ref_forecast
Usage
ref_forecast(dynamic, history, times, outcome = NULL)Returns a dist_conditioned predictive distribution at each requested time. The dynamic process only changes the predictive distribution; scores and graphics then follow from the existing contract.
Arguments
Section titled “Arguments”| Argument | Description |
|---|---|
dynamic |
A ref_dynamics object. |
history |
Subject history data frame. Must contain the time column used in ref_dynamics() and the outcome. |
times |
Future times. |
outcome |
Outcome name. |
Details
Section titled “Details”The forecast refuses to run when the process for outcome is not
identified (see ref_dynamics()): a history-conditioned distribution
from an unestimated kernel would be silently wrong. History rows with a
non-finite time or outcome are dropped and the number actually used is
returned as history_n. The exact marginal uncertainty estimand stored by
ref_dynamics() is reused. Kernel parameters remain plug-in estimates;
their uncertainty is explicitly recorded but not propagated.
A ref_forecast with $dist (a distribution vector, one element per time),
$summary (median and 90% interval), $history_n, and
$lag_support ("in" or "extrapolated_lag" per forecast time,
judged from the last history visit against the reference lag range).