BKP 0.3.1 (2026-07-13)
- Reordered optional arguments in
fit_BKP() and
fit_DKP() to group kernel specification, length-scale
settings, loss/ESS options, and optimization controls more consistently.
Named-argument usage is unaffected.
- Updated probability-scale plot labels across
plot.BKP(), plot.DKP(),
plot.TwinBKP(), and plot.TwinDKP() to use
“Posterior Mean,” “Posterior Variance,” and “95% Credible Interval”
instead of “Predictive Mean,” “Predictive Variance,” and “95% CI.” This
change clarifies that these panels summarize posterior uncertainty for
the latent probability surface rather than the predictive distribution
of future responses.
- Added links to the companion
BKP-paper
reproducibility repository in the package metadata and README. The
repository contains the manuscript, replication code, data-processing
scripts, and materials used to generate the examples and figures in the
software paper.
BKP 0.3.0 (2026-07-01)
- Added
fit_TwinDKP() and associated S3 methods for
scalable global-local Dirichlet Kernel Process modeling.
- Added
fit_TwinBKP() with full S3 support for scalable
Twin Beta Kernel Process modeling.
- Implemented Twinning-based global subset selection and kd-tree
local-neighbour search.
BKP 0.2.4 (2026-06-16)
- Improved computational efficiency by implementing kernel evaluation,
prior construction, loss evaluation, and hyperparameter optimization
routines in C++.
- Added support for the compactly supported Wendland kernel via
kernel = "wendland".
- Added optional Shepard effective-sample-size calibration for
fit_BKP(ess = "shepard") and
fit_DKP(ess = "shepard"), while keeping the default
ess = "none" behavior unchanged.
- Added ggplot2 support for plotting;
plot.BKP(..., engine = "ggplot") now produces ggplot2-based
visualizations.
- Added a package-wide
isotropic argument, defaulting to
TRUE, for isotropic kernels with a shared length-scale
across dimensions. Set isotropic = FALSE to use anisotropic
kernels with dimension-specific length-scales.
BKP 0.2.3 (2025-09-22)
- Removed vignettes to avoid redundancy with the arXiv paper and to
resolve CRAN thread limit issues during vignette building.
BKP 0.2.0 (2025-09-16)
- Added
fitted(), parameter(), and
quantile() methods.
- Updated
predict() and simulate() methods:
both now return results for the training data by default when
Xnew is not provided.
- Extended
plot() method with new dims
argument for higher-dimensional inputs.
- Added Section 5 to the vignette, presenting a real-data application
on Loa loa parasite infection in North Cameroon.
- Added argument checking with informative error messages.
BKP 0.1.1 (2025-08-19)
- Added a vignette introducing the package.
- Fixed minor bugs and improved stability.
BKP 0.1.0 (2025-07-23)