seqwrap

• New dispersion_evaluation() is an evaluation function for glmmTMB models that records the log dispersion, its standard error and the log mean count of each target, which is what seqwrap_priors() needs to build the dispersion prior.
• A list passed as targetdata can now be named by target identifier, in which case elements are matched to targets by name regardless of order, and the list may hold more targets than data (for example priors built from a full run applied to a subset). seqwrap_priors() returns such a list. An unnamed list is still matched by position, as before, but seqwrap() now warns that it does so, since a misordered list is otherwise silent. A named list that lacks a target is rejected with a message naming the missing targets.

Improvements and bug fixes

• seqwrap() now reports, in its verbose header, how many times the progress bar will advance and in what step size. The bar moves once per batch of chunks (one chunk per core), so with the default chunk size it advances in about four steps; the message points to chunk_size for finer updates.
• Target-wise data (targetdata) was paired with the wrong targets when the target identifiers in data were not in sorted order. Targets are grouped by sorted identifier internally, while target-wise data was taken by row, so row j of targetdata ended up with the j-th target in sorted order rather than with row j of data. Target-wise data now follows the same reordering as the data, for data frames and lists, with and without chunking or subset.