basic workflow for sparse-smooth spatially varying
coefficient quantile regression.

<<setup, include=FALSE>>=
knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
@

<<fit-example>>=
library("sssvcqr")

dat <- simulate_sssvcqr_data(n = 90, q = 1, p = 3, seed = 1)
dat$active

fit <- ss_svcqr(
  y = dat$y,
  Z = dat$Z,
  X = dat$X,
  u = dat$u,
  tau = 0.5,
  lambda1 = 5,
  lambda2 = 0.1,
  k_nn = 8,
  control = list(max_iter = 80, warn_nonconvergence = FALSE)
)

fit
summary(fit)
selection_recovery_table(fit, dat)
head(predict(fit))
head(predict(fit, type = "coefficients"))
@

The spatial plot types use the first two columns of the coordinate matrix saved
in the fitted object. Their colorbars show the numeric scale for deviations,
local total coefficients, or residuals.

<<plot-example, fig.width=7, fig.height=6>>=
plot(fit, type = "deviation", index = 1)
plot(fit, type = "coefficient", index = 1)
plot(fit, type = "residual")
@

Penalty parameters can be selected by spatially blocked cross-validation. The
small grid below is for demonstration; empirical applications should use a
broader grid and more iterations.

<<cv-example>>=
cv <- cv_ss_svcqr(
  y = dat$y,
  Z = dat$Z,
  X = dat$X,
  u = dat$u,
  tau = 0.5,
  lambda1_seq = c(1, 2),
  lambda2_seq = c(0.5, 1),
  K_folds = 3,
  adaptive_weights = FALSE,
  control = list(max_iter = 25, warn_nonconvergence = FALSE)
)

cv
cv$best
@

\end{document}