K0stIa K0stIa wiking --- Log opened Tue Nov 27 00:00:15 2012 -!- zxtx [~zv@ool-45750cfe.dyn.optonline.net] has quit [Ping timeout: 256 seconds] 00:19 -!- zxtx [~zv@ool-45750cfe.dyn.optonline.net] has joined #shogun 00:24 -!- gsomix [~gsomix@188.168.2.95] has quit [Ping timeout: 276 seconds] 00:34 -!- ptizoom [~christian@79-71-89-182.dynamic.dsl.as9105.com] has quit [Quit: Ex-Chat] 03:47 -!- ptizoom [~christian@79-71-89-182.dynamic.dsl.as9105.com] has joined #shogun 03:47 -!- gsomix [~gsomix@188.168.2.95] has joined #shogun 04:55 -!- gsomix [~gsomix@188.168.2.95] has quit [Ping timeout: 248 seconds] 05:16 -!- blackburn [~blackburn@188.168.2.95] has quit [Quit: Leaving.] 08:18 -!- gsomix [~gsomix@46.20.65.185] has joined #shogun 10:07 -!- gsomix [~gsomix@46.20.65.185] has quit [Ping timeout: 255 seconds] 10:53 -!- gsomix [~gsomix@37.61.180.167] has joined #shogun 12:31 -!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has joined #shogun 13:36 hello, everybody! 13:40 the example so_multiclass_director_BMRM.py stopped working. It does segmentation fault if you run the script for several times without exiting from python environment. The same problem appears if I am trying to extend DirectorStructuredModel, it does also segmentation fault if you run the same script for several times. 13:40 -!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has left #shogun [] 14:01 -!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has joined #shogun 14:02 K0stIa: mmm 14:25 K0stIa: which version of shogun r u  usinf 14:25 using? 14:25 last version from github 14:47 K0stIa: can u share the script? 14:48 you made make file ? 14:49 which? 14:49 sorry :) 14:49 you want me to show make file ? 14:50 i'd like to test it but i need an example to test this problem with 14:50 I see :) 14:50 so if u can share me your script that does segfault 14:50 yes :) 14:50 wait a minute please 14:50 that'd help me to debug the problem 14:50 https://code.google.com/p/shogun-toolbox/source/browse/examples/undocumented/python_modular/graphical/so_multiclass_director_BMRM.py?r=8be38323f2dc590295a2913cc5fc944d1fc1333c 14:51 I run this file several times in ipython 14:51 after third it throws exeption 14:51 mmm 14:52 ok i'll try that 14:52 but why r u running this several times? :) 14:52 well, the same problem appear if you want to run it for different lambda 14:52 to make cross-validation 14:52 ok i'll try it now 14:53 this is just example which simulates my problem 14:53 thanks 14:53 btw: did you implement the director class for structure model ? 14:53 noup 14:53 do you know who did it ? 14:53 but i've rewritten lately couple of parts of bmrm 14:54 I need now bmrm :) 14:54 and it was ok before... 14:54 well maybe i've introduced a bug :( 14:54 so maybe the problem is there ? 14:54 i'll check it for you 14:54 thanks a lot :) 14:54 i've done tests with it 14:54 ok :) 14:54 but i haven't done it with neither python 14:54 nor trying to run it several times 14:55 although i run it several times 14:55 in a different setup 14:55 and i haven't noticed any problems 14:55 please, check this script :) 14:55 yeah i'll do 14:55 i use bmrm via c++ 14:55 but that shouldn't matter in this case anyways 14:56 well, maybe it could 14:56 soon we'll see.. i'm just recompiling shogun with python support 14:56 please, recompile it also with director classes support 14:57 btw 14:59 u just start ipython and start fomr ther the script? 14:59 *from 14:59 i started ipythin, then did %run script.py several times 15:00 and it did segmentation fault 15:00 the same things appears in my working script 15:00 ok 15:00 which I can not share :( 15:00 secret? :) 15:01 kind of :P 15:01 how come? :) 15:01 no :) it will take a time to make it independent :) 15:02 and I don't have it know 15:02 I detected the problem 15:02 ? 15:02 independent? 15:02 yep :) 15:02 dont understand :) 15:02 well, i would have to share some files in addition for you :) and that I don't want :) 15:03 i see 15:04 what context are u using this for? 15:04 i mean shogun 15:04 or that's unshareable information 15:04 I use it in my work on paper now :) 15:05 for learning :) 15:05 ah so u r using it for research 15:05 yes 15:06 I also have some suggestion for some improvements for structure output learning 15:07 we are listening 15:07 the reason i'm running it several times is because i use these algos to do latent structural classification... 15:08 mmm i'm getting this error: SystemError: [ERROR] FATAL: Parameter::add_type(): Double parameter m_TolRel'! 15:08 i'm having some patches already prepared to for a bit better API of SO learning 15:08 it would be nice to provide in director class possibility to override the the risk funtion 15:08 well the risk function is soon to be changed 15:09 as i've just commited a non-convex risk minimizer 15:09 so that u don't have to do funky things if u want to minimize an concave risk function 15:09 will it be possible to override the risk function ? 15:09 well yeah 15:10 that's nice 15:10 what's your risk function now? 15:10 or that's secret as well? 15:10 no :) 15:10 ok i have totally different problem now: SystemError: [ERROR] FATAL: Parameter::add_type(): Double parameter m_TolRel'! 15:10 it's risk approximation for supervised learning 15:11 which is convex 15:11 i see it's cool 15:11 hope you'll share it 15:11 when u have the paper done 15:11 btw: which algorithm is implemented in your non-convex ? 15:11 it's simply solving 15:11 1/2 \lambda ||w||^2 + R(w) 15:12 where R(w) could be anything 15:12 it's based on libqp 15:12 it's known approximation... you can find it in SO related papers 15:12 is there some paper about it ? 15:12 not yet 15:12 i'm preparing one 15:12 I see :) 15:12 what is the class of functions that algorithm works for ? 15:13 libncbm.cpp 15:14 it's in structure 15:14 I mean for with type of risk functions it could be applied ? 15:14 no python api for it 15:14 *which 15:14 any risk function 15:14 currently it's bundled to structured model 15:15 so you would need to give a structured model with your desired risk function 15:15 really ? for any nonconvex function it will find a global optima ? 15:15 should 15:15 worked for me 15:15 at least for 2 problems 15:16 that are non-convex 15:16 can you say what they problems are ? 15:16 well i'm applying this in computer vision 15:16 so the problems are from that field 15:17 ok i think ishould switch to master branch 15:18 do you have any proof about guarantees ? 15:18 as i already have some other patches in my local branch and i get totally different 15:18 errors 15:18 K0stIa: you mean about the convergence 15:18 yes 15:18 it would be interesting to see you r paper 15:21 well it's not published 15:21 it's not even submitted 15:21 :D 15:21 and now can you do something with that bugs ? 15:21 well i'm trying to reproduce it first 15:21 but i'm getting very different errors 15:22 but i'm guessing it's because of the local patches i've done 15:22 well, could you try to install version from github  ? 15:24 without patches ? 15:24 maybe there are some another bugs ? 15:24 yeah that's what i'm trying 15:25 good :) 15:25 K0stIa: well i'm running now for the 5th time 15:42 no segfault yet 15:42 really ? 15:43 you tried the same example ? 15:43 several times ? 15:44 yes 15:45 %run so_multiclass_BMRM.py 15:46 should be %run so_multiclass_director_BMRM.py 15:47 ok 15:48 this one segfaulted 15:48 i'll gdb it 15:48 thanks 15:48 wiking: how it goes ? 16:43 wiking, is there any reason for this ::vec1_plus_scalar_times_vec2 function in SGVector 17:16 sonney2k: hi, who is responsible for doing director class for structured model ? 17:20 could he change oracle for structured model in the following way, the oracle should be represented as an interface  and user should provide function which takes point and returns function value and subgradient ? 17:20 K0stIa, n4nd0 did this. 17:21 K0stIa, and gsomix is the master who made it possible... 17:22 sonney2k: now it is a bit confusing... one has to pass some loss function to the svm, however loss is not needed, which is not that crucial for me, but I guess it would be better to get rid of it. 17:23 K0stIa, well ask on the mailinglist if he doesn't appear here before you leave. 17:27 sonney2k: for structured learning now it is not that convenient as it could be, 17:29 1. the summation in case of learning more than one example is done inside shogun, I think it would be better if user do this summation, because this is something with is hard to compute and could be easily paralyzed. One can do it now of course, however is not convenient as it could be. that's why I think this director class should be as follows, just an interface with risk function that takes point from R^n space and returns function va 17:29 -!- blackburn [~blackburn@37.61.180.167] has joined #shogun 17:41 sonney2k, K0stIa hey 17:55 gsomix: hey 17:58 -!- K0stIa [~kostia@2001:718:2:1634:4a5b:39ff:fe99:cc49] has quit [Remote host closed the connection] 18:43 -!- K0stIa [~kostia@alt2.hk.cvut.cz] has joined #shogun 19:16 sonney2k: well just that because there's a cblas optimized code for that 19:25 K0stIa: u know that u can do it and submit a PR, right? : 19:38 :) 19:38 K0stIa: pr = pull request 19:39 sonney2k: i think ccsosvm broke... aint' working anymore for me :( 19:39 wiking:  I haven't been working on it. It could take time for me which I don't have now. 19:40 K0stIa: well yeah that's a usual problem for everybody :( 19:40 time i mean 19:40 wiking: I was reading logs and don't you find strange you can find global maxima of non-convex function? 19:40 blackburn: why? :) 19:41 and not global maxima 19:42 but minima 19:42 it is the same 19:42 yeah indeed 19:42 argh 19:47 now i have to cherry pick this stuff :( 19:47 bisect is my friend 19:55 K0stIa: maybe you could use bisect to find out which commit introduced this bug if it used to work before... 19:56 would be great if we would have unittests for this :) 20:00 could run bisect automatically 20:00 ahahah http://www.theregister.co.uk/2012/11/27/google_raters_manual/ 20:10 -!- gsomix [~gsomix@37.61.180.167] has quit [Ping timeout: 246 seconds] 20:28 sonney2k: that patch of yours for std::vector is not working 20:29 so there's something wrong with DynArray afaik 20:29 sonney2k: curious question; was there a particular reason you settled on buildbot as a CI tool? 20:49 I don't necessarily have a better suggestion or anything, just, curious, is all 20:50 -shogungit:#shogun- [shogun] lisitsyn pushed 1 new commit to master: https://github.com/shogun-toolbox/shogun/commit/315b058291d6d0ad9ddc1e8b791f53d6e7938615 21:01 -shogungit:#shogun- shogun/master 315b058 Sergey Lisitsyn: Flexible director risk 21:01 naywhayare, no I just tried it out first and it was python so I could easily extend it to my needs. 21:04 wiking, what happens? 21:05 wiking, re SGVector I meant why not do template specializations instead of overloading? I am currently changing the code to do exactly that 21:05 wiking, I am a bit fed up with the get_uint_matrix etc functions and try to do overloading for that stuff now too... 21:06 sonney2k: dunno what happens currently... i'm just saying that std::vector -> DynArray introduced a bug that basically the whole CCSOSVM doesn't work 21:06 wiking, well if you had I test or at least an example then I could check 21:06 this way you have to figure out what is wrong 21:07 ok.  thanks for the answer :) 21:07 sonney2k: do u have mosek? 21:08 no 21:09 naywhayare, he - well and all issues I had were resolved in #buildbot - IIRC tomprince helped a lot and basically immediately 21:10 or at the GSoC mentors summit FWIW 21:10 ah, that's convenient 21:10 I use Jenkins and have had decent success with it, but it's one of those things where I haven't had the time to set it up properly 21:11 I'll keep buildbot in consideration for the future when I have infinite free time 21:11 which unfortunately seems infinitely far away... 21:11 sonney2k: i've just changed back to std::vector and it works perfectly... :S 21:14 so i dont know what happens with DynArray actually 21:14 as i really just changed 3 things. definition of the variable 21:15 and resize_array -> resize 21:15 so most probably there's a bug in DynArray :( 21:16 naywhayare, I also have good experience with jenkins 21:16 wiking, what does valgrind say? 21:17 sonney2k: i cannot valgrind it, as valgrind is broken currently on my os :( 21:17 wiking, this makes us a perfect team :-/ 21:18 hehhe 21:19 build #443 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/443  blamelist: Sergey Lisitsyn 21:19 wiking, the only idea I have is that we add some small example in examples/undocumented/library_dynarray.cpp that uses SGVectors... 21:20 which I guess might be the issue 21:20 ok i've pushed a pr 21:22 -shogungit:#shogun- [shogun] vigsterkr opened pull request #839: CCSOSVM refactoring (master...so-dev) https://github.com/shogun-toolbox/shogun/pull/839 21:22 it's a code refactor anyways of CCSOSVM 21:22 and currently switches back to std::vector 21:22 so don't merge it until we don't find out the dynarray stuff 21:22 and then i'll do a little commit for this PR 21:23 that switches back to dynarray 21:23 I am currently deep in refactoring hell 21:23 that's ok 21:23 ;) 21:23 this can sit there for some time 21:23 hah it looks like it compiles 21:24 mmm 21:24 wiking, so now we can just use SGVector.load 21:25 what happened to travis? 21:25 SGMatrix.load etc 21:25 and no more need for tons of defines 21:25 build #680 of deb3 - modular_interfaces is complete: Failure [failed compile csharp_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/680  blamelist: Sergey Lisitsyn 21:29 -shogungit:#shogun- [shogun] sonney2k pushed 2 new commits to master: https://github.com/shogun-toolbox/shogun/compare/315b058291d6...b7a678696238 21:32 -shogungit:#shogun- shogun/master b4322ff Soeren Sonnenburg: Refactoring... 21:32 -shogungit:#shogun- shogun/master b7a6786 Soeren Sonnenburg: compile fix for csharp (directors & SO) 21:32 build #444 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/444  blamelist: Soeren Sonnenburg 21:46 sonney2k: does buildbot do something like jenkins' "configuration matrix"? 21:56 where you can build your project against a combination of different compilers, libraries, and so forth 21:57 naywhayare: no 21:57 okay 21:57 build #445 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/445  blamelist: Soeren Sonnenburg 21:57 I couldn't seem to find anything about that in the documentation I was perusing 21:57 -!- heiko [~heiko@46.17.162.114] has joined #shogun 21:59 build #681 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/681  blamelist: Soeren Sonnenburg 22:12 build #682 of deb3 - modular_interfaces is complete: Failure [failed compile csharp_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/682  blamelist: Soeren Sonnenburg 22:12 wat? 22:15 crazybrokenness 22:17 :> 22:18 blackburn, you broke it all :) 22:26 sonney2k: can't really understand the error 22:26 blackburn, I am fixing the csharp one 22:26 ahh 22:26 but the other? 22:26 other like? 22:26 other error 22:27 cygwin! 22:27 which one/ 22:27 ah 22:27 http://shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/445/steps/compile/logs/stdio 22:27 it makes no sense to me 22:27 at all 22:27 at least it has to do with your commit :) 22:28 12 people in room 22:29 wow 22:29 heiko: ! 22:29 ugh csharp compiles here 22:29 blackburn, hi :) 22:29 heiko: long time no see as sonney2k likes to say 22:30 blackburn, yes he would be right 22:30 busy weeks here 22:30 how are you? 22:30 heiko, trying to rescue the world from evil? 22:30 sonney2k, I did, but usually nobody notices 22:31 everything going alright here? 22:31 thats the destiny of super heros 22:31 hehe ;) 22:31 sonney2k, blackburn, no just first few weeks of new job, and its all a bit different than expected so I had lots of things to do 22:32 currently in Cambridge 22:32 next week in Warsaw 22:32 heiko, we are tying hard to break all code - but it is not as easy ;-) 22:32 project ends with this year thank god 22:32 sonney2k, sounds good ;) 22:32 yeah I use axe 22:32 to break everything 22:32 but it still works sometimes 22:32 heiko, but back in germany next year then? 22:32 blackburn, csharp just compiles here!? 22:33 sonney2k, no, I will stay in London 22:33 oh different job? 22:33 not the GP classification one? 22:33 sonney2k, well currently I have to finish a EU project with a company where GPs are used, but since so little time left, I just need to collect stuff - so no time for shogun implementation 22:34 but next year I will continue working in the stats department, but doing research 22:34 I start on new job next week starting from monday 22:34 -shogungit:#shogun- [shogun] sonney2k pushed 1 new commit to master: https://github.com/shogun-toolbox/shogun/commit/1d57638fa6365234e9b1639c4e5f8e73e711b8cd 22:34 -shogungit:#shogun- shogun/master 1d57638 Soeren Sonnenburg: include lib common to make data types know 22:34 and then, the cool MCMC stuff with GPs will happen - hopefully 22:34 heiko, sounds more like you won't have time either :/ 22:35 no I think it will be better since no grant project deadlines 22:35 sonney2k: have you ever thought it will be better some day? ;) 22:35 they basically did nothing during the summer and now I am supposed to do all the stuff 22:35 heiko, poor slave 22:36 so rescuing the world indeed! 22:36 sonney2k, yeah a bit annoying, but at least I can do traveling and get a salary - and its interesting stuff in any way, so learning... 22:37 sounds like my thought when I got into netcracker :D 22:37 yo 22:38 heiko, I hope you will find the time to add GPC to shogun - now that you've done this stuff once you should probably know best how to do it right :) 22:38 where should be the structure machines be defined? 22:38 sonney2k: ^ 22:38 wiking, what? 22:38 sonney2k, yes I will do that for sure, implemented a lot of stuff in matlab recently 22:38 sonney2k: currently structured machines doesn't have a type 22:38 multi-class GPs, approximate inference, approximate exact inference, cool stuff 22:38 sonney2k: EMachineType 22:38 but currently only hacks, 22:38 wiking, shogun/machines/StructuredOutputMachine.h ? 22:39 ahh 22:39 also there 22:39 sonney2k: they just have problem type 22:39 sonney2k: but i cannot differentiate the different SO machine implementations 22:39 only if i try try catch dynamic csat 22:39 cast 22:39 sonney2k: or of course get_name + strcmp ;) 22:40 currently seeing christians pydata talk 22:41 wiking, blackburn - http://shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/573/steps/compile%20libshogun/logs/warnings%20%284%29 22:45 there are still a few warnings 22:45 sonney2k: so? 22:46 sonney2k: (this is not a so about the warning but the machines..) 22:46 wiking, in your code. 22:47 sonney2k:  i know that i see that :) 22:47 wiking, feel free to just add to EMachineType in shogun/machine/Machine.h 22:48 build #446 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/446  blamelist: Soeren Sonnenburg 22:50 sonney2k: there u go 22:51 -shogungit:#shogun- [shogun] vigsterkr opened pull request #840: Fix warnings in libncbm and StructuredModel (master...master) https://github.com/shogun-toolbox/shogun/pull/840 22:51 -shogungit:#shogun- [shogun] sonney2k pushed 2 new commits to master: https://github.com/shogun-toolbox/shogun/compare/1d57638fa636...11acf9002c3f 22:52 -shogungit:#shogun- shogun/master 46a1f28 Viktor Gal: Fix warnings in libncbm and StructuredModel 22:52 -shogungit:#shogun- shogun/master 11acf90 Soeren Sonnenburg: Merge pull request #840 from vigsterkr/master... 22:52 -shogungit:#shogun- [shogun] vigsterkr opened pull request #841: Define EMachineType for the SO learners (master...master) https://github.com/shogun-toolbox/shogun/pull/841 23:02 * sonney2k ZZzzz... 23:03 build #447 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/447  blamelist: Viktor Gal 23:09 build #683 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/683  blamelist: Soeren Sonnenburg 23:16 build #448 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/448  blamelist: Soeren Sonnenburg 23:20 build #684 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/684  blamelist: Soeren Sonnenburg , Viktor Gal 23:46 --- Log closed Wed Nov 28 00:00:15 2012