I’m first training using a model and a guide as defined in the Bayesian Regression code (the one on github), I would like to compute the Log Likelihood of some test data. I know how to compute the ELBO loss, either during training with `svi.step(train_data)`

or after training using `loss`

function of an `ELBO`

object (I have not figured out yet how to do it, but I imagine it is possible)

The ELBO is by definition a lower bound of the Log Likelihood. Is the best possible Log Likelihood approximation the ELBO approximation itself ?

Thanks for your time