3 ms·
How to be a Bayesian in Python
- new_test 12y agoIs that prior ever used in practice? How does it generalize to several predictors and interaction terms?
- BoppreH 12y agoFor lower-level use of Bayes I have a library that can be helpful: https://github.com/boppreh/bayesian https://github.com/boppreh/bayesian , or "pip install bayesian". It has simple belief updates and even some classification methods.
- glamp 12y agoanother great post from jakevdp. thanks jake!
- tel 12y agoStan is definitely the project you'll want to follow if you're going to be doing this sort of thing a lot. It's built by Andrew Gelman out of Columbia who is an expert in hierarchical Bayesian methods and has done lots of big models for social sciences (and I believe was responsible for the original NYT 538 model). Stan uses the really clever No-U-Turn sampler algorithm which will help a lot in highly correlated models (where sampling tends to take much, much longer to converge). Edit: Also, if anyone is interested in learning more about MCMC samplers then reading up on NUTS is a good idea. The basic material there does a great job both getting ideas about Hamiltonian MCMC out in the air and also talking about how to do some tricky optimizations to the algorithm while retaining its probabilistic properties.