NEWS
flocker 1.1-1
- Fixed single-visit event formatting dropping the dimensions of its index table.
- Added deterministic characterization tests for packing and reconstruction ahead
of the version 2 schema rewrite.
flocker 1.1-0 (2026-10-16)
- Added the
twolevel_single model family for single-season occupancy models
in which groups of closure units may themselves be structurally absent.
Users specify the grouping factor with level2_group, may supply a separate
table of group-level predictors with level2_covs, and model group-level
occupancy with f_meta.
- Refactored the data-augmented model as a special case of the general
two-level model, with shared data formatting, likelihood, fitting, and
post-processing machinery.
- Breaking change for augmented models: every species supplied in the original
obs array must have at least one detection. Never-observed pseudospecies
must instead be added through n_aug.
- Breaking change for augmented models:
get_Z() now returns a list containing
unit states for species-site combinations and level2 states for species.
Generic two-level models use the same return structure. Joint samples from
the two levels are mutually consistent.
- Two-level outputs from
get_Z(), log_lik_flocker(), and loo_flocker() now
retain user-supplied group names. Augmented models similarly retain supplied
species names and assign stable names to augmented pseudospecies.
- Improved the numerical stability of two-level log-likelihood calculations by
evaluating occupancy mixtures in log space.
- Corrected backward smoothing and backward sampling of latent states from
multiseason models so that both use the appropriate transition probabilities
and future-data likelihoods.
- Corrected history-conditioned
get_Z() calculations for augmented models
and when using new_data, including rep-constant models. Augmented
group-level probabilities now marginalize over unit occupancy and incorporate
the fitted group-level occupancy probability.
- Fixed expansion of user-supplied unit and event covariates in
simulate_flocker_data() so that values are shared across species at each
unit or visit, and allowed make_flocker_data() to accept data-frame
observation inputs as documented.
- Fixed reconstruction of maximally ragged observation and covariate arrays,
including multiseason data with a final season that is unobserved in every
series, and made post-processing robust when exactly one posterior draw is
requested.
- Fitted models and formatted data objects are now tagged with the version of
flocker that created them. Two-level packing metadata is stored separately
from the underlying brmsfit data.
flocker 1.0-2 (2026-06-22)
- fixed a bug that resulted in incorrect log likelihoods returned by log_lik_flocker for the data augmented model. Model fitting was not affected, and posteriors were correct.
- log-likelihood computations now enabled over new data.
- dramatically improved efficiency in certain post-processing functions for multiseason models
- CI runs properly
- backend hygiene improvements
- updated compatibility with upcoming changes to
loo_compare() output
structure in the loo package (> 2.9.0), which now returns a data frame
instead of a matrix and includes additional diagnostic columns.
flocker 1.0-1
- fixed a bug that was throwing an uninformative error when making predictions in multiseason models where history_condition is TRUE
- dramatic efficiency improvements to
get_positions as applied to multiseason models.
- groundwork laid for enabling log-likelihood computations over new data.
flocker 1.0-0 (2024-02-05)
- Initial CRAN submission.
- Fixed bug in mixed predictive checking.
- Fixed bug that prevented post-processing with new data with different
numbers of visits/seasons than original data.
- Substantially more thorough unit testing.
- Removed testing dependency on
cmdstanr (not on CRAN)
- Refactored to avoid shipping large objects as package data