Klara.jl
Klara.jl copied to clipboard
JuliaStats
Convergence Diagnostics
Would there be interest in some code for a couple of convergence diagnostics?
Yes, there would; which convergence diagnostics would you be interested in coding?
Yes, Julia provides all the required machinery for implementing r-hat (it is not one of those mcmc diagnostics relying on AR processes), so go for it. Here is some sample R code for it:
http://www.stat.cmu.edu/~hseltman/rube/rube0.3-7/R/Rhat.R
Sorry, I've caused the thread to look confusing by leaving a comment suggesting Rhat and then changed my mind and deleted it right away (since I figured you guys are all familiar with that diagnostic and I had second thoughts about butting in).
There is nothing wrong with your suggestion, it would be very helpful if you would like to code Rhat and open a PR; we haven't implemented it (we haven't simply taken the time to do it), so it would be a nice contribution.
We need to figure out what license that code was released under before we can use. I suspect it's not MIT-licensed.
I am not familiar with licensing issues when it comes to coding as the boundaries are sometimes blur; would it make sense for example to simply reference Gelman's paper that introduces Rhat? Otherwise we could simply ask for the aurhors' permission to use their code as a reference point to reproduce the routine in Julia perhaps.
I was thinking about r-hat, since there seems to be some interest I'd be happy to start on that. I'm not sure about the licencing either, though I know PyMC has a version. I can look into that.
Ok, keep us posted here once you know more about the licensing issue and you have progressed with coding - I am sure you won't be writing to us from prison next time we hear from you.
It's definitely fine if you work based on the paper or book instead of the code.