Skip to content

Happy things! Not yet 1.0 but close!
Compare
Choose a tag to compare
@Naereen Naereen released this 28 Sep 13:03
· 438 commits to master since this release
v0.9.3
This tag was signed with the committer鈥檚 verified signature.
Naereen Lilian Besson
GPG key ID: 01AACDB9C108F8A0
Learn about vigilant mode.
2e6f49d
This commit was signed with the committer鈥檚 verified signature.
Naereen Lilian Besson
GPG key ID: 01AACDB9C108F8A0
Learn about vigilant mode.

Summary of the last 6 months

I have opened this project since more than 6 months, I wrote a companion research paper to present it, its documentation is now handled automatically by ReadTheDocs (see it live here), Travis CI is used to test every commit, etc.

SMPyBandits has been used in more research articles, it received its first other contributor (thanks @guilgautier !), and many other cool things happened.
I have kept SMPyBandits up-to-date with my (small and partial) point of view on the state-of-the-art research in classical or multi-player multi-armed bandit algorithms and heuristics.

I met researchers in a workshop in Rotterdam (Netherlands) and in a workshop in Toulouse (France) who knew my library and my work but didn't know me, and told me that they found this work useful.
Colleagues in Inria Lille has used SMPyBandits, for teaching or research, and I was happy to help them and learn from them.

Directions for next release

  • I want to stay as up-to-date as possible for multi-player MAB as my library is most surely the only one being open-source that implements these models and algorithms. I need to work on #145 and #139 mainly, in October and November. Maybe #120 but I won't have time for all.
  • I want to work on more general non-stationary problems, see #17 with for instance the algorithm from #100, or even harder models like #123, #124.
  • I need to finish my work on the documentation, mainly #138 (@guilgautier thanks the idea).
  • With more time, I want to work on #140, #135 too.

Disclaimers

Please keep in mind that this is only meant as a research framework: easy to interact with, easy to modify, and easy to do some small or medium-sized simulations and get nice figures for research paper.

It is not meant as an industry package for multi-armed bandits. If you want to use any MAB algorithms for real-world content optimization, you should rather implement them yourself to better suit your needs.

Please help and contribute!

With that being said, I am still excited to share this project on GitHub, now on its own organization instead of my personal profile.
If you have any suggestion on how I could improve this project, I would be delighted to here them! Contributions like issues, pull requests, questions etc are welcome.

Assets 2