Hello devs

Suppose I fit two different family of functions (linear and polynomial function) to the same dataset using two different NumPyro models. Is there a (Bayesian) way to determine numerically which of the two models better fit my data? I’m new to Bayesian Model Selection, so any guidance will be much appreciated. Thank you!

1 Like

I recommend applying the PSIS-LOO-CV approach from Vehtari, Gelman, & Gabry – implemented & documented in R as package `loo`

.

In Python, module ArviZ implements it – see their model comparison plots. An example of applying it, along with the ArviZ `loo()`

& `compare()`

functions, from PyMC is here.

1 Like

@apollostream thanks. I’m following the pymc3 approach.