Ahead of next week’s Insight meeting I pushed out a new epinowcast release. I thought it was worthwhile sharing here.
The headline additions are print(), summary(), and improved plotting for both preprocessed data and fitted model objects, so inspecting what you have before and after fitting is much less friction.
On the modelling side, epinowcast can now be used for retrospective count estimation without any nowcasting — the same shape of model as EpiNow2 — thanks to support for max_delay = 1.
See the single-timeseries Rt vignette: Estimating the effective reproduction number in real-time for a single timeseries with reporting delays • epinowcast
A new inference methods vignette compares NUTS sampling, standalone pathfinder, and pathfinder used to initialise NUTS.
Useful if you’re weighing posterior accuracy against speed, or looking for something fast enough for real-time production use.
The getting started vignette has been rewritten to fit and compare several models end-to-end rather than just running the default:
And there’s a new data visualisation vignette walking through plotting before you fit:
Plus the usual bug fixes and documentation improvements.
Full changelog: Changelog • epinowcast