[00:43:45] o/ [14:18:22] o/ [17:21:52] https://github.com/wiki-ai/revscoring/issues/234 [17:22:13] ^ To address my struggles with effectively measuring and improving performance in our feature extraction process. [21:27:49] Hello everyone [21:48:54] o/ [21:48:57] halfak: o/ [21:52:27] I've been making progress with this sknn, reading examples for using it and stuff, but still have quite a bit more to learn. [21:52:48] o/ Amir1 [21:52:56] Showed violetto your work on the ORES ui. [21:53:16] I'm working on slides for a presentation I'll give tomorrow. As soon as that is ready, I'll merge your work into ORES. [21:53:21] great [21:53:26] hello Amir1 [21:53:32] violetto: hey :) [21:53:42] aah, im running into a meeting now, but i'd love to check out what you did [21:53:50] sure [21:53:56] I'm here for a while [21:54:05] I woke several hours ago [21:54:10] *woke up [21:55:49] halfak: in the mean time what should I do? [21:56:07] I was trying to read ORES wsgi code to see how I can add test results [21:56:20] but didn't have time to examine it more [21:56:21] Amir1, updated "balanced" datasets for all of our reverted models. [21:57:07] Amir1, also, thoughts on how we could generate a balanced dataset for Wiki labels. [21:57:13] (which uses API calls much more) [21:57:38] hmmm [21:57:39] * halfak should write down the rules that are encoded into `editquality prelabel` [21:57:48] using the dump reader? [21:57:54] Amir1, maybe. [21:58:01] Maybe we shouldn't use a dump reader at all? [21:58:16] I was talking about " updated "balanced" datasets for all of our reverted models." [21:58:19] Maybe the dump reader should just flag reverted edits and a secondary process can use that. [21:59:07] about the second one I agree making API calls ~20K edits isn't that bad [22:00:58] "Maybe the dump reader should just.." +1 +1 [22:02:03] Cool. I haven't had time to think it through, but I am imagining that we can draw from different subsets of edits purposefully. E.g. edits by non-trusted users (reverted or not), reverted edits and edits by blocked users. [22:03:27] I think in these cases we should act case-by-case [22:03:55] e.g. we should only consider edits by not trusted users in bot-pedias [22:04:42] but when it comes to wikidata even though it's a bot-pedia we should check both edit by not trusted users and reverted [22:06:43] halfak: btw: can you check this and tell me what Adam meant? I have trouble understanding it: https://phabricator.wikimedia.org/T122684 [22:08:48] It seems that sknn does have regularization and not like I said. It was probably n00b of me to think it doesn't. :/ Anyway, halfak you mentioned that you'd be happy to me get the labeled data to experimetn with - IIUC I could get it from editqually (or sth) myself. Right? [22:09:44] *editquallity [22:10:06] Amir1, I think he means to say that we don't know if users will want/understand a variable setting or not yet. [22:10:15] Maybe we should do user testing first? [22:10:20] Not sure. [22:10:36] pipivoj, will send an email with fun labeled data in a moment. [22:10:53] Actually...wait.. I don't think I have your email. [22:11:04] ok, I see [22:11:06] :) [22:11:09] thanks halfak [22:17:51] I've just confirmed my address through wikiemdia. Send me a msg through wiki interface and I'll reply so you get my address. My user account is [[:meta:User:PIPIVOJ]] [22:17:58] 10[5] 04https://meta.wikimedia.org/wiki/:meta:User:PIPIVOJ [22:19:34] halfak, I've actually posted it here but now I reconsidered and don't want it make public. If anyone sees it through logs, then they'll have it :/ [22:49:25] pipivoj, ^ that user link doesn't work. [22:50:21] Asimov, bad dog, you bad. :) [22:51:59] tsting : [[:en:User:PIPIVOJ]] [22:51:59] 10[6] 04https://meta.wikimedia.org/wiki/:en:User:PIPIVOJ [22:52:19] again tsting : [[:User:PIPIVOJ]] [22:52:30] 10[7] 04https://meta.wikimedia.org/wiki/:User:PIPIVOJ [22:53:15] Now it should [22:53:45] Ah ha! You have no user page yet. :P [22:54:08] But I see that I can look at your contribs and email you so Meta knows your username [22:54:09] :) [22:55:22] Now is it AsimovBot's fault or ... ;) [22:56:11] Nah, it's mine, I didn check the syntax :( [22:56:20] of the url [22:57:17] I mean that of rubber ducking :) [23:43:30] Ahh! It's already the end of the day :( [23:43:40] * halfak finishes slides so he can work on ORES UI [23:51:36] \o/