[11:02:23] 06Revision-Scoring-As-A-Service, 06Operations, 06Research-and-Data-Backlog, 10Research-management, and 3 others: [Epic] Deploy Revscoring/ORES service in Prod - https://phabricator.wikimedia.org/T106867#2238896 (10mark) [13:26:42] o/ [13:35:08] halfak: nlwiki damaging is like russian? (5k edits needing review and needs to be mergred with other 15K?) [13:47:01] Good Q. That might be one of the *new style* [13:47:38] In the new style, we mix 2.5k "needs review" and 2.5k "probably good" so that we can get some observations of "probably good" going wrong. [13:47:50] * halfak checks [13:48:44] Nope. Old style. So yeah. Mix the 5k needing review with 15 remaining. [13:49:30] It should be about 4160 + 15782 [13:50:12] oh, okay [13:50:15] thanks halfak [13:50:20] going to do it [13:50:32] Once we get to one of the new style ones, we'll learn some interesting stuff. [13:50:52] * halfak wants to know how many "probably good" edits are damaging. [13:50:56] Or bad faith! [14:00:00] We need a good way to convert a TSV into JSON. [14:00:07] rather JSON-lines [14:00:21] * halfak imagines a utility in json2tsv called tsv2json [14:00:30] Formatting is going to be a bitch :/ [14:01:47] Amir1, see https://github.com/wiki-ai/wikilabels/pull/112 [14:01:54] I just did some updates and merging. [14:02:15] awesome [14:02:20] thank you halfak [14:02:25] I have a news [14:02:44] it seems six edits in nlwiki is not labeed [14:02:49] *labeled [14:03:04] and it fails to catch them [14:03:09] what do you suggest? [14:06:38] Ignore them [14:10:19] but the fetch_label utility fails [14:10:21] with error [14:10:36] I can make it ignore them [14:23:21] halfak: ran prelabeling for nlwiki, right now we have 1019 cases of "True" for damaging [14:23:29] Is it expected? [14:23:38] I think so but double check won't make any harm [14:29:05] halfak: I want to close all tasks in done column, is it okay? [14:50:01] halfak: around? [14:54:02] 06Revision-Scoring-As-A-Service, 10wikilabels: i18n for API errors in wikilabels - https://phabricator.wikimedia.org/T133561#2235853 (10Ladsgroup) https://github.com/wiki-ai/wikilabels/pull/113 [14:55:47] o/ Amir1 [14:56:01] hey! [14:56:29] halfak: please read above comments [14:56:45] * halfak looks. [14:56:53] Oh! So, when you say "prelabel".... [14:58:21] Do you mean the "fetch_labels" script? [14:58:23] Amir1, ^ [14:58:34] Also, +1 for closing the tasks in the done column. [14:58:47] no, the prelabel [14:58:47] I can do them all in batch if you want. [14:58:51] since I don't have the dataset [14:58:53] Weird. [14:58:58] * halfak checks [14:59:10] prelabel is fine [14:59:22] true cases in 5K datasets are rare [14:59:27] (20%) [14:59:50] that is the part that needs your attention [14:59:53] Why are you running relabel on a 5k dataset? [14:59:59] *prelabel [15:00:20] of course I don't run prelabel on the 5K dataset [15:00:27] I ran it on the 20K one [15:01:05] I'm confused. So when you run prelabel on the 20k sample, you get only 1k that need review? [15:02:05] https://www.irccloud.com/pastebin/kBxlFDta/ [15:03:00] halfak: no, when I do grep "True" in nlwiki.rev_damaging.20k_2016.tsv (which is 5K review and 15K other) it returns only 1K [15:03:12] Oh! That's great! [15:03:23] We only have like 0.5k damaging in enwiki [15:03:29] ladsgroup@ores-compute-01:~/editquality/datasets$ grep "True" nlwiki.rev_damaging.20k_2016.tsv | wc -l [15:03:29] 1019 [15:03:39] Sorry. 807 in enwiki [15:03:44] oh, okay [15:03:46] :) [15:03:53] Ran feature extraction [15:04:00] running tuning reports right now [15:04:03] Great! [15:04:16] BTW, do you want me to batch resolve the Done column? [15:04:25] Or do you have batch rights in phab? [15:04:40] * Amir1 prays that gradient boosting comes up [15:04:43] BRB dog needs attention [15:04:44] I have th right [15:04:48] lol @ AMir1's RF models [15:04:51] I do it right now [15:04:56] kk back in a bit [15:05:52] 06Revision-Scoring-As-A-Service, 10rsaas-articlequality : Train/test `reverted` model for huwiki - https://phabricator.wikimedia.org/T133382#2239707 (10Ladsgroup) 05Open>03Resolved [15:05:55] 06Revision-Scoring-As-A-Service, 10rsaas-edittypes, 10wikilabels: Load 200 & 5k samples into wikilabels - https://phabricator.wikimedia.org/T132593#2239708 (10Ladsgroup) 05Open>03Resolved a:03Ladsgroup [15:05:56] 06Revision-Scoring-As-A-Service, 10wikilabels: Setup labels-experiment.wmflabs.org - https://phabricator.wikimedia.org/T132588#2239710 (10Ladsgroup) 05Open>03Resolved [15:05:59] 06Revision-Scoring-As-A-Service, 10wikilabels: DB performance improvements on wikilabels - https://phabricator.wikimedia.org/T132436#2239711 (10Ladsgroup) 05Open>03Resolved [15:06:01] 06Revision-Scoring-As-A-Service, 13Patch-For-Review: Add IRC logging to this board and -backlog - https://phabricator.wikimedia.org/T132359#2239712 (10Ladsgroup) 05Open>03Resolved [15:06:03] 06Revision-Scoring-As-A-Service, 07I18n: Complete the message documentation (qqq) for Revision Scoring - https://phabricator.wikimedia.org/T132208#2239715 (10Ladsgroup) 05Open>03Resolved [15:06:13] 06Revision-Scoring-As-A-Service, 10Beta-Cluster-Infrastructure, 07Puppet, 03Scap3: deployment-((sca|aqs)01|ores-web) puppet failures due to scap3 errors - https://phabricator.wikimedia.org/T132267#2239713 (10Ladsgroup) 05Open>03Resolved a:03Ladsgroup [15:06:15] 06Revision-Scoring-As-A-Service, 10rsaas-editquality: Train/test `damaging` and `goodfaith` model for ruwiki - https://phabricator.wikimedia.org/T131999#2239718 (10Ladsgroup) 05Open>03Resolved [15:06:17] 06Revision-Scoring-As-A-Service, 10ores, 10rsaas-editquality: Update RTRC to work with new ORES models - https://phabricator.wikimedia.org/T131870#2239719 (10Ladsgroup) 05Open>03Resolved [15:06:19] 06Revision-Scoring-As-A-Service, 10crosswatch, 10ores, 10rsaas-editquality: Update crosswatch to work with new ORES models - https://phabricator.wikimedia.org/T131869#2239720 (10Ladsgroup) 05Open>03Resolved [15:06:21] 06Revision-Scoring-As-A-Service, 10ores, 10rsaas-editquality: Update ScoredRevisions for new models - https://phabricator.wikimedia.org/T131867#2239722 (10Ladsgroup) 05Open>03Resolved [15:06:23] 06Revision-Scoring-As-A-Service, 10ores, 10rsaas-editquality: Update ra-un for new models - https://phabricator.wikimedia.org/T131868#2239721 (10Ladsgroup) 05Open>03Resolved [15:06:25] 06Revision-Scoring-As-A-Service, 10rsaas-editquality: Train/test ruwiki `reverted` model - https://phabricator.wikimedia.org/T131666#2239724 (10Ladsgroup) 05Open>03Resolved [15:06:27] 06Revision-Scoring-As-A-Service, 10revscoring: Language assets for Hindi - https://phabricator.wikimedia.org/T131670#2239723 (10Ladsgroup) 05Open>03Resolved [15:06:29] 06Revision-Scoring-As-A-Service, 10wikilabels: Edit quality campaign for Arabic Wikipedia - https://phabricator.wikimedia.org/T131505#2239725 (10Ladsgroup) 05Open>03Resolved [15:06:31] 06Revision-Scoring-As-A-Service: [Discuss] ORES machine learning as a service for Wikis - https://phabricator.wikimedia.org/T131434#2239728 (10Ladsgroup) 05Open>03Resolved [15:06:33] 06Revision-Scoring-As-A-Service, 10revscoring: Gather language assets for hungarian - https://phabricator.wikimedia.org/T131443#2239727 (10Ladsgroup) 05Open>03Resolved [15:06:35] 06Revision-Scoring-As-A-Service, 10wikilabels: Edit quality campaign for Hungarian Wikipedia - https://phabricator.wikimedia.org/T131446#2239726 (10Ladsgroup) 05Open>03Resolved [15:06:37] 06Revision-Scoring-As-A-Service, 06Research-and-Data: Reach out to CE for Revscoring support - https://phabricator.wikimedia.org/T130893#2239729 (10Ladsgroup) 05Open>03Resolved [15:06:39] 06Revision-Scoring-As-A-Service, 10wikilabels: [Investigate] Intermittent performance issues with wikilabels - https://phabricator.wikimedia.org/T130872#2239730 (10Ladsgroup) 05Open>03Resolved [15:06:41] 06Revision-Scoring-As-A-Service, 10wikilabels: i18n wikilabels error messages - https://phabricator.wikimedia.org/T130774#2239733 (10Ladsgroup) 05Open>03Resolved [15:06:43] 06Revision-Scoring-As-A-Service, 10rsaas-edittypes, 10wikilabels: Add "no selection made" message to Semantic(s|Operations)Selector form - https://phabricator.wikimedia.org/T130871#2239731 (10Ladsgroup) 05Open>03Resolved [15:06:45] 06Revision-Scoring-As-A-Service, 10rsaas-edittypes: Add training campaigns to wikilabels for enwiki and itwiki - https://phabricator.wikimedia.org/T130860#2239732 (10Ladsgroup) 05Open>03Resolved [15:06:47] 06Revision-Scoring-As-A-Service, 10wikilabels: wikilabels JS should report on server status - https://phabricator.wikimedia.org/T130773#2239734 (10Ladsgroup) 05Open>03Resolved [15:06:49] 06Revision-Scoring-As-A-Service, 10ores: Selective precaching for ORES - https://phabricator.wikimedia.org/T130609#2239735 (10Ladsgroup) 05Open>03Resolved [15:06:51] 06Revision-Scoring-As-A-Service, 10ores: ORES-ui broken - https://phabricator.wikimedia.org/T130536#2239737 (10Ladsgroup) 05Open>03Resolved [15:06:54] 06Revision-Scoring-As-A-Service, 10MediaWiki-extensions-ORES, 13Patch-For-Review: implement an option in preferences to make hidenondamaging default - https://phabricator.wikimedia.org/T130560#2239736 (10Ladsgroup) 05Open>03Resolved [15:06:56] 06Revision-Scoring-As-A-Service: Create puppet role for compute nodes - https://phabricator.wikimedia.org/T130461#2239738 (10Ladsgroup) 05Open>03Resolved [15:06:59] 06Revision-Scoring-As-A-Service, 10MediaWiki-extensions-ORES: Move ORES extension test MediaWiki install to `ores-staging` project - https://phabricator.wikimedia.org/T130408#2239739 (10Ladsgroup) 05Open>03Resolved [15:07:01] 06Revision-Scoring-As-A-Service, 10wikilabels, 07I18n: Message description needed for Wikimedia:Wiki-ai-wikilabels-form-et-ref-label - https://phabricator.wikimedia.org/T130348#2239740 (10Ladsgroup) 05Open>03Resolved [15:07:03] 10Revision-Scoring-As-A-Service-Backlog, 10rsaas-editquality: Deploy edit quality models for nlwiki - https://phabricator.wikimedia.org/T130290#2239744 (10Ladsgroup) [15:07:05] 10Revision-Scoring-As-A-Service-Backlog, 10rsaas-editquality: Deploy edit quality models for ruwiki - https://phabricator.wikimedia.org/T130293#2239742 (10Ladsgroup) [15:07:07] 06Revision-Scoring-As-A-Service, 10rsaas-editquality, 10wikilabels: Complete ruwiki edit quality campaign - https://phabricator.wikimedia.org/T130270#2239741 (10Ladsgroup) 05Open>03Resolved [15:07:09] 06Revision-Scoring-As-A-Service, 10rsaas-editquality, 10wikilabels: Complete nlwiki edit quality campaign - https://phabricator.wikimedia.org/T130267#2239743 (10Ladsgroup) 05Open>03Resolved [15:07:11] 06Revision-Scoring-As-A-Service, 10rsaas-edittypes, 10wikilabels: Load 500 revisions for new edit_type campaign pilot - https://phabricator.wikimedia.org/T127081#2239747 (10Ladsgroup) 05Open>03Resolved [15:07:13] 06Revision-Scoring-As-A-Service, 10ores, 13Patch-For-Review, 10Scap3 (Scap3-Adoption-Phase1): Move to using scap3 for deployment for ORES service - https://phabricator.wikimedia.org/T128670#2239745 (10Ladsgroup) 05Open>03Resolved [15:07:25] 06Revision-Scoring-As-A-Service, 10ores: Deploy precaching daemon for ORES - https://phabricator.wikimedia.org/T106638#2239754 (10Ladsgroup) 05Open>03Resolved [15:07:25] 06Revision-Scoring-As-A-Service: Russian language utilities - https://phabricator.wikimedia.org/T106836#2239753 (10Ladsgroup) 05Open>03Resolved [15:07:28] 06Revision-Scoring-As-A-Service, 10wikilabels: Move Wiki Labels javascript to the wiki - https://phabricator.wikimedia.org/T104071#2239756 (10Ladsgroup) 05Open>03Resolved [15:07:29] 06Revision-Scoring-As-A-Service, 10wikilabels: Implement abandoning tasks in Wikilabels - https://phabricator.wikimedia.org/T105521#2239755 (10Ladsgroup) 05Open>03Resolved [15:07:31] 06Revision-Scoring-As-A-Service, 10wikilabels: Log events in the Wiki Labels server - https://phabricator.wikimedia.org/T102347#2239757 (10Ladsgroup) 05Open>03Resolved [15:12:32] "| GradientBoostingClassifier | 0.961 | 0.003 | learning_rate=0.1, n_estimators=300, max_features="log2", max_depth=3 " The giant spaghetti monster answered my prayers, hallelujah [15:12:37] ramen [15:14:30] \o/ [15:14:35] That's a nice score too [15:14:51] We should consider switching to PR-AUC as our scoring metric. [15:15:03] I'll add a phab task for exploring that [15:16:53] 10Revision-Scoring-As-A-Service-Backlog, 10revscoring, 10rsaas-editquality: [Spike] Explore using PR-AUC to score when tuning - https://phabricator.wikimedia.org/T133698#2239792 (10Halfak) [15:18:57] https://github.com/wiki-ai/wikilabels/pull/113 looks good. [15:19:10] Amir1, did you test it to make sure there's not some weird syntax error? [15:19:23] yup [15:19:27] I use jshint [15:19:36] kk [15:19:47] {{done}} [15:19:47] How cool, halfak! [15:19:48] (We need to make wikilabels pass jshint too) [15:21:01] thanks halfak :) [15:21:26] halfak: PR-AUC: 0.637 [15:21:31] bad or good? [15:21:42] ROC-AUC: 0.956 [15:22:52] Really good [15:23:48] I need to change my location, wifi is really bad here [15:34:10] 06Revision-Scoring-As-A-Service, 07I18n: Complete the message documentation (qqq) for Revision Scoring - https://phabricator.wikimedia.org/T132208#2239860 (10Amire80) Thank you! [15:40:26] halfak: I think I made a mistake for goodfaith in Russian [15:40:35] I will fix it with my PR [15:41:11] 06Revision-Scoring-As-A-Service, 10wikilabels: i18n for API errors in wikilabels - https://phabricator.wikimedia.org/T133561#2239889 (10Ladsgroup) a:03Ladsgroup [15:41:14] Sounds good. [15:41:19] 06Revision-Scoring-As-A-Service, 10rsaas-editquality: Train/test 'damaging' and 'goodfaith' model for nlwiki - https://phabricator.wikimedia.org/T133563#2239890 (10Ladsgroup) a:03Ladsgroup [16:48:23] halfak: nlwiki goodfaith [16:48:27] PR-AUC: 0.998 [16:48:27] ROC-AUC: 0.952 [16:48:56] too good to be true? [17:23:01] 06Revision-Scoring-As-A-Service, 10wikilabels: Move Wiki Labels javascript to the wiki - https://phabricator.wikimedia.org/T104071#2240395 (10He7d3r) @Ladsgroup, @Halfak: Are we supposed to use https://meta.wikimedia.org/wiki/MediaWiki:Gadget-WikiLabels-loader.js https://meta.wikimedia.org/wiki/MediaWiki:Gadge... [18:10:34] Amir1, it looks like that because of the high percentage of positive values. [18:10:40] But that is really good. [18:11:06] oh great [18:11:11] Looks like enwiki :) [18:11:23] See https://ores.wmflabs.org/v2/scores/enwiki/goodfaith/?model_info [18:11:47] PR-AUC is 0.997 and ROC-AUC is 0.908 [18:19:07] https://github.com/wiki-ai/editquality/pull/29 [18:19:17] it seems github bot is down now [18:20:41] 06Revision-Scoring-As-A-Service, 10rsaas-editquality: Train/test 'damaging' and 'goodfaith' model for nlwiki - https://phabricator.wikimedia.org/T133563#2240632 (10Ladsgroup) https://github.com/wiki-ai/editquality/pull/29 [18:21:02] {{merged}} [18:21:02] 10[5] 04https://meta.wikimedia.org/wiki/Template:merged [18:21:07] \o/ [18:21:21] thanks halfak [18:21:21] Kick ass! We need to do a deployment soon :) [18:21:35] I'm going to try to do that first thing in the morning tomorrow. [18:21:38] yeah [18:21:57] I will be working on staging in wikilabels [18:22:04] great [18:22:21] We should do a deployment there soon too. Do you want to try deploying both ORES and wikilabels to staging? [18:22:57] yeah [18:23:07] that should be straightforward [18:23:10] :) [18:23:27] btw. I reviewed staging process in wikilabels [18:23:46] As far as I checked everything looks okay except testing it in frontend [18:23:56] http://labels-staging.wmflabs.org/gadget/loader.js [18:23:59] loader is fine [18:24:13] but how to load it in a test enviroment [18:24:28] that part is tricky [19:00:57] Amir1, looks like I get a server error when loading the gadget [19:01:00] http://labels-staging.wmflabs.org/gadget/ [19:01:21] I have 30 mins, so I'm going to work on getting ORES in staging. [19:01:40] let me do it [19:01:53] kk will do. [19:02:03] * halfak looks for other things that can happen in 30 minutes. [19:04:09] regarding the wikilabels issue [19:04:25] it seems I have ssh connection issues [19:05:40] no, I'm dumb [19:06:19] "no python application found, check your startup logs for errors" [19:06:30] okay. Fixing it should be simpel [19:06:33] *simple [20:23:22] Amir1, just about to take off to run some errands. [20:23:29] How's ORES-->Staging going? [20:34:09] * halfak --> away for a few hours [20:45:32] halAFK: hey, please check telegram [20:46:03] I was afk for some offline work [23:01:58] 06Revision-Scoring-As-A-Service, 10wikilabels: Review staging protocol for Wikilabels - https://phabricator.wikimedia.org/T133557#2241625 (10Ladsgroup) Here's my review. - puppet - There is no issue with puppet, it's generalized enough - Since both wikilabels and wikilabels-staging coexist in a pro...