Hello all,
I am fitting quite large Stan models with large data and I am trying to use the cross validation as described in the link with loo_subsample() :
Now since my models are quite big, I cannot really fit it locally so I do not have access to the model fit object as described in the documentation. Instead, I work with the Stan output csv files that I process externally and extract the point-wise log-likelihood values.
Is there a way to make loo_subsample() work with such input file when I provide the log-likelihood values directly without having access to the model fit object?
Hope my question makes sense,
Thank you!
Update - I attach a sample csv file with point-wise log-likelihood values concatenated across 4 chains of 1000 samples each.
log_lik_sample.csv (709.9 KB)