Look into model selection using 'reversible jump mcmc' about pints HOT 3 CLOSED

Yep - I looked into this very briefly, but some/all of the below might be wrong!

This might be sensible for "nested models" (where most of the parameters map across). The difficulty is the line "provided you have a suitable map between them". For instance, you have an ion channel model C-C-O and jump to the model C-O. I think you need to get to an "equivalent" set of parameters once you've jumped model. I can see how you might be able to go down to smaller models by lumping/averaging/combining parameters somehow, but the leap up to add more parameters seems like another identifiability problem where there would be infinitely many options.

But definitely worth someone looking at more carefully...!

from pints.

I am very late to the boat here (getting up to speed with github!), but don't think you'll want to do this in PINTS. RJMCMC is notoriously difficult to get to work in even modestly complex models. I honestly think that WAIC will be the best bet for model selection (I am also not a fan of marginal likelihoods!).

from pints.

Closing this now until we have a more concrete idea of how/if to approach this

from pints.