Implementing Model Averaging

[nels]

Thank you @avehtari for your feedback.

@avehtari

Then from yrep you can compute what you need. @nels, is this what you have done?

Yes, taking y_rep samples from each model in proportion to the weights.
(just using rstan::sampling directly, with y_rep as a generated quantity, instead of the rstanarm::posterior_predict interface in your code sample.)

@avehtari

I have also computed the weighted average of the log-likelihoods from the posterior and fed that back into loo to obtain a WAIC value that I compare to the individual models’.

This is wrong.

Thanks, I suspected it wouldn’t be that simple.
Abandoning my attempt at loo-comparing the averaged model to individual models for time being.

Thanks again.