[ANN] Announcing TinyGibbs.jl

TinyGibbs.jl came out of my needs for a simple but yet fully functional Gibbs sampler. I basically noticed that most Bayesian sampling in my area of work uses Gibbs sampling and involves statements like: Sample a from p(a|y, b), then sample b from p(b|y, a). I therefore wanted an easy way to implement such samplers in Julia. Other packages, such as Turing, seemed a bit overkill. TinyGibbs therefore only introduced one new macro, the @tiny_gibbs macro which translates a sampling statement often found in papers into a full AbstractMCMC valid sampler. It therefore supports all the methods of AbstractMCMC.jl and MCMCChains.jl.

At this point, TinyGibbs is still rather experimental, but I am curious to hear your feedback about it. What do you think of this, and are there any things that could be improved?


One tiny bit of bikeshedding: I would maybe just call the macro @gibbs instead of @tiny_gibbs.


Thanks! This is actually a good idea. Not sure why I didn’t have it. Will incorporate it into the next version.

1 Like