[02:13:34] YuviPanda: hey, around? [02:13:36] https://phabricator.wikimedia.org/T118039 [02:19:24] thanks :) [02:32:14] legoktm: hey, around? [02:32:57] Amir1: what's up? [02:33:37] we are testing ORES extension and can't make it work [02:33:52] legoktm: that would be great if you tae a look at it [02:34:05] Just enough to get us going would be great [02:34:10] So long as we get traction :) [02:34:12] https://phabricator.wikimedia.org/T118039 [02:35:27] ok, I'm a bit behind on everything right now, so I'll take a look after dinner [02:35:40] legoktm: I added you the ores service group in tools [02:35:42] Thanks legoktm [02:35:48] Thanks [02:36:07] Amir1: wait you're doing MW dev in tools? [02:36:09] don't do that [02:36:16] it'll only cause pain and suffering [02:36:49] YuviPanda: Should I change it to vagrnat? [02:36:50] Amir1: use https://wikitech.wikimedia.org/wiki/Help:MediaWiki-Vagrant_in_Labs [02:36:51] yes [02:37:01] awesome [02:37:02] tools project is terribly setup for mediawiki [02:37:03] okay [02:37:07] you can get it to work, but only with a lot of pain [02:37:10] and suffering [02:37:13] and human sacrifice [02:37:17] :D [02:37:59] YuviPanda: this is about instances [02:38:04] yes [02:38:08] create an instance in the ores project [02:38:09] Can we use vagrant in tools? [02:38:10] and use it [02:38:12] no [02:38:16] don't use tools for MW development [02:38:19] as I thought [02:38:22] it will cause you pain [02:38:27] and suffering [02:38:30] and require human sacrifice [02:38:44] (I already this at least three times :D) [02:38:51] :))))) [02:38:55] okay [02:38:58] as many times as it takes to stop people using tools for MW development :D [02:39:04] often times I end up being the human sacrifice. [02:39:25] Okay [02:40:30] halfak: can you make another instance in https://wikitech.wikimedia.org/wiki/Nova_Resource:Revscoring? [02:40:52] Amir1, what size? [02:41:00] YuviPanda: ^ [02:41:03] IDK [02:42:15] Amir1: 'small' is enough [02:42:17] halfak: ^ [02:42:27] can use 'medium' too [02:42:27] Weird.... The wikitech sidebar changed. I'm looking for the "projects" link [02:42:40] always remember: if it feels wrong try logging out and back in [02:42:56] Choose Image type ubuntu-14.04-trusty. MediaWiki-Vagrant will not run on the default debian-8.1-jessie base image. [02:42:57] Goddamn wikitech [02:42:58] Choose a security group that allows access on port 8080. (This can only be done while creating the instance!) [02:42:59] I just did that [02:43:01] OK [02:43:38] Yup. Still can't find the projects link [02:44:08] http://wikitech.wikimedia.org/wiki/Special:NovaInstance for instance creation [02:44:14] * halfak figures out URL [02:44:17] http://wikitech.wikimedia.org/wiki/Special:NovaProject for project [02:44:36] halfak: ^ [02:45:08] Yay wikis! [02:45:32] YuviPanda, security groups? [02:46:01] halfak: [02:46:14] https://wikitech.wikimedia.org/wiki/Help:MediaWiki-Vagrant_in_Labs [02:46:17] read this [02:46:23] halfak: it's importannt [02:46:27] *important [02:46:35] once you created it, I'll do the rest [02:46:54] halfak: create one called 'web' maybe? if it doesn't exist already? [02:46:58] https://wikitech.wikimedia.org/wiki/Special:NovaSecurityGroup [02:47:12] WHY IS THIS SO HARD [02:47:16] Thanks YuviPanda [02:47:22] Ensure your project has a security group that allows access on port 8080, the default port for MediaWiki-Vagrant HTTP requests. The security group may be named "Web" or "web-server", or the "default" security group for your project may allow on port 8080. [02:47:24] Create your instance [02:47:24] np, because we're stuck in about... 2009 [02:47:25] Choose Image type ubuntu-14.04-trusty. MediaWiki-Vagrant will not run on the default debian-8.1-jessie base image. [02:47:26] Yeah. Looks like webserver will do it. [02:47:27] Choose a security group that allows access on port 8080. (This can only be done while creating the instance!) [02:47:28] Apply the role role::labs::mediawiki_vagrant. You can do this by clicking on the "configure" next to your instance in the table of all instances. [02:47:31] You sure we only need 8080? [02:47:31] (sorry for flooding) [02:47:33] halfak: yes [02:47:39] kk [02:47:59] next quarter's goal is to switch to horizon.wikimedia.org [02:48:01] which sucks a lot less [02:48:37] building [02:48:50] thanks halfak and YuviPanda [02:48:58] is Amir1 part of the project, halfak? [02:49:06] I'll be back in twenty minutes [02:49:15] https://wikitech.wikimedia.org/wiki/Nova_Resource:Revscoring [02:49:19] mediawiki.revscoring.eqiad.wmflabs [02:49:20] It seems so [02:49:26] thanks [02:49:44] halfak: for next time, always use a prefix (revscoring-mw) or something, since we suck and names are still global [02:49:48] even though they don't look it [02:50:02] mediawiki.revscoring is in the name though [02:50:03] ok [02:50:08] * halfak face palms [02:50:08] Permission denied (publickey). [02:50:10] amir@amir-275E4E-275E5E:~$ [02:50:16] Amir1, might still be building [02:50:20] Or it's crazy [02:50:22] halfak: no, mediawiki is the name and 'revscoring' is the project name. mediawiki.eqiad.wmflabs also works [02:50:25] halfak: takes 2-3 mins [02:50:29] okay [02:50:38] Every other time I make and instance I have to have abogott delete it and recreate it 5 times :D [02:50:40] but yes, to be clear: this is way behind the times [02:50:50] thanks [02:51:02] What takes 2-3 times? [02:51:11] the initial puppet run [02:51:17] which sets up ssh key setup and stuff [02:51:31] OK. I'll leave it be. [03:13:09] Alright I'm off for the night. Have a good one! [03:13:10] o/ [03:13:36] o/ halfak [03:23:29] awight: o/ [03:40:44] Amir1: hey! [03:41:01] I'm building a vagrant in labs [03:41:07] so we can test it together [03:41:13] Great, thank you. [03:41:21] and see logs [03:41:26] There's no vagrant role yet, but that's not a big deal. More so for the ORES server. [03:41:39] it'll be done in a ten min. [03:41:50] However, I'm pretty sure I understand the problems. [03:42:56] There were some small things (ORES API changed, typos, etc), but at this point the remaining blocker is that halfak's pseudomodel is connected to the "reverted" model, but we're querying the "damaging" model now. [03:43:29] can we change it hard-codeed? [03:43:33] for now [03:43:44] Faking that is probably not worth the effort--we would have to manipulate ORES response data in a crazy way, I'd rather just wait for the "damaging" model to return data. [03:44:28] There is another trick we can do, I've added a "wgOresWikiId" config which lets us pull enwiki scores, for example. [03:44:56] So I'll just point to enwiki, then fast-forward the rev_id autoincrement key to something that exists in enwiki. [03:45:14] hmm [03:45:37] we cna use this method until tomorrow that Aaron wakes up [03:45:43] awight: there's a 'test' model now [03:45:46] that always returns data [03:45:53] no matter what [03:46:02] YuviPanda: testwiki, or a new model? [03:46:09] new model [03:46:53] Tricking the extension into using the wrong model is not a pretty hack--most of the extension is set up to treat model data very generically and pass it through untouched. [03:46:59] http://ores.wmflabs.org/scores/testwiki/ [03:47:03] only reverted [03:47:26] oh [03:47:28] no [03:47:29] http://ores-staging.wmflabs.org/scores/testwiki/ [03:47:29] We'd have to change all the code that references "damaging" to use "reverted" instead. Which is fine, but I think it's easiest to just pull enwiki for now [03:47:30] it's implemented as a wiki [03:47:32] http://ores.wmflabs.org/scores/testwiki/ [03:47:46] * awight races to do my dirty hack [03:47:53] :D [03:47:55] ok [03:47:57] we have damaging in ores-staging [03:50:03] awight: we only need to change URL of ores.wmflabs.org to ores-staging.wmflabs.org [03:51:09] looking... [03:51:17] oh, great! Thanks [03:51:56] That's better, cos it'll be easier to get a "damaging" edit [03:51:56] Amir1: I'm going to announce PAWS today or tomorrow https://etherpad.wikimedia.org/p/paws-announce [03:52:09] Amazing [03:52:11] awesome! [03:52:16] let me check the announcement :D [03:52:29] I'm still testing to make sure this one bug I was hunting is gone [03:54:30] toollabs went down. one moment, please [03:54:44] oh hah, sorry. It was me [03:55:04] ";" fail [03:55:06] ? [03:55:19] awight: also no using (WMF) accounts [03:55:20] I syntax errored out the LocalSettings.php [03:55:27] ah [03:55:29] heh [03:55:31] unrelated [03:55:33] also no using tool labs [03:55:35] for MW Development [03:55:41] k! [03:56:18] yeah, I'm paranoid enuf from fundraising that I would never consider using a WMF account there [03:56:31] hehe :D [03:56:38] it breaks when user accounts have '(' in them [03:56:46] good :) [03:56:54] those are an abomination anyway [03:57:03] yeah I agre [03:57:12] but I'm worried it'll break for users with non-ascii scripts [03:57:16] which is more important [03:58:55] No dice getting ores_classification data, yet. [03:59:24] I need to wolf dinner, then will reproduce this new endpoint config on my dev box and see what's going on. [04:02:23] I too should eat dinner [04:02:29] and go to the gym and then go watch a movie [04:02:31] hmm [04:02:33] lotsa things to do [04:10:34] That sounds so civilized. [04:19:40] rats, now I'm back where I was yesterday--works locally but not on the toollabs wiki. Software! [04:20:14] YuviPanda: So you were saying, bd808 might know about toollabs logging? [04:23:25] awight: why we don't do it in localhost and check logs there [04:23:33] in my localhost it doesn't work either [04:23:52] and apply fixes to the extension in labs [04:24:02] Amir1: did you try the latest config? [04:24:03] $wgOresBaseUrl = 'http://ores-staging.wmflabs.org/'; [04:24:03] $wgOresDamagingThreshold = 0.30; [04:24:03] $wgOresWikiId = 'testwiki'; [04:24:03] $wgOresModels = 'damaging'; [04:24:11] Pretty arcane... but it's working locally [04:24:16] (for me :) [04:24:24] let me try again [04:24:32] maybe I'm missing something [04:25:02] There's one more subtle gotcha, often the FetchScoreJob won't run until the next pageview [04:25:50] awight: no I'm saying: do not use tool labs for mw development. [04:26:04] YuviPanda: ah. okay that explains some things. [04:26:17] it will produce pain, suffering and capitalism. [04:26:24] haarrrr [04:26:32] and requires human sacrifice [04:26:40] It certainly feels that way so far. [04:26:42] Amir1 is setting up a mediawiki-vagrant-on-labs thing [04:26:56] Friends don't let friends setup MW on tool labs [04:27:24] I'm going to go to a gym now [04:27:28] I'll cya guys later [04:27:29] good idea! [04:27:38] https://wikitech.wikimedia.org/wiki/Help:MediaWiki-Vagrant_in_Labs [04:27:41] is appropriate hel page [04:27:56] thx! I did it maybe twice, was quite enjoyable, as well as profitable. [04:28:12] It's finished [04:28:19] but 1- I didn't set a proxy yet [04:28:24] I don't have access [04:29:03] 2- I couldn't install ORES extension yet [04:29:40] Can we tunnel instead of set up the proxy? [04:29:58] Amir1: send me the hostname and I can set up + configure the extension [04:30:51] awight: mediawiki.revscoring.eqiad.wmflabs [04:31:32] Looks like I need perms [04:31:49] yeah [04:31:56] I hope still YuviPanda is around [04:32:12] I might be able to help with this part... [04:32:56] Did you create the labs instance under a project? [04:33:51] yeah [04:33:58] the vagrant is up now [04:35:23] Amir1: I see what you're saying--you don't have admin rights to https://wikitech.wikimedia.org/wiki/Nova_Resource:Revscoring [04:35:38] exactly [04:35:47] halfak: You still around + could add me to ^ ? [04:36:02] I don't think he's around [04:36:08] I hope not ;) [04:37:25] Well, if it's not too annoying for you I'd be happy to help debug your local installation. [04:39:54] Here's some additional logging: https://phabricator.wikimedia.org/P2424 [04:39:54] I re-worked it and it's working now [04:39:57] OH [04:40:02] tell me more [04:40:10] (04:33) root@localhost:[wiki]> select * from ores_classification; [04:40:12] +----------+------------+--------------------+------------+------------------+-------------------+ [04:40:13] | ores_rev | ores_model | ores_model_version | ores_class | ores_probability | ores_is_predicted | [04:40:15] +----------+------------+--------------------+------------+------------------+-------------------+ [04:40:16] | 6 | damaging | 0.0.3 | false | 0.4000000000 | 0 | [04:40:18] | 6 | damaging | 0.0.3 | true | 0.6000000000 | 1 | [04:40:20] +----------+------------+--------------------+------------+------------------+-------------------+ [04:40:22] 2 rows in set (0.00 sec) [04:40:23] * awight throws hat in the air [04:40:25] (I love flooding) [04:40:45] Looks like you even got a "damaging" edit to poke at [04:41:07] http://imgur.com/d6WW74Q [04:41:51] I've sort of been waiting for this moment: a simple "r" making so many people this happy ;) [04:42:15] awesome [04:42:17] awesome [04:42:21] * Amir1 dances of joy [04:42:36] maybe it works only in vagrant machines [04:42:42] which is super strange [04:42:43] hehehe thank you for the incredible tenacity [04:42:48] hahaha [04:43:05] My guess is that toollabs is not fit for MediaWiki [04:43:18] okay [04:43:31] It was impossible to tell if the save hook job was even firing [04:43:41] Did you try "hide good edits"? [04:43:41] I will have the the test wiki instance working in several hours [04:43:48] waiting for someone to wake up [04:44:20] my machine is super slow [04:44:21] That sounds very painful... You sure you don't want to focus on the labs one? [04:44:22] .... [04:44:43] I meant the vagrant one [04:44:52] ah, perfect [04:45:02] I will probably delete this service group in toollabs [04:45:08] once I got this one ready [04:45:44] I still haven't found the labs server... it doesn't show up in https://wikitech.wikimedia.org/wiki/Nova_Resource:Revscoring #Instances [04:46:22] "hiding good edits" works very well [04:46:38] I don't know why it's not showed up there [04:46:42] good to hear [04:46:50] we just created it (one or two hours) [04:47:04] sorry, what isn't showing up? [04:47:30] oh, right--the instance [04:47:54] yeah [04:48:41] Okay, I've convinced myself there's nothing we can do to set up the public proxy address, without a project admin. [04:49:10] I agree [04:49:19] I'll make some patches very soon [04:49:32] We need these patches merged asap [04:49:40] That was a huge deal to set up the demo server, thanks again. I'll drift back to day job projects for a bit, but will be online for a few hours if I can help with anything. [04:49:53] Agreed--the mw-core patches are especially irksome. [04:50:40] Luckily, jdlrobson is interested in the second-to-last RecentChanges patch, cos it will also be useful for some mobile things he wants to do. [04:51:35] amazing [04:52:08] I need to go, my class starts in ten min and I'm busy for the next right hours [04:52:14] Unfortunately, 4 years later I still don't have any good strategies for encouraging people to do code review of stuff I feel is urgent. [04:52:17] k bye! [04:52:27] but once I'm back and someone is awake I'll finish all of them [04:52:39] feel free to kick the patches, +1 and stuff [04:52:44] sure [04:53:06] code review in mw is a ill process IMO [04:53:13] anyway [04:53:17] o/ awight [04:53:24] thanks again for this work [04:53:43] you too! [05:58:43] awight: yes, tool labs on MW is pain and suffering [06:00:29] YuviPanda: hey. Make me an admin on labs Revscoring when you get the chance. Otherwise, I will ask you to help create a web proxy. [06:01:45] awight: sure. what instance? [06:02:11] mediawiki.revscoring.wmflabs.eqiad, I believe [06:02:31] mediawiki.revscoring.eqiad.wmflabs [06:02:45] awight: http://ores-test.wmflabs.org/ [06:03:02] That's probably a decent public name, too: mediawiki.revscoring.wmflabs.org [06:03:09] wat is ores-test [06:03:11] yeah, except we all suck and can't get that [06:03:23] oh. mediawiki-revscoring, whatev [06:03:25] in fact even mediawiki.revscoring.eqiad.wmflabs ia a lie [06:03:33] lolol [06:03:40] it's actually mediawiki.eqiad.wmflabs only [06:03:45] O_o [06:03:48] which means no other project can have an instance named mediawiki now [06:03:52] see also: everything sucks [06:03:56] wait. [06:03:58] d'oh [06:04:11] http://mw-revscoring.wmflabs.org/ works now [06:04:16] <_< [06:04:24] or does not [06:04:30] the proxy works :) [06:04:31] but the URL is a thing, at least! [06:04:56] indeed [06:04:56] Thanks, um so I don't know if I'm asking you to do any work or if it's already done, but I did want to clarify that I don't need this tonight. [06:05:13] what is 'this' [06:05:20] the proxy? that was only like 2 clicks :) [06:05:23] If there's any ... [06:05:28] ok thanks [06:05:33] <3 [06:05:48] I thought that's what had happened, but I totally lost track of which cup it was under [06:06:09] Can you now anoint me a project member? [06:06:47] sure [06:06:50] what's your wikitech username [06:06:57] excellent. awight [06:07:09] done [06:07:28] ssh awight@mw-revscoring.eqiad.wmflabs [06:07:28] channel 0: open failed: administratively prohibited: open failed [06:07:38] <_< [06:07:43] * awight retries hostname [06:08:23] ssh awight@mediawiki.revscoring.eqiad.wmflabs [06:08:41] Will this "everything sucks" hostname continue to exist? [06:08:50] yes since that's the real hostname [06:08:53] (success logging in, btw) [06:08:55] mw-revscoring is just a magical web proxy [06:08:59] that only works for http [06:09:02] and nothing else [06:09:08] But didn't we bust the "mediawiki" party for everyone? [06:09:13] it's not just the hostname issues that suck, but *everything* [06:09:14] yes but that's ok [06:09:18] ookay [06:09:25] other people have busted: 'database', 'proxy', 'server', 'cache' [06:09:28] I suppose the squatting will end one day son [06:09:30] soon [06:09:31] hahahaha [06:09:35] That's beautiful [06:09:52] :) [06:11:18] YuviPanda: Here's a small cause for tonight's celebration. http://imgur.com/d6WW74Q [06:11:45] awight: \o/ [06:11:46] Amir1 was able to reproducibly cause that lower-case "r" to appear [06:11:57] yay teamwork! :D [06:11:58] Such a happy letter. [06:12:01] :D [06:16:21] * awight scratches head over mw-revscoring instance serving MediaWiki without running Apache or Nginx [06:16:40] Read the Docs, Luke: https://wikitech.wikimedia.org/wiki/Help:MediaWiki-Vagrant_in_Labs [06:17:03] * awight bookmarks YuviPanda [06:19:17] * awight baffles further cos I thought we were running the puppet files directly on labs hosts rather than use the additional layer of containers [06:19:44] * awight learnings [06:22:08] * awight bristles at vagrant ssh 24e34e9 [06:23:02] awight: that was how it used to be (labs-vagrant) but that's deprecated now [06:23:22] vagrant@mediawiki-vagrant:/vagrant$ grep -r ORES settings.d/ [06:25:25] whoa [06:26:08] ok I'm not the only one losing it here: head /vagrant/mediawiki/extension.json [06:26:11] { "name": "ORES", [06:26:21] hehe, alternative directory structure [06:26:38] so at this point I've no idea what's happening except it's similar / same as whatever mediawiki-vagrant does [06:27:50] oh--nbd. The ORES extension was somehow rsync'd on top of mw-core/ rather than under extensions/ORES [06:28:05] Which also explains the 'only "(" in Hack' error [06:28:27] that seems wrong [06:28:43] that seems hacked in while the 'right' thing should probably be to add a 'ores' role to vagrant [06:29:19] that's something I've been meaning to do, anyway. [06:29:53] * awight makes more important-sounding noises and then proceeds to immediately "git clean -f" in home directory [06:30:03] own goal. [06:36:05] :) [06:47:40] (03CR) 10Ladsgroup: [C: 031] Update field name in API response [extensions/ORES] - 10https://gerrit.wikimedia.org/r/258937 (owner: 10Awight) [06:53:46] (03CR) 10Ladsgroup: [C: 031] "Tested it in my localhost, worked correctly" [extensions/ORES] - 10https://gerrit.wikimedia.org/r/258916 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [06:55:05] (03CR) 10Ladsgroup: [C: 031] "Correct wording" [extensions/ORES] - 10https://gerrit.wikimedia.org/r/258042 (owner: 10Awight) [06:56:21] (03CR) 10Ladsgroup: [C: 031] Simplify down to a single threshold [extensions/ORES] - 10https://gerrit.wikimedia.org/r/257856 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [06:56:33] * awight blushes [06:56:51] (03CR) 10Ladsgroup: [C: 031] Switch from the "reverted" to "damaging" model [extensions/ORES] - 10https://gerrit.wikimedia.org/r/257851 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [07:00:53] (03PS1) 10Awight: [WIP] Generalize thresholds [extensions/ORES] - 10https://gerrit.wikimedia.org/r/259212 [07:02:23] (03CR) 10jenkins-bot: [V: 04-1] [WIP] Generalize thresholds [extensions/ORES] - 10https://gerrit.wikimedia.org/r/259212 (owner: 10Awight) [07:15:23] Wow, every second--/var/log/mediawiki/jobchron.log [07:35:56] (03PS1) 10Awight: Log API requests [extensions/ORES] - 10https://gerrit.wikimedia.org/r/259214 [07:41:49] (03PS2) 10Awight: Log API requests [extensions/ORES] - 10https://gerrit.wikimedia.org/r/259214 [08:03:23] awight: hey [08:03:37] It seems it's not working yet [08:03:40] maybe I did something [08:03:52] Amir1: hi! ah? I was about to crow that it was [08:04:08] http://mw-revscoring.wmflabs.org/wiki/Special:RecentChanges [08:04:10] yes [08:04:38] we have "r" but no edit has it [08:04:48] * awight double-takes [08:05:03] maybe caching... [08:05:18] also I think some patches are not there yet [08:05:19] I see lots of "r" changes [08:05:29] e.g. changing "r" to "D" [08:06:25] awight: http://imgur.com/BrvlUX9 [08:06:57] Can you try hard refresh? [08:07:42] hehe, next I will ask us to both reboot, cos I see "r"s all day [08:10:25] strange [08:11:35] table is populated [08:12:48] using curl in labs works fine [08:12:49] r [08:13:42] but hard refresh doesn't work [08:25:45] * awight mourns comrade eaten by misbehaving software [14:13:08] o/ Amir1 [14:13:15] I just saw Helder's bug report [14:13:20] o/ halfak [14:13:24] I changed the name "probability" to "probabilities" [14:13:35] could that be the source of your frustration with the testwiki model? [14:13:50] It's okay now [14:14:02] we have everything running [14:14:03] http://mw-revscoring.wmflabs.org/wiki/Special:RecentChanges [14:14:06] Oh! I'm still going to fix it. Is that okay? [14:14:12] yeah [14:14:17] that would be amazing [14:14:31] there is a bug when "grouping" is disabled [14:14:33] Any other requests while I'm working on it? [14:14:40] but otherwise it works as expected [14:14:52] hmmm [14:15:06] Can you push the damaging model to ores.wmflabs.org for testwiki [14:15:08] ? [14:15:19] it's in ores-staging now [14:15:44] Yeah. It's on ores-staging now, but it still had the old "probabilities" [14:15:58] I'll be pushing a new version in a moment and then moving to prod. [14:17:43] thanks :) [14:22:27] Woops. Need to increment the version so that the old testwiki scores get blown out of the cache. [14:25:37] Well... something is weird. [14:25:50] yeah... [14:26:06] hmmm... Definitely pulling new ores. [14:26:28] * halfak runs an update [14:27:39] ha. OK. I'm still confused about where our dependencies are coming from, but this works. [14:29:44] OK. Looks like we're good to go for a deploy. [14:30:04] thanks [14:30:18] once it's there tell me to change the settings [14:30:39] halfak: btw Do you have any ideas about how we can make use of this extension? [14:31:36] Amir1, not sure what Q you are asking. [14:31:58] But I see it as a sort of https://en.wikipedia.org/wiki/Foot-in-the-door_technique [14:32:40] in general any idea of improvement, feature requests, etc. [14:32:42] It will at least bring basic ORES functionality to MediaWiki and also maybe encourage a more substantial design project. [14:32:48] Oh! Gotcha. [14:33:43] So, I've been trying to recruit the design researchers at the WMF to take a look at this with me and go through a few design sprints around quality control and curation work. [14:33:49] That's more long term. [14:34:06] In the short term, I think that we should nearly -- if not totally -- replicated ScoredRevisions functionality. [14:34:34] E.g. we should have an option where the user can have whole rows highlighted in a color rather than the tiny "r" [14:35:04] I'm guessing that the extension doesn't work on Special:UserContribs, right? [14:35:13] no [14:35:21] not anywhere except rc [14:35:35] I think that would be good if we can sort rc [14:36:47] e.g. bring revisions with highest damaging probability to the top? [14:36:50] +1 for that [14:37:08] We should really be able to sort RC in a bunch of ways. This is just one of them. [14:39:38] Uh oh. Looks like web-01 is having trouble. [14:40:12] I don't think I had anything to do with that. [14:40:23] okay [14:40:30] let me write all of them down [14:40:40] finally I'm happy we have something ready to show [14:40:49] :D [14:42:00] OK.. So I'm getting an error that suggests that git can't download our repo because there isn't enough space on the hard drive. [14:42:45] But we have 8.2 GB available. [14:42:46] Hmmm [14:44:20] OK. Trick is to run fabricate from python 2.7 [14:44:25] *fabric [14:45:23] * halfak updates workers and web nodes in prep for deploy [14:55:04] * halfak watches a worker go down and freaks out [14:55:09] worker comes back up [14:55:09] :) [15:01:17] Amir1, ORES is now updated with a "damaging" model for testwiki [15:01:25] thanks [15:01:27] :) [15:01:28] I also fixed some issues regarding itwiki and the use of "ha" [15:01:53] I'm fixing the ORES extension in wmflabs [15:01:58] Woot. [15:02:06] I'll get back to the new text processing features. [15:02:26] Actually I owe our phab board some cleanup. [15:02:34] I made some big changes yesterday. [15:02:46] We are now RevisionScoringAsAService [15:03:01] And "revscoring" only describes one of our sub-projects. [15:03:13] Along-side ORES and Wiki labels [15:03:14] :) [15:16:38] amazing [15:17:11] Amir1, I'd like to get wb-vandalism merged fully into editquality and revscoring soon too. [15:17:30] It's going to take some doing, but I've got the basic wikibase stuff merged into my current work on revscoring too. [15:17:36] *too --> already [15:17:42] * halfak drinks more coffee [15:17:47] sure :) [15:18:06] I'll start working on it once I'm finished on some businesses [15:18:16] the blog post, things like that [15:18:20] +1 [15:18:39] Sure. I think this new branch of revscoring will need some substantial testing when it is ready because I'm gutting everything :) [15:18:55] I want to make sure that we can rebuild and use all of our models before we merge. [15:19:08] But then I want to start thinking about cleaning up for 1.0.0 :) [15:19:32] I think we can go into "no backwards incompatible stuff" for a while [15:29:33] great [15:37:13] Amir1, did you make an announcement about the new random forest model for WikiData? [15:37:21] The one with .97 AUC? [15:37:27] I told about it to Lydia [15:37:38] Great. [15:37:44] but not an official announcement yet [15:37:49] We can cover it in the blog post then. [15:37:49] in wikidata-l or WD:PC [15:38:06] but if you want I can do it, depends on you [15:38:07] This could be the *reason* for us posting a blog. [15:38:15] great [15:38:17] "reason" [15:38:17] :) [15:38:44] I love that we got this high accuracy [15:38:52] tuning was a great idea [15:39:06] *ingenious [15:39:33] heh. You know, Ellery did some tuning for us about a year ago, but we weren't ready to take advantage of it yet. [15:39:48] It was his old ipython notebook that I reviewed while I developed our tuning utility. [15:40:04] :) [15:40:19] We should make sure to ack him in the blog post. [15:40:24] And Yuvi [16:29:49] o/ harej [16:29:55] it's aaron halfaker [16:30:16] I've got some researchers from UMN who want to experiment with recommending WikiProjects to newcomers. [16:30:21] !!!!!! [16:30:30] * halfak realizes he should be in the WikiprojectX channel [16:30:32] is morten one of them? [16:30:35] Want to hop over there? [16:30:42] harej, na. He's in thesis mode. [16:30:53] But he might be a good advisor to their project. [16:30:59] So I'll suggest that to him. [16:31:35] Anyway, I'm curious if you have the bandwidth to talk to them? [16:31:47] They are CS grad students so you could rely on them to build some stuff. [16:32:06] Like mini Nettroms [16:32:09] :) [16:32:15] Except -- not actually smaller. [16:32:18] Just greener [19:11:26] halfak: I started an ORES vagrant role, btw. We can build the ores server in there. [19:12:10] I can use testwiki + pseudomodels for the first pass, but I assume we eventually want a functional server that can be trained and stuff? [19:12:39] awight, it turns out that the server is fully functional as it is. [19:12:50] All that is needed is reconfiguration to use a different model. [19:13:07] I'm not sure why you would do ORES/revscoring dev in MW vagrant though. [19:13:15] ok, fair enough! [19:13:40] It's just a convenience, so people can spin up your environment without README'ing [19:13:49] So, we'll need to install 'ores', provide a config file like the one in the ORES repo, and presumably make a startup script for loading the ORES server. [19:14:31] What's the default vagrant dbname? [19:14:48] meh, I won't push for an ORES server under mw-vagrant unless the need arises [19:15:00] I think the default Mediawiki db is "wiki" [19:42:06] awight, so what will the ORES role hit by default? [19:42:38] (sorry. Got lost in trying to solve floating point error in my test cases. The trick is to round at the end!) [19:44:45] floats are such a bugbear. [19:45:23] Um, so far the ORES role is only the MediaWiki extension, and it's configured to hit staging, use the testwiki ID, and only fetch the "damaging" model [19:45:36] https://gerrit.wikimedia.org/r/#/c/259211/6/puppet/modules/role/manifests/ores.pp,unified [19:45:38] That's probably OK. [19:45:56] I imagine that it won't be much trouble for new development to take advantage of different models, right? [19:46:18] e.g. the wp10 model [19:46:36] It wouldn't make sense to run that model against recent changes. [19:46:39] At least not obviously. [19:46:49] definitely not. The extension handles models in a pretty general way already. [19:47:12] The save hook fetches data from all models for each revision. [19:47:24] I'm only restricting to the "damaging" model because the others aren't available for testwiki, yet. [19:47:37] Gotcha. [19:47:44] Makes sense. [19:48:24] token frequency is fully tested! [19:48:25] WOOOO [19:48:26] I wanted to ask you--do you agree that we should default to fetching and caching data from all available models? [19:49:19] awight, long term, we might not want to do that. Right now, I think it makes sense. [19:49:36] It would be great if the RC user could filter on an arbitrary model. [19:49:44] But I think substantial UI changes would be needed for that [19:51:11] It would be easy to surface additional models, but yeah not to switch between them [19:51:57] Right now, let's focus the MVP on pulling all of the models. [19:52:19] In the future, I think it might make sense to add some taxonomy information to the model info. [19:52:30] E.g. "edit quality" would be pulled in for the RC feed. [19:52:47] "page quality" might be pulled in only for the most recent revision of pages. [19:54:29] cool. Well, for now the list of models is set by a configuration variable, and the default is everything available. I'll leave it alone. [19:57:47] o/ ShiveringPanda [19:57:58] * ShiveringPanda does as shivery wave [23:04:25] halfak: You have a minute to talk about MVP and priorities for the MW extension? [23:05:00] I'm tempted to add some more functionality to the MVP, but we should probably make our goals clear if nothing else. [23:08:48] halfak: On second thought, don't worry about it for now. I'm just going to attach tasks as blocking https://phabricator.wikimedia.org/T120923 [23:13:03] We're hard up for code review, unfortunately. Anyone you want to add as a reviewer for https://phabricator.wikimedia.org/T112856 ? [23:13:38] awight: I usually just bug Leah [23:13:39] legoktm, ^ have any time for this? [23:13:39] err [23:13:42] legoktm: [23:13:59] awight, will make some time to discuss MVP [23:15:06] halfak: thanks! [23:33:47] ShiveringPanda, halfak: sorry, I'm swamped right now, probably won't have time until the weekend. [23:34:03] legoktm, I think that'd be reasonable. [23:34:32] I'll reach out to the commtech team to see if they are interested in sharing some cycles as well. [23:38:50] legoktm: Take your time! This is a volunteer gig after all :p but if you think anyone else has eyeballs to spare, please add them to the CR list or mention to us.