[00:39:28] YuviPanda: around? [00:39:30] http://semantic-ui.com/introduction/getting-started.html [00:39:52] this says "Gulp is an NPM module and must be installed globally" [00:42:15] Amir1: you can't really globally install things without root, and so things that require npm -g won't work on tools. http://stackoverflow.com/a/22115441 might be of help [00:42:40] Amir1: it looks like gulp only wants a global install to put gulp commandline in your PATH... [00:43:11] yeah, I imagined so [00:43:27] I'm thinking of a way to bypass that [00:44:57] Amir1: I think if you just install it locally via package.json and npm install, and specify the full path to gulp under node_modules it should work [00:45:05] unless they've explicitly hardcoded it to make that *not* work [00:45:39] let me give it a try [02:58:41] YuviPanda: Hey, is it okay to use this in labs? http://jpgraph.net/download/ [02:58:58] "JpGraph is released under a dual license. QPL 1.0 (Qt Free License) For non-commercial, open-source or educational use" [02:59:10] Amir1: yup [02:59:24] great [02:59:26] Thanks [05:27:36] (03CR) 10Awight: Actually implement the ORES RC filter (032 comments) [extensions/ORES] - 10https://gerrit.wikimedia.org/r/256641 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [05:33:16] (03PS11) 10Awight: Flag reverted risk rows using the recentChangesFlag [extensions/ORES] - 10https://gerrit.wikimedia.org/r/247185 (https://phabricator.wikimedia.org/T112856) [05:33:57] (03Abandoned) 10Awight: Actually implement the ORES RC filter [extensions/ORES] - 10https://gerrit.wikimedia.org/r/256641 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [05:34:52] (03PS12) 10Awight: Flag reverted risk rows using the recentChangesFlag [extensions/ORES] - 10https://gerrit.wikimedia.org/r/247185 (https://phabricator.wikimedia.org/T112856) [05:35:30] (03CR) 10Awight: "PS11: Squash with I0602aea8a9ef." [extensions/ORES] - 10https://gerrit.wikimedia.org/r/247185 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [10:22:15] (03CR) 10He7d3r: Actually implement the ORES RC filter (031 comment) [extensions/ORES] - 10https://gerrit.wikimedia.org/r/256641 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [10:30:08] (03CR) 10He7d3r: Actually implement the ORES RC filter (031 comment) [extensions/ORES] - 10https://gerrit.wikimedia.org/r/256641 (https://phabricator.wikimedia.org/T112856) (owner: 10Awight) [14:29:43] o/ [14:49:12] o/ halfak [14:49:22] I'm currently working on something cool [14:49:27] I'm about to finish [14:49:34] after that I'll show it to you [14:50:51] yay :) [14:50:56] Also hey! :) [16:25:31] I got to go, talk to you later [17:05:28] Amir1, sorry to be in meetings when you could talk :( [17:09:05] np halfak [17:09:13] have you seen those pages? [17:10:49] Yeah. It's pretty cool. :) [17:12:54] o/ YuviPanda [17:13:08] I'm gearing up for a new analysis that will work well with iPython. [17:13:29] I'd like to take advantage of your recent work presenting notebooks on-wiki. Any protips? [17:14:03] Amir1, forgot to ask. Any progress on the comparison of old and new Wikidata models? [17:14:38] halfak: do you mean report of improvements? [17:14:48] Amir1, yeah. [17:15:01] I wrote that in the page several days ago [17:15:10] https://www.wikidata.org/wiki/Wikidata:ORES/Report_mistakes [17:17:39] Oh! Great! [17:18:10] We should discuss the new features and improved "damaging revert" detection that we are employing to make the improvements. [17:18:22] But it looks like we are ready to try deploying the model now. [17:34:50] Amir1, new model is live [17:35:04] amazing [17:35:06] thanks [17:35:23] I'm writing the announcement re. the analyst [17:35:38] Cool! [18:08:21] Hi, anyone around? [18:29:10] Hi, anyone around? [18:29:24] Hi Jem. In a meeting now. Back soon. [18:30:16] Hi, halfak [18:30:25] Ok, please ping me when back :) [18:50:45] o/ jem [18:50:49] Just got back [18:50:54] :) [18:51:02] Good [18:51:18] I just wanted to ask a little about ORES [18:51:37] I have been testing it with some edits in es.wikipedia [18:51:45] (Well, testing the API results) [18:52:39] My main doubt is about the algorithms... are there any "bad words lists" used, or similar customizable things? [18:53:08] I have found https://gist.github.com/Ladsgroup/cc22515f55ae3d868f47#file-eswiki but I'm not sure if that is used in the code [18:53:41] And is there any kind of self-learning involved? [18:53:48] See https://github.com/wiki-ai/revscoring/blob/master/revscoring/languages/spanish.py [18:54:00] jem, not sure what you mean by self-learning. [18:54:11] Ok, I try to explain [18:54:51] halfak: \o/ [18:55:05] * YuviPanda lets his current conversation finish before barraging halfak [18:55:33] Is the algorithm capable of modify its scores based on previous results? [18:56:08] Or maybe it would have to be by modifying the configuration only? [18:56:44] jem, our process is to train the model on a dataset containing human judgement and then to try to fix the model to address mistakes that it makes. [18:57:19] Our "fixing" process involves looking for better sources of signal (e.g. fixing the badwords list), re-training the model and then comparing the old and new scores on mistakes. [18:57:35] See our work with wikidata here: https://www.wikidata.org/wiki/Wikidata:ORES/Report_mistakes [18:58:03] Ok [18:58:26] So, by reporting false-postives and false-negatives to us, we can then try to figure out why the model made those mistakes and make improvements. [18:58:43] It's especially helpful if you notice trends. [18:58:51] E.g. if all edits of a certain type tend to get flagged. [18:59:10] Good, so there's no self-learning in the way I was thinking of it [18:59:47] Just adjustments as a consequence of human input [19:00:10] And is there a similar report page for eswiki? [19:01:46] halfak ? [19:01:50] jem, we haven't figured out the *right* way to set up such a page, but we've had a few volunteers do this on their own wiki. [19:01:59] * halfak digs for itwiki's example. [19:02:40] Here: https://it.wikipedia.org/wiki/Progetto:Patrolling/ORES [19:03:15] If you would be interested in setting up a space on eswiki and helping make sure we understand major trends, I'd really appreciate the support :) [19:03:24] And I promise to keep working to make the model better. [19:03:44] :) [19:04:40] In fact I was thinking about setting up a patroller bot which would revert based on the points of the reverted model [19:05:22] But of course it's important that the model is tested in our eswiki [19:05:59] Yes. I'd like to get a Wiki labels campaign done in eswiki too. [19:06:19] * halfak gets link and info [19:06:29] See https://es.wikipedia.org/wiki/Wikipedia:Etiquetando [19:06:45] By going here and labeling a random sample of edits, we'll be able to train better models. [19:06:54] (since a lot of reverts happen for reasons other than damage) [19:07:59] * jem checks [19:11:37] OK, let's see I have understand [19:11:43] understood* [19:13:54] Installing the Wiki labels accesory would be a better way to mark false positives or negatives for the algorithms to be improved, compared to a page similar to the one in itwiki with a list of diffs... [19:14:07] Is that it? [19:14:34] jem, regretfully we can't directly use wiki labels for misclassifications. [19:14:46] However, we can train *better* models with wiki labels. [19:14:47] Ah [19:15:13] These models are less likely to make classification mistakes because we don't have to make so many assumptions (all reverted edits are bad) when training the models. [19:16:36] Ok, so you mean it's better to check those edits than to just check all reverted edits, as you don't know why are reverted [19:17:12] (They could even be good edits by a puppet of a blocked user) [19:17:22] Indeed. [19:17:24] That's right [19:17:54] jem, but we should still talk about misclassifications. There are two roads to high quality models. [19:18:03] 1. better training data and 2. better signal [19:18:24] Wiki labels gets us better training data. Your misclassification reports will help us thing of way to get better signal. [19:19:11] I've got to run for a bit, but I'll be around later if you want to pick up where we left off. [19:19:32] Ok [19:19:43] For sure [19:19:50] Thanks for your interest. I look forward to working with you, jem. :) [19:19:51] And yes, I have already a false positive [19:20:02] I say the same :) [19:20:15] Thank you and all the team for the work [19:21:24] * jem still has to understand better about the "signal" concept [19:26:34] I have permanent connection so I'll stay here to keep on the collaboration if there's no problem [19:38:43] jem, no problem. You are very welcome :) [19:49:32] :) [19:50:57] So, halfak, when back, tell me what would be the best way to report false positives, for the moment [20:11:40] jem, for now, I moved our misclassification reports to https://meta.wikimedia.org/wiki/Research:Revision_scoring_as_a_service/Misclassifications/Edit_quality [20:11:59] I'll likely want to do some re-organizing, but that'll be a good place to record them for now. [20:15:03] Ok [20:20:20] I see two eswiki diffs where mixed in the enwiki section... moving with mine [20:25:32] Woops. Thanks. [20:25:39] It was a quick copy-paste :/ [20:25:51] (Horray for wikis!) [20:27:53] :) [20:27:54] {{done}} [20:38:25] [[m:ORES]] [20:38:49] ... if a linking bot is useful here, I can offer mine :) [20:40:35] linking bot? [20:41:09] Yes [20:41:17] Not sure what you mean by that. [20:41:23] For example [20:41:38] 21:41:07 < jem> [[m:ORES]] [20:41:59] 21:41:07 < AsimovBot> [1] https://meta.wikimedia.org/wiki/Objective_Revision_Evaluation_Service - Redirección desde https://meta.wikimedia.org/wiki/ORES?redirect=no [20:42:31] Indeed. It is true that [[:m:ORES]] redirects to [[:m:Object Revision Evaluation Service]] [20:43:05] My bot in particular solves redirects that way, just in case the user really wants to edit the redirection [20:45:57] It also answers "You rule!" and the like to {{done}}... apart from other useful things related to the projects, of course [20:46:14] Oh! I see a bot that posts the full link when the Wikilink syntax is included in a message in the IRC chatroom? [20:46:42] That's it, halfak [20:47:10] I suppose you aren't very used to other Wikimedia IRC channels [20:47:51] :P Most of the wikimedia channels I'm in don't have linking bots. [20:47:59] But yeah, we used to have one in #wikipedia-snuggle [20:48:06] I just didn't know it by the name "linking bot" :P [20:48:29] Anyway, feel free to invite your bot :) [20:49:20] :) [20:50:01] My bot is much more than a linking bot, but it doesn't get offended, anyway :) [20:50:18] I'll configure the channel and bring it here in a moment [20:52:32] wiki-ai/revscoring#373 (meta_features_refactor - 74a6310 : halfak): The build failed. https://travis-ci.org/wiki-ai/revscoring/builds/95664273 [21:27:43] -info [21:27:44] 10[1] User -jem- in 07meta - Registered: 17 Apr 2009, 16:57 - Edits: 360 (0,15 ed./day) - First: 17 Apr 2009, 19:53 in «Steward requests/SUL requests» - Last: 038 Dec 2015, 20:27 in «Research:Revision scoring as a service/Misclassifications/Edit quality» - Groups: autopatrolled, user, autoconfirmed - Sex: N/D - 10https://meta.wikimedia.org/wiki/Special:Contributions/%2Djem%2D13 => [21:27:47] Associated nicks/cloaks: jem-, jem [21:27:56] [[ORES]] [21:27:57] 10[2] 1010https://meta.wikimedia.org/wiki/Objective_Revision_Evaluation_Service - Redirección desde 10https://meta.wikimedia.org/wiki/ORES?redirect=no [21:28:00] -time [21:28:01] Mar 2015-12-08 21:27:40 UTC [21:28:06] Ok, working :) [21:28:49] I've configured meta as default project instead of enwiki, I guess it's more appropiate (but it's easy to change) [21:29:29] Can be invoked with @ - = or ! (also easy to change) [21:30:16] And I take note to translate to English the redirections and days of week [21:31:02] [[ORES]] [21:31:02] 10[3] 1010https://meta.wikimedia.org/wiki/Objective_Revision_Evaluation_Service - Redirección desde 10https://meta.wikimedia.org/wiki/ORES?redirect=no [21:31:05] :D [21:31:05] (Well, the days should be... that's probably a bug to check) [23:06:41] -info [23:06:47] 10[4] User EpochFail in 07meta - Registered: 26 Aug 2010, 17:25 - Edits: 2484 (1,29 ed./day) - First: 26 Aug 2010, 17:40 in «Research:Committee» - Last: 038 Dec 2015, 20:11 in «Research talk:Revision scoring as a service» - Groups: autopatrolled, oauthadmin, user, autoconfirmed - Sex: N/D - 10https://meta.wikimedia.org/wiki/Special:Contributions/EpochFail [23:07:04] Hi AsimovBot [23:07:20] AsimovBot, do a barrel roll [23:07:21] 04Error: Command “do” not recognized. Please review and correct what you’ve written. [23:07:28] AsimovBot, barrelroll [23:07:28] 04Error: Command “barrelroll” not recognized. Please review and correct what you’ve written. [23:07:32] awww [23:08:33] halfak: how does it know you are EpochFail [23:08:51] The masky thingie [23:09:11] oh [23:09:15] -info [23:09:16] 10[5] User YuviPanda in 07meta - Registered: 2 Jul 2015, 20:23 - Edits: 138 (0,87 ed./day) - First: 2 Jul 2015, 20:23 in «Steward requests/Username changes» - Last: 037 Dec 2015, 22:31 in «Grants:IdeaLab/Teahouse question query tool» - Groups: oauthadmin, user, autoconfirmed - Sex: N/D - 10https://meta.wikimedia.org/wiki/Special:Contributions/YuviPanda [23:09:18] Mine is halfak@wikipedia/EpochFail [23:37:22] * halfak sets up model building and tuning reports for ALL THE WIKIS. [23:37:32] We're going to have some beautiful reports. [23:37:48] Also, I can get 0.975 AUC with a GradientBoost classifier in wikidata. [23:53:55] awight, :P I did say "tomorrow" [23:55:21] seriously, thanks for spelling that out in big lowercase [23:55:50] I had everyone in the fundraising channel |afk'ing with promises to jump back online at 10pm Eastern and stuff. It could have been fun. but bad.