The default prior hyperparameters for PBclassifier()
and miso() have changed from
a0 = 0.25, b0 = 0.25 to a0 = 0.5, b0 = 0.5
(Jeffreys prior). Results from previous versions can be reproduced by
explicitly passing a0 = 0.25, b0 = 0.25.
mcmiso() is now an exported function. Previously it
was internal. Its interface has also changed: the future
parallel plan must now be set by the caller via
future::plan() before calling mcmiso(), rather
than being set internally. This follows best practices for the
future package.
The return value of miso() now has class
"miso", enabling a dedicated print method.
Code that tested class(fit) == "list" will need
updating.
misoN(): multivariable isotonic regression for
continuous outcomes using a Normal-Inverse-Chi-Squared
conjugate model.
mcmisoN(): parallel computing wrapper for
misoN() (requires the future package and a
parallel plan set by the caller).
mcPBclassifier(): parallel computing wrapper for
PBclassifier().
boundary(): extracts the decision boundary (minimal
positive set) from a fitted "pbc" object.
New print methods: print.pbc(),
print.miso(), print.misoN(),
print.boundary().
Comprehensive input validation has been added to all exported functions.
getScenesV3() has been rewritten from a brute-force
expand.grid(2^K) enumeration to a recursive backtracking
algorithm. This substantially reduces memory usage and computation time
for large numbers of unique feature combinations.
Vectorized computation replaces inner loops in
SweepCombTogBinom(), SweepMcCombBinom(),
SweepCombTogNorm(), and
SweepMcCombNorm().
dplyr has been removed as a dependency (no longer
used internally).
future has been moved from Imports to
Suggests. It is only required for the parallel computing
wrappers (mcmiso, mcPBclassifier,
mcmisoN). These functions check for the package at run time
and give an informative error if it is not installed.