Consider models with planet identifiability

[j-faria]

That's probably not the right word, but we might want to assign specific priors for a given planet (for the orbital period if detected with transits for example). I started working on this adding a integer id argument to the RVConditionalPrior::from_uniform and to_uniform but:

• not sure if there is a better way to go at it
• log_pdf also needs access to id
• this might make setting the priors very complicated...

[eggplantbren]

IMO, a much better way to go about this is to

• Identify regions of planet parameter space, after the fact, that the user is happy to use to identify a planet.
• Allow the parameters of that planet to be extracted by conditioning on one planet being in that region.

[j-faria]

By "after the fact" you mean a posteriori, after running the analysis? Doesn't that assume that you somehow managed to find the planet already?

I have to say that I've been debating this option for quite some time. A few people asked me if kima could do this, and I'm always inclined to tell them "no because it's probably not a good idea". But then the answer is "the planet transits, we know it's there and we know its period!"

[eggplantbren]

Yes, I meant after running kima.

If there is prior information about specific identified planets, I could imagine incorporating that into the kima prior by modulating the existing prior by conditioning on N>=1 and modulating the existing prior by something like exp(-0.5*pow(period[0] - p, 2)/var).

[eggplantbren]

I don't think this should be for the current release. That level of customisability will require a lot of effort put into how we are going to implement it and make it user friendly.

[j-faria]

I agree