There’s a PyStan 3 alpha available. It’s not ready for general use. It works well enough for me that I’m going to start using it for my day-to-day work. You can install the alpha release on most systems with python3 -m pip -pre install pystan
. (No Windows support yet.)
I’d appreciate some help with the following items:
- Try out eight schools and confirm things work on your machine.
- Try out the new API and propose changes. The “spec” for the new API (which covers RStan and PyStan) is on the stan-dev wiki.
What’s new in PyStan 3
PyStan 3 is a rewrite. There’s a new user-facing API. Models and fits are automatically cached. Python 3.6 and higher is supported. The code has been simplified, making contributing more inviting.
User-facing changes
- New interface.
- Automatic caching of Stan models and fits.
- Uses the same argument names as CmdStan (e.g.,
num_samples
,num_warmup
). -
-DSTAN_THREADS
by default. - Python 3.6+.
Developer-facing changes
- Easier to maintain. 60% fewer lines of code.
- Split into “front-end” and “back-end” packages,
pystan
andhttpstan
. -
pystan
is pure-Python, enabling faster development. (Running tests is ~80% faster.) -
httpstan
calls Stan C++ services functions and returns raw output. - ISC license.