IRC logs of #shogun for Tuesday, 2012-10-23

« Monday, 2012-10-22 Index Wednesday, 2012-10-24 »
--- Log opened Tue Oct 23 00:00:17 2012
@sonney2kblackburn, I just recruited 2 mentors for next year :D00:47
blackburnsonney2k: ha! who?00:47
@sonney2ksomeone from google and someone who was in tuebingen but is now in CA00:47
@sonney2kboth excellent researchers00:48
blackburnsecret names? :)00:48
@sonney2kquoc and olivier :D00:48
@sonney2kthey wanted to recruit me but I won :P00:48
blackburnrecruit to?00:48
@sonney2kg00:49
@sonney2kblackburn, btw I think we should have some acknowlegdements on the shogun homepage00:49
@sonney2kand then thank all the authors/projects we stole code from00:49
@sonney2klike we have in our contributions file00:49
blackburnyeah but it is pretty hard to not lost track00:49
@sonney2kI know but still00:49
blackburnsonney2k: that was kinda embarrassing to get that mail, I feel guilty00:51
@sonney2kblackburn, you should :(00:52
@sonney2kblackburn, do the acknowlegdement / contributions page on the website and send these guys an additional sorry email00:53
@sonney2kthat can really cause trouble00:53
blackburnsonney2k: okay added have to fill it now01:00
blackburnsonney2k: LibSVM, LibLinear, SLEP, MALSAR, LaRank, GPBT, LibOCAS, SVMLin, SGDSVM, VW01:08
blackburnanything else?01:08
blackburnsonney2k: http://shogun-toolbox.org/page/about/contributions01:16
blackburnsonney2k: okay another sorry message01:22
blackburnI next one I should include some kitten photos01:23
shogun-buildbotbuild #575 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/575  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>01:27
blackburnIf I could sing like elton john I would send them a song01:28
blackburnwhat do I do to make you love me01:29
blackburn:D01:29
blackburn*I've got to do01:29
shogun-buildbotbuild #576 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/576  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>01:42
n4nd0the HMSVM toolbox should probably be in acknowledgements too01:50
n4nd0time to sleep now01:51
n4nd0good night!01:51
blackburnn4nd0: do you have admin acc at website?01:51
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Quit: Page closed]01:51
-!- cwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun01:53
-!- heiko [~heiko@host86-179-39-38.range86-179.btcentralplus.com] has left #shogun []02:35
-!- ptizoom [~christian@79-71-85-99.dynamic.dsl.as9105.com] has joined #shogun02:48
shogun-buildbotbuild #124 of nightly_none is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/12403:01
-!- blackburn [~blackburn@188.168.14.140] has quit [Quit: Leaving.]03:14
shogun-buildbotbuild #141 of nightly_default is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/14103:41
-!- cwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has quit [Quit: Page closed]05:11
-!- ptizoom [~christian@79-71-85-99.dynamic.dsl.as9105.com] has quit [Quit: Ex-Chat]08:30
-!- ptizoom [~christian@79-71-85-99.dynamic.dsl.as9105.com] has joined #shogun08:33
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun09:16
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun10:47
-!- ptizoom [~christian@79-71-85-99.dynamic.dsl.as9105.com] has quit [Remote host closed the connection]10:56
wikingblackburn: ping11:26
blackburnwiking: pong11:26
wikingblackburn: i think we could get a dataset from uricamic for so svm11:26
blackburnwhich?11:27
blackburnand what for  :)11:27
wikingminst char recognition11:27
wikingand for the pylab11:27
blackburnI do not really understand how it is a SO11:27
blackburnwhat is structured out there?11:27
wiking1-9 char11:28
blackburnit is just multiclass I'd say11:29
wikingyes11:34
wikingbut i think i could create an latent-sosvm dataset as well11:35
wikingi'll try to do this till friday11:35
wikingdoes it have to work via python? :)11:35
wikingmmm new git11:44
wikingblackburn: btw when to define director classes for swig?11:46
blackburnwiking: what do you mean 'when'?11:51
wikingblackburn: well i've tried to figure out what is at all director class11:56
wikingfrom swig doc11:56
blackburnahh11:57
blackburnso when is like what is the case11:57
blackburnokay11:57
blackburnthe case is:11:57
wikingyeah11:57
blackburnyou want to train a model11:57
blackburnstructured model using risk that is11:57
blackburnbut you are totally lazy to write C++ risk function11:57
blackburnso you inherit from DirectorRisk or how it is called11:58
blackburnand implement risk function out there11:58
wikinginherit in python you mean?11:58
blackburnyes11:58
wikingok11:58
blackburnand implement in python as well11:58
wikingso basically if i have latentmodel11:58
blackburnthen you just plug it in11:58
wikingthat is totally abstract11:58
blackburnand voila11:58
wikingand i want to support that u can implement it in python11:59
wikingi need to provide a director class for it?11:59
wikingright?11:59
blackburnwell you'd need to mark it as director in swig interfaces file, right11:59
wikingok11:59
blackburnand you'd need to declare *ALL* virtual methods from upper classes11:59
wikingyeah i see that11:59
wikingthere's a directorstructuremodel.h12:00
blackburnyes12:00
wikingbtw: i hate nerdtree12:00
wikinganything better than that?12:00
blackburnI do not use it12:00
blackburnI remember all paths12:00
blackburn:D12:00
wikinglolface12:00
blackburnwiking: there is an example12:00
blackburnexamples/undocumented/python_modular/graphical/12:00
blackburnsomething with word director dot py12:01
blackburnthere is a multiclass model wrote using python12:01
wikingexamples/undocumented/python_modular/graphical/so_multiclass_director_BMRM.py12:01
wikingi guess12:01
blackburnhttps://github.com/shogun-toolbox/shogun/blob/master/examples/undocumented/python_modular/graphical/so_multiclass_director_BMRM.py12:01
wikingok cool12:01
wikingyeah12:01
wikingi'll then create the director class for latent12:02
wikingi have to give some extra options for ./configure to support director classes12:03
wikingor that's automagically detected12:03
wikingwhether i support or not12:03
wikingbtw: if i dont define a function in a director class then i won't be able to override that function in the target language?12:07
wikinglike i have a function in LatentModel that is virtual but not pure virtual12:07
blackburnwiking: no, you'd have to explicitly define it still12:09
blackburnand call base class' method12:09
blackburnunfortunately12:09
blackburnin other way you will got infinite recursion12:09
wikingokok12:10
wikinggot it12:10
wikingblackburn: will it use the default implementation though if it's not redefined?12:13
blackburnI am not sure but we did it explicitly12:14
blackburnIIRC it worked only that way.12:14
-!- heiko1 [~heiko@dhcp-160-59.internal.eduroam.ucl.ac.uk] has joined #shogun12:28
heiko1sonney2k, blackburn, just got a job offer :)12:54
blackburnheiko1: nice, where?12:54
heiko1blackburn, at uni, with the statistics department12:55
blackburnahh12:55
heiko1only a little 6-month project, but ...12:55
blackburnis it what you wanted?12:55
heiko1the cool thing is: I will be implementing things into shogun for that12:55
blackburnoh so we've got shogun employee12:55
heiko1yeah kind of, to bridge a gap between now and phd start12:55
blackburn:D12:55
blackburnnice12:55
heiko1GP-classification12:56
heiko1so an extension to jacobs work12:56
wiking:>12:56
wikingcongratz heiko112:56
heiko1thanks, wiking :)12:56
blackburnand I am still waiting for decision :D12:57
heiko1fingers crossed!12:57
n4nd0yeah, congrats heiko :)12:57
wikingn4nd0: i have a new so solver12:58
wikingbased on mosek12:58
wikingbut i don't know12:58
n4nd0wiking: I read sth about that yesterday night12:58
wikingi thnk we should talk about your mosek wrapper12:58
wikingcan u check it out some time?12:58
n4nd0yes12:59
wikingnowish?12:59
wiking:)12:59
n4nd0not really :S12:59
wikinghehe ok12:59
n4nd0I have to go in a few minutes12:59
wikingi'll send u the code in email?12:59
wikingis that good?12:59
n4nd0but just tell me, what would you like to do with the mosek wrapper?12:59
n4nd0wiking: yeah sure, send it12:59
wikingn4nd0: check out the code12:59
wikingand then we'll talk12:59
n4nd0ok13:00
n4nd0wiking: give me also a clue of wthat is the code doing if you think it is a bit hard to understand from only the code13:00
wikingn4nd0: ok13:01
n4nd0ok gtg now13:06
n4nd0tty later13:06
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]13:06
wikingblackburn: https://gist.github.com/393841214:07
blackburnwiking: specialization?14:08
wikingye14:09
wikingso that we use optimized version of that stuff14:09
wikingif available14:09
blackburnis there real impact from that?14:16
wikingwell it all depends if that funciton has any real impact :)14:17
wikingi mean vec1_plus_scalar_times_vec2 itself14:17
-!- heiko1 [~heiko@dhcp-160-59.internal.eduroam.ucl.ac.uk] has left #shogun []14:44
-!- heiko [~heiko@dhcp-191-208.internal.eduroam.ucl.ac.uk] has joined #shogun16:31
-!- yoo__ [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun17:19
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun17:33
wikingn4nd0: ping17:33
n4nd0wiking: pong17:34
n4nd0I have not checked it yet17:34
n4nd0I will do it later, during the evening17:34
wikingokey cool17:41
-!- yoo__ [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Quit: Page closed]17:54
-!- heiko [~heiko@dhcp-191-208.internal.eduroam.ucl.ac.uk] has left #shogun []18:06
n4nd0sonney2k: around?18:06
@sonney2kn4nd0, yes18:15
n4nd0sonney2k: I read yesterday you said sth about that you had recruited a couple of menthors for gsoc18:18
n4nd0sonney2k: I think it would be very cool if Georg Zeller could be menthor in an hypothetical next gsoc18:19
n4nd0I was really happy with him while he helped me last summer18:19
n4nd0only as a suggestion :)18:19
@sonney2kn4nd0, I will check with him but there is still plenty time18:21
n4nd0sure18:21
n4nd0I just mentioned it since the topic poped up18:22
wikinghehe i'm happy to sign up for mentoring ;P18:23
@sonney2kwiking, I will ask next spring :D18:26
@sonney2kbut it seems like we can easily get ~5 mentors18:26
wikingwhy's the but? :)18:27
@sonney2kno but18:27
@sonney2kI did expect it to be harder18:28
-!- ptizoom [~christian@79-71-85-99.dynamic.dsl.as9105.com] has joined #shogun18:29
@sonney2kptizoom, hey!18:29
ptizoomhi18:29
@sonney2kdo you have a minute?18:29
ptizoomyep18:29
@sonney2kwanted to ask about the (surprise) perl interface18:30
@sonney2kI think the pdl variant is a good idea to support indeed18:30
@sonney2knot sure if anyone would use it from core perl anyways...18:30
ptizoomwhy surprise?18:30
@sonney2kjust joking :D18:31
ptizoomyea, PDL was the best equivalent to pynum I thought.18:32
ptizoomdo you mean "core perl"  as opposed to PDL?18:35
ptizoomwhy shogoonies always joke around?18:37
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]18:40
-!- sonne_ [~sonne@adsl-75-5-123-182.dsl.pltn13.sbcglobal.net] has joined #shogun18:41
ptizoombut PDL is also a mess...! one shall see there busy mailing list of despair...18:41
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun18:47
n4nd0wiking: ping18:59
wikingn4nd0: pong19:00
n4nd0wiking: I am reading the code ATM19:00
n4nd0I see you are playing with fire ... using std::vector :-O19:01
n4nd0haha19:01
wikingn4nd0: didn't have time19:01
wikingwanted to finish the code19:01
wiking:D19:01
n4nd0wiking: so what is that you wanted to discuss?19:05
n4nd0I have taken an overall look19:05
wikingn4nd0: well Mosek.h19:05
wikingbecause as u see now19:05
wikingi'm directly using mosek.h19:06
wikingit'd be cool to at least share some part of that mosek_qp_optimize with your Mosek.h wrapper19:06
n4nd0definitely19:06
sonne_ptizoom, why also :D19:06
wikingn4nd0: so this is what i wanted to discuss19:07
wikinghow and when19:07
wiking:>19:07
n4nd0wiking: my idea was do one is able to do the stuff through the wrapper19:07
n4nd0that* one, sorry for the type19:07
n4nd0typo19:08
n4nd0haha, I am idiot19:08
n4nd0anyway19:08
n4nd0what I wanted is that one can get rid of mosek if it comes the time we have alternative solvers in shogun for that19:08
wikingheheh well yeah19:08
wikingas you can see in the cutting_plane function19:09
wikingthere's a lot of uncommented code19:09
wikingthat part is for using SVM^light optimizer19:09
wikingthe problem with that is that the SVM^light in shogun does not have those functions ported19:09
wikingthat i need19:09
n4nd0I am not sure I see the uncommented code you're talking about19:10
wikingjust a sec19:10
wikingn4nd0: check line 33219:10
wikingin https://gist.github.com/393820319:11
n4nd0what does CC in CCSOSVM stanfs for btw?19:11
wikingcopy-cat19:11
n4nd0so what's up with line 332?19:12
wikingso that line19:12
wikingis basically if we would have a full svm^light port in shogun19:12
wikingwe could just use svm^light in this SO-SVm19:12
wikinginstead of mosek19:12
n4nd0it looks good to do that then19:13
wikingwell the problem is that one needs to get svm^light into shogun19:13
n4nd0too problematic?19:13
wikingwell it's just timeconsuming19:13
n4nd0and what isn't? :D19:14
wikingwell i'm happy that we have already this one19:14
wikingthe question is19:14
wikingwhat to do now with Mosek.h19:14
n4nd0what do you want to do with it?19:15
wikingwell how to change it19:15
wikingso that it could be shared19:16
n4nd0I would say that it could be good to do a similar wrapper19:16
wikingbetween PrimalMosek... and this19:16
n4nd0so for example19:17
n4nd0I wonder why is it that you couldn't just use CMosek::optimize19:17
wikinggivemeapatch19:18
wiking:)19:18
n4nd0of what?19:18
n4nd0that code is already in shogun19:18
n4nd0wiking: do you know what I am talking about?19:19
wikingi know19:20
n4nd0good :)19:20
wikingshould i just add the wrapper functions i need?19:21
n4nd0for me that makes more sense19:21
n4nd0to use the wrappers to put the Q, A, etc matrices19:22
n4nd0to use the wrapper to solve the problem19:22
n4nd0and so19:22
n4nd0if you need other mosek functions, put other wrappers maybe19:22
n4nd0but idk, maybe that doesn't make sense for you or you don't like it so19:22
n4nd0gtg for dinner, I'll be back later19:23
n4nd0re19:35
-!- sonne_ [~sonne@adsl-75-5-123-182.dsl.pltn13.sbcglobal.net] has quit [Ping timeout: 248 seconds]19:37
-!- blackburn1 [~blackburn@83.234.54.235] has joined #shogun20:00
blackburn1ptizoom: we are usually drunk - that's why we've got to joke around :(20:02
ptizoomblackburn: aaahhhh the spirited easterners of the tundras, islanders of the white goulag...we drink and fill our glasses for a bit of fun...21:01
-!- blackburn1 [~blackburn@83.234.54.235] has left #shogun []21:34
-!- blackburn1 [~blackburn@83.234.54.235] has joined #shogun22:45
-!- ptizoom [~christian@79-71-85-99.dynamic.dsl.as9105.com] has quit [Ping timeout: 268 seconds]23:06
n4nd0wiking: ping23:12
-!- travis-ci [~travis-ci@block630322-r7m.blueboxgrid.com] has joined #shogun23:17
travis-ci[travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/290553723:17
-!- travis-ci [~travis-ci@block630322-r7m.blueboxgrid.com] has left #shogun []23:17
shogun-buildbotbuild #468 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/468  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:17
n4nd0haha travis-ci is quite funny23:18
blackburn1finally he has visited us23:19
-!- travis-ci [~travis-ci@block630322-r7m.blueboxgrid.com] has joined #shogun23:20
travis-ci[travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/290558423:20
-!- travis-ci [~travis-ci@block630322-r7m.blueboxgrid.com] has left #shogun []23:20
shogun-buildbotbuild #462 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/462  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:22
shogun-buildbotbuild #463 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/463  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:29
shogun-buildbotbuild #9 of ubu1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/ubu1%20-%20libshogun/builds/9  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:32
shogun-buildbotbuild #59 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/59  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:36
shogun-buildbotbuild #577 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/577  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:46
shogun-buildbotbuild #10 of ubu1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/ubu1%20-%20libshogun/builds/10  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:50
shogun-buildbotbuild #578 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/578  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:51
shogun-buildbotbuild #469 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/469  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:53
wikinghehehehe23:53
wikingtravis came :D23:53
wikingblackburn1: :D23:53
wikingblackburn1: u broke it!23:54
wikingn4nd0: pong23:55
shogun-buildbotbuild #60 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/60  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>23:58
n4nd0wiking: I am trying to prepare a simple example of image segmentation for Christian's talk23:59
n4nd0using the so-svm23:59
--- Log closed Wed Oct 24 00:00:00 2012
« Monday, 2012-10-22 Index Wednesday, 2012-10-24 »

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