[08:57:46] <dcausse>	 Trey314159: question related to japanese on the cirrus help board https://www.mediawiki.org/wiki/Topic:Ygx4garhqu80uxnx
[10:42:58] <dcausse>	 errand+lunch
[14:53:09] <inflatador>	 <o/
[14:53:28] <gehel>	 o/
[14:54:34] <dcausse>	 o/
[15:07:46] <ebernhardson>	 \o
[15:08:07] <dcausse>	 o/
[15:13:22] <inflatador>	 still getting alerts for categories on graph split hosts ;( I need to c lean that up
[15:17:26] <ebernhardson>	 puppets un-install things story is indeed prettty meh
[15:21:28] <inflatador>	 True, it's possible the checks didn't get removed on the icinga pollers or whatever they come from
[15:29:42] <Trey314159>	 dcausse: I will take a look—thanks!
[16:00:46] <ryankemper>	 Missing first 15’ of triage
[17:30:38] <ebernhardson>	 hmm, re-reading the comment about deepsearch user is asking to be able to apply a suffix-based filter to graph traversal. So for example it wouldn't decend into a category suffixed 'in fiction' or 'in art'
[17:36:48] <dcausse>	 yes, that seems way more complicated than what we anticipated by asking if controlling the dept param would be useful
[17:36:59] <dcausse>	 s/dept/depth
[17:37:26] <ebernhardson>	 indeed. I did notice though poking at blazegraph, there is a gas:maxVisited argument we don't use. Not sure if we need to, but might be useful somewhere
[17:37:34] <ebernhardson>	 maybe limit is same in the case anyways
[19:02:01] * ebernhardson is always a little worried when issuing a query that should have taken <1s and is still running a minute later...
[19:56:54] <ebernhardson>	 randomly curious information, mjolnir enwiki has been reasonably consistently training to ndcg@10 of .870 to .895.  If anything high values might suggest training data is too easy
[20:00:12] <inflatador>	 lunch, back in ~40
[20:22:56] <inflatador>	 back
[20:46:43] <Trey314159>	 dcausse & ebernhardson: I pinged you two on the Japanese question David mentioned above. There are some language things they could do to improve, but they seem to be running old code or at least have old configs, and they might be using out-of-date query profiles and such. I'm not sure what to look at or ask, so... you're up!
[20:49:45] <ebernhardson>	 :P