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

## -----------------------------------------------------------------------------
library(gp3bayes)

sim <- simulate_advanced_pupil_timecourse(
  n_participants = 18,
  trials_per_participant = 4,
  time_points = 41,
  time_range = c(-200, 1800),
  family = "student",
  residual_scale = 0.08,
  seed = 2050
)

plot_advanced_pupil_simulation(sim)

## -----------------------------------------------------------------------------
spec <- specify_advanced_pupil_timecourse_model(
  sim$data,
  temporal_structure = "gaussian_process",
  family = "student",
  residual_scale = "condition_time",
  gp_spec = create_pupil_gp_spec("matern32", basis = "approximate", k = 25),
  autocorrelation = "none",
  predictive_target = "future_segment"
)

audit_advanced_pupil_identifiability(spec)
pupil_model_card(spec)

## ----eval=FALSE---------------------------------------------------------------
# fit <- fit_advanced_pupil_model_cmdstanr(spec)
# traj <- predict_advanced_pupil_trajectory(fit, ndraws = 1000)
# 
# d1 <- estimate_pupil_trajectory_derivative(traj, order = 1)
# plot_pupil_trajectory_derivative(d1)
# 
# contrast <- estimate_pupil_dynamic_contrast(
#   traj,
#   contrast = c("treatment", "control"),
#   threshold = 0.05
# )
# plot_pupil_dynamic_contrast(contrast)
# 
# estimate_pupil_threshold_duration(
#   contrast,
#   direction = "absolute",
#   threshold = 0.05
# )

## ----eval=FALSE---------------------------------------------------------------
# cal <- audit_pupil_predictive_calibration(
#   fit,
#   newdata = held_out_trials,
#   ndraws = 1000,
#   probability = 0.90,
#   allow_new_levels = FALSE
# )
# 
# as.data.frame(cal)
# plot_pupil_predictive_calibration(cal)