This release has been a long timing coming - mostly due to distraction on my part. A lot of work from a lot of people in here.
What’s new
The headline addition is a new Model Features Summary • epinowcast that helps users understand when and how to apply epinowcast to different problems.
There have also been documentation enhancements throughout, with particular improvements to the formula interface guidance.
Performance and new features
Performance improvements include optimised Stan functions, support for sparse design matrices for memory-intensive models, and tightened priors to improve run times. Experimental pathfinder support has been added for rapid prototyping and informing initialisation of HMC runs, allowing faster iteration during model development. Model enhancements include a negative binomial observation model with linear mean-variance relationship and improved probability aggregation support.
Breaking changes and bug fixes
The release includes one breaking change that fixes an off-by-one error in `enw_filter_reference_dates()`. Previously, `include_days = n` incorrectly returned `n + 1` dates instead of exactly `n` dates. Users relying on the previous behaviour will need to adjust their `include_days` arguments by subtracting 1. Functions deprecated at version 0.4.0 or earlier have been removed.
Several bugs have been fixed, including issues with `enw_aggregate_cumulative()` counting and IDate storage mode compatibility with dplyr workflows.
See the Changelog • epinowcast for complete details on all changes in this release.
Also, we had a cheeky title change to: “A Bayesian Framework for Real-time Infectious Disease Surveillance” what do people think?
Many claude agents died in the creation of this release (see: @seabbs.bsky.social on Bluesky)