[04:26:32] (03PS1) 10Ammarpad: Reword messages shown on prefs page [extensions/ORES] - 10https://gerrit.wikimedia.org/r/541137 (https://phabricator.wikimedia.org/T228951) [13:33:53] o/ [15:14:42] o/ codezee [15:14:46] how are you doing? [15:18:24] halfak: o/ [17:09:22] 10ORES, 10Scoring-platform-team, 10Product-Analytics: Controlled experiment for enabling ORES filters by default - https://phabricator.wikimedia.org/T234468 (10kzimmerman) @Halfak Product Analytics is tagged on this; is that as a heads up or a request? Moving to tracking for now. [17:15:24] 10ORES, 10Scoring-platform-team, 10Product-Analytics: Controlled experiment for enabling ORES filters by default - https://phabricator.wikimedia.org/T234468 (10Halfak) Yes. Just a heads up. We might come back to you for advice when we can get this prioritized. Thanks! [17:29:33] 10ORES, 10Scoring-platform-team, 10Product-Analytics: Controlled experiment for enabling ORES filters by default - https://phabricator.wikimedia.org/T234468 (10kzimmerman) Sounds good! [18:56:24] 10Scoring-platform-team, 10editquality-modeling, 10artificial-intelligence: Implement hunspell dictionary for euwiki article quality model - https://phabricator.wikimedia.org/T223788 (10Halfak) [18:59:28] 10Scoring-platform-team, 10editquality-modeling, 10artificial-intelligence: Implement hunspell dictionary for euwiki article quality model - https://phabricator.wikimedia.org/T223788 (10Halfak) @Theklan, do you think it's likely we'll find examples of articles that are long enough to be high quality but cont... [19:31:33] wikimedia/revscoring#1750 (basque_minimal - 57eb7de : halfak): The build was fixed. https://travis-ci.org/wikimedia/revscoring/builds/594765687 [19:41:01] * halfak does the build-fixing dance [20:39:33] 10MediaWiki-extensions-ORES, 10Scoring-platform-team (Current), 10MW-1.34-notes (1.34.0-wmf.25; 2019-10-01), 10Patch-For-Review, and 2 others: Remove SpecialContributions::getForm::filters hook call - https://phabricator.wikimedia.org/T219238 (10Volker_E) [21:10:41] halfak, Nettrom: it looks like article quality models are only available on a few projects, is that correct? https://tools.wmflabs.org/ores-support-checklist/ But my sense is that since many/most of the features the model uses are structural, you could make meaningful guesses about the quality of an article on a wiki without a locally-trained model, based on similarity to, say, English articles with those features. Is this a thing? [21:11:39] It's been a perennial idea, but not something we have implemented. [21:11:42] In theory it would work. [21:12:06] But one of the big issues is that the model would have unknown performance characteristics in a new context. [21:12:15] I'd support the work of someone to explore that. [21:12:18] J-Mo: the notion of quality does to some extent vary from wiki to wiki (ref Stvilia's work, I can look up the paper if you want) [21:12:26] We have the data. We have some good ways to move forward [21:12:52] what wikirank.net does is use FAs as the benchmark on a given wiki and compare against that, because a lot of Wikipedias have FAs [21:13:14] Nettrom makes a solid point. Not only do the definitions and the classes sometimes change, but what is good enough for any given class also shifts. [21:13:40] (if my understanding of wikirank.net is correct based on reading their papers) [21:13:43] There are some local concerns about feature extraction too. E.g. the {{sfn}} template in enwiki adds a tag and that's hard to know without local knowledge of the wiki. [21:13:44] 10[1] 10https://meta.wikimedia.org/wiki/Template:sfn [21:13:50] Shh AsimovBot [21:14:26] yeah, I was thinking that template differences might not always localize [21:14:31] have you thought about wikilabels campaigns? Maybe use the scores-based-on-Enwiki as a first pass, then asking locals to verify? [21:14:42] Sounds like a good idea. [21:15:06] * halfak looks for someone else to do the hard work while he advises and gets middle author credit. [21:15:08] no shame [21:15:09] ;) [21:15:20] hah. you do plenty of hard work :) [21:15:25] <3 [21:16:42] I'm taking a stab at a research team agenda for knowlege gaps. Seems like a natural next step would be trying to extend our AQ models across wikis. I'll advise we do this (who 'we' is still TBD) [21:18:00] J-Mo, if I may, I think there's a triangulation here between topic, quality, and demand. [21:18:14] So +1 for pushing on the cross-wiki quality dimension. [21:18:33] cross-wiki topic is a drum I want to beat. [21:18:58] I think we (essentially) get cross-wiki demand for free. [21:19:09] agree. per the white paper we're also going to need to account for gaps in supply vs. demand, and coverage by topic. [21:19:14] Actually wait. I take that back. Demand is too low level. We want "importance" but that is complicated. [21:20:43] Agree. I just finished a close read of https://www.frontiersin.org/articles/10.3389/fphy.2018.00054/full and now I'm thinking we need to revisit the idea of importance in terms of "cultural context" [21:22:39] btw halfak (totally unrelated). One of my students is trying to install ores in python3 (anaconda). And getting some incompatibility errors: "There were multiple errors. For example, one was with mmh3 hashing library. It is a known issue and there is a pull request waiting to be accepted which resolves the bug with mmh3, but the current version is incompatible. There are also some error with the pyyaml and another library" [21:22:47] you don't need to triage this. just noting it [21:26:10] Hmm. I'm surprised that is an issue unless they are working from Windows -- in which case, things are crazy. [21:26:22] Mac Mojave [21:26:40] But that said, there are a lot of dependencies for ores that are unrelated to API access. [21:26:45] I should make a separate package. [21:27:13] All it would depend on is requests -- and that works everywhere. [21:27:17] I'll have a quick look at that. [21:27:26] I'll pass the word along, and encourage him to file a bug. He can always use the API [21:28:34] Hmm. I think I can actually whip this together in 10 minutes. [21:28:40] * halfak tries [21:29:16] wow [21:40:19] OK it works and I have a cute example but how do I register a python package again [21:40:21] ? [21:41:17] https://github.com/halfak/oresapi [21:43:27] Upload is stuck at 82%. WHY [21:43:35] Yay! OK [21:43:50] J-Mo, `pip install oresapi` should work in about 2 minutes. [21:44:52] So it took 13 minutes. [21:44:57] Or 15 if we're still waiting on pypi [21:45:10] I've been meaning to do this for a while. [21:45:17] I think I want to move some other stuff into this package. [21:45:30] I just pinged the student. Told him to give it a try in a couple minutes. [21:47:08] Noice [21:47:20] * halfak goes back to regularly scheduled work :) [21:48:29] halfak "it installed with some warnings". Thank you! Darshan (student) said he'd file a bug if he gets blocked. [21:48:40] OK cool. :) [21:49:05] I set the requirements to be really general so hopefully that's why there are warnings. I figured that I wanted to leave the student unblocked if at all possible. [21:49:14] We can get specific with the requirements later :) [21:56:44] halfak: so I found something else to debug in the historical scoring step [21:56:50] so a little more delay on that :( [21:57:37] halfak: but I've also been thinking about using hierarchical models to pool data from the different wikis in a simpler way [23:10:35] halfak: sorry about icinga2 downtime earlier that was due to the cloud services oopsie [23:11:45] I ought to see about getting you guys away to maybe exec a restart of icinga2-wm somehow, (not really needed but a quality of life thing) [23:11:57] a way*