Packaging self-contained pylearn2 for the PyPI release?

[alexjc]

Since the goal of the project is to make deep neural networks and PyLearn2 more accessible, I'm wondering about packaging up the library along with the PyPI release of sknn. It would be self-contained and sknn would load that up as a fallback... What do you think?

It's not something that the PyLearn2 developers are considering right now (due to the overheads):
lisa-lab/pylearn2#1493

... but could be worthwhile as part of this project. We have some automated tests, and along with PyLearn2's test suite it can be pretty safe.

If there's already a PyLearn2 version installed then I think sknn should use that, maybe printing a debug message: "This is the packaged version of sknn but it's using a version of pylearn2 that's already installed."

Thoughts welcome!

[stonebig]

Until PyLearn2 as a release exists on pypi, your suggestion seems the only meaningfull solution to get a safe/stable one.

Ideal situation would be:

• PyLearn2 exists as a wheel package with a perfectly meaningfull release commit,
• you don't need (then) to re-include PyLearn2 in scikit-neuralnetwork wheel.

[ssamot]

I think I kind of agree with both comments above - packaging PyLearn2 makes the use of the library transparent.

[alexjc]

OK, this seems to work, from the wrap branch. Some links I found useful:
http://peterdowns.com/posts/first-time-with-pypi.html
https://tom-christie.github.io/articles/pypi/

[alexjc]

Assuming there's a symlink to the pylearn2 within sknn, release looks like this now:

python setup.py sdist
twine upload -r test dist/*.tar.gz

(This applies to testpypi.python.org.)