I gifted myself a #Probula Friday.
The library has gained an ability to perform #posterior #inference via grid approximation - on top of the already existing importance sampling.
Had some major fun on the DSL, forcing the type system to track models for which grid approximation is allowed (only models for which we can compute posterior density in given point).
I bundled a sub-language for ranges of doubles: `50 doubles (0.0 -> 1.0)`
I have spent some time cleaning up my home-grown #Bayesian inference library for public consumption. Enjoy:
https://codeberg.org/wasowski/probula
The story goes that I needed a pure Scala3 replacement for #Figaro, that I can use for teaching purposes. The status is:
- Probula can handle regression models
- Importance sampling, monadic style implementation
- Very basic descriptive stats built in.
- CVS export for arviz, to perform posterior analysis.
Andrzej Wąsowski ☑️ 🟥