TyXe-BDL / TyXe

MIT License
144 stars 33 forks source link

Development status #16

Closed icfly2 closed 2 years ago

icfly2 commented 2 years ago

TyXe is recomended to be used instead of Pyro HiddenLayers, but it looks like it is pinned to an older version of Pyro. What is the development status of this library?

I'd be happy to contribute, there are some marked first issues, but before adding work I'd like to find out what the original authors have in mind with this package.

hpplyt commented 2 years ago

Development has been somewhat dormant (as per the repo activity, there's nothing secret in the works :) ). This is more due to a lack of time than to a lack of interest though, so if you'd like to contribute I'm more than happy to support you in this.

The older pyro version is mostly due to one of the tests and the PretrainedInitializer breaking with the next pyro version, the rest of the library should still work. It's a fairly minor change in the way the BNN variables get named by pyro, I haven't tracked down the exact source of this, but it should just be a small fix.

Are you interested in the library from the BNN side or do you have specific applications which need more support? In the former case I can give you pointers to some smaller features I have in mind that could be fun to implement. In the latter case just let me know what your're working on and what functionality you're missing.

icfly2 commented 2 years ago

Thansk for the quick and detailed reply. I was hoping to initially use it for some classification problems on tabular data with different prior distributions (some Normal, some Poisson, some Binomial etc).

I'd be happy to help with makign it compatible with the latest pyro and pytorch, but probably won't be able to do it alone. Also hosted docs somewhere and publishing to pypi would be nice, and something I'm happy to help with.

I assume from the lack of development you aren't using this package anywhere anymore?

hpplyt commented 2 years ago

Thanks for your work on #18 ! I'll look into setting up continuous integration and publishing to pypi this week.

Regarding hosting docs, we have set up basic API docs on readthedocs. Obviously missing tutorials and would probably benefit from spending a bit more time with sphinx to make them look prettier.