[11:28:37] (03PS1) 10Hashar: Define npm peer dependency explicitly [extensions/ORES] - 10https://gerrit.wikimedia.org/r/368695 (https://phabricator.wikimedia.org/T172094) [13:02:25] o/ [13:44:35] 10Scoring-platform-team, 10editquality-modeling, 10User-Ladsgroup, 10artificial-intelligence: Flagged revs approve model to fiwiki - https://phabricator.wikimedia.org/T166235#3486152 (10Halfak) I think that we did not fully understand the meaning of "approved" and that has resulted in us not using that par... [14:59:21] o/ halfak [14:59:35] o/ Zppix [15:00:08] anything you need merged halfak? [15:00:17] Nope. Nothing right now Zppix. [15:00:25] I'm working on https://etherpad.wikimedia.org/p/meta_ores_schema [15:01:40] halfak interesting [15:15:53] 10Scoring-platform-team, 10DBA, 10articlequality-modeling, 10artificial-intelligence: [Discuss] Hosting the monthly article quality dataset on labsDB - https://phabricator.wikimedia.org/T146718#3486484 (10Halfak) 05Resolved>03Open Looks like I'm failing to create one important index on the table. ```... [15:19:02] 10Scoring-platform-team, 10DBA, 10articlequality-modeling, 10artificial-intelligence: [Discuss] Hosting the monthly article quality dataset on labsDB - https://phabricator.wikimedia.org/T146718#3486489 (10Marostegui) >>! In T146718#3486484, @Halfak wrote: > Looks like I'm failing to create one important in... [15:31:05] 10Scoring-platform-team, 10DBA, 10articlequality-modeling, 10artificial-intelligence: [Discuss] Hosting the monthly article quality dataset on labsDB - https://phabricator.wikimedia.org/T146718#3486557 (10Halfak) Maybe? ``` $ host enwiki.labsdb enwiki.labsdb has address 10.64.4.11 ``` [15:31:43] halfak if you got a moment mind helping me with a project of my own i cant find out why my ircbot is spamming with a certain command i have... [15:31:48] (its python) [15:32:28] 10Scoring-platform-team, 10DBA, 10articlequality-modeling, 10artificial-intelligence: [Discuss] Hosting the monthly article quality dataset on labsDB - https://phabricator.wikimedia.org/T146718#3486575 (10Marostegui) >>! In T146718#3486557, @Halfak wrote: > Maybe? > > ``` > $ host enwiki.labsdb > enwiki... [16:06:29] 10Scoring-platform-team, 10MediaWiki-extensions-FlaggedRevs, 10ORES: Decrease FlaggedRevs backlog by using ORES predictions models - https://phabricator.wikimedia.org/T165848#3486765 (10Halfak) [16:06:33] 10Scoring-platform-team, 10editquality-modeling, 10User-Ladsgroup, 10artificial-intelligence: Flagged revs approve model to fiwiki - https://phabricator.wikimedia.org/T166235#3486764 (10Halfak) 05Resolved>03Open [16:15:55] awight hi,upstream now have a patch for removing gwtui from gerrit https://gerrit-review.googlesource.com/#/c/gerrit/+/116790/ [16:16:35] paladox: lol, I love it [16:16:39] i see alot of merge conflicts with that in the future [16:16:40] heh [16:17:19] that is probaly the biggest change in gerrit's history. Due to gwtui being used since well when gerrit was avilable [16:18:57] awight i've started creating a simple plugin to brand polygerrit with wmf branding [16:19:07] it's done through a polygerrit plugin now [16:19:13] + i followed how they did it for chromium [16:19:25] http://gerrit-new.wmflabs.org/r/ [16:40:34] github is down [16:41:02] and now it's back. It seems to be slow + itermittent [16:43:21] 10Scoring-platform-team, 10DBA, 10articlequality-modeling, 10artificial-intelligence: [Discuss] Hosting the monthly article quality dataset on labsDB - https://phabricator.wikimedia.org/T146718#3486863 (10Marostegui) 05Open>03Resolved ``` mysql:root@localhost [u2041__ores_p]> CREATE INDEX page_idx ON m... [16:54:14] 10Scoring-platform-team, 10Wikilabels, 10editquality-modeling, 10artificial-intelligence: Change "yes/no" in damaging_goodfaith form to "damaging/good" and "good-faith/bad-faith" - https://phabricator.wikimedia.org/T171493#3486886 (10Halfak) a:03Natalia [16:54:31] 10Scoring-platform-team, 10ORES, 10Operations, 10Patch-For-Review, 10User-Joe: Stress/capacity test new ores* cluster - https://phabricator.wikimedia.org/T169246#3486887 (10Halfak) a:03Halfak [17:01:11] fajne, is it OK if we delay our call for 15 minutes? [17:01:25] sure [17:01:58] 10Scoring-platform-team, 10DBA, 10articlequality-modeling, 10artificial-intelligence: [Discuss] Hosting the monthly article quality dataset on labsDB - https://phabricator.wikimedia.org/T146718#3486911 (10Halfak) <3! [17:02:00] Amir1: are you available? [17:02:04] cool. Thanks [17:02:36] fajne: yeah, but I have a meeting with Ops, they haven't got on call yet [17:02:41] but they will be [17:02:56] okey then [17:05:44] fajne: o/ Sumit here, what are your thoughts regarding incorporating sentiment beyond lexic sentiments as given here https://etherpad.wikimedia.org/p/natalia_aaron ? [17:08:20] codezee: beyond lexic sentiments? [17:09:29] fajne: meaning supervised methods or including domain specific info [17:10:05] fajne: like my current experiment involved use of only sentiment wordnet which is kind of a global db for polarity scores [17:10:16] oh, I thing accounting for domain when doind sentiments is a very sane ides [17:10:21] *idea [17:11:32] then the same revereted edits could be used as a training set for any new feature developed around setiments i think [17:16:38] codezee: the problem with reverted edits is that it's not 100% reliable: not all edits that are reverted should have been reverted. Maybe using handtagged Goodfaith set would be better? [17:17:50] fajne, I'm ready when you are [17:25:10] yes that is right... [18:05:40] Heads up, I’m taking a friend out for maybe 1.5h, and another 30 min gap at 2pm Pacific [19:11:38] halfak: how do i enable memcached on my computer? [19:12:59] Good Q. Can't look it up right now. Ask google :) [19:16:34] halfak: whatever google is suggesting seem not work for Bash on Ubuntu on windows 10. Can I use an existing memcached running on wiki servers? [19:17:14] Regretfully not. [19:39:31] OK Back now. [19:39:36] fajne, any luck with memcached? [19:40:03] halfak: getting through [19:40:12] OK. I'm looking at docs for memcached now. [19:40:17] Let me know if you need a hand. [19:41:10] what version of windows 10 are you running? [19:44:59] halfak: ok, i got there [19:45:37] to the campaigns page (http://localhost:8080/ui/enwiki/) i mean [19:46:30] is that ok that i can see 4 campaigns: Draft notability (raw), Draft notability, Edit Type -- 2015 january sample and Edit Quality -- 2014 10k sample ? [19:46:52] Yup [19:47:00] Super old campaigns we re-used for example data :) [19:47:09] i expected to see something like edit quality 2015 sample.. [19:47:20] ok [19:48:16] daaaaamn [19:51:08] no, it works [19:51:22] I'm back [19:52:33] \o/ [19:52:39] o/ awight [19:52:45] How goes the fiwiki stuff? [19:53:27] Just starting to dig in now, but I haven’t been able to touch a thing since our meetings. [19:54:04] Gotcha. Cool. Let me know if you want to chat about it. [19:54:24] will do! [19:54:41] BTW, I was thinking that we should not use the "reverted_for_damage" field but rather the "needs_review": "reverted" [19:54:44] As the indicator. [19:54:56] Because sometimes things are reverted but not "reverted_for_damage" [19:55:03] And we should exclude them regardless. [19:55:09] I think I have two things to do in the approved query: 1) filter for one-revision chains only, and 2) remove anything that was later reverted. [19:55:16] +1 [19:55:21] huh [19:55:35] I didn’t know there was any such reverted_for_damage field [19:56:02] That's what we use to train the "reverted" models. [19:56:12] ah ok I’ll read the Makefile [20:01:01] O_O I thought we had a 1pm meeting. Canceled? [20:02:44] halfak: ^? [20:03:32] Yeah, sarah sent an email to me earlier. Sorry I should have forwarded. [20:03:46] I was supposed to review her work this AM and asked her about it. [20:04:01] She wasn't quite ready and said she wanted to work with you for a bit this week. [20:04:05] So I moved the meeting to Weds. [20:07:11] ok rad. We had planned to cowork that day so we could do multimodal knowledge transfer :). too bad it’s just me on the Left Coast, and I have about one month worth of knowing anything. [20:10:03] halfak: it doesn’t really matter at this point, but I’m wondering if you saw akosiaris’s comment about industry-standard load testing tools, e.g. apache bench? [20:10:27] awight, I did. [20:11:04] these industry standard tools aren't doing anything very difficult and they wouldn't be able to read our results and provide reports. [20:11:06] I would be curious to run that side-by-side with the custom tool to verify results [20:11:18] I don't think that is necessary [20:11:19] sure, that make sense. [20:11:22] ah ok [20:11:27] me neither ;-) [20:11:33] We have several things that record request rates and the proportion of errors. [20:12:50] yeah but the details could be tricky, e.g. managing sockets on the benchmarking host, and a fine-grained breakdown of net timings might be helpful. meh [20:13:08] * awight wanders back to fiwiki [20:14:00] * halfak wonders if awight is trying to add more work to his plate [20:14:01] :P [20:14:14] lol. procrastinating, yeah maybe [20:14:33] Seems wrong that in https://phabricator.wikimedia.org/T169246#3481494 the error breakdowns are the same for every model. [20:14:57] and do not add up to 239 [20:15:09] also, successes are the same for every model [20:15:29] Why does that seem wrong again? [20:15:30] I guess that’s just requests - errors [20:15:40] why would you get the same number of timeout errors for each model [20:16:01] Because most timeouts occur in feature extraction and most models use similar features. [20:16:18] The most time-intensive features involve tokenizing the text of the page. [20:16:38] aha you were using the same revision range for each model [20:16:40] When feature extraction fails for one model in the request, it's very likely to fail for all. [20:16:57] Um... the request expects a score for all of the models at the same time [20:16:58] So yea [20:17:06] oh the different models were all requested in a batch [20:17:09] gotcha [20:17:12] yeah [20:17:35] might be worthwhile to either break that out or just request one model, no? [20:18:07] cos for the reasons you’re listing, ORES can’t respond with a success for one model and failure for another. [20:18:23] It can. It's just rare [20:18:41] The multi-model request is probably closest to real traffic [20:18:47] Right [20:18:50] kk [20:18:51] precached [20:18:54] yep [20:19:04] hmm although that uses other wiring [20:19:23] actually, I’m not sure what state that’s in. [20:19:40] Is that still served using ?precache=1 ? [20:19:54] /v3/precache [20:20:08] ok [21:00:14] halfak: https://usercontent.irccloud-cdn.com/file/0Zm8UCRD/image.png [21:00:37] \o/ [21:00:54] I think it looks good. :) [21:02:06] i put a space into "damaging-label" but i can remove it all together if we are sure we won't need it ever later [21:03:01] and you were right, it did not require changes in the css) [21:04:15] \o/ [21:04:43] I have no comment about the space. It seems like we might want to delete the message key though so that people don't need to translate a space in translatewiki.net [21:05:28] Amir1, do you know if we delete a message key in en.json, and update our code appropriately, that will work as expected? [21:05:48] halfak: yup [21:06:08] halfak: remove the key from qqq and we are fine [21:06:14] Cool. fajne, I think you'll need to change damaging_and_goodfaith.yaml in forms/available/ [21:06:15] ok. also, why i don't see goodfaith-yes-tooltip and the likes. you decided not to show them? [21:06:16] *from qqq too [21:06:37] fajne, there should be one. I'm not sure why there isn't [21:07:30] https://github.com/wiki-ai/wikilabels-wmflabs-deploy/blob/master/forms/i18n/damaging_and_goodfaith/en.json#L14 [21:07:45] https://github.com/wiki-ai/wikilabels-wmflabs-deploy/blob/master/forms/available/damaging_and_goodfaith.yaml#L33 [21:08:22] yes, i am talking about this tltip hm.. does it show up when you hover over a button? [21:09:40] Aha! I doesn't [21:09:46] Must be a regression in oojs-ui [21:09:48] * halfak checks. [21:10:38] https://doc.wikimedia.org/oojs-ui/master/js/#!/api/OO.ui.ButtonOptionWidget [21:10:47] Looks like that is *supposed* to work [21:14:01] Got it! Replace "indicatorTitle" with "title" [21:14:09] ok [21:14:11] If you add that to your commit, I'll confirm it :) [21:16:15] 10Scoring-platform-team, 10revscoring, 10artificial-intelligence: Improve test coverage and cleanup docs for revscoring 2.0 - https://phabricator.wikimedia.org/T172151#3487687 (10Halfak) [21:16:20] https://phabricator.wikimedia.org/T172151 [21:16:30] Amir1, ^ can you give me a quick merge :DD [21:16:39] awight|afk already +1'd [21:17:49] halfak: this is a little bit big [21:17:54] let me read through it [21:17:58] It is? [21:20:53] not that big [21:21:31] Ahh yeah. I guess any change to docs gets a lot of lines [21:21:36] \o/ thanks dude [21:22:47] Just looked and I *did* review https://github.com/wiki-ai/draftquality/pull/9 [21:22:59] There's more I can do but it shouldn't be in the Review column :/ [21:23:03] * halfak does the "more" [21:24:49] halfak: what if i change the content of the tooltips as it doesn't make much sense now to have "Yes, this edit appears to have been made in good-faith" popping up at the button saying "Good-faith". I would incorporate an info explanation into a tooltip instead. Like "It appears as though the author of this edit was trying to contribute productively" for good-faith and so on. Hm? [21:25:04] +1 [21:26:02] all: objections? [21:26:36] 3, 2, 1. ok, i am editing it. [21:27:49] \o/ [21:33:33] * halfak starts a long running process. [21:33:44] Trying to replicate adam's draft quality model building work :) [21:50:15] halfak: I left a pretty scant crumb trail for that. Any questions I can answer? [22:32:31] awight, not yet. I'm mostly just going to replicate building the model and check the stats against your final version. [22:32:43] If there's alignment, then I'll build the old model and check on that too. [22:32:57] sounds good, ty [22:36:47] Happen to know if there’s a scratch db on quarry? [22:36:57] I want to make a temporary table... [22:37:46] I see nothing in https://wikitech.wikimedia.org/wiki/Help:Toolforge#Database_access [22:57:23] halfak: just opened a PR. if all is well with the labels, i am ready to move on [22:57:41] awight, there is. How big? [22:57:49] fajne, link? [22:58:00] halfak: not too crazy, just 50k rows [22:58:25] I’m doing the query on (flaky-ass) stat1006 for now, but it would be nice to try the final thing on Quarry for repeatability’s sake. [22:58:44] awight, OK so do you have a tool labs account? [22:58:56] Sure I can dig that up [22:59:03] OK. Log into tool labs. [22:59:12] Then type "sql enwiki" to connect to labsdb1001 [23:00:05] fyi I am async, cannot follow your directions right now, but I’m listening. [23:00:41] Do I create. database and grant write perms to 's52788'@'%' ? [23:00:55] You're going to want to follow the instructions that are near this diff: https://wikitech.wikimedia.org/w/index.php?title=Help%3AToolforge%2FDatabase&type=revision&diff=1761524&oldid=1761521 [23:01:02] Which is lame that I don't have a better anchor. [23:01:13] Oh wait. here: https://wikitech.wikimedia.org/wiki/Help:Toolforge/Database#Steps_to_create_a_user_database_on_tools.labsdb [23:01:30] Maybe sure that when you name your database, you put a "_p" on the end to make it public. [23:01:57] something like https://wikitech.wikimedia.org/w/index.php?title=Help%3AToolforge%2FDatabase#Steps_to_create_a_user_database_on_tools.labsdb [23:02:02] Wait. This is the best link: https://wikitech.wikimedia.org/wiki/Help:Toolforge/Database#Steps_to_create_a_user_database_on_the_replica_servers [23:02:10] You do not have to "become toolaccount" [23:02:26] kk thanks, I’m prepared to use that if necessary :) [23:03:35] 10Scoring-platform-team, 10Wikilabels, 10editquality-modeling, 10artificial-intelligence: Change "yes/no" in damaging_goodfaith form to "damaging/good" and "good-faith/bad-faith" - https://phabricator.wikimedia.org/T171493#3488010 (10Natalia) https://github.com/wiki-ai/wikilabels-wmflabs-deploy/pull/39 [23:05:47] fajne, just dropped one note in there/ [23:06:29] i saw [23:12:12] Nice work :) [23:12:18] * halfak runs away. [23:12:20] Have a good night! [23:15:56] temprary tables ftw. This flaggedrevs query takes just over a minute, down from an untold duration. [23:16:31] lol, and the output is empty. Back to the drawing board.