Multi-scale temporal structure

Specify a temporal random effect that decomposes variation into separate scales: a smooth trend, an optional seasonal cycle, and a short-term component. Each scale uses its own prior, letting slow and fast dynamics be modelled jointly.

Usage

temporal_multiscale(
  time_var,
  trend = c("rw2", "rw1", "none"),
  seasonal = NULL,
  short_term = c("ar1", "iid", "none"),
  group_var = NULL,
  shared = NULL
)

Arguments

time_var

Single character string naming the time variable in the data.

trend

Prior for the smooth long-term trend. One of "rw2", "rw1", or "none".

seasonal

Optional integer period (>= 2) of a seasonal cycle, e.g. 12 for monthly data with an annual cycle. NULL (default) omits the seasonal component.

short_term

Prior for the short-term component. One of "ar1", "iid", or "none".

group_var

Optional character string naming a grouping variable for group-specific temporal effects.

shared

Whether the effect is shared across processes in a multi-process model. NULL (default) shares it; FALSE fits process-specific effects and emits a warning.

Value

A tulpa_temporal_multiscale object.

Details

At least one of trend, seasonal, or short_term must be active.

See also

temporal_rw1(), temporal_rw2(), temporal_ar1() for single-scale temporal priors.

Examples

# Trend + annual seasonal cycle + AR1 short-term component on monthly data
temporal_multiscale("month", trend = "rw2", seasonal = 12, short_term = "ar1")