Changelog • STbayes

STbayes 1.2.0

Package now supports generating multinetwork models where the strength of social transmission is constrained to a single value for all networks.
Using STb_summary() on an OADA model now includes $s'$ summary statistics, rather than only $log(s')$ .
Package now supports datasets with different numbers of individuals in each trial.

Fixed bug where supplying non-character ILV names would throw an uncaught error.
Fixed bug where supplying asymmetric networks with self-loops would clobber imported edge-weight values.
Fixed bug where %ST was not calculated correctly and would exceed 1 when using multiplicative ILVs.
Fixed bug where hi-rez mode would throw an uncaught error if users did not supply transmission weights.

Updated guidance for model comparison in the Getting Started vignette, include about p_worse feature in loo for model comparison. Please update loo, if you haven’t recently!

Package now supports PSIS-LFO-LOOCV as another tool for for model comparison (inspiration from Bürkner, Gabry & Vehtari 2020).
Package now supports time-varying and trial-varying edge-weights when networks are given as posterior draws. Before, varying networks were only supported when edge-weights were point estimates.
Estimating acquisition times in generated quantities block has been optimized, now supports when using float event times and when supplying posterior draws of edge-weights.

If importing network edge-weights as draws from posterior distributions, arrays must have named dimensions for my sanity and yours. Please see this vignette for more details.
The example code for plotting PPCs in the Getting Started vignette is now more generalisable, addresses cases where trials may have different numbers of individuals, and clarifies how to handle demonstrators.

Fixed bug where demonstrators were not excluded from prediction set when using generate_STb_model(est_AcqTime=T)
Fixed bug when estimating acquisition times using Weibull shaped hazards

Initial release accompanying the published paper.
Note: this release was tagged as V1.0 on GitHub, but the package version field was mistakenly left as 0.0.0.9000.