bgoodri
bgoodri
The parameters are estimated under an orthogonal rotation (the Q matrix from the X=QR decomposition) and then after the initial posterior distribution has been obtained the coefficients are re-rotated back...
Now that the CRTP stuff is in the version of RStan on CRAN, we might be able to get this into rstanarm in about a month. I had to not...
I think the 32bit Windows thing is not a major concern anymore. On Sun, Jul 26, 2020 at 2:48 AM Sam Brilleman wrote: > ok, cheers @bgoodri . I have...
We still can't afford to run tests or examples on 32bit Windows, but we can compile almost anything within the address space limitations as long as we use LTO. LTO...
Possibly ran out of RAM. Possibly need to first install `StanHeaders` from source.
The `decov` prior is scaled by the error standard deviation in Gaussian models (and by the overdispersion in negative binomial models) like lme4 does. If we were to scale by...
Yeah, also in things like that sleepstudy example, you can and probably should rescale the outcome variable to not be in milliseconds. But that is something I think researchers should...
It isn't a big bottleneck to do the QR decomposition once in doubles, but if we could do it faster I wouldn't object. But I had assumed GPU stuff was...
@jburos I think the GitHub actions were already enabled for the survival branch but were not being executed due to the merge conflicts. I fixed those and now `R CMD...
My impression was that the unit testing was not as extensive as for the rest of rstanarm. It would be good if SBC has been used to make sure that...