EpiNow2
Estimate Realtime Case Counts and Time-varying Epidemiological Parameters
EpiNow2 unified its model interface, then went back to deepen the estimators behind it
◆Recent moves
- 1mo ago
estimate_truncation gains full delay and observation options
`estimate_truncation()` catches up with the rest of the package: all `dist_spec` delay families, `obs_opts()` observation model selection, and the `likelihood`/`return_likelihood` wiring that enables prior-only fits and loo-compatible output. Its previously hardcoded noise prior becomes a specifiable `dist_spec`, it warns when irrelevant `obs_opts()` settings are supplied, and a new `estimate_dist()` fits delay distributions from interval-censored linelist data.
View source ↗ - 6mo ago
Unified return objects and shared accessors across all models
⚡ SPARKThe release that gave the package one shape instead of four: all main modelling functions return consistent S3 objects reached through shared accessors, discretisation moves to the primarycensored package, and Rt estimates can account for susceptible depletion. The subsequent releases extend this base outward rather than reworking it.
View source ↗ - 1y ago
Patch for an upstream rstan issue
A patch released in response to an upstream `rstan` issue flagged in CRAN checks. Maintenance forced from outside the package.
View source ↗ - 1y ago
Accumulation for irregularly reported data; unified priors
Introduces accumulation, letting models fit data reported at regular or irregular intervals via `fill_missing()` and an `accumulate` column, with fixed frequencies detected automatically. All priors move onto the internal distribution interface — the groundwork the 1.8.0 unification then built on.
View source ↗ - 1y ago
Matern kernel spectral density fix and GP prior revert
Fixes a bug in the spectral densities of Matern kernels of orders other than 3/2, which silently affected results for anyone using them, and reverts the Gaussian Process magnitude prior to HalfNormal(0, 0.05) after users reported unexpected results. Users are now told explicitly that `NA` observations are treated as missing rather than zero.
View source ↗ - 1y ago
Gaussian Process model improvements and explicit defaults
Gaussian Process model improvements alongside a broad documentation pass, including making implicit argument defaults explicit and clarifying what settings apply when `rt = NULL`. The delay distribution interface work here precedes the wider prior unification in 1.7.0.
View source ↗