PyStan 3 with non-copyleft license?

@ariddell You mention on this thread that you closed that PyStan 3 has a non-copyleft license:

Is there a PyStan 3 that users can use?

I think it’s time for one or more of us to write lightweight, non-copyleft Python wrappers that call CmdStan. Same for R. That would allow people in these languages to access Stan without all the hassles of our heavy, restrictively licensed PyStan and RStan releases.

A working prototype of PyStan 3 is available. Could use some more tests. It works right now for sampling using the default sampler.

(My sense was that this should be rolled out with RStan 3 and potentially the new Stan language.)

Is there doc on how to build it and use pystan-next?

We haven’t even designed all of the Stan 3 language. So it’s going to be at least a year but I wouldn’t bet under two. Is that in line with your timing?

If you don’t see PyStan 3 being released for two years, then we really need to prioritize a non-GPL replacement for PyStan 2. I think we can find the resources to build a lightweight replacement and I think it’s valuable in its own right. (I mentioned we were thinking about this when it came up at StanCon and the response was enthusiastic, but then it’s easy to get an enthusiastic response when talking about something abstractly.)

At the moment, industrial users with Python need to cobble together their own interfaces on top of CmdStan (which they’re telling me they’re doing). I understand supporting industry isn’t a priority for all of our developers, but it’s important for the project if we expect to get revenue that way.

FWIW, there’s already PyCmdStan, which is under the (non-copyleft) Apache License 2.0.

(looks like solves the immediate

A beta version of PyStan 3 could be ready to go within a month or so. I
hesitate on releasing it if RStan 3 isn’t ready to go though because
they both use a similar API. It makes sense to release them together.

Perhaps I’m being overcautious here. It wouldn’t cause any harm to
release something for early adopters.