[00:16:50] halfak: https://github.com/yuvipanda/ores-contrib is ores-contrib, has a thing for translating titles and pageids into revscoring! [00:17:18] halfak: some of the underlying architecture is still up in the air and I want to talk to you about how the user based one would work, so I can make sure the underlying setup is generic enough without being too generic [00:17:26] I'll also do the puppet and whatnot work for this after we've discussed [00:17:32] (and maybe move this to wikiemdia-ai too) [00:28:22] halfak: the output format is up in the air too. we can merge output of ores with augmented info instead of what we're doing now [00:28:36] halfak: so think of it as a super-first draft than anything else :) [00:28:38] * YuviPanda goes afk [14:02:23] halfak: o/ [14:02:30] Hey! [14:02:36] Sorry. running a bit late [14:02:40] Will be in the hangout in 1 min [14:03:08] ok, I'm away for 1 min [14:05:07] I'm in. [17:06:17] * PlasmaFury waves at halfak [17:06:23] did you see my messagse [17:06:29] o/ I did. [17:06:34] good morning :) [17:06:42] :D [17:08:44] PlasmaFury: wait I mean my messages here, not in PM, etc. [17:08:47] * PlasmaFury stays in meeting [17:09:04] Oh yes. [17:09:10] * halfak goes back to thinking about that. [17:09:35] Had an 8AM meeting, so I read your messages and then promptly forgot to think more. [17:09:40] ah heh [17:09:42] 8AM meetings boo [17:09:53] Timezones! [17:10:06] Gotta meet with Amir1 at sane hours sometimes :) [17:10:40] PlasmaFury, I'm wondering if we should let each ORES tool be it's own project/repo. [17:10:56] It seems like you are working on an ores-contrib framework for all ORES tools. [17:11:01] Is that right? [17:11:10] not for 'all', but for a specific subset [17:11:15] that will get some form of support [17:11:24] Yeah. Gotcha. [17:11:40] * halfak thinks. [17:12:11] so we'll do CR in some form, and run it 'ourselves' and I can maybe convince people to run it in prod too at some point [17:12:31] and we'll set a scope for what is and isn't going to be there [17:12:51] Gotcha. [17:13:32] so the user stuff should definitely be in it and I'm looking to find out the shape of it so we can make sure the framework supports that too [17:14:06] Looks like I should try to implement a user mapper and see what we get. [17:15:41] halfak: yeah [17:15:51] halfak: you can tell me what the output should look like and I can implement it [17:16:22] halfak: there's also questions about wether it should be augmenting the ORES output directly (merge the outputs) or have it be separate like now [17:16:27] * PlasmaFury likes it being separate [17:16:36] +1 [17:17:09] so that is more flexible [17:17:18] Yeah. [17:17:27] and also if I get page titles,, I can key using the page title I have and then key into the revid [17:17:55] Yeah... the user stuff might not work for this. [17:17:59] right [17:18:06] so can you show me example output from it? [17:18:08] User scores are not 1:1 with ORES scores. [17:19:06] {"user-goodfaith": {"prediction": true, "pobability": {"true": 0.998, "false": 0.002}}} [17:19:29] In the case of this scoring, I'd need to (1) build a model for goodfaithyness [17:19:46] And (2) process multiple ORES scores with that model to arrive at one goodfaithyness score. [17:21:06] right [17:21:10] so that can already work with this [17:21:17] there needs be no 1:1 [17:21:23] but you probably don't want the output [17:21:24] halfak: hello [17:21:35] to have all ORES responds [17:21:44] o/ aetilley [17:22:19] PlasmaFury, I see. Anyway, I think the ball is in my court now to see what limitations exist and make a proposal. [17:23:19] hmm [17:23:34] halfak: I can modify it to be a bit more generic and send a PR [17:23:44] and see how that is [17:23:49] Sure. You'll likele beat me to it. [17:23:57] Hard to prioritize this right now :\ [17:23:59] yeah [17:24:02] * halfak manages himself [17:24:09] halfak: that's why I wanna drive this and not put this on your head [17:24:17] :) Cool then. Rock on. [17:24:32] I'll plan to get some snippets together that demonstrate how I do good-faithiness detection. [17:24:37] You'll love how simple it is. [17:24:52] Maybe I can have bearloga review my simple bayesian strategy and make suggestions. :) [17:24:58] +1 [17:25:07] the output is good enough for me to make sure that's usable [17:25:16] halfak: would love to! [17:25:38] bearloga, I'll have a summary, but if you can't wait, see "Desirability sorting" in https://www.researchgate.net/profile/Rstuart_Geiger/publication/263115950_Snuggle_Designing_for_efficient_socialization_and_ideological_critique/links/00b49539f028de7640000000.pdf [17:26:04] BaH! stupid re-hosting. See http://www-users.cs.umn.edu/~halfak/publications/Snuggle/halfaker14snuggle-preprint.pdf [17:26:08] for better linkiness. [17:26:52] halfak: is it computationally intensive? [17:27:14] or is it just IO mostly? [17:33:08] IO mostlyu [17:33:21] kkk nice [17:33:23] err [17:33:25] kk nice [17:58:59] halfak: hey, I can't make it to the talk [17:59:04] :( [17:59:05] :( [17:59:12] They don't let me in [17:59:16] No worries. Will have notes. Thanks for letting me know :) [17:59:40] I will definitely help [18:00:10] Ok [18:00:28] Did we cant to call or IRC? [18:00:39] *want [18:00:42] halfak: ^ [18:00:56] Let's do a call :) [18:01:00] k [18:38:27] hello halfak [18:38:33] lookıng ınto the chınese stuff now [18:38:52] Just got home, had to dodge police action at the train station [18:39:02] which involved me ignoring them and waiting in the train [18:39:18] bah! [18:39:25] ToAruShiroiNeko, international spy [18:39:27] ;) [18:39:33] (in the James Bond sense) [18:39:54] I love spying as much as I love budwiser lime :p [18:40:04] :D [18:40:24] James Bond is less into martinis and more into computers these days [18:40:26] * halfak barfs [18:40:44] next movie will probably have just geeks, they may as well shoot it during the wikimania hackathon [18:41:03] I had some bug lime once and it was better than expected. But I did have pretty low expectations. [18:41:35] with that you can enjoy almost anything [18:41:43] even cleaning the kitty litterbox [18:43:40] Not saying I enjoyed it. But if there was nothing else to drink, I'd prefer it to death. [18:45:00] halfak: hehe at 'bug lime' [18:45:21] Apt. typo [18:45:29] heh [18:46:01] yes death tastes not so good [18:46:08] typo? [18:46:09] aw [18:46:34] so bud lime is bug juice to you? [18:46:52] Frued would think so [18:47:02] Freud [18:47:09] Freeeuuuude [18:47:16] like if you typo his name to 'Fraud' [18:47:19] he would probably cease to think after an episode with bud lime [18:47:24] is that a Fraudian Typo? [18:47:39] Do you know who drunk bud lime as a challenge to fix FIFA? [18:47:43] there's no trailing "e". https://en.wikipedia.org/wiki/Sigmund_Freud [18:47:46] And that it worked? [18:47:59] Bud lime works in mysterious ways [18:51:29] halfak do we have the budget to offer free bud lime to our volunteers? [18:51:52] lolno [18:53:08] "Complete worksets or we WILL send you bud lime and you will drink it or all your edits will get deleted" [18:53:14] >:D [18:53:18] Anti-motivation FTW [20:34:40] Looks like we serve between 35k and 70k scores per day to non-precached requests. [20:34:50] Depending on whether someone is doing a batch job that day. [21:18:20] o/ Amir1 [21:18:45] Do you think you could get that pywikibase issue with wb-vandalism resolved? I'd like to test out a new model.