Scikit learn examples as LIVE Jupyter notebooks #12075
Conversation
|
There is an option in the recent versions of sphinx-gallery to do this. We already use sphinx-gallery in scikit-learn, maybe it would be good using this option? PS: we should explore interplay between sphinx gallery and Jupytext, but that discussion should happen on the sphinx-gallery tracker. |
There is a #11221 to do just that. Now that 0.20 has been released, maybe it is a good time to review it. The main thing to decide is whether we want binder to use scikit-learn stable (0.20 right now) or from master. |
|
Hello @lesteve , good to see you here! I am not sure I completely got the point about the scikit-learn version in binder. Do you mean the sphinx gallery binder is able to run an unpublished version of scikit-learn (is that what you mean by master)? Second question, do you have any idea why the jupytext binder starts faster than the sphinx gallery one? Is that related to having the jupytext binder using pypi's version of scikit-learn? |
master is the development version yes.
I suspect this is because of the caching mechanism in binder. The first time is quite slow because you need to build the docker image but next time the docker image is found and it is a lot faster. Having said that, I think our scikit-learn.github.io repo is very big and this may be a problem to use it for binder. I'll put a bit more details in #11221 to try to keep the conversation in a single place. |
I haven't really understood the binderhub infrastructure. Does it fetch the repo through repo2docker via I also don't know if a partial checkout could be applicable here. We should not need to check out anything but |
|
In fact, it's a bit weird that it relies on repos at all, if it does not make use of history. All it should really need is something equivalent to read-only FTP! But if a feature request to binderhub or repo2docker is appropriate, I'd appreciate help proposing it precisely. |
|
Ping @choldgraf |
|
A few of Joel's questions are answered in #11221 (comment)
If you run
Correct. Being able to use sources other than git repositories has been on our wish list for a while. Progress is slow because "Too many things to do" :-/ |
|
what @betatim said :-) we'd love for repo2docker to support non-git code artifacts, PRs and scoping via issues welcome! |
|
So we should be closing this in favour of #11221? |
|
👍 for that. |
|
Sure, you should do as best fit the project. I had proposed the PR as I liked the idea of contributing only two short files. And the resulting binder starts quickly (link in first comment). But do I understand that it is not as well integrated with the sphinx gallery as #11221 will be, and that my PR does not provide a fine control on the sk-learn version. |
|
But do I understand that it is not as well integrated with the sphinx
gallery as #11221 will be, and that my PR does not provide a fine
control on the sk-learn version.
But maybe this could be done in sphinx-gallery? ie a PR on sphinx-gallery
that enables binder via jupytext?
|
|
@GaelVaroquaux could you explain a bit more what you mean by this? Do you mean replacing the sphinx-gallery |
|
@GaelVaroquaux could you explain a bit more what you mean by this? Do you mean
replacing the sphinx-gallery .py -> .ipynb code with using jupytext?
At least plugging jupytext on the .py files.
|
|
@GaelVaroquaux I opened this back in October :-) sphinx-gallery/sphinx-gallery#424 maybe we can discuss implementation there? |
|
Good! I will close this PR, and instead subscribe and contribute to the two other trackers if I there is anything I can help with. See you there! |
The binder project and Jupytext can turn the beautiful scikit-learn example library into a collection of interactive notebooks.
See the proposed change on the README, and try it now:
.