NEWS
MRS 1.3.2 (2026-07-22)
- Standardized reproducibility seeds across examples, demos, tests, and
archived simulation scripts to
12345.
- Consolidated the former standalone examples into three non-redundant,
reproducible visual demos. They cover one-dimensional Bayesian FDR,
observed-versus-shuffled two-dimensional scans, pooled versus replicate-aware
fitting across small, medium, and large replicate counts, both ANDOVA
integration methods, baseline effects, and representative multiresolution
trees.
- Restored current CRAN metadata and registered all native routines.
- Replaced deprecated Armadillo index-returning
max() calls and fixed an
uninitialized-variable path in the C++ tree traversal.
- Added input validation that prevents invalid dimensions, labels, bounds, and
tuning parameters from reaching the C++ core.
- Restored the version 1.2.6 ANDOVA midpoint-quadrature default; deterministic
and randomized compatibility suites now reproduce legacy numerical and tree
results.
- Preserved the legacy default sample-space bounds for ordinary data while
safely supporting constant and non-positive dimensions.
- Documented posterior sampling, effect-size baselines, numerical integration,
and Bayesian-FDR plotting introduced on the development branch.
- Fixed posterior sample matrix shaping and direction probabilities,
multidimensional sampling, sampling without a representative tree,
Bayesian-FDR matrix indexing, and a Riemann-quadrature second-moment error.
- Kept sampled ANDOVA probabilities inside their valid domain and removed
undefined behavior in dyadic bin conversion and region reconstruction.
- Added numerical compatibility, posterior-sampling, invariant, validation,
plotting, and native-routine regression tests.
- Removed the unused
optimx dependency and replaced the igraph-backed
representative-tree renderer with a deterministic base-graphics renderer.
- Added explicit return-value documentation for all exported plotting and print
methods, including the meaning of their graphical or printed side effects.