Changes and New Features in 1.5 (2023-01-06):
* streamlining the interface to nft2/nft/tsvs2/tsvs
* for code maintainability, nft is just a wrapper of nft2
(and predict.nft is just a wrapper of predict.nft2)
* for non-FPD scenarios, calculate lower/upper just like FPD
* harden bMM for P=1
Changes and New Features in 1.4 (2022-08-25):
* New option: mask unstable sd(.) estimates to calculate soffset
* Use hardened DPM LIO variance prior setting
* Return both the LPML and the CPO
* add bMM (a la bartModelMatrix) to compute dummies/etc.
* predict produces draws from the predictive distribution of log t
to facilitate C index calculations
* predict.aftree provided for producing survival/pdf/hazard
function estimates (or their Friedman's partial dependence
estimates if requested) for AFT BART like predict.nft does
for NFT BART
also draws from the predictive distribution of log t (as above)
* experimental support for differing X matrices for f and sd
see nft2 function
* Thompson Sampling Variable Selection functions
tsvs for the same X matrix and tsvs2 for two different matrices
* new function bMM() to replace the deprecated bartModelMatrix()
* new function Cindex() to compute Harrell's C-index
Changes and New Features in 1.3 (2022-03-26):
* Trap error code returned from rcat
Changes and New Features in 1.2 (2022-02-01):
* Allow DPM LIO parameters to be altered for
debugging/sensitivity analysis
* Return inital AFT fit
Changes and New Features in 1.1 (2021-12-17):
* Fix documentation bug
Changes and New Features in 1.0 (2021-12-16):
* First release after almost 3 years of testing