[00:06:11] https://ar.wikipedia.org/wiki/%D8%AE%D8%A7%D8%B5:ORESModels [00:06:21] ^ is something wrong with arwiki? [00:15:24] zZ08u3QqFG [00:15:26] oops [08:26:56] 10Jade, 10Scoring-platform-team: Can endorse two different labels for the same facet via jadeproposeandendorse. That should be impossible. - https://phabricator.wikimedia.org/T247974 (10kevinbazira) a:03kevinbazira I have pushed a patchset that solves this problem. Now as required, User A is not able to cre... [13:53:05] 10Scoring-platform-team, 10VisualEditor, 10edittypes-modeling: Change from source code editing to visual editing: minor edit property not kept - https://phabricator.wikimedia.org/T250388 (10tomert) [15:24:37] https://github.com/wikimedia/draftquality/pull/35 [15:25:00] here is the work we've been doing on ptwiki draftquality [15:25:10] just requires some cleanup [15:25:15] halfak ^ [15:29:01] chtnnh, I just left a couple of notes about grabbing regexes from our articlequality work for template/link names. [15:29:19] just gonna check that out [15:42:32] 10ORES, 10Scoring-platform-team (Current), 10Operations: ORES uwsgi consumes a large amount of memory and CPU when shutting down (as part of a restart) - https://phabricator.wikimedia.org/T242705 (10Halfak) I made this stack overflow post: https://stackoverflow.com/questions/61130651/memory-available-free-pl... [15:43:23] 10Scoring-platform-team (Current), 10revscoring, 10artificial-intelligence: revscoring fetch_text login should be optional - https://phabricator.wikimedia.org/T250311 (10Halfak) 05Open→03Resolved [15:52:48] chtnnh, your changes look good. You'll need to re-extract features and rebuild the model now. [15:53:10] right almost forgot about that XD [15:59:54] chtnnh, you'll need to delete the "with_cache" file and then you can tell it to make the model again. [16:00:14] already on it [16:07:04] halfak should i rebuild the tuning reports as well? [16:07:12] i think yes [16:07:32] Hm. That's less likely to make a change, but it's not bad practice to rebuilt the tuning. [16:07:58] okayy let me rebuild [16:14:22] 10Scoring-platform-team (Current), 10Wikilabels, 10articlequality-modeling, 10artificial-intelligence: Build article quality model for ptwikipedia - https://phabricator.wikimedia.org/T246663 (10Halfak) Forgot to ping here. This is ready for final review. [16:16:02] wikimedia/articlequality#336 (chtnnh-ptwiki-features - d5a8e3c : Aaron Halfaker): The build was fixed. https://travis-ci.org/wikimedia/articlequality/builds/675824157 [16:22:52] still building tuning report halfak [16:23:22] Cool. I'm in meeting anyway :| [16:24:10] let me know when youre out [16:30:32] pushed new model halfak [16:37:00] posting our async update notes -- [16:37:18] halfak - [16:37:20] Y: Dry run of tuning session. Got haksoat set up with calendar and ssh. Unblocked chtnnh for draftquality. Worked on capex and a bit of backlog cleanup. I also did some security related work in wikilabels and got a deployment out. I didn't quite get to compressing our topic models. [16:37:22] T: Met with the head of Pivot for Humanity and discussed manifestos about responsible technology. Might turn into an essay. We'll see. I'll be presenting at the tuning session today. I'm meeting with Isaac to discuss candidates for outreachy project (adding a wikidata-based topic model to ORES). I'll also sync up with the grad student experimenting with auditing methods for ORES. I have a meeting [16:37:24] with a future Jade user about working with our API and I'll be talking with chtnnh and other about NSFW image modeling. If I find time between all of that, I'll work on topic model/vectors memory usage. [16:37:42] kevinbazira - [16:37:45] Y: [16:37:47] Fixed endorsement delete bug [16:37:49] User A was able to delete an endorsement created by User A but was not able to delete an endorsement created by User B. [16:37:51] This patch fixes this issue. Now User A is able to delete an endorsement created by User B. [16:37:53] T: [16:37:55] Fixed double endorsement on facet [16:37:57] User A was able to endorse a proposal in a facet, then was also able to create and endorse a new proposal. [16:37:59] This caused User A to endorse 2 proposals in the same facet. [16:38:01] This patch fixes this issue. Now User A is not able to create and endorse a new proposal if they've already endorsed a proposal in the same facet. [16:38:07] haksoat - [16:38:09] Y: [16:38:11] I continued studying the tokenization module and the deltas package. I had a meeting with Aaron to help setup calendar and ssh. I did research on tokenization and discovered other analyzers which allow matching of tokens with regex. [16:38:13] T: [16:38:15] I will take a deeper look at the analyzers found from yesterday, see how they will work with the regex generated by the deltas package and continue studying revscoring's tokenization module to understand it better. [16:38:17] and me - [16:38:31] Y: Did some code review for Kevin & Aaron, tried out Kevin's simplified MW Vagrant install docs (everything seems to work well, we should add this to our team onboarding docs), also worked on storing label data in our link table, bitmask solution is a no-go, we should probably do it relationally instead, will bring this to sync tomorrow [16:38:34] T: Clean up Link Summary classes and continue re-enabling Jade hooks, will review Kevin's patchset for double endorsements and will work on a couple of Jade maintenance scripts to load the jade_facet table on install. [16:40:07] accraze o/ [16:40:23] Do you have a minute to jump on a short video call? [16:41:15] kevinbazira yeah sounds good, im free now, call me when ready [16:48:23] 10Scoring-platform-team, 10Discovery-Search, 10drafttopic-modeling: super-detect-noop: Support recognizing and updating subsets within an array - https://phabricator.wikimedia.org/T250237 (10EBernhardson) [17:16:10] kevinbazira -- looks like the annotation fix worked, but I think there are some issues with that approach that are causing two tests to fail [17:16:32] i'll write up what i think is happening in a review real quick [17:16:43] hold on [17:16:54] I'm pushing something [17:16:58] cool [17:29:06] accraze please add the write up [17:32:53] will do, still looking through the code, there might be a couple of things happening here [17:34:04] cool [19:17:30] accraze the issue is still there :( It's getting late here. I'll continue working on this tomorrow. Good day 👋 [20:02:59] 10Scoring-platform-team (Current), 10drafttopic-modeling: Why does loading the drafttopic models take so much memory? - https://phabricator.wikimedia.org/T250435 (10Halfak) [20:04:00] 10Scoring-platform-team (Current), 10drafttopic-modeling: Why does loading the drafttopic models take so much memory? - https://phabricator.wikimedia.org/T250435 (10Halfak) In my analysis of T249917, I learned that it requires 1.2GB to load each topic model into memory. This is surprising because the models a... [20:13:18] 10Scoring-platform-team (Current), 10drafttopic-modeling: Why does loading the drafttopic models take so much memory? - https://phabricator.wikimedia.org/T250435 (10Halfak) When I work from a clean python interpreter, `ps` reports that it uses about 9MB of RES. After `from revscoring import Model` we are usi... [20:50:57] 10Scoring-platform-team, 10Wikilabels: Allow privileged users to label deleted revision in Wikilabels - https://phabricator.wikimedia.org/T209960 (10He7d3r) I just run into this problem while labeling edits for ptwiki: `lang=json { "code": "missingcontent", "info": "Missing content for revision ID 47013389... [21:29:24] 10Scoring-platform-team, 10Wikilabels: Allow privileged users to label deleted revision in Wikilabels - https://phabricator.wikimedia.org/T209960 (10Halfak) When I look this up in the UI, I get "Não pode ver esta diferença entre revisões porque uma das revisões foi eliminada. Encontrará detalhes no registo de... [22:34:44] halfak: I've spent sometime checking out the pattern tokenizer for elastic, it does give tokens matched by regex. [22:35:35] Trying to replicate the regex from the wikitext_split though, and it's not a seamless transition as Java's regex support is quite different to Python's. [22:36:44] Currently trying to get it to work though, but I'm struggling with the unicode values.