spatiotemporal_effects: Extract spatiotemporal effects from fitted model

In tulpa: Templated Unified Library for Posterior Approximation in Bayesian Hierarchical Models

Extract spatiotemporal effects from fitted model

Description

Extract posterior distributions of spatiotemporal interaction effects from a fitted tulpa model.

Usage

spatiotemporal_effects(
  object,
  format = c("array", "long", "summary"),
  probs = c(0.025, 0.5, 0.975),
  ...
)

## Default S3 method:
spatiotemporal_effects(
  object,
  format = c("array", "long", "summary"),
  probs = c(0.025, 0.5, 0.975),
  ...
)

## S3 method for class 'tulpa_st_summary'
print(x, ...)

Arguments

object	A fit carrying a ⁠$spatiotemporal⁠ Knorr-Held interaction block (n_spatial, n_times) and its draws in ⁠$.internal$spatiotemporal_draws⁠. tulpa() itself has no ⁠spatiotemporal =⁠ argument; such a block comes from a model package's own fitter (see spatiotemporal() for what the engine itself fits), and the default method reads any fit, of any class, that stores it this way.
format	Output format: "array" (default, S x T x draws), "long" (data frame with s, t, draw, value columns), or "summary" (posterior summaries).
probs	Quantiles to compute if format = "summary".
...	Ignored
x	A tulpa_st_summary table.

Value

For format = "array", an S x T x draws numeric array carrying n_spatial, n_times and n_draws attributes; for "long", a data frame with columns s, t, draw and value; for "summary", a tulpa_st_summary data frame with columns s, t, mean, one column per entry of probs and sd, which has print() and plot() methods.

Examples

# A Knorr-Held interaction block is attached by a model package's own
# fitter (see spatiotemporal() for what the engine fits). The accessor reads
# its layout and an n_draws x (S * T) draw matrix stored s-fastest: column
# (t - 1) * S + s holds region s at time t. Here both are simulated for
# S = 3 regions over T = 4 times.
set.seed(1)
S <- 3L; n_t <- 4L
st_mean <- outer(c(-0.5, 0, 0.5), seq(-0.3, 0.3, length.out = n_t))
draws <- matrix(rnorm(100 * S * n_t, mean = rep(c(st_mean), each = 100),
                      sd = 0.1), nrow = 100)
fit <- structure(
  list(spatiotemporal = list(n_spatial = S, n_times = n_t),
       .internal = list(spatiotemporal_draws = draws)),
  class = "tulpa_fit")

st_effects <- spatiotemporal_effects(fit, format = "summary")
head(st_effects)
dim(spatiotemporal_effects(fit, format = "array"))

tulpa documentation built on Oct. 19, 2026, 5:09 p.m.