loo_R2 and further arguments passed to ‘posterior_epred’ and ‘log_lik’

[avehtari]

loo_R2() allows "Further arguments passed to ‘posterior_epred’ and ‘log_lik’,
which are used in the computation of the R-squared values."

The code has

    ypred <- do_call(posterior_epred, args_ypred)
    ll <- do_call(log_lik, args_ypred)

I have a case where it makes sense to examine LOO-R^2 for posterior_epred(..., re_formula=NA). Now that re_formula=NA is passed also to log_lik() which makes the LOO part to be wrong, and the computed LOO-R^2 values are nonsense (e.g. -0.7, when the correct value is 0.08)

It seems some "further arguments" make sense to be passed to both posterior_epred() and log_lik(), like ndraws and draw_ids, so not passing any "further arguments" to log_lik() is not the solution.

[paul-buerkner]

I think we would have to add new list arguments args_epred and args_log_lik which would then only be passed to the individual functions. I will add this to my todo list for the next release. If you have time, you may of course already make a PR earlier.