restore alpha prior SD to 0.05

NEWS.md

@@ -3,6 +3,7 @@
 ## Model changes
 
 - A bug in the spectral densities of Matern kernels of order other than 3/2 has been fixed and the vignette updated accordingly. By @sbfnk in #835 and reviewed by @seabbs.
+- Changed the prior on the (square root of the) magnitude alpha of the Gaussian Process back to HalfNormal(0, 0.05) based on user feedback of unexpected results. By @sbfnk in #PR and reviewed by.
 
 ## Package changes
 
@@ -25,7 +26,7 @@ A release that introduces model improvements to the Gaussian Process models, alo
 - When defining probability distributions these can now be truncated using the `tolerance` argument
 - Ornstein-Uhlenbeck and 5 / 2 Matérn kernels have been added. By @sbfnk in #741 and reviewed by @seabbs.
 - Gaussian processes have been vectorised, leading to some speed gains 🚀 , and the `gp_opts()` function has gained three more options, "periodic", "ou",  and "se", to specify periodic and linear kernels respectively. By @seabbs in #742 and reviewed by @jamesmbaazam.
-- Prior predictive checks have been used to update the following priors: the prior on the magnitude of the Gaussian process (from HalfNormal(0, 1) to HalfNormal(0, 0.1)), and the prior on the overdispersion (from 1 / HalfNormal(0, 1)^2 to 1 / HalfNormal(0, 0.25)). In the user-facing API, this is a change in default values of the `sd` of `phi` in `obs_opts()` from 1 to 0.25. By @seabbs in #742 and reviewed by @jamesmbaazam.
+- Prior predictive checks have been used to update the following priors: the prior on the magnitude of the Gaussian process (from HalfNormal(0, 0.05)^2 to HalfNormal(0, 0.01)^2), and the prior on the overdispersion (from 1 / HalfNormal(0, 1)^2 to 1 / HalfNormal(0, 0.25)). In the user-facing API, this is a change in default values of the `sd` of `phi` in `obs_opts()` from 1 to 0.25. By @seabbs in #742 and reviewed by @jamesmbaazam.
 - The default stan control options have been updated from `list(adapt_delta = 0.95, max_treedepth = 15)` to `list(adapt_delta = 0.9, max_treedepth = 12)` due to improved performance and to reduce the runtime of the default parameterisations. By @seabbs in #742 and reviewed by @jamesmbaazam.
 - Initialisation has been simplified by sampling directly from the priors, where possible, rather than from a constrained space. By @seabbs in #742 and reviewed by @jamesmbaazam.
 - Unnecessary normalisation of delay priors has been removed. By @seabbs in #742 and reviewed by @jamesmbaazam.

R/opts.R

@@ -458,7 +458,7 @@ backcalc_opts <- function(prior = c("reports", "none", "infections"),
 #' deviation of the Gaussian process (logged Rt in case of the renewal model,
 #' logged infections in case of the nonmechanistic model).
 #'
-#' @param alpha_sd Numeric, defaults to 0.1. The standard deviation of the
+#' @param alpha_sd Numeric, defaults to 0.05. The standard deviation of the
 #' magnitude parameter of the Gaussian process kernel. Can be tuned to adjust
 #' how far alpha is allowed to deviate form its prior mean (`alpha_mean`).
 #'
@@ -509,7 +509,7 @@ gp_opts <- function(basis_prop = 0.2,
                     ls_min = 0,
                     ls_max = 60,
                     alpha_mean = 0,
-                    alpha_sd = 0.01,
+                    alpha_sd = 0.05,
                     kernel = c("matern", "se", "ou", "periodic"),
                     matern_order = 3 / 2,
                     matern_type,