Skip to content

Request for inclusion: metric-learn #40

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
6 tasks done
wdevazelhes opened this issue Jul 22, 2019 · 18 comments
Closed
6 tasks done

Request for inclusion: metric-learn #40

wdevazelhes opened this issue Jul 22, 2019 · 18 comments

Comments

@wdevazelhes
Copy link
Member

wdevazelhes commented Jul 22, 2019

Request for project inclusion in scikit-learn-contrib

  • Project name: metric-learn
  • Project description: A scikit-learn compatible library for metric learning, including learning a metric from labels, from pairs, from quadruplets, and more to come
  • Authors: CJ Carey, Yuan Tang, William de Vazelhes, Aurélien Bellet, and Nathalie Vauquier
  • Current repository: https://github.com/metric-learn/metric-learn
  • Requirements:
  • scikit-learn compatible (check_estimator passed): check_estimator passed for supervised learning estimators, but not for weakly supervised learning ones, since they take as input 3D arrays (or 2D arrays of "indicators", of size (n_samples, 2) for pairs, or (n_samples, 4) for quadruplets for instance), therefore they don't pass the tests on toy examples (we could elaborate on which tests don't pass if needed)
  • Documentation (guide, API reference, example gallery) http://metric-learn.github.io/metric-learn/
  • Unit tests (coverage: 96%)
  • Python3 compatible
  • PEP8 compliant: it is not fully PEP8 compliant, this is indeed something we need to improve. It is now PEP8 compliant, except for the indentation
  • Continuous integration: the continuous integration checks tests and coverage
@wdevazelhes wdevazelhes changed the title metric-learn Request for inclusion: metric-learn Jul 22, 2019
@adrinjalali
Copy link
Member

Nice, this looks like an easy one to include. I'd be happy to have it the contrib.

@rth
Copy link
Contributor

rth commented Jul 25, 2019

Thanks @wdevazelhes, The documentation and package quality is very nice! A few comments,

  • it could be indeed useful to improve PEP8 compatibility. I have not seen that many flake8 errors, apart from the 2 space indentation. It's a (regrettable :)) choice some projects make, I wouldn't consider it a blocker.
  • it would be good to add Python 3.7 to CI tests.
  • for the weakly supervised models that take as input another data structure than a 2D matrix, the same would apply to ML libraries for modeling time series, hybrid recommendations etc. As long as it's clearly documented, I don't think it should exclude them from scikit-learn-contrib.

Overall I'm +1 for inclusion as well. Let's wait a few more days in case there are other comments.

@agramfort
Copy link
Member

+1 for me

@wdevazelhes
Copy link
Member Author

Thanks for the review and comments @adrinjalali, @rth, and @agramfort, we just fixed the PEP8 errors (except the indentation), and added a check in travis so ensure future PRs are PEP8 compliant. We also added a test to check Python 3.7 compatibility.

@rth
Copy link
Contributor

rth commented Aug 1, 2019

@wdevazelhes I have created the metric-learn team in this org and invited you there. You should have the permission to move metric-learn under this org, but if it's not the case please let me know. Also please let me know if other contributors should be added to that team.

@wdevazelhes
Copy link
Member Author

@rth thanks a lot ! I'll try the transfer. Yes, could you add the other members from the metric-learn org ? Namely @perimosocordiae, @bellet, @nvauquie, and @terrytangyuan

@rth
Copy link
Contributor

rth commented Aug 1, 2019

Sure, invites were sent for https://github.com/orgs/scikit-learn-contrib/teams/metric-learn/members

@wdevazelhes
Copy link
Member Author

Thanks @rth
I did the transfer, though I don't know why we don't have edit access to the repo scikit-learn-contrib/metric-learn anymore
Did I do something wrong ? When transferring, github asked me which teams should have write access to the package, and I put the metric-learn team

@rth
Copy link
Contributor

rth commented Aug 1, 2019

Great that it worked! It looks like default permissions are read. Changed them so that @scikit-learn-contrib/metric-learn members have admin permissions to https://github.com/scikit-learn-contrib/metric-learn. You would likely need to re-setup CI for the repo (unless they now handle it automatically).

Closing as resolved. If there are any other issues, please let me know by commenting in this issue :)

@rth rth closed this as completed Aug 1, 2019
@terrytangyuan
Copy link
Member

Thanks @rth! @wdevazelhes Seems like the documentation page is not available anymore: http://metric-learn.github.io/metric-learn/. Should we redirect it to http://contrib.scikit-learn.org/metric-learn/?

@bellet
Copy link
Member

bellet commented Aug 1, 2019

Yes if we could redirect it (ideally, with HTTP status 302: Moved permanently), that would be great

@bellet
Copy link
Member

bellet commented Aug 1, 2019

@terrytangyuan do you know how to do that?

@rth
Copy link
Contributor

rth commented Aug 1, 2019

https://help.github.com/en/articles/redirects-on-github-pages might help. However as far as I understood that would require to create a metric-learn/metric-learn repo meaning that automatic redirects from https://github.com/metric-learn/metric-learn to the new location will stop to work.

If you do that, maybe worth forking the repo back to https://github.com/metric-learn/metric-learn, explcitly writing in the readme that the location of the repo moved, and adding redirects to Github pages could work.

@bellet
Copy link
Member

bellet commented Aug 1, 2019

Thanks, @rth. Not sure we should redirect then - forking the repo is likely to introduce confusion?
I think we also lost the doc url when we moved the package from all-umass to metric-learn organization last year - seems like Google picked up the new address rather quick

@terrytangyuan
Copy link
Member

I would say we can probably leave it for now. I'd prefer to keep the auto-redirect to the new Github repo. I am sure users would be able to do a quick Google search and find out where the documentation is.

@wdevazelhes
Copy link
Member Author

wdevazelhes commented Aug 1, 2019

Also, there seems to be a possible solution here: https://gist.github.com/domenic/1f286d415559b56d725bee51a62c24a7
Creating a Github Pages user metric-learn (and not a Github user) as well as a Github Page, so that the repo metric-learn/metric-learn is still empty (hence redirection), but it allows to edit the Github Page with a redirection strategy

@bellet
Copy link
Member

bellet commented Aug 1, 2019

Good job @wdevazelhes, looks like it is working (for the root page anyway). Not a 301 redirection but better than nothing anyway.

@terrytangyuan
Copy link
Member

Nice work @wdevazelhes!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

6 participants