[13:26:05] Community of Persian Wikipedia is deciding to use clustering on ArbCom elections (which always was controversial) [13:26:18] and in the other hand we have German Wikipedia [15:00:23] I hope halfak_ is around [18:10:03] YuviPanda: hi! [18:10:12] hey awight! [18:10:16] <3 on python3-sklearn [18:10:25] Holler if I can help kick it down the road! [18:10:45] I wasn't sure if I should bite off python3-revscoring quite yet [18:11:47] There's a lot of dust left to settle, for some reason I'm worried about the glue between my pointers to branches I was able to build into a package, and actually building the packages. I think I mentioned this already and you said, don't trip? [18:12:16] lmk when you're doing that work and if you hit any obstacles. [18:13:09] awight: so step 1 there is to move requirements.txt to match the versions [18:13:42] awight: that we have in packages, both our packaged versions and the ones we're using from debian [18:13:46] cool. And I guess I'm waiting for consensus on what version of py3-scipy we want to target [18:13:57] awight: whatever is in jessie, I'd say [18:14:04] (or jessie-backports) [18:14:09] I definitely don't think we want to be buildign it ourselves [18:14:12] that doesn't work, IIR. lemme check [18:14:32] awight: 14.0 doesn't wokr? [18:14:44] What I found was that scipy 0.14.0 breaks the sklearn unit tests [18:14:46] awight: we can request a backport from stretch, 14.1 [18:14:54] yep, that's my suggestion too [18:16:09] awight: yeah, I think that'll work [18:16:22] ok, perfect [18:16:56] awight: so that's the kink right? [18:17:00] and everything else is ok? [18:17:10] so we just switch requirements.txt to the versions we want [18:17:21] and then run tests, and what not [18:17:24] and then just switch to debs [18:17:24] k, doing that now. [18:17:37] and ahem the build all debs step [18:17:53] awight: ah, that. might not be too hard - so we have ores-misc-01 set up with wikitech.wikimedia.org/wiki/Aptly [18:19:05] awight: so we build it there and put the debs / sources there, and they'll automatically be available to all the project [18:21:54] YuviPanda: Know if we're compatible with mwparserfromhell 0.5? [18:22:36] awight: there's no mwparserfromhell 0.5, last version is 0.4.2 [18:22:40] 0.5 is master [18:22:46] right. hrm [18:22:56] I think we still need to backport the debian files [18:22:59] awight: mentors.debian.net/package/mwparserfromhell is the latest package for that :) [18:23:06] oh k sure [18:23:26] awight: I've conned moritzm into being a sponsor for it :) [18:23:29] once I fix some python2 issues [18:24:22] We require nltk 3.0.1 but jessie only has 3.0.0 . Should I assume we can use 3.0.4 from stretch? [18:24:36] why do we need the .1? [18:24:46] donno. It's specified in current requirements [18:24:48] is that just what was on pypi or is there a reason? [18:24:59] so I'd say - let's put up what's in jessie, and see if things break, and if they do fix them [18:25:08] like, scipy has an actual issue we need to fix... [18:25:32] k [18:25:50] numpy is scarier. we require 1.9.2, but jessie has 1.8.2 [18:25:57] you have any context on that? [18:27:08] why do we require 1.9.2? [18:27:18] I think that was just what the latest version was... [18:27:33] yes, 1.9.2 is the latest version [18:27:38] I think 1.8.2 shuld mostly work [18:28:27] I'll back it up and see what breaks. scipy and sklearn will be ok with 1.8.0 [18:29:18] halfak_: Do you know anything about this ^? I'm going to roll us back to requiring numpy 1.8.2 unless you know of a reason these versions should not be wedded. [18:33:37] YuviPanda: wowza. https://packages.qa.debian.org/p/python-setuptools.html [18:33:47] setuptools is 5.5.1 on jessie, we require 15 [18:33:58] I'm setting the needle back per "break shit" protocol [18:34:19] awight: haha :D [18:34:28] awight: +1, I think let's just set the needle back and see what breaks [18:36:03] On that note, I'm planning to use the newer sklearn, which will require model regeneration. k? [18:37:53] arargh, that might not work. It looks like I built the deb packaging on 0.17.dev0 [18:38:06] I'll backport to 0.15.2... [18:38:21] ok! [18:38:37] awight: yeah, I guess it should be ok. I've no idea how to generate models tho [18:38:39] we need the halfak_ [18:39:46] naw, I'll shoot for backporting all the way to the version we're currently using, while I'm at it [18:42:03] awight: I wonder if we backport to the version we're currently using we'll not need a newer scipy [18:42:08] https://github.com/wiki-ai/revscoring/pull/175 [18:42:18] good point! [18:43:08] awight: I commented! also, <3