Open in new window / Try shogun cloud
--- Log opened Wed Apr 30 00:00:28 2014
-!- x4i [~A2o@unaffiliated/x4i] has joined #shogun01:16
-!- x4i [~A2o@unaffiliated/x4i] has quit [Quit: irc]01:23
-!- HeikoS [] has joined #shogun02:32
-!- mode/#shogun [+o HeikoS] by ChanServ02:33
-!- khalednasr [~k.nasr92@] has quit [Quit: Leaving]02:45
shogun-buildbotbuild #696 of nightly_none is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>, lambday <>03:08
shogun-buildbotbuild #689 of nightly_all is complete: Failure [failed compile test]  Build details are at  blamelist: Heiko Strathmann <>, lambday <>03:14
-!- abinash_panda [0e8b2998@gateway/web/freenode/ip.] has joined #shogun04:21
-!- abinash_panda [0e8b2998@gateway/web/freenode/ip.] has quit [Ping timeout: 240 seconds]04:28
-!- HeikoS [] has quit [Ping timeout: 240 seconds]04:46
-!- soumyaC [uid15286@gateway/web/] has quit [Quit: Connection closed for inactivity]05:06
-!- abinash_panda [d2d43d97@gateway/web/cgi-irc/] has joined #shogun06:20
-!- iglesiasg [] has joined #shogun06:37
-!- mode/#shogun [+o iglesiasg] by ChanServ06:37
-!- shogun-buildbot [] has quit [Ping timeout: 252 seconds]06:37
-!- shogun-buildbot_ [] has joined #shogun06:37
-!- iglesiasg [] has quit [Quit: Leaving]07:09
-!- abinash_panda [d2d43d97@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]09:38
-!- lisitsyn [~qdrgsm@] has left #shogun []10:53
-!- thoralf [] has joined #shogun10:58
thoralfFunny facts about code on github:
-!- lambday [67157f4f@gateway/web/freenode/ip.] has joined #shogun11:39
-!- lisitsyn [] has joined #shogun11:48
-!- soumyaC [uid15286@gateway/web/] has joined #shogun12:13
-!- lambday [67157f4f@gateway/web/freenode/ip.] has quit []13:08
-!- HeikoS [] has joined #shogun13:30
-!- mode/#shogun [+o HeikoS] by ChanServ13:30
-!- khalednasr [~k.nasr92@] has joined #shogun14:04
-!- lambday [67157f4f@gateway/web/freenode/ip.] has joined #shogun14:16
lambdayHeikoS: there??14:42
@HeikoSlambday:  hi14:46
lambdayHeikoS: hey!14:46
lambdayHeikoS: just upgraded my RAM and did the benchmark with vectorized Eigen3 sum calls14:46
lambdayHeikoS: sent a PR - results are remarkable14:46
@HeikoSlambday: cool!14:47
lambdayHeikoS: according to this - we should definitely override these sum methods in CCustomKernel14:47
@HeikoSlambday: I see14:47
@HeikoSlambday: how does it do the magic?14:47
lambdayHeikoS: I'm not sure! Eigen3 says its operations are vectorized14:48
lambdayso that could be it14:48
lambdayHeikoS: compiled the benchmark with " g++ -std=c++0x -O3 -fopenmp benchmark.cpp -I/usr/include/eigen3 -lshogun"14:49
-!- shogun-notifier- [] has joined #shogun14:49
shogun-notifier-shogun: lambday :develop * dd56776 / benchmarks/kernel_matrix_sum_benchmark.cpp:
shogun-notifier-shogun: added kernel matrix sum benchmark - CKernel::kernel calls vs Eigen3::sum14:49
shogun-notifier-shogun: Heiko Strathmann :develop * 6c71771 / benchmarks/kernel_matrix_sum_benchmark.cpp:
shogun-notifier-shogun: Merge pull request #2196 from lambday/feature/quadratic_time_mmd14:49
shogun-notifier-shogun: added kernel matrix sum benchmark - CKernel::kernel calls vs Eigen3::sum14:49
@HeikoSlambday: I see, so thats vector operations on hardware level14:49
lambdayHeikoS: yep!14:50
@HeikoSlambday: super cool14:50
@HeikoSlambday: ah man, we need linear algebra framework to be working14:50
@HeikoSlambday: can you push that a bit?14:50
lambdayHeikoS: sunil is having exams14:50
@HeikoSat least for basic things such as summing up matrix values, factorisations etc14:50
@HeikoSlambday: yeah14:50
@HeikoSah I mean ok, it will come :)14:50
lambdayHeikoS: khaled wanted to work on that CPU/GPU stuff14:50
@HeikoSlambday: in the mean time, feel free to override the CCusotmKernel things14:50
@HeikoSlambday: but could you already put summing over matrices in the linear algebra backend?14:51
@HeikoSso that the stuff you are adding there is available for others14:51
lambdayHeikoS: he wanted to replace current SGVector with GPUVector instead - although I don't think that its a good diea14:51
@HeikoSlambday: should be straight forward not?14:51
lambdayHeikoS: absolutely14:51
@HeikoSlambday: no, no replacement14:51
lambdayHeikoS: umm do we need summing in the linear algebra backend?14:52
@HeikoSlambday: why not?14:52
@HeikoSlambday: why should it rely on eigen?14:52
lambdayumm yeah it would be cool!14:52
lambdayyeah you're right14:52
@HeikoSlambday: if we change the linalg lib then we have to change everything again14:52
@HeikoSthat should not be the case14:52
lambdayseeing how it performs with GPU would be actually interesting14:52
@HeikoSwould in fact like to get rid of most eigen3 calls in algorithms and rather call shogun classes that do the eigen3 stuff inside (or any other lib)14:53
@HeikoSlambday: I gotta go now, out for a hiking trip14:53
@HeikoSso see you later today14:53
lambdayHeikoS: yo man! have fun! :D14:53
lambdayHeikoS: see you!14:53
lambdayI'll send a PR with overridden CCustomKernel calls soon and then start the backend stuff14:54
-!- HeikoS [] has quit [Quit: Leaving.]14:56
khalednasrlambday, I didn't  want to replace SGVector, I meant handling both GPU and CPU memory inside SGVector14:57
lambdaykhalednasr: ah sorry man I didn't get you completely14:57
lambdaywanted to talk to you regarding this14:57
lambdaykhalednasr: what's wrong with having 2 different classes for 2 different implementations?14:58
khalednasrIf you have an algorithm that can run on both cpu or gpu, you'd have to have instances of both SGVector and GPUVector for every vector you have14:59
lambdaykhalednasr: we can always manage that using templates, right?15:00
khalednasrpossible, but it might be complicated if you wanted more control on what data is stored on the gpu and what data is stored on the gpu15:03
lambdaykhalednasr: I am not getting the problem properly - may be we can discuss this in more details when heiko is here?15:04
shogun-buildbot_build #624 of FCRH - libshogun is complete: Success [build successful]  Build details are at
khalednasrlambday, yeah sure15:05
lambdaykhalednasr: but if you think that its problematic for some specific instance, could you please post an example on that thread? it would be really helpful15:05
khalednasrlambday, ok, I'll see what I can do15:06
lambdaykhalednasr: thanks man!15:06
* lambday afk15:06
-!- travis-ci [] has joined #shogun15:49
travis-ci[travis-ci] it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []15:49
-!- thoralf [] has quit [Ping timeout: 265 seconds]16:31
-!- thoralf [] has joined #shogun16:32
shogun-buildbot_build #282 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at  blamelist: Heiko Strathmann <>, lambday <>17:13
-!- lisitsyn [] has quit [Quit: Leaving.]17:22
-!- shogun-notifier- [] has quit [Quit: transmission timeout]17:49
-!- abinash_panda [0126158c@gateway/web/cgi-irc/] has joined #shogun18:09
abinash_pandathoralf: Hi !18:10
thoralfHey abinash_panda18:12
abinash_pandathoralf:So finally my exams are over :)18:12
abinash_pandathoralf:I can now devote my full time for shogun18:12
thoralfabinash_panda: That's great :)18:13
thoralfabinash_panda: I guess you already have ideas on this, right?18:13
abinash_pandathoralf: You mean ideas regarding my project, right?18:14
abinash_pandathoralf: Yeah, I have been going through various papers and tutorials18:14
abinash_pandathoralf: There is an implementation of multi-label learning in scikit-learn. I was playing with it to get an idea18:15
-!- abinash_panda [0126158c@gateway/web/cgi-irc/] has quit [Quit: - A hand crafted IRC client]18:53
-!- abinash_panda [0126158c@gateway/web/cgi-irc/] has joined #shogun18:56
-!- jd_tx [80538d58@gateway/web/freenode/ip.] has joined #shogun19:36
jd_txhi, I'm using LibSVM through the python swig interface. I've trained a classifier, and I'd like to now get margins(decision values) for my task. apply() only returns the hard +1/-1 outputs. How can I access the decision values?19:38
-!- iglesiasg [] has joined #shogun19:41
-!- mode/#shogun [+o iglesiasg] by ChanServ19:41
-!- lambday [67157f4f@gateway/web/freenode/ip.] has quit [Ping timeout: 240 seconds]19:50
-!- thor|alf [] has joined #shogun20:08
thor|alfabinash_panda: Sorry, had to leave...20:09
thor|alfabinash_panda: I played with scikit-learn's multilabel stuff as well, but back then it implemented intractable methods like label-combination.20:10
thor|alfabinash_panda: I checked and it changed a bit...20:14
-!- thor|alf [] has quit [Ping timeout: 265 seconds]20:45
-!- HeikoS [] has joined #shogun21:29
-!- mode/#shogun [+o HeikoS] by ChanServ21:29
-!- soumyaC [uid15286@gateway/web/] has quit [Quit: Connection closed for inactivity]21:46
-!- HeikoS [] has quit [Ping timeout: 240 seconds]21:51
-!- iglesiasg [] has quit [Quit: Leaving]21:51
-!- khalednasr [~k.nasr92@] has quit [Quit: Leaving]22:11
-!- lambday [67157f4f@gateway/web/freenode/ip.] has joined #shogun22:47
-!- jd_tx [80538d58@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:57
--- Log closed Thu May 01 00:00:29 2014