[02:12:36] o/ guillom [02:12:43] hey halfak [02:12:49] DO you know who I would talk to about researchy things in Search and Discovery? [02:12:50] What's up? [02:12:55] Hmm. [02:12:57] Other than Ironholds [02:13:10] Bleh. He was my only lead. [02:13:23] Hokay no worries. [02:13:25] I don't think they have anyone else. [02:13:27] Thanks :) [02:13:34] But I think they're hiring a minion for Ironholds? [02:13:34] Soon, they will. [02:13:38] Yup [02:13:39] :) [02:15:07] (And by "minion", I mean "adorable yellow creature as seen in "Despicable me", not "A subordinate official, especially a servile one.") [02:16:01] :D [02:16:20] Never seen it, but I laugh at the trailers :D [02:17:13] halfak: You should watch them. Despicable Me, and Despicable Me 2. Then you'll be ready for the upcoming Minions movie. [02:17:47] Despicable Me was awesome. The sequel was good for a sequel. [02:17:57] I have unrealistically high hopes for the next one. [02:18:29] Is it pixar? [02:20:04] Wikipedia says no. [02:21:24] Interesting. I usually abuse that as a heuristic. [02:21:40] OK. I'm outa here. Have a good night, sir! :) [02:21:45] G'night! [02:21:56] * guillom dives again into JavaScript promises. [10:58:29] * Nemo_bis a bit confused by https://meta.wikimedia.org/w/index.php?title=Research_talk%3AIncreasing_article_coverage&type=revision&diff=12552856&oldid=12552811 [14:02:39] Hello. [14:11:56] Good morning guillom! [14:12:06] Hey halfak [14:37:55] AAAARGHPANDA. [14:38:28] A different panda for every state. Collect them all! [14:38:40] Gotta catch'em all! [14:39:05] Can we have a Wikipedia page for each of them? ARE THEY NOTABLE????!!!11 [14:39:30] is the original YuviPanda notable? [14:58:29] * halfak takes notes [15:31:32] guillom I think our pandas all got an article in a parallel universe [15:31:52] ToAruShiroiNeko: :) [15:32:25] cross parallel universe import would be my proposal for the next mediawiki addon [15:32:58] since we are only importing information and not actual matter it should not violate quantum physics constraints [15:33:09] Interesting. [15:33:31] But how is information conveyed? [15:34:08] By light? Electrons? Either way it's matter. [15:34:28] quantum superposition would be my guess [15:34:37] light isnt matter [15:34:49] photons! [15:35:16] Well, it doesn't have mass while it is a photon. [15:35:51] lack of mass means not matter :p [15:35:52] I'm guessing that energy and mass are roughly equivalent in their nontranferability. [15:36:11] idea here is to copy the information [15:36:17] Well, we could just convert all of the mass to energy, transfer and then convert back to matter. [15:36:24] Thus we can transfer matter. [15:36:27] My script is working again! [15:36:38] halfak you made guillom's script work [15:36:48] No problem. Happy to help :) [15:37:39] halfak: idea: work on priority prediction at wikimania? [15:37:50] To be exact, it was working agin, then I completely rewrote it, and now not only is it working, but files are organized properly, everything is documented, and it handles errors. [15:37:50] halfak the problem is the question of what would happen if we remove matter from a parallel universe [15:38:07] would that not violate energy conservation law? [15:38:25] harej, I'm hoping to work on a UI to help WP:Medicine et al self-serve article quality predictions. [15:38:31] halfak I still got no response from wikimania people [15:38:36] But I don't need to work on just that. [15:38:40] halfak: would importance be part of that? :) [15:38:48] We can maintain conservation if we exchange matter. I can think of a few people to send to that other universe :) [15:39:23] harej, it isn't for the current classifier, but I'm down for adding an importance classifier. [15:39:33] guillom, lol! [15:40:38] I made the decision not to ship priority prediction for now, since I think it will take a significant amount of time to make it into a useful tool. [15:41:42] +1. But still backburner? [15:42:04] In the short term, I think sorting articles by pageviews and inlink counts could be very valuable. [15:42:36] SuggestBot kind of already does that. [15:42:42] And it is back burner. [15:42:47] Everything is back burner [15:44:52] Yup. :) The trick is keeping the back burners from getting cold. [15:46:55] Hence my page of half-baked ideas you once commented on. [16:29:03] no RG meeting today, right? [16:29:23] oh wait there is someting in the pad [18:13:40] hi Nemo_bis. Did you get my copy-pasted response in the other channel about SUL tables? [18:15:30] guillom: here is the response from Santhosh about the SUL tables: "for logged in users the preferences are stored in user_properties table with key as 'uls-preferences'. The preferences is a json serialized datait has language preferences- previous language choices, input method preferences etc." [18:17:32] leila: good to know; thanks :) [18:18:40] np. :-) [18:37:54] Nettrom, I get better AUC with a SVM than RF for article quality, but worse accuracy. https://gist.github.com/halfak/9c74317dac3457e174b6 [18:39:27] Hmm... that could be because of balancing. Let me try without balancing the labels. [19:03:19] ouch gone [19:03:43] guillom: what do you think filing a ULS feature request for an API that provides language preferences using info like that [19:12:14] Looks like re balancing doesn't matter. [19:15:23] Nemo_bis: I'm not sure what you mean. [19:21:56] halfak: file w/model params https://github.com/nettrom/Wiki-Class/blob/master/examples/train_model.py [19:22:14] guillom: I mean that we currently have extensions and tools inventing language guessing mechanisms on their own. [19:22:47] guillom: instead of scraping babel and so on, there should be a single "official" API where to fetch this info, which would assemble it from various sources. [19:23:32] Nemo_bis: My guess is that it's considered to be private information [19:24:15] guillom: then the API could expose the public information only :) [19:24:44] Or be restricted to the user. [19:25:16] At least Wikibase wouldn't need to reinvent the wheel and projects like this (which eventually is supposed to go into ContentTranslation anyway?) would be able to reuse the method in PHP server-side at least [19:25:35] But maybe I'm not making any sense [19:27:20] Sorry, I'm eating and discussing with others at the same time :) [19:31:30] Sure [19:32:20] In any case, I agree reusing APIs and preferences is a good thing. [19:44:31] halfak: if you look at https://github.com/nettrom/Wiki-Class/blob/master/wikiclass/features/content_and_infonoise.py you'll see it has "num_citetemplates" and "lognoncitetemplates", so I log one but not the other [19:44:47] somehow that worked slightly better than the alternative [19:44:49] IIRC [19:44:57] (log vs non-log, that is) [19:44:58] Cool. Thanks. I'll include it. [19:45:29] halfak: great! then the models should be comparable, makes it easier to check the results [19:46:02] Nettrom, I got up to .581 accuracy with the param changes we made \o/ [19:46:12] I'll work some more on filtering out A class. [19:46:47] halfak: \o/ Yeah, I suspect removing the A class is the remaining difference [19:47:58] Nettrom, the trick to grep was using perl regex instead of extended regex :S [19:48:37] * halfak builds new model without 'A' class [19:48:45] halfak: Perl saves the day, again! :D [19:48:46] it's a nettrom! [19:49:17] o/ harej [19:49:59] Ahhh! Phabricator looks weird [19:50:08] Disruption! [19:52:20] halfak: It has a Wikimedia logo now, instead of the Eye of Sauron! [19:53:48] https://upload.wikimedia.org/wikipedia/commons/thumb/3/3b/Phacility_phabricator_logo.svg/1024px-Phacility_phabricator_logo.svg.png [19:54:16] Nettrom, I added the last results to the gist. it looks like the AUC is comparable to SVC now. :) [19:54:20] https://gist.github.com/halfak/9c74317dac3457e174b6 [19:54:39] Really need that table to output in a consistent order. [19:55:38] Nettrom: I am encouraged that everyone will be getting The Table now. For WikiProjects, I'll probably want the output to look different, but I am not done formalizing the WPX UI design requirements. So for now, output as a table, and I will submit a pull request later? [19:56:45] halfak: cool, nice results, but yeah, list them in order ;) [19:57:46] harej: WPX will have its own template on-wiki, so it should be fairly easy to change it as necessary, and if there's something that needs coding, pull requests are nice [19:58:12] Nettrom, sorry for all the pings. Does this change look OK to you? https://github.com/wiki-ai/ores/pull/58 [19:58:12] Sounds great. Thank you for your effort toward the cause. [20:00:07] halfak: yep, looks good! [20:00:11] Thanks dude [20:08:26] Nettrom, Accuracy: 0.5947568087877244 [20:08:33] Gist updated [20:08:36] https://gist.github.com/halfak/9c74317dac3457e174b6 [20:39:07] * halfak hops on bike [20:39:10] back in ~ an hour [21:01:01] Any R experts available soon for a few minutes of tutoring on R fundamentals (basic graphing, area graphs)? Happy to do assigned homework first; looking to reduce learning curve from hours to less [21:26:43] * guillom does the little dance of the code that compiles. [21:26:46] ┏(-_-)┛┗(-_- )┓┗(-_-)┛┏(-_-)┓ [22:05:58] DarTar, I just got a ping from ggellerman, but I'm in the call waiting for you guys [22:10:21] halfak: sorry, coming - all my meetings today are running over [22:10:35] Hokay [23:31:32] halfak: dammit, I’m 30 min late now, you talking to Wes? [23:32:16] I’ll try and join the last part if this is still running