If anyone’s interested, and people approve (whoever they are), it might be a good idea to take all of the default models in the documentation and add them to posteriorDB. Right now there’s only about 15, I think, from my last PR, but to have basic, commonly used linear/multilevel models and mixture models/AR models (whatever) on benchmarking Stan’s HMC might be a good idea. (The benchmarks seem to vary on each run with small changes, which is a more complicated question).
This would take:
- scraping the Stan documentation.
You could parse the text file that generates the documentation, I think it’s doxygen. You’d have to make sure the scripts you parse compile with Stan’s latest compiler after parsing. - simulating data and making sure the models recover parameters of simulated data (it might be there, it might not be).
- opening a PR and making sure each model is validated.
- Things I’m missing
I’d be willing to review/supervise. Right now we have, kind of, randomly selected models, so it’s not representative of what’s commonly used. I don’t do a lot of text parsing but some people are into this, people that build programming languages, and they like a lot of functional programming. This may not require that level of programming. Probably just for loops in python and if statements but this could be cool. But people who built the new compiler and Stan language parser may be cool to interact with.
Open to objections/rejections, but if anyone’s interested I can supervise. This is pretty straightforward.
~ regards