RtForecastR 0.1.1
- Bug fix:
rt_forecast()’s
forecast_quantiles element (used for weighted interval
score calculations via wis()) was computed from a
disconnected in-sample calculation that did not correspond to the
genuine out-of-sample forecast reported in fit$forecast. It
now derives directly from the same forecast distribution, so e.g.
forecast_quantiles[["0.025"]] is guaranteed identical to
c(fit$forecast$lo95, fit$forecast$hi95). Also fixes a
duplicate "0.25" entry that could appear in
forecast_quantiles when the default
quantile_levels was used.
RtForecastR 0.1.0
- Initial CRAN release.
rt_forecast(): filtered (real-time) and smoothed
(retrospective) R_t estimation, one-step-ahead in-sample predictions, a
genuine out-of-sample forecast, and elimination probability P(R_t <
1).
plot.rtforecast(): R_t, forecast, and
observed-vs-predicted plots.
- Forecast calibration metrics:
mae(),
mse(), rmse(), coverage(),
interval_score(), wis().
score_batches(): score archived weekly forecast output
against realized case counts.
- Two bundled example datasets:
measles_cdmx,
jalisco.