[04:00:31] executing fab -H 127.0.0.1 initialize_staging_server on my new lab instance [14:58:21] o/ Amir1 [14:58:34] I'm considering moving our sync meeting. [14:58:41] o/ halfak [14:58:44] good morning [14:58:47] to when? [14:58:50] Good afternoon :) [14:59:35] Naively, it looks like 1700 UTC on Mondays would work. [14:59:44] This is so that I can pull in DarTar for a few meetings. [14:59:56] But it will make it easier for schana to join us too. [14:59:58] yeah, that would be awesome [15:00:06] great [15:00:11] Would that time work for you? [15:01:18] is there any chance of it being sooner? [15:02:18] my university closes at 10 PM (UTC+3:30, it'll be in +4:30 after March 21) [15:02:31] Amir1, regretfully, I can't convince anyone in SF to get on a call before 9AM. [15:02:38] And that is 1700 UTC for them [15:02:55] I can join after 10 using my home connection but that means no hangout [15:03:00] I wonder if the end of the day for SF would be easier. [15:03:15] How's 2300 UTC? [15:03:42] it'll be 3:30 AM here, so not very good ;) [15:03:49] Bah. Hmm. [15:04:18] give me some days and I'll figure a way out :) [15:05:13] Oh! Good news. We have daylight savings coming up. [15:05:23] So next week 9AM PST is 1600 UTC [15:06:57] next week would be 1930 for me which is good and after March 21, it'll be 2030. [15:07:00] both are good [15:07:04] my bad [15:07:29] I'm pretty okay with both of them [15:07:36] Great! Let me move the event. [15:09:15] OK. *done* [15:09:18] ToAruShiroiNeko, ^ FYI [15:09:37] the revscoring sync is now moved to Mondays at 1600 UTC [15:14:02] halfak: check your email :) [15:21:14] halfak: I think I can help with some tasks regarding moving to prod [15:21:26] yuvipanda just made some tasks [15:21:30] they are fairly easy [15:32:33] Amir1, just got to the email with the work plan. [15:32:48] \o/ [15:32:52] We should turn this into a public proposal of sorts for your continued funding. [15:33:00] And also so that we can track it easier. [15:33:16] I'd like to make sure each of these lines has an phab task with a bunch of blocking sub-tasks. [15:33:28] Except for maybe "rescoring accuracy" [15:33:28] hmmm [15:33:33] Since that should be more specific. [15:33:59] when I say accuracy I mean semi-supervised ML :D [15:34:06] but yeah [15:34:14] Ahh. yeah. Let's make that a task :) [15:34:24] We should start with a spike and then a process proposal. [15:34:34] Then we can decide if/how to implement. [15:42:18] halfak: I think we are good to go for i18n [15:42:27] I checked both of them in my local host [15:42:30] it worked [15:42:34] Oh yeah. Let me check those out again. [17:22:54] Amir1, hey dude. I'm seriously considering switching our models back to SVCs just to be able to do a new deployment of ORES without surprising everyone. [17:22:58] How do you feel about that? [17:24:07] halfak: how much is difference between AUCs? [17:24:22] * halfak checks. [17:24:26] I can't recall correctlly [17:26:03] This'll take a while. I'll set up a new version of our makefile to generate the linear_svc files. [17:30:24] okay [17:30:36] feel free to change it back if things are alike [17:30:59] one question: halfak why do you want to change back? Is it faster? [17:42:35] Amir1, because the score range (probability) changes when we change modeling strategies. [17:42:46] So, all the tools that use ORES will need to be updated. [17:42:53] https://phabricator.wikimedia.org/T129230 [17:43:05] I think we need to work this out and integrate it into v2 of the UI. [17:44:21] hmm, okay [17:45:48] halfak: In the mean time I work on packaging stuff for migration to prod [17:46:25] Amir1, the bdist wheels? [17:46:31] yeah [17:46:37] the fab files [17:46:49] and https://phabricator.wikimedia.org/T129110 [17:46:59] pretty straightforward I guess [17:47:13] Amir1, https://github.com/wiki-ai/wikilabels/pull/97/files#r55142625 [17:47:32] Amir1, +1 for picking that one up. [20:59:46] Hmm... Bad news. Looks like we dropped a lot of the code for "balancing labels" for SVC models. [20:59:56] So, it'll be a pain to replicate the old scores. [21:00:00] * halfak digs into that. [21:22:30] OK. I think I have the old sample balancing code into a nice structure. I'll run some tests with building models now. [21:55:57] * halfak waits forever for linear_svc model to build. [21:55:58] ARG [21:58:00] GOGOGOGO