## ----setup, include = FALSE---------------------------------------------------
has_stan <- requireNamespace("rstan", quietly = TRUE)
eval_fits <- identical(Sys.getenv("DCVAR_EVAL_VIGNETTES"), "true") && has_stan
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  eval = eval_fits
)

## ----mixed-margins------------------------------------------------------------
# library(dcvar)
# 
# sim_mix <- simulate_dcvar(
#   n_time = 200,
#   rho_trajectory = rho_constant(200, rho = 0.6),
#   margins = c("normal", "exponential"),
#   skew_direction = c(1, 1),
#   seed = 11
# )
# 
# fit_mix <- dcvar_constant(
#   sim_mix$Y_df,
#   vars = c("y1", "y2"),
#   margins = c("normal", "exponential"),
#   skew_direction = c(1, 1),
#   seed = 11
# )
# 
# # Each dimension is reported under its own family: sigma_eps for the normal
# # variable, sigma_exp for the exponential variable.
# coef(fit_mix)[c("sigma_eps", "sigma_exp")]

## ----tv-fit-------------------------------------------------------------------
# library(dcvar)
# 
# # A series with declining coupling to illustrate the time-varying fits
# sim_tv <- simulate_dcvar(
#   n_time = 150,
#   rho_trajectory = rho_decreasing(150, rho_start = 0.7, rho_end = 0.3),
#   seed = 3
# )
# 
# # Only the autoregressive coefficients evolve over time
# fit_ar <- dcvar(
#   sim_tv$Y_df,
#   vars = c("y1", "y2"),
#   tv_phi = "ar",
#   seed = 4
# )
# 
# # Time-varying coefficients and time-varying scales together, with a
# # right-skewed second margin (the exponential dimension uses the soft barrier)
# fit_full <- dcvar(
#   sim_tv$Y_df,
#   vars = c("y1", "y2"),
#   margins = c("normal", "exponential"),
#   skew_direction = c(1, 1),
#   tv_phi = TRUE,
#   tv_sigma = TRUE,
#   seed = 5
# )

## ----tv-extract---------------------------------------------------------------
# # Posterior summaries of phi11(t), phi12(t), phi21(t), phi22(t)
# head(phi_trajectory(fit_ar))
# 
# # Per-variable residual-scale paths
# head(sigma_trajectory(fit_full))
# 
# # Facetted trajectory plots
# plot_phi_trajectory(fit_ar)
# plot_sigma_trajectory(fit_full)

## ----fit-all------------------------------------------------------------------
# library(dcvar)
# 
# # Simulate data with a step-function trajectory
# sim <- simulate_dcvar(
#   n_time = 200,
#   rho_trajectory = rho_step(200, rho_before = 0.7, rho_after = 0.3),
#   seed = 42
# )
# 
# # Fit all three models on the same data
# fit_dc  <- dcvar(sim$Y_df, vars = c("y1", "y2"), seed = 1)
# fit_hmm <- dcvar_hmm(sim$Y_df, vars = c("y1", "y2"), K = 2, seed = 2)
# fit_con <- dcvar_constant(sim$Y_df, vars = c("y1", "y2"), seed = 3)

## ----compare------------------------------------------------------------------
# dcvar_compare(dcvar = fit_dc, hmm = fit_hmm, constant = fit_con)

## ----tidy---------------------------------------------------------------------
# # Full parameter summary as a data frame
# param_df <- as.data.frame(fit_dc)
# head(param_df)
# 
# # Predictions with marginal intervals
# pred_df <- predict(fit_hmm)
# head(pred_df)

## ----hmm----------------------------------------------------------------------
# # State posteriors, Viterbi path, transition matrix
# states <- hmm_states(fit_hmm)
# 
# # State-specific rho values
# states$rho_state
# 
# # Transition matrix
# states$A
# 
# # Visualise state posteriors
# plot(fit_hmm, type = "states")
# 
# # Transition matrix heatmap
# plot(fit_hmm, type = "transition")

## ----interpret----------------------------------------------------------------
# interpret_rho_trajectory(fit_dc)
# interpret_rho_trajectory(fit_hmm)
# interpret_rho_trajectory(fit_con)