What is the process for obtaining the LOO-metric for a model in Pymc5?

What is the goto method to retrieve the loo for a model in pymc5? does not seem like the trace include the log likelihood per default. Seems like the stats module got dropped and running an arviz.loo(idata, model) does return an “no log likelihood found…”-error. Should i explicitly compute an variable accounting for the log likelihood or how does it work, didnt find any info about this in the docs for v5. Obviously we dont want to refit the whole/sequentially thing every time but do it in the “smart” way as in v3.


1 Like