[00:26:33] * halfak notes aetilley lurking :) 
[00:35:28] <halfak>	 ewulczyn, since I've convinced you to join, I could give some quick intros. 
[00:35:55] <halfak>	 aetilley is a math PhD who is working with us on bias detection in ORES/revscoring
[00:36:36] <halfak>	 ToAruShiroiNeko, (you probably met at Wikimania) is also on the IEG grant and working on community engagement & error correcting output codes. 
[00:36:43] <halfak>	 ToAruShiroiNeko == Gediz
[00:37:19] <halfak>	 I figure you know YuviPanda and awight. 
[00:37:57] <halfak>	 ewulczyn, is Ellery W., one of the researchers I work with at the WMF.  Also involved in article recommendations and reader behavior stuff.  notably released the clickstream dataset that's been picked up by a lot of researchers. 
[00:38:21] <halfak>	 Anyway, likely a good reference for #wikimedia-ai :) 
[00:38:23] <YuviPanda>	 what nobody knows who I am!
[00:38:25] * YuviPanda hides
[00:38:28] <halfak>	 :D 
[00:38:38] <halfak>	 You changed your name back from PlasmaPants or whatever 
[00:38:39] <halfak>	 :P 
[00:45:36] <ewulczyn>	 thanks for the intros halfak. I also do a bit of AB testing an bayesian stats in case those topics ever come up in a project.
[00:46:19] * halfak wonders if bear-loga has "bayesian" as a ping word. 
[00:52:55] <halfak>	 I'm off.  Have a good one, folks!
[18:24:32] <Amir1>	 halfak: o/
[18:24:39] <halfak>	 o/ Amir1 
[18:24:54] <Amir1>	 I had a crazy idea re. wikidata
[18:25:08] <halfak>	 Oooh.  I like crazy ideas :) 
[18:25:09] <Amir1>	 we can extract features from API
[18:25:34] <Amir1>	 so let's introduce some bias and use user.age
[18:25:44] <Amir1>	 to have better AUC
[18:26:07] <Amir1>	 what do you think
[18:26:28] <halfak>	 Well... we *are* using it in other models, but I'm thinking that we want to get rid of user.age. 
[18:26:38] <halfak>	 But I'm interested in running some tests. 
[18:27:36] <Amir1>	 that would be great
[18:27:54] <Amir1>	 we can implement other API-based features too
[18:28:19] <Amir1>	 things we couldn't do before
[18:28:28] <halfak>	 +1
[18:28:49] <Amir1>	 I also made a PR for siglcust
[18:28:53] <Amir1>	 *sigclust
[18:31:56] <halfak>	 Awesome. 
[18:32:37] <halfak>	 it looks like we should be able to (1) generate docs and (2) make it easier to access the functions.  Then the library will be ready for use by others. 
[18:32:45] <halfak>	 I suppose we should also have some unit test. 
[18:36:07] <Amir1>	 The docs is a mess
[18:38:09] <halfak>	 We should have some examples too. 
[18:45:07] <Amir1>	 I spend some time today or tomorrow to finish this
[18:45:49] <halfak>	 Great.  Thanks for picking this up.  :)  Be bold. 
[18:50:31] <Amir1>	 you're welcome :)
[18:50:36] <Amir1>	 I hope I can do more
[18:50:53] <Amir1>	 halfak: Do you have something in mind? as another task
[18:52:42] <halfak>	 I think we should (1) keep trying to drive up the AUC for wikidata's revert model and (2) get a wiki labels campaign set up to help us get better data to train/test with. 
[18:54:29] <halfak>	 Is there a Wiki labels page on wikidata.org?
[18:54:38] <Amir1>	 Not yet
[18:54:44] <Amir1>	 but 
[18:55:00] <Amir1>	 Building something to show diffs in js is pretty complicated
[18:55:15] <Amir1>	 Am I wrong?
[18:55:28] <halfak>	 Oh.  I think we can just use what the wikibase API gives us. 
[18:55:37] <halfak>	 It should look like what you get on a wiki page, right?
[18:56:22] <Amir1>	 I'm not sure
[18:56:51] <Amir1>	 what API action do you call in "traditional" wikilabels
[18:57:09] * halfak gets URL
[18:58:31] <halfak>	 https://www.wikidata.org/w/api.php?action=query&prop=revisions&titles=Q238&rvdiffto=prev&rvprop=content
[18:59:03] <halfak>	 Looks like we get the table structure.  It's just that the lines of the table are replaced with "property" etc. 
[18:59:10] <halfak>	 That means our current CSS should work OK too :) 
[19:00:05] <Amir1>	 \o/
[19:00:13] * Amir1 dances of joy