[21:20:02] <halfak>	 o/ Amir1, if you're around, I have an easy review for you.  https://github.com/wiki-ai/editquality/pull/69
[21:20:14] <Amir1>	 Yup
[21:20:19] <Amir1>	 working on wikidata stuff
[21:20:24] <halfak>	 Gotcha.  No rush
[21:20:24] <Amir1>	 but I can spare a minute
[21:21:33] <Amir1>	 halfak: why we have even rev ids for enwiktionary 
[21:21:49] <halfak>	 Because the 200k sample was *way to big* 
[21:21:54] <halfak>	 I needed to cut it in half 
[21:22:06] <halfak>	 And this is a way to do so fairly and deterministically. 
[21:22:08] <halfak>	 :D
[21:24:25] <Amir1>	 :)))
[21:24:29] <Amir1>	 Do shuffle dude
[21:25:24] <halfak>	 Not deterministic.  I would have had to check in the file. 
[21:26:21] <Amir1>	 hmm, still I don't have faith in numbers 
[21:26:22] <Amir1>	 :D
[21:26:27] <Amir1>	 it's not that bad
[21:26:39] <Amir1>	 but I don't like it
[21:41:37] <halfak>	 Amir1, yeah, either that or we (1) have a totally separate query or (2) we check in the dataset
[21:41:45] <halfak>	 Why no faith in #s?
[21:42:03] <Amir1>	 They might not be totally random 
[21:49:28] <halfak>	 Amir1, they are auto incremented. 
[21:49:42] <halfak>	 Under what conditions would non-uniformity occur?
[21:50:16] <Amir1>	 someone makes too  much edit I guess
[21:50:59] <halfak>	 But how would that be a problem?
[21:51:27] <halfak>	 I've been in this debate before. https://meta.wikimedia.org/wiki/Research_talk:VisualEditor%27s_effect_on_newly_registered_editors/June_2013_study#Observational_study_with_haphazard_non-random_assignment_to_VE_or_Wikitext
[21:51:58] <halfak>	 I ran a ton of analysis to show that the strategy does not produce a bias.
[21:52:06] <halfak>	 Round robin sampling is pretty common
[21:52:07] <Amir1>	 Interesting 
[21:52:55] <Amir1>	 I will read it