search

baggepinnen / MonteCarloMeasurements.jl

Propagation of distributions by Monte-Carlo sampling: Real number types with uncertainty represented by samples.
https://baggepinnen.github.io/MonteCarloMeasurements.jl/stable/
MIT License
261 stars 17 forks source link

Fun use case: Variational inference #23

Closed cscherrer closed 5 years ago

cscherrer commented 5 years ago

Your package is making it a lot easier to think about this sort of thing. I'd like to get this automated for Soss models, but for now it's hard-coded and seems to work pretty well: https://gist.github.com/cscherrer/72062a3b9f264a328f00159d84a61b98

baggepinnen commented 5 years ago

Cool! I can't say I understand exactly what is going on.. You're maximizing the elbo of a linear regression model with normal priors on the coefficients using a VIS (?) algorithm?

baggepinnen commented 5 years ago

If you think this example can turn into something that others might find useful, we can add it to the examples :)

cscherrer commented 5 years ago

Yeah I really need to write it up. It was just a crazy idea that ended up working surprisingly well, and being able to think and work in terms of particles made it all much easier. Pretty great abstraction once you get used to it :)

Hoping to get through a blog post today giving some details

baggepinnen commented 5 years ago

Wow yeah that sounds nice, I can link it from the Readme. It's great to hear that the package is general enough to work for something completely unforseen!

baggepinnen commented 5 years ago

I added a link to the blogpost in the readme :D