MansMeg
commented
3 years ago This is definitely of interest. Especially models that has been published (and break algorithms). So astrophysics examples are excellent!
here are some instructions on how to add data, model and posterior objects. Im happy to help if you encounter any problem. https://github.com/MansMeg/posteriordb/blob/master/doc/CONTRIBUTING.md
The only issue now is that we need to implement the model as a stan model since we yet have not implemented PyMC3 models. But if you have both, I would be happy to collaborate and find a way to get both into the database.
JohannesBuchner
Dear all,
I am wondering if there is interest to incorporate some problems from astrophysics. On the one hand, I have assembled a few real problems which use specialized libraries, in exoplanet, cosmic microwave background and gravitational wave analysis. I assume they cannot be easily accessed from Stan, but getting their posterior right is important. A perhaps easier example is the radial velocity challenge data by Nelson+2020 (with data and problem specification available at https://github.com/EPRV3EvidenceChallenge/Inputs/ ), which only requires Gaussian processes and a Kepler orbit solver; the GP kernel and measurement uncertainties given. I understand there is at least one package which allows doing this through PyMC3.
On the other hand, there are some analytic problems that have been useful for breaking algorithms, such as the LogGamma mixture of Beaujean & Caldwell (2013). Also, I have found that a d-dimensional product of beta distributions with a, b drawn uniformly between -1 and 1 seems to break some simple algorithms already when d is between 10 and 20.
Cheers, Johannes