[21:12:44] <dsaez>	 isaacj, yt?
[21:12:57] <isaacj>	 dsaez: yep
[21:13:31] <dsaez>	 I'm looking to your code and outputs.. ...
[21:13:38] <isaacj>	 uh oh
[21:13:50] <dsaez>	 === Human ===
[21:13:50] <dsaez>	                  label      prop
[21:13:50] <dsaez>	 0                human  0.268812
[21:13:50] <dsaez>	 2977             child  0.000023
[21:13:50] <dsaez>	 6146  displaced person  0.000008
[21:13:52] <dsaez>	 === 0 ===
[21:13:54] <dsaez>	   label      prop
[21:13:56] <dsaez>	 1     0  0.063171
[21:13:59] <dsaez>	 8     0  0.011147
[21:14:01] <dsaez>	 === Film ===
[21:14:03] <dsaez>	                     label      prop
[21:14:07] <dsaez>	 2                    film  0.030739
[21:14:10] <dsaez>	 31                3D film  0.002277
[21:14:12] <dsaez>	 55  animated feature film  0.001395
[21:14:16] <dsaez>	 not sure how to interpret that
[21:14:39] <dsaez>	 3Dfild is a subset of film?
[21:14:49] <dsaez>	 *3D film
[21:15:37] <isaacj>	 sure -- in between the "===" is the wikidata class that met a given threshold and then listed under it are the top-three (by pageviews) subclasses that were aggregated into that class
[21:16:10] <isaacj>	 so Film has under it "film", "3D film", "animated feature film", and probably a bunch more subclasses
[21:16:35] <isaacj>	 toy story: https://www.wikidata.org/wiki/Q171048 
[21:17:00] <dsaez>	 ok
[21:17:01] <isaacj>	 so in the data any views to toy story would be associated w/ "animated feature film" initially but based on this taxonomy would be considered "Film"
[21:17:49] <isaacj>	 the numbers on the far left in the printout are just the pandas index numbers, which happen to correspond with their pageview rank
[21:17:55] <dsaez>	 the 100% is computed out of the sum of pageviews? items? subclass-of relationships?
[21:18:17] <isaacj>	 100% is sum of pageviews
[21:18:39] <isaacj>	 i think the numbers are 1.9M wikidata items -> 32K unique instance-of properties
[21:19:22] <isaacj>	 and then depending on the thresholds you use, those 32K instance-of properties are aggregated into somewhere betweeen like 20 and 100 classes
[21:20:38] <isaacj>	 and its 7.5M pageviews across those items
[21:20:44] <dsaez>	 interesting distribution ... it's a super long tail
[21:21:40] <isaacj>	 yeah because you have instance-of values like "GDR Badminton Championships" and "Canadian Forces base"
[21:21:59] <dsaez>	 subclass-of or instance-of or both?
[21:22:36] <isaacj>	 those are values for instance-of
[21:23:02] <isaacj>	 as far as unique subclasses...let me check
[21:23:53] <isaacj>	 80k different subclasses across all of wikidata
[21:24:21] <dsaez>	 got you.. 
[21:24:44] <dsaez>	 we need some sort of hierarchical clustering 
[21:24:46] <isaacj>	 and the 32k unique instance-of values are only based on the articles that i'm analyzing so there are probably way more instance-of values out there
[21:24:57] <isaacj>	 yeah, hence this aggregation technique