Open in new window / Try shogun cloud
--- Log opened Tue Jul 24 00:00:17 2012
gsomixgood night guys00:07
gsomixshogun-buildbot, good work!00:07
shogun-buildbotWhat you say!00:07
-!- gsomix [~gsomix@] has quit [Ping timeout: 260 seconds]00:12
shogun-buildbotbuild #203 of deb3 - modular_interfaces is complete: Failure [failed compile ruby_modular]  Build details are at  blamelist: Evgeniy Andreev <>, Soeren Sonnenburg <>00:20
CIA-18shogun: Sergey Lisitsyn master * ree6b9c2 / src/shogun/lib/malsar/malsar_clustered.cpp : A few fixes for malsar clustered multitask solver -
-!- heiko1 [] has joined #shogun00:29
shogun-buildbotbuild #204 of deb3 - modular_interfaces is complete: Failure [failed compile ruby_modular]  Build details are at  blamelist: Sergey Lisitsyn <>00:31
-!- needsch [] has quit [Ping timeout: 255 seconds]00:47
emreceliktenGood night all00:57
-!- heiko1 [] has quit [Quit: Leaving.]01:00
-!- emrecelikten [~emre@] has quit [Ping timeout: 240 seconds]01:02
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]01:27
wikingsonney2k: eeeey!!!01:38
wikingsonney2k: just realised why submodule will be really bad for unit testing01:38
wikingsonney2k: it's gonna be way too complicated to keep the unit testing in synch with shogun itself01:39
wikingsonney2k: say you want change some classes in shogun, you push it01:39
wikingthen you have to push separately in the submodule01:39
wikingsince if you not, and most of the developers won't as they will not be able to put in 1 PR the new code + the unit tests for it01:40
wikingthe unit testing will keep on failing...01:40
wikingthis my 2 cents ;P01:45
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]02:32
shogun-buildbotbuild #32 of nightly_none is complete: Failure [failed compile]  Build details are at
shogun-buildbotbuild #38 of nightly_default is complete: Failure [failed compile]  Build details are at
shogun-buildbotbuild #28 of nightly_all is complete: Failure [failed compile]  Build details are at
-!- romi_ [] has joined #shogun03:45
-!- romi_ [] has quit [Ping timeout: 264 seconds]04:07
-!- romi_ [] has joined #shogun04:07
-!- Cheng [80fa500f@gateway/web/freenode/ip.] has joined #shogun05:36
-!- Cheng [80fa500f@gateway/web/freenode/ip.] has quit [Client Quit]05:40
-!- n4nd0 [] has joined #shogun06:20
n4nd0wiking: around?06:21
-!- gsomix [~gsomix@] has joined #shogun07:26
-!- uricamic [~uricamic@2001:718:2:1634:4c95:e117:3805:c6f] has joined #shogun07:30
-!- gsomix [~gsomix@] has quit [Client Quit]07:31
-!- gsomix [~gsomix@] has joined #shogun07:31
-!- romi_ [] has quit [Quit: Leaving]07:50
-!- pluskid [~pluskid@] has joined #shogun08:24
-!- n4nd0 [] has quit [Ping timeout: 245 seconds]08:39
-!- gsomix [~gsomix@] has quit [Remote host closed the connection]09:12
-!- pluskid [~pluskid@] has quit [Ping timeout: 248 seconds]09:23
-!- pluskid [] has joined #shogun09:24
-!- pluskid [] has quit [Ping timeout: 264 seconds]10:12
-!- pluskid [~pluskid@] has joined #shogun10:32
-!- heiko [] has joined #shogun11:29
-!- needsch [] has joined #shogun11:42
-!- blackburn [~blackburn@] has joined #shogun11:45
-!- pluskid [~pluskid@] has quit [Ping timeout: 252 seconds]11:49
-!- pluskid [~pluskid@] has joined #shogun11:49
-!- zxtx [] has joined #shogun12:12
-!- needsch175 [] has joined #shogun12:25
-!- needsch [] has quit [Ping timeout: 255 seconds]12:26
-!- heiko [] has quit [Quit: Leaving.]12:28
wikingn4nd0 yes12:42
wikingblackburn: ping?12:55
wikingblackburn: would u consider using this lib:
blackburnwhat for?12:56
wikingblackburn: well behind SGVector12:57
wikingand SGMatrix12:57
blackburnwell we have eigen3 for that I think12:57
wikingblackburn: mmm that's not replacement :)12:57
blackburnwiking: I don't understand what is it then :)12:58
wikingblackburn: fuckshit i'm wrong12:58
blackburnso if we optimize inner loops with SIMD12:58
wikingi mean half12:59
blackburneigen3 does the same via any available SSE12:59
wikingso orc is runtime optimizer12:59
blackburnrun-time? oh12:59
wikingblackburn: does eigen do it runtime? afaik it's compile time no?12:59
blackburnfor variable size arrays it does splitting to 128 bits long parts13:00
wikingmmm that makes me wonder13:07
wikingwhere's the -DHAVE_EIGEN3 flag13:07
wikingwhen i compile :S13:07
blackburnwiking: was it detected on configure?13:08
-!- uricamic1 [~uricamic@2001:718:2:1634:b07b:e85b:8940:fa6f] has joined #shogun13:08
wikingblackburn: fuck it's not detected13:08
wikingblackburn: i'll fix the configure script :P13:08
blackburnwiking: config script is ok I think13:08
wikingblackburn: one flag is missing :S13:09
blackburnwhich one?13:09
wikinginclude flag ismissing there13:09
wikingblackburn: btw why dont u detect it via13:09
wikingpkg-config ?13:09
-!- uricamic [~uricamic@2001:718:2:1634:4c95:e117:3805:c6f] has quit [Ping timeout: 245 seconds]13:10
wikingblackburn: as eigen3 supplies pkg-config files13:10
blackburnwell it is something common to do that this way in our configure13:10
blackburnI don't know13:10
wikingblackburn: configure script checks for pkg-config13:10
wiking 522     die "pkg-config not detected (required)"13:11
wikingblackburn: mmm13:17
wikingblackburn: where do u include eigen3 apart from Math.h?13:19
blackburnwiking: only in math13:19
blackburnit should be included only once to support preprocessor directives13:19
wikingblackburn: eigen headers should be included only by Eigen/...13:20
wikingblackburn: i've just checked the docs13:20
wikingof eigne13:20
blackburnyeah I think so too13:20
wikingit works now13:20
blackburnwell probably there is a need to add proper include13:21
-!- pluskid [~pluskid@] has quit [Ping timeout: 264 seconds]13:23
CIA-18shogun: Viktor Gal master * ref423ad / (src/configure src/shogun/mathematics/Math.h): Use the supplied pkg-config script to detect eigen3 -
CIA-18shogun: Sergey Lisitsyn master * re37e6fe / (src/configure src/shogun/mathematics/Math.h): Merge pull request #670 from vigsterkr/master -
blackburnI like it more13:27
blackburnthan the other way13:27
wikingthis way it's detected on my system as well13:27
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]13:29
-!- heiko [] has joined #shogun13:35
shogun-buildbotbuild #205 of deb3 - modular_interfaces is complete: Failure [failed compile ruby_modular]  Build details are at  blamelist: Viktor Gal <>13:37
-!- heiko [] has left #shogun []13:37
-!- pluskid [~pluskid@] has joined #shogun13:43
shogun-buildbotbuild #206 of deb3 - modular_interfaces is complete: Failure [failed compile ruby_modular]  Build details are at  blamelist: Sergey Lisitsyn <>13:48
-!- gsomix [~gsomix@] has joined #shogun14:36
gsomixgood day14:46
-!- n4nd0 [] has joined #shogun15:27
n4nd0wiking: I wanted to comment something about the PR for the testing unit15:28
n4nd0wiking: but let me first to read the last comments in the PR15:28
wikingn4nd0: just a sec i'm on skype15:29
wikingi'll be back in 1015:29
n4nd0ok, heiko and you have already been talking about my concern15:31
n4nd0I think the best option is to use another branch, as you suggested in the PR15:32
wikingn4nd0: ok sorry i was awway15:43
wikingbut now back15:44
wikingyeah just put there your opinion as well about submodule vs having it in the main repo15:44
wikingn4nd0: i have a question related to PrimalMosekSVM15:45
wikingi'll dig it out now just a sec15:45
n4nd0wiking: ok15:45
n4nd0wiking: I am updating that class right now by the way15:45
wikingn4nd0: ok let me know how15:45
wikingmy concern is about argmax in StruturalModel15:46
n4nd0there was something that is required for the HMSVM but was not regarded at the beginning15:46
n4nd0ok, tell me about it first15:46
wikingis that the function for you that finds \hat{y} for you?15:46
wikingif i'm not clear what i'm talking about here. just a sec to dig out the equation15:47
n4nd0yes, it is15:47
n4nd0you can think of it too as the function you use to make predictions (the common apply we have in other machines)15:48
n4nd0CLinearStructuredOutputMachine::apply_structured is simply calling this argmax15:49
wikingn4nd0: but when you do the learning15:50
wikingi mean training15:50
wikingit's called as well afaik15:50
n4nd0since it is a cutting plane algorithm15:51
wikingthen we'll only need to do a new EnumType for strutural learners :P15:51
n4nd0yeah why not15:52
wikingi'll need it in LatentSOSVM15:53
wikingso that user can choose between solvers15:53
n4nd0the LatentSOSVM inherits from?15:54
wikingi thought first about relying on user to supply the so solver15:54
wikingbut then i thought that it's going to be too complicated.. and just go with simply setting the solver type15:54
wikingand i'll do internally everything for the 2 different solvers15:54
wikingit's inherited from LinearLatentMachine15:54
wikingso has nothing to do in a way with StructuredOutputMachine15:55
wikingit'll just contain a CLinearStructuredOutputMachine15:55
wikingbut that's going to created runtime15:55
n4nd0aham, no instantiation of it in the constructor you mean?15:56
wikingit'll be created in the train_machine function15:56
-!- pluskid [~pluskid@] has quit [Quit: Leaving]16:39
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has joined #shogun16:45
yoohi all16:45
n4nd0hi yo16:50
n4nd0yoo, sorry :)16:51
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun17:22
-!- uricamic1 [~uricamic@2001:718:2:1634:b07b:e85b:8940:fa6f] has quit [Quit: Leaving.]17:32
-!- heiko [] has joined #shogun17:40
-!- n4nd0 [] has quit [Read error: Operation timed out]17:46
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:55
-!- blackburn [~blackburn@] has joined #shogun18:05
blackburnwiking: I am now having a lot of shadow warnings18:37
CIA-18shogun: Evgeniy Andreev master * r8879634 / (3 files in 3 dirs): fixed compilation error -
CIA-18shogun: Sergey Lisitsyn master * rdf82997 / (3 files in 3 dirs): Merge pull request #671 from gsomix/buffer_protocol -
CIA-18shogun: Heiko Strathmann master * rc908469 / (src/shogun/lib/SGMatrix.cpp src/shogun/lib/SGMatrix.h): added new helper method to get an allocated matrix -
CIA-18shogun: Heiko Strathmann master * ra3cbd47 / (4 files in 2 dirs): make data generator use new matrix allocation method -
CIA-18shogun: Heiko Strathmann master * r497cd01 / (6 files in 3 dirs): Merge pull request #672 from karlnapf/master -
-!- Netsplit *.net <-> *.split quits: shogun-buildbot19:00
-!- Netsplit over, joins: shogun-buildbot19:02
shogun-buildbotbuild #207 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Evgeniy Andreev <>19:16
CIA-18shogun: Evgeniy Andreev master * rd0a0875 / examples/undocumented/python_modular/ : fixed example for old version numpy -
CIA-18shogun: Sergey Lisitsyn master * r9458ec7 / examples/undocumented/python_modular/ : Merge pull request #673 from gsomix/buffer_protocol -
-!- alexlovesdata [] has joined #shogun19:33
-!- heiko [] has left #shogun []19:38
shogun-buildbotbuild #208 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>19:45
-!- romi_ [] has joined #shogun19:48
shogun-buildbotbuild #209 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Heiko Strathmann <>20:21
wikingblackburn: what should we do with the header?20:41
blackburnwiking: which header?20:41
blackburnargh I think I need eigen3 include to some other file20:41
wikingblackburn: eigen320:42
wikingas sonney2k told that it's not good that it's included in Math.h20:42
blackburnit is being included in every file20:43
blackburnwiking: but where did he say it?20:44
blackburnooh yeah and actually he is right20:51
shogun-buildbotbuild #210 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at
blackburnwiking: I am now confused where to put it in20:53
wikingblackburn: :)20:54
wikingblackburn: solve it :)20:54
blackburnwiking: well straightforward solution is to put it into math/eigen.h20:55
blackburnbut it may be a little silly20:55
-!- alexlovesdata [] has left #shogun []20:59
CIA-18shogun: Sergey Lisitsyn master * r50f8b64 / src/shogun/converter/LocallyLinearEmbedding.cpp : Made LLE thread-safe -
-!- n4nd0 [] has joined #shogun21:44
-!- test [cf60c47a@gateway/web/freenode/ip.] has joined #shogun22:09
-!- gsomix [~gsomix@] has quit [Ping timeout: 240 seconds]22:09
-!- test [cf60c47a@gateway/web/freenode/ip.] has left #shogun []22:10
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]22:19
-!- gsomix [~gsomix@] has joined #shogun22:21
-!- alexlovesdata [] has joined #shogun22:22
n4nd0good night guys22:47
n4nd0tomorrow I'll be testing the hm-svm22:47
n4nd0and I need a new day to face all the errors that are to come :D22:47
CIA-18shogun: Sergey Lisitsyn master * r629ccb8 / (6 files in 3 dirs): Get eigen3 out of math.h -
blackburnn4nd0: night!22:48
blackburnn4nd0: didn't you test it before?22:49
blackburnit could become a nightmare I have been in a weeks already - comparing outputs step by step :)22:49
n4nd0blackburn: I did some tests yet ... but the specification was wrong22:49
n4nd0there were some errors + things missing in the theory22:50
blackburnI see22:51
-!- gsomix [~gsomix@] has quit [Ping timeout: 276 seconds]22:52
n4nd0I had to introduce changes on top of the SO framework22:53
n4nd0not just in the hm-svm part22:53
n4nd0I am going to call it the so patch 1.0 :)22:53
n4nd0well not really on top to tell the truth, in the so-svm22:54
n4nd0anyhow, good night!22:57
-!- n4nd0 [] has quit [Quit: leaving]22:57
wikingblackburn: yooouuuu motherfucker :)23:31
wiking :>23:31
wikingyou went with eigen3.h23:31
blackburnwiking: why not?23:31
wikingwell it's putting the shit on another shelf23:31
wikingbut yeah i couldn't solve it better23:31
wikingi'm just pulling your leg23:31
wikingthere you go23:32
blackburnwell  actually I don't think it is so brute23:32
* wiking runs away in shame :S23:34
wikingi mean it's just that it was said that 'no lapack and such in .h'23:34
blackburnwhat's the other way?23:35
wikingstraight into .cpp23:35
blackburnto each cpp?23:36
wikingwell yep23:37
wikingif we strictly speaking about this 'no lapack and such ifdefs'23:37
wikingthen yeah23:37
wikingeach time u use it somewhere23:37
wikinginclude it23:37
blackburnproblem is that in case of some cross-dependency or so23:38
blackburnand custom options23:38
blackburnit won't work23:38
wikingi see that of course23:38
blackburnI am now trying to test blas vs loops vs eigen323:38
blackburnon dot23:39
wikingfor the order of the ifdefs ? ;)23:39
wikingbut yeah23:39
wikinggreat idea23:39
blackburnactually I do a standalone example23:39
blackburnwiking: I am confused how to link with blas :D23:39
wikingjsut take into the consideration23:39
wikingsince it's really fucking killer23:39
wikingwhen i did some benchmarking about vectors ;)23:40
wikingthe cache in the cpu is really fucking good actually ;P23:40
blackburnyeah I noticed that when element-by-element kernel matrix computation was 2 times faster than jumps from one side of diagonal to the other one23:41
-!- yoo [575b08cb@gateway/web/freenode/ip.] has joined #shogun23:54
yoogood evening all23:54
yoocould someone give me plz a small example for iterate over CFeatures or print CFeatures.23:55
yooI have seen get_next_feature and get_feature_iterator but ..23:56
yooI try something like :23:57
yoovoid *iterator; float64_t val; int32_t j;             wdfeats->get_feature_iterator(j);             while (wdfeats->get_next_feature(j, val, iterator)){                std::cout << val << std::endl;             }             wdfeats->free_feature_iterator(iterator);23:58
yooouch sorry for that23:58
yoomore like that :
blackburnyoo: well iterators are pretty crappy way here23:59
blackburnthey are here just for some compatibility23:59
blackburnwhat exactly do you need?23:59
yooyes thats why I ask :p23:59
yoojust iterate and print features23:59
--- Log closed Wed Jul 25 00:00:05 2012