[00:07:03] <legoktm>	 yuvipanda: ?
[00:07:14] <yuvipanda>	 bah
[00:07:17] <legoktm>	 :P
[00:07:19] <yuvipanda>	 was for lzia
[15:30:17] <bearloga>	 good morning
[15:37:45] <halfak>	 o/ bearloga 
[15:37:49] <halfak>	 I have a thing to show you. 
[15:38:02] <halfak>	 https://www.mediawiki.org/wiki/User:Halfak_%28WMF%29/mediawiki-utilities/Index
[15:38:20] <halfak>	 I'm still working on the docs, but this is the rough layout of an index to all the utilities. 
[15:38:51] <halfak>	 (Re. your comment about having a way to find out what's available in the utilities I'm building) 
[15:39:11] <halfak>	 A few of the doc links are broken still, but most of them work. 
[15:40:14] <bearloga>	 oooooh, very neat, halfak!
[16:12:08] <subbu>	 statistic geeks .. can someone explain why parsoid's latency charts have this behavior .. i look at a 6-month trend in grafana .. smooth out the metrics using a movingMedian of 100 values .. and it turns out p95 latency = PI * mean latency.
[16:12:11] <subbu>	 http://grafana.wikimedia.org/#/dashboard/db/parsoid-load-latency-correlation?panelId=14&fullscreen 
[16:13:45] <subbu>	 easier to see if you hide the traffic graph there (which happens to be a mirror reflection of the latency graph).
[16:14:36] <subbu>	 is it a statistical feature of these metrics or is it an artefact of smoothening?
[16:15:40] <halfak>	 subbu, What is PI?
[16:15:42] <subbu>	 I am being somewhat gimmicky with the choice of PI .. but it looks like mean and p95 seem to be identical scaled versions and 
[16:16:01] <subbu>	 i picked 3.1415 just for the heck of it. ignore the pi part of it.
[16:16:08] <subbu>	 unless it is relevant
[16:16:23] <halfak>	 Oh!  
[16:17:26] <subbu>	 click on that link and play around :)
[16:17:49] <halfak>	 subbu, do you have a histogram of latency timings?
[16:17:54] <halfak>	 I want to know how they are distributed. 
[16:18:20] <subbu>	 the data is coming from graphite .. and grafana is just charting them.
[16:18:28] <subbu>	 so, i guess you could get raw data from graphite?
[16:18:45] <subbu>	 not worth spending a lot of time on this .. was just curious mostly.
[16:20:28] <halfak>	 subbu, one possibility is that the distribution of latency timings is skewed in a way that results in this relationship between the mean and 95%ile
[16:20:48] <halfak>	 I'm simulating a lognormal now to see how likely of an explanation that is., 
[16:21:37] <halfak>	 What does scale() do?
[16:21:43] <subbu>	 i've forgotten most of my statistics material in 20 years .. mostly from disuse .. so i'll go look up wikipedia for what lognormal, etc. mean once you find out more :)
[16:21:49] <subbu>	 N * value .. is all
[16:21:54] <subbu>	 for every point in the graph
[16:22:06] <halfak>	 What is N?
[16:22:11] <halfak>	 What is value?
[16:22:13] <subbu>	 I picked 3.1415 
[16:22:29] <halfak>	 oh. 
[16:22:50] <subbu>	 it charts a "3.1415 * mean(t)" vs "p95(t)" for every t on the x-axis
[16:22:50] <halfak>	 Yeah... looks like I'm getting similar results with my lognormal simulation
[16:25:35] <subbu>	 you mean .. parse latencies in parsoid seem to have a lognormal distribution?
[16:25:44] <halfak>	 Yeah.  
[16:25:48] <halfak>	 That'd make sense. 
[16:25:50] <subbu>	 interesting ... 
[16:25:57] <subbu>	 say more.
[16:26:06] <subbu>	 because i don't know why that would be.
[16:26:21] <halfak>	 If I'm right, then you should find that the geometric mean generally corresponds to the median. 
[16:26:42] <halfak>	 mean = sum(x)/length(x)
[16:27:27] <halfak>	 geo.mean = exp(log(x)/length(x))
[16:29:15] <subbu>	 ok .. will play aroudn with that after i eat breakfast which i forgot to eat since i spent the last 2 hours messing with charts and numbers ..
[16:29:16] <subbu>	 thanks!
[16:29:19] <subbu>	 https://en.wikipedia.org/wiki/Log-normal_distribution#Occurrence 
[16:30:02] <halfak>	 no prob.  :)  fun stuff. 
[17:52:52] <Emufarmers>	 bearlognormal
[20:03:13] * halfak upgrades bearloga 
[20:03:40] <bearloga>	 halfak???
[20:03:45] <halfak>	 :D
[20:17:08] <bearloga>	 halfak: I am so confused
[20:17:25] <halfak>	 Oh!  I just gave you voice in the channel. 
[20:18:10] <halfak>	 Everyone who is a regular that I would expect to be able to either answer a question or find out who can gets voice. 
[20:19:09] <bearloga>	 Oh. RAD
[20:19:12] <bearloga>	 Thank you :)
[20:20:08] * halfak looks at EGalvez_ 
[20:21:14] <EGalvez_>	 @halfak, what was the convo? i just logged on a few minutes ago...
[20:21:31] <halfak>	 halfak>30 Everyone who is a regular that I would expect to be able to either answer a question or find out who can gets voice. 
[20:21:41] <halfak>	 EGalvez_, want some voice?
[20:22:34] <EGalvez_>	 @halfak; confused.. what do you mean voice? as in.. do I have a question for the channel?
[20:23:05] <halfak>	 Na.  Voice is a mode in IRC.  Move clients will put you at the top of the list of people in the channel and put some cool symbol by your name. 
[20:23:19] <halfak>	 It's basically a symbol for "I'm a regular.  Ask me what's up."
[20:23:27] <EGalvez_>	 ah cool, sure!
[20:23:39] <halfak>	 Could you set up one of these? https://meta.wikimedia.org/wiki/IRC/Cloaks
[20:25:34] <Krenair>	 The actual ability it grants is the right to talk while the channel is moderated (+m)
[20:25:44] <Krenair>	 along with the ops
[20:25:54] <halfak>	 Yeah.  Which we never do. 
[20:26:01] <halfak>	 So it's just a badge
[20:26:27] <Krenair>	 it's the bottom in a chain of different IRC channel permissions: q, a, o, h and v
[20:26:41] <Krenair>	 I don't know if FreeNode really does q/a/h
[20:28:40] <Krenair>	 I don't actually remember what those do
[20:33:09] <yuvipanda>	 halfak: I've some time today for Quarry
[20:33:21] <yuvipanda>	 I'll be doing the importing stuff and import some datasets and setup the process
[20:33:32] <yuvipanda>	 I guess you won't be happy if I picked to use go for this. I've been trying to learn it...
[20:33:58] <halfak>	 If you're going to be maintaining it.  Then I'll be happy. :P