You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, @bgroenks96, thanks for reporting:)
There are at least two bugs here, unfortunately, and maybe a third one if you consider the fact that prefix isn't exported to be a bug. We'll push fixes for these next week, hopefully.
As for where you report it, please don't fret about it, it's trivial for us to transfer the issues to the right repository. As it happens the error message that you get is a DynamicPPL problem but there's also a broader Turing/AbstractMCMC problem about the prefixes in the chain (it turns out it's actually in DynamicPPL; I've now put that here #788).
The new
to_submodel
implementation does not work correctly with prefixes specified viaprefix
.Here is an MWE using Turing:
The parameter named tuple returned by
rand(m)
is correct:But the call to
sample
throws an error:This does not occur if autoprefixing is enabled. However, in that case, the manual prefix seems to be ignored in the resulting chain:
Maybe this is actually a bug in Turing...? Feel free to move this issue if so.
The text was updated successfully, but these errors were encountered: