SeBR 1.1.0
Improvements to previous
functionality
- Added
bb() to sample from the Bayesian bootstrap (BB)
posterior more efficiently.
- Added a
fixedX case for when the covariates are fixed
(not random), which also improves computing time for all semiparametric
regression functions.
- Since location (intercept) and scale (error standard deviation) are
not identifiable in the general transformed regression model, these are
no longer reported as coefficients/parameters.
- The posterior draws of the transformation
post_g now
report (g - intercept)/scale instead of g,
which properly corresponds to the transformation under the
location-scale identified model. Now, post_g can be
compared directly to the “true” transformations from simulated data
without any further location-scale matching.
Fewer dependencies
fields and GpGp are only needed for
sbgp() and bgp_bc().plyr is only needed for
sblm_modelsel().statmod is only needed for sbqr() and
bqr().quantreg is only needed for sbqr().
spikeSlabGAM is only needed for sbsm() and
bsm_bc().
New functions
- Added
sblm_hs() for semiparametric regression with
horseshoe priors.
- Added
blm_bc_hs() for Box-Cox transformed regression
with horseshoe priors.
- Added
sblm_ssvs() for stochastic search variable
selection for semiparametric regression with sparsity priors.
- Added
sblm_modelsel() for model/variable selection for
semiparametric regression with sparsity priors.
- Added
hbb() function to sample from the hierarchical BB
(HBB) posterior. concen_hbb() samples from the marginal
posterior distribution of the HBB concentration parameters.