adrian-lison/EpiSewer: EpiSewer v0.0.4

EpiSewer is an R package and Bayesian generative model to estimate epidemiological parameters from wastewater concentration measurements over time.

EpiSewer v.0.0.4 comes with important new functionality and model extensions but also a number of breaking changes:

• Rt smoothing priors: Gaussian process smoothing (R_estimate_gp) has been added and is now the new default smoothing approach. This enables faster sampling and more robust real-time behaviour. Two other smoothing priors based on changepoint modeling have been added (R_estimate_piecewise and R_estimate_changepoint_splines). Existing priors (R_estimate_splines, R_estimate_rw, and R_estimate_ets) have been refactored to better support changes in variance over time. Note that various arguments of these functions have been refactored. See #53 and #61 for details.
• Shedding load distribution uncertainty: New function shedding_dist_estimate now supports parametric shedding load distributions with parameter uncertainty (via priors). It is even possible to mix several distinct priors (e.g. from different studies) together. See #54 for details.
• Sedding: An improved seeding model based on growth rates has been added and is now the new default. Credits for the general idea go to @SamuelBrand1 and the EpiAware authors, suggested by @seabbs in https://github.com/adrian-lison/EpiSewer/issues/5. See #56 for details.
• Forecasting: Now supports dampening of forecasts.
• Sampling: Added support for pathfinder sampling. Pathfinder is now used by default to initialize parameters before MCMC sampling. It can also be used for sampling directly (via set_fit_opts()), however performance is mixed, so this is currently only recommended for a quick inspection. See #55 for details.
• Plotting: Small improvements to various plotting functions. Spaghetti-plots of Rt and infections now support coloring by chain to check mixing.