conda-forge / conda-forge/staged-recipes
Package vowpal wabbit
- Dominant language
- Python
- Stars
- 846
- Forks
- 6.3k
- Avg merge
- 2d 8h
- Merged PRs (30d)
- 186
Description
It would be useful to package [Vowpal Wabbit](https://github.com/JohnLangford/vowpal_wabbit) ML library
It is a C++ library with Python, R and Java bindings.
I have a few questions regarding the packaging,
* I imagine we first need to add the `vowpal-wabbit` dynamic library, then build extensions as separate packages ? The issue is that from what I can see `python/setup.py` assumes that the `bin/*.so` is located one folder up [[1](https://github.com/JohnLangford/vowpal_wabbit/blob/master/python/setup.py#L20)], [[2](https://github.com/JohnLangford/vowpal_wabbit/blob/master/python/setup.py#L108-L114)] and would probably requires some work to find a way around it
* Alternatively could it be acceptable, to just build a python package e.g. `py-vowpal-wabbit` ?
* The [Linux / Mac OS Python build](https://github.com/JohnLangford/vowpal_wabbit#osx-python-binding-installation-with-anaconda) seems fairly straightforward when compared to the [Windows build](https://github.com/JohnLangford/vowpal_wabbit/blob/master/README.windows.txt). Is it acceptable to skip Windows in the initial submission?
cc @eugene-yang
Essentially, my question is that I can commit to package and maintain the Python VW wrapper on Linux / MacOS, but not so much the rest. What should be done in such cases: wait until complete packaging is implemented by someone or start small?
Contributor guide
Assessment
This issue has not been assessed yet.