VsusP 1.1.0
Gaussian fitting and
summaries
- Add
FitRidge(), FitHorseshoe() and
FitSpikeSlab() with explicit priors, intercept draws and
separate chain axes. Existing bayesreg defaults
remain.
- Add coefficient summaries, posterior and sparse prediction,
PosteriorDraws() and ChainDiagnostics(). Add
coefficient, inclusion, trace and autocorrelation plots that return the
displayed values invisibly.
Selection methods
- Add
TwoMeansBeta() for basic two-means selection and
SpikeSlabSelection() for exact-zero inclusion summaries and
visited-model frequencies.
- Add SAVS and the adaptive-L1 DSS approximation as separate methods
that can return an empty selection. Document scaling, tuning and
numerical limits.
- Add exact one-dimensional clustering, recursive-pair refinement and
S2MSensitivity(). Randomized kmeans and legacy small-set
stopping remain the defaults; sensitivity summaries do not choose an
optimal threshold.
- Allow all-zero S2M draws with
zero.draws = "empty";
default rejection remains.
Fixes
- Retain the saved S2M noise set at recursive stopping, following Li
and Pati (2017), Section 2.2.1, doi:10.1016/j.csda.2016.10.008. The previous count used
the smaller cluster from the stopping split and could select too many
signals. The strict center-gap comparison is unchanged.
- Handle distinct initial pairs and indivisible recursive candidates.
Reject unsupported constant-magnitude draws; accept explicit
H = 0 selections.
- Validate dimensions, finite values, scalar counts and tuning
ranges.
- Return consistent selection summaries and use every returned sampler
draw.
- Accept prior aliases
rr, hs and
hs+; preserve predictor names without allowing a predictor
name to replace the response.
- Correct citations, draw orientation and tuning-plot descriptions.
Document tie conventions and empty-model limits.
Documentation and checks
- Add eight reproducible figures and worked fitting, selection and
prediction examples. Retain the exploratory pilot’s outcomes and
limitations.
- Remove tracked build products and duplicate documentation. Run one
package check per CI job and retain strict test-count and coverage
verification.