Modeling human response time data with an exponential model: model comparison issues

Hi all,
This should be fixed on ArviZ most recent version: pointwise log likelihood values from multiple variables are now stored automatically.
The only case where no log likelihood is stored is when logp_elemwise cannot be evaluated (for example with models modified after sampling or when variables have been removed from trace). There should be a warning in these cases that log likelihood cannot be evaluated.

Note that IC functions do not work yet – some extra work has to be done. The workaround is to manually combine log likelihood data into a single array and store it in sample_stats as log_likelihood variable.

Hope this helps clarify things :vulcan_salute:
cc @OriolAbril

1 Like