IRC logs of #shogun for Wednesday, 2012-09-26

« Tuesday, 2012-09-25 Index Thursday, 2012-09-27 »
--- Log opened Wed Sep 26 00:00:17 2012
-!- hoijui [~hoijui@dslb-088-074-107-101.pools.arcor-ip.net] has quit [Ping timeout: 248 seconds]00:36
-!- heiko1 [~heiko@host86-181-80-213.range86-181.btcentralplus.com] has quit [Quit: Leaving.]01:16
shogun-buildbotbuild #114 of nightly_default is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/11403:47
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has left #shogun ["Leaving"]03:59
-!- hoijui [~hoijui@dslb-088-074-107-101.pools.arcor-ip.net] has joined #shogun07:28
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun09:11
wikingblackburn: yo09:22
blackburnwiking: hey09:22
blackburnwiking: ha! I bet these slep leaks blocks you from sleeping :)09:47
wikingblackburn: hhehehe no not really just dont want to start with other latent stuff09:48
wiking:)09:48
blackburnthanks, I merged that09:49
wikingso i've started up fixing the static analyzer stuff09:49
shogun-buildbotbuild #545 of deb3 - modular_interfaces is complete: Failure [failed test lua_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/545  blamelist: Viktor Gal <viktor.gal@maeth.com>10:16
wikingheheh wazntme :)10:21
blackburnwiking: yes, I didn't manage to fix up lua yet10:23
shogun-buildbotbuild #546 of deb3 - modular_interfaces is complete: Failure [failed test lua_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/546  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>10:37
sonne|workblackburn: the build braker is in da hous10:38
sonne|worke10:38
blackburnsonne|work: lua is undebuggable10:38
sonne|workI see10:39
sonne|workThis thing sporadically pops up on the bb10:39
sonne|workbut never locally...10:39
blackburnsonne|work: no, it actually hangs on my machine10:41
sonne|workbut then valrind / gdb and voila!10:43
blackburnsonne|work: I've got recursive calls of typemaps10:44
blackburnneither gdb nor valgrind pukes :D10:44
blackburnit is just eats all my 8gb of memory and dies10:45
blackburnit just*10:45
sonne|workwell ok then check the typemaps :)10:46
blackburnyeah I did (twice!!!!!) :)10:49
blackburnwell, I'll get back to that soon10:49
-!- heiko [~heiko@host86-182-167-91.range86-182.btcentralplus.com] has joined #shogun11:18
blackburnhttp://en.wikipedia.org/wiki/Troxler%27s_fading11:59
heikoblackburn, wiking, I solved all problems in another way :)13:15
heikoblackburn, no streaming kernel needed, I just extended streaming features and use normal kernel13:16
heikowiking, I have removed this method I mentioned. Its possible to let the caller decide at another point in the code13:16
shogun-buildbotbuild #438 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/438  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>13:25
heikodont worry buildbot, the last build of the 50 works :)13:26
blackburnheiko: it doesn't, lua is still broken :)13:31
heikoblackburn, lol :)13:31
-!- hoijui [~hoijui@dslb-088-074-107-101.pools.arcor-ip.net] has quit [Ping timeout: 240 seconds]13:35
wikingheiko: :)13:36
heikothis streaming two-sample test is really cool now: 500 millions samples and more13:36
heikogotta go, see you later guys13:36
blackburn500 millions??13:36
blackburnhow?13:36
heikoblackburn, yeah :)13:36
blackburnwhy? :D13:36
heikowell streaming13:37
heikosome problems are so hard that they need this many samples13:37
blackburnhow does that work?13:37
blackburnI thought you create NxN matrix13:37
heikostreaming data, so does not have to be stored in mem13:37
heikono13:37
blackburn2 examples by step?13:37
heikojust need trace of kernel matrix13:37
blackburn2x2?13:37
blackburnah13:37
heikokind of yeah13:37
blackburninteresting13:37
blackburnremarkable thing to mention13:38
blackburnlike13:38
blackburnhey guys we are working with data with 500 millions examples13:38
blackburnwhat's up with your dlib/scikits/weka?13:38
blackburn:D13:38
heikoindeed, thats what I did in my thesis ;)13:38
heikomore interesting is: there are problems that *need* that much data13:38
heikootherwise you cannot solve them since they are too hard13:38
blackburnwhat is the problem?13:39
wikinglol13:39
heikohttp://herrstrathmann.de/images/blobs.png13:39
blackburnthat's like such a big data one can estimate probability distribution pretty precisely13:39
heikoto distinguish between the red and the blue13:39
blackburnis that possible?13:39
heiko(with other parameters than in the image, but thats the principle)13:39
wikingheiko: and what machine do u use with the kernel?13:39
heikoCLinearTimeMMD, its not a machine13:40
heikostatistical test13:40
wikingah yeah13:40
heikobut I mean two-sample testing and binary classification is similar13:40
heikonature13:40
blackburnso H0 is like blue13:40
blackburnand H1 is red13:40
blackburn?13:40
wikingheiko: so u reckon it should work with a binary machine as well?13:40
wiking*binary classifier...13:40
heikoblackburn, yes13:41
heikono13:41
heikoehm, no13:41
heikoH0: p=q13:41
heikoH1: p!=q13:41
heikowhere p is read and q is blue13:41
heikowiking, well I dont know a machine that can handle that much data13:41
heikowe have online lib-linear right?13:42
wikingheiko: apart from online13:42
wikingor like pegasus-svm13:42
heikoah ok13:42
wikingso yeah it would only work with that ...13:42
heikoyear, easily classifiable data is also easy to distinguish in a two-sample test13:42
heikothe dataset is quite interesting13:42
heikosince the differences are hidden inside, at a different scalling that the overall data has13:43
blackburnheiko: so we only say it is the same distribution or it is not13:43
heikoblackburn, yes, but one tries to find a difference, that is "pull the distributions apart", so quite similar to classification13:43
heikowhen you can pull them apart, you can put a hyperplane in between13:43
blackburnI see13:43
heikoI gotta go now, see you later :)13:44
wikinglaterz13:44
shogun-buildbotbuild #547 of deb3 - modular_interfaces is complete: Failure [failed test lua_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/547  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>13:44
-!- heiko [~heiko@host86-182-167-91.range86-182.btcentralplus.com] has quit [Quit: Leaving.]13:44
shogun-buildbotbuild #439 of deb2 - static_interfaces is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/43913:51
shogun-buildbotbuild #548 of deb3 - modular_interfaces is complete: Failure [failed test lua_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/548  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>13:59
-!- heiko [~heiko@host86-182-167-91.range86-182.btcentralplus.com] has joined #shogun14:27
-!- heiko [~heiko@host86-182-167-91.range86-182.btcentralplus.com] has quit [Quit: Leaving.]16:38
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed]17:16
-!- hoijui [~hoijui@dslb-088-074-107-101.pools.arcor-ip.net] has joined #shogun17:46
-!- in3xes [~in3xes@122.174.80.141] has joined #shogun18:00
* wiking bangs the head into the wall... found a bug! :D18:35
-!- blackburn [~blackburn@188.168.14.95] has joined #shogun18:57
-!- romi__ [~mizobe@187.57.9.8] has quit [Ping timeout: 246 seconds]19:36
-!- romi__ [~mizobe@187.101.115.113] has joined #shogun19:50
shogun-buildbotbuild #440 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/440  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>19:57
shogun-buildbotbuild #434 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/434  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>19:58
-!- in3xes_ [~in3xes@122.174.90.177] has joined #shogun19:58
-!- in3xes [~in3xes@122.174.80.141] has quit [Ping timeout: 265 seconds]20:01
wikingyes !!! SO-BMRM: 100.0020:12
wiking\o/20:12
wikingso it actually worx !!!20:14
shogun-buildbotbuild #31 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/31  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>20:16
shogun-buildbotbuild #549 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/549  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>20:26
blackburnwiking: great20:33
wikingblackburn: yep in the meanwhile i even invented a new way of training the latent so-svm ;)20:42
wikingto have a better convergence20:42
blackburnwithout a train set?20:42
wikingi have a train set20:42
wiking:P20:42
blackburnbad20:42
blackburnI'd like a method without train set20:42
blackburn:)20:42
wikingnocando20:43
wiking:)20:43
blackburnwiking: wanna try to beat guys on digit recognition? http://www.kaggle.com/c/digit-recognizer20:44
blackburnkind of test20:45
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun21:04
wikingblackburn: kagggle and no prize? :)21:12
blackburnwiking: I do not expect any prize to receive without experience..21:12
blackburnbtw21:13
blackburncheck that21:14
blackburnhttps://gist.github.com/378995021:14
wikingwhat's the time? :)21:14
blackburnlooks like useful thing for ml, I am now thinking about a way to register local context variables and save them21:14
blackburntime.time?21:15
blackburnwell, time21:15
blackburn:)21:15
blackburnwith thing looks like pretty compact way to designate scopes where you need to save something or measure21:15
wikinghttp://worrydream.com/LearnableProgramming/21:54
wikinghehe21:54
blackburntl;dr21:57
blackburn??21:57
blackburnoops :D21:57
wiking:>21:58
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: Lost terminal]22:40
-!- blackburn [~blackburn@188.168.14.95] has quit [Quit: Leaving.]23:44
-!- hoijui [~hoijui@dslb-088-074-107-101.pools.arcor-ip.net] has quit [Ping timeout: 248 seconds]23:48
--- Log closed Thu Sep 27 00:00:17 2012
« Tuesday, 2012-09-25 Index Thursday, 2012-09-27 »

Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!