-
Notifications
You must be signed in to change notification settings - Fork 24
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Can Laplace's getNodeNamesVec return mismatched length? #1426
Comments
@perrydv I wonder if they used a Dirichlet or something that was transformed? A reproducible example would be helpful. |
@weizhangstats @perrydv @paciorek
|
@paul-vdb Thanks for the example. I can run the code fine on my machine - the last line does not lead to crashes. The results do not look good though. I will take a deeper look into this. |
@paul-vdb @paciorek @danielturek I took at look at this and there are two issues, one solved and the other pointing to more investigation. (Also @paul-vdb, just making sure: in this example there are two entirely independent sub-models and the part with the multinomial data does not have any random effects. I think you were working on that and must have stumbled into this issue.)
Here is the working version:
|
@perrydv Yes you are correct. I was just forcing a dimension reduction with the parameter transformation in an otherwise simple model to see if it would potentially create any issues. |
@perrydv @paul-vdb @weizhangstats I am wondering how to close this issue given the awkwardness of having Laplace results depend on a prior that we don't otherwise use. At the least I suppose we might mention in the user manual that we suggest not using uniform priors for variance/sd/precision parameters when using Laplace. One step beyond that is that we emit a [Note] warning users they may want to use a different prior (one that has domain extending to Inf) in this situation. Though I'm not sure how systematically we can detect it. |
@paciorek In the updated Laplace code, we've added a findMAP option which will make use of the priors. |
Perry and I decided that for now I will just put a note in user manual and roxygen to be cautious about using priors with a bounded domain for variance component parameters. |
@perrydv @paciorek @weizhangstats Hey guys, going back to the original issue from this thread. I noticed in my current updates that This issue is resolved now with making sure I think other parts of this thread are still valid though, and worth discussing. |
@paul-vdb @perrydv If there is only one param/random effects node, we expected one to use |
@weizhangstats I've done that. |
We have a report from a user (without reproducible example) that the Laplace method
$getNodeNamesVec
can return a vector whose length does not match the length of the input parameter vector. I am noting this here to investigate.The text was updated successfully, but these errors were encountered: