Open in new window / Try shogun cloud
--- Log opened Wed Jul 31 00:00:54 2013
-!- FSCV [] has joined #shogun00:02
-!- lisitsyn [~lisitsyn@] has quit [Read error: Connection reset by peer]00:10
shogun-notifier-shogun: Soeren Sonnenburg :develop * 3c689be / src/shogun/mathematics/Math.h:
shogun-notifier-shogun: define M_PI if not available00:23
-!- travis-ci [] has joined #shogun00:31
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []00:31
shogun-buildbotbuild #1128 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at  blamelist: Soeren Sonnenburg <>00:41
-!- iglesiasg [] has joined #shogun00:42
-!- mode/#shogun [+o iglesiasg] by ChanServ00:42
-!- FSCV [] has quit [Quit: Leaving]00:54
shogun-notifier-shogun: Viktor Gal :feature/CMake * 8ff0791 / / (9 files):
shogun-notifier-shogun: Define CPack components,preliminary CPack settings00:58
shogun-notifier-shogun: Fix FindOctave cmake script00:58
-!- travis-ci [] has joined #shogun01:04
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []01:04
@iglesiasgwiking: hey! So feature/cmake is ready to start trying it out?01:04
@wikingiglesiasg: it was already ready before01:10
@wikingbut yeah01:10
@wikingalthough it's not rebased yet01:10
@iglesiasgwiking: I just played around with ccmake and I am building right now!01:10
@wikingand it was branched quite some time ago01:10
@wikingyeah cool01:10
@wikinglet me know wha's happening01:10
@iglesiasgwiking: I thought it was very WIP before and could not be yet used01:11
@wikinglibshogun + python_modular was working already before the WS01:13
@iglesiasgwiking: I see, cool01:13
@iglesiasgwiking: out of curiosity, why do you think rebase will cause problems?01:13
@wikingiglesiasg: i've tried yesterday01:14
@wikinghad to abort it01:14
@wikingalthough i havent really checked how serious that manual merge would be01:14
@wikingbuuut it wasnt working automatically01:14
@wikingso i just postponed it01:14
@iglesiasggot it01:14
@wikingas there's still some stuff to be done01:14
@iglesiasgsuch as?01:15
@wikingmake tests target01:16
@wikingthat's completely missing01:16
@wikingas well as matlab static01:16
@wikingand kind of like that's it01:16
@iglesiasgwiking: btw I got this at the beginning, but it continued01:16
@iglesiasg[  1%] Generating version header01:16
@iglesiasgfatal: Not a valid object name master01:16
@wikingmmm yeah i saw that on travis as well01:17
@wikingno fucking idea wtf is that01:17
@iglesiasgwiking: it is working good here01:33
@iglesiasgjust tried libshogun for the moment01:33
-!- travis-ci [] has joined #shogun01:34
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []01:34
-!- travis-ci [] has joined #shogun01:39
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []01:39
@iglesiasggood night!02:29
-!- iglesiasg [] has quit [Quit: Leaving]02:45
shogun-buildbotbuild #1438 of deb3 - modular_interfaces is complete: Failure [failed test octave_modular]  Build details are at  blamelist: Soeren Sonnenburg <>03:22
-!- shogun-notifier- [] has quit [Quit: transmission timeout]03:58
-!- gsomix [~gsomix@] has quit [Quit: Leaving]03:59
-!- zxtx [] has quit [Ping timeout: 264 seconds]04:00
shogun-buildbotbuild #474 of nightly_default is complete: Success [build successful]  Build details are at
-!- foulwall [~user@] has joined #shogun05:14
-!- shogun-notifier- [] has joined #shogun05:27
shogun-notifier-shogun: Viktor Gal :feature/CMake * 8afb189 / .travis.yml:
shogun-notifier-shogun: travis: add headers pkg of octave05:27
-!- travis-ci [] has joined #shogun05:37
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []05:37
@wikingsonney2k: ideas:  ??05:38
-!- shogun-notifier- [] has quit [Quit: transmission timeout]08:27
-!- foulwall [~user@] has quit [Remote host closed the connection]08:51
-!- zxtx [] has joined #shogun09:00
-!- az_de [82954e22@gateway/web/freenode/ip.] has joined #shogun10:27
-!- votjakovr [] has joined #shogun10:28
-!- lisitsyn [] has joined #shogun10:30
-!- lisitsyn1 [] has joined #shogun10:48
-!- lisitsyn [] has quit [Ping timeout: 264 seconds]10:51
-!- van51 [] has joined #shogun11:00
-!- van511 [] has joined #shogun11:04
-!- van51 [] has quit [Ping timeout: 264 seconds]11:06
-!- van511 is now known as van5111:07
-!- az_de [82954e22@gateway/web/freenode/ip.] has left #shogun []11:29
van51I'm having some trouble with cmake11:36
van51I'm getting stuff like "/shogun/mathematics/eigen3.h:17:24: fatal error: Eigen/Eigen: No such file or directory compilation terminated."11:37
van51while with make it is doing ok11:37
van51maybe I'm using it wrong11:38
van51if someone has an idea let me know :)11:38
-!- iglesiasg [] has joined #shogun11:43
-!- mode/#shogun [+o iglesiasg] by ChanServ11:43
@iglesiasggood morning!11:43
van51iglesiasg: hi!11:44
van51iglesiasg: you used cmake yesterday, right?11:47
@iglesiasgvan51: yeah!11:47
@iglesiasgvan51: I compiled libshogun with it11:48
@iglesiasgeverything went pretty smooth11:48
van51iglesiasg: I'm getting some errors, like clapack.h: No such file or directory :S11:48
@iglesiasgvan51: I see11:49
van51iglesiasg: while with the other way it was fine11:49
@iglesiasgvan51: can you take a look with ccmake maybe what is recognized in your system?11:49
@iglesiasgvan51: I had for instance to set the gmock path by hand using ccmake11:50
van51iglesiasg: and how did you do that?11:53
van51iglesiasg: I can't really tell if something is off tbh :)11:53
@iglesiasgvan51: let me check about lapack in particular11:53
@iglesiasgvan51: all right, this is a doubt I always have but I think arpack and lapack are actually the same11:54
@iglesiasgvan51: so go to the directory <your shogun root dir>/build11:54
@iglesiasgvan51: and do ccmake ..11:55
@iglesiasgvan51: not the two cs, what do you see?11:55
@iglesiasgnot -> note*11:55
van51iglesiasg: <shogun>/build or <shogun>/src?11:56
@iglesiasgvan51: I build with cmake from shogun/build11:56
@wikingplease add here your error11:56
@wikingotherwise i cannot help u11:57
@wikingcmake needs testing11:57
@wikingand i just test it on travis and my own machine11:57
@wikingso maybe some things are not yet working smoothly on other systems11:57
@iglesiasgwiking: here libshogun worked fine too, I am to test python_modular now11:57
van51maybe I need to initialize something? because I don't see a build directory11:58
@wikingiglesiasg: that should work too as i've tested that intensively11:58
@iglesiasgvan51: yeah sure, mkdir build11:58
@wikingvan51: # Instructions:11:58
@wiking# $ mkdir build11:58
@wiking# $ cd build11:58
@wiking# $ cmake ..11:58
@wiking# $ make11:58
@iglesiasgvan51: sorry, I didn't mention that11:58
@wikingas it says11:58
@wikingin the beginning of the main cmakelists.txt11:58
@wikinganyhow i'm back to finish up some of the cmake shit11:58
@iglesiasgvan51: with cmake you normally do those steps wiking just copied from the CMakelist.txt11:58
@wikinghopefull i'll have make tests within some hours11:59
@wikingif u really need help11:59
@iglesiasgwiking: all right, I will go with another interface too as well then11:59
@wikingwith cmake11:59
@wikingiglesiasg: yeah test other interfaces plz11:59
@wikingpython should really work11:59
@wikingvan51: if u got really really stuck or something is not working add your comment to that bug11:59
@wikingthat way i know what i should fix11:59
@wikingi'm now not looking at irc that reguraly12:00
@wikingas i really want to finish up this one12:00
van51wiking: ok! if I don't manage to do something I will comment there12:00
-!- HeikoS [] has joined #shogun12:32
-!- mode/#shogun [+o HeikoS] by ChanServ12:32
thoralfHeikoS: Hey.12:32
thoralfHeikoS: I'm unable to add a label to my ticket.  Maybe you have got the permissions to do so?12:33
@HeikoSthoralf: yeah I can do this12:34
@HeikoSthoralf: sorry for not getting back, NIPS reviews got in :=)12:34
@HeikoSthoralf: labelled12:34
thoralfHeikoS: That's okay.  Issues are patient. ;)12:34
@sonney2kwiking, use a recent octave! octave and clang don't work well together12:36
@sonney2kshogun-buildbot, force build cyg1 - libshogun --branch=develop12:37
shogun-buildbotno such builder 'cyg1'12:37
thoralfHeikoS: My eclipse (kepler) does not work any more... it dies with a java.lang.StackOverflowError, even after clean shogun-checkout and removing caches/workspaces/etc.12:37
thoralfHeikoS: Ever seen this on shogun?12:37
@sonney2kshogun-buildbot, force build 'cyg1 - libshogun' --branch=develop12:37
shogun-buildbotbuild forced [ETA 16m09s]12:37
shogun-buildbotI'll give a shout when the build finishes12:37
lisitsyn1hahah stackoverflow12:37
-!- lisitsyn1 is now known as lisitsyn12:37
thoralflisitsyn: Yes. :)12:37
lisitsynwhat a programmers12:37
thoralflisitsyn: The logfile is 100KiB big.12:38
thoralf(for one crash)12:38
lisitsynthoralf: that's okay12:38
lisitsynI used to dig into 10MB logs :D12:38
thoralflisitsyn: I won't dig... imagine I would do so, then I am fixing eclipse to fix my shogun problems to fix my ML problems.12:39
thoralflisitsyn: What's next?  Fixing Java?12:39
lisitsynthoralf: yeah indeed you don't have to12:40
lisitsynthoralf: I mean 100kb is not the thing to crash anything12:41
lisitsynrecursion or something like that12:41
thoralflisitsyn: No, 100kb is the stack trace. ;)12:41
lisitsynthoralf: then you know what happened12:41
thoralflisitsyn: Yes.  Recusion. ;)12:41
lisitsynHeikoS: your evaluations are the only to go12:42
@HeikoSlisitsyn: ok I will do them today12:44
@HeikoSwanted yesterday but had too many meetings12:44
@sonney2kHeikoS, bad excuse ... takes 2 minutes12:45
@sonney2khurray all builds except cyg1 are green since last night!12:45
@HeikoSsonney2k: okok Ill do them now12:45
lisitsynHeikoS: sonney2k: have you decided anything on doc camp thing?12:46
@sonney2klisitsyn, on travis is still failing - any updates on that?12:46
@HeikoSsonney2k, lisitsyn we should write a good applicatoin, but not recycle the one from last year, we should have  amore clear vision what we will get12:46
lisitsynsonney2k: if you can live with removing it for now - lets do that12:47
lisitsynit takes a lot of time to compare with matlab impl I made it from12:48
@sonney2klisitsyn, I would prefer you fixing it and adding a test12:48
lisitsynHeikoS: I think it should be something like that 'X for hackers' or whatever12:48
lisitsynsonney2k: I am afraid I won't be able to do that next days12:49
@sonney2klisitsyn, when?12:49
lisitsynsonney2k: I have time this weekend for sure12:49
-!- shogun-notifier- [] has joined #shogun12:50
shogun-notifier-shogun: Roman Votyakov :develop * 4c15b21 / src/NEWS:
shogun-notifier-shogun: update news12:50
shogun-notifier-shogun: Soeren Sonnenburg :develop * f7d66ef / src/NEWS:
shogun-notifier-shogun: Merge pull request #1337 from votjakovr/feature/numerical_integration12:50
shogun-notifier-shogun: Update news12:50
@iglesiasghi sonney2k! how is it going?12:50
@wikingiglesiasg: mmm12:51
@wikingiglesiasg: travis gave me this12:51
@iglesiasgsonney2k: I saw that the SGMatrixList patch fixed the problem in the buildbot, however valgrind still complains about python examples here :S12:51
@iglesiasgsonney2k: let me know if you want to have a look at the trace12:51
@wikingiglesiasg: what is your distrib?12:52
@iglesiasgwiking: linux mint12:52
@wikingiglesiasg: do you have the HEAD of the feature/CMake?12:52
@wikingcommit 8ff0791e0d026e8144d83d00bad38ae69895535512:53
shogun-buildbotbuild #1129 of cyg1 - libshogun is complete: Success [build successful]  Build details are at
@wikingiglesiasg: since there's a fix for octave detection12:53
@wikingtry to do a git pull on that branch12:53
@wikingand rerun cmake12:53
@HeikoSsonney2k, lisitsyn done12:54
@sonney2klisitsyn, ok then this weekend is what we could still bear but we really need it then12:54
@sonney2kHeikoS, thanks!!12:54
@sonney2kiglesiasg, w/o traces you could say anything :P12:55
@sonney2kman all green12:55
@HeikoSsonney2k, wiking we should enable valgrind to break the unit tests at some point, same for the libshogun examples12:56
@HeikoSthoralf: ^12:56
lisitsynsonney2k: ok I'll either fix it or remove ti12:56
@wikingHeikoS: yeah we should... :) no time sorry want to finish cmake ;P12:56
@HeikoSwiking: yeah dont worry12:56
@wikingsonney2k: what is a good octave version? :)12:56
@HeikoSwiking: just to keep in mind :)12:56
@sonney2kwiking, the latest and greatest but I am not sure any works with clang yet12:57
@iglesiasgsonney2k: that is for structure_plif_hmsvm_bmrm.py12:57
@sonney2kwiking, try it with g++ rather12:57
thoralfHeikoS: I'm valgrinding every code I write - if I find something, I'll fix it.12:57
@sonney2kvan51, hey how is it going?12:57
@wikingsonney2k: 3.2?12:57
@wikingsonney2k: as this is what we have by default on travis :(12:57
@sonney2kwiking, would work with gnu12:57
@sonney2kwiking, no way with clang12:57
@iglesiasgwiking: I just saw you made a commit 8h ago for octave package, I was missing that one12:58
@wikingsonney2k: ok will switch in travis12:58
@wikingiglesiasg: cool try that one12:58
@wikingiglesiasg: u using gcc?12:58
@iglesiasgwiking: my bad I didn't check it first12:58
@iglesiasgwiking: yeah, I think that one is set by default, I have clang too12:58
@wikingiglesiasg: ok try with gcc13:00
@wikinglets see if it fails for u13:00
@wikingsonney2k: ideas?13:02
@wikingi have jblas 1.213:04
@wiking1.2.0 to b precise13:04
shogun-buildbotbuild #1130 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Soeren Sonnenburg <>13:05
@iglesiasgwiking: same error after rebase13:06
@wikingiglesiasg: did you: rm -rf build13:06
@wikingbecause if u just rebased13:06
@wikingand rerun cmake13:06
@wikingyou might have some cached values there13:06
@wikingso just rm -rf build && mkdir build etc...13:07
@iglesiasgwiking: I didn't, thanks! trying atm13:08
@iglesiasgwiking: still the same, unfortunately13:09
@wikingwell have fun with the FindOctave.cmake script ;)13:11
@sonney2kiglesiasg, the bug is with ParameterMapElement being stored in a Dynarray...13:14
@sonney2kHeikoS, ^13:14
@sonney2kiglesiasg, so this is sth else13:14
@sonney2kHeikoS, do you need ParameterMapElement to be virtual?13:16
@HeikoSsonney2k:  I did not write that13:16
@sonney2kHeikoS, any ideas who did?13:16
@sonney2kHeikoS, what is this for?13:17
@HeikoSof wow13:17
@HeikoSI wrote it13:17
@HeikoSno idea let me check13:17
@HeikoSsonney2k: ah thats migration13:17
@sonney2kHeikoS, if it would work as a CSGObject one could use DYnamicObjectArray13:17
@HeikoSsonney2k: I want to drop that anyways13:17
@HeikoSmigration is useless13:17
@HeikoSwhat problem does it cause?13:18
@sonney2kHeikoS, errm but this means we have no serialization13:18
@HeikoSsonney2k: no13:18
@HeikoSsonney2k: why that?13:18
@HeikoSsonney2k: migration never worked so when did we have serialization?13:18
@sonney2kHeikoS, because when you save data with one version of shogun you won't be able to load it in the next13:18
@HeikoSsonney2k: I think it is way better to just ignore new elements13:18
@HeikoSsonney2k: since that is what usually changes, new members are added13:19
@sonney2kHeikoS, and renames are done...13:19
@HeikoSsonney2k: nobody used the migration framework for something else ever13:19
@HeikoSsonney2k: we cannot support that, it is too much hassle13:19
@HeikoSsonney2k: check the code and you will see, it is horrible13:20
@HeikoSand it is also horrible to use13:20
@HeikoSso I dont know about that13:20
-!- travis-ci [] has joined #shogun13:20
travis-ci[travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []13:20
@HeikoSwe are open-source so why support old formats, companies can do that but I think we cannot13:20
-!- foulwall [~user@] has joined #shogun13:20
@HeikoSor re-write parameter framework with migration in mind13:20
@sonney2kHeikoS, well if you cannot maintain it then we better drop it13:21
@HeikoSsonney2k: strongest reason to drop it is that it is not used13:21
@sonney2kit is only a convenience for users13:21
@HeikoSsonney2k:  and that it doesnt really work13:21
@HeikoSyeah, I mean it would be nice to have13:21
@HeikoSbut not in the way it is currently done13:21
@HeikoSsonney2k:  and I would also drop the integration tests for that reason13:22
@sonney2kHeikoS, no way13:22
@HeikoSsonney2k: since they always break and are hard to replace13:22
@HeikoSrather unit test results13:22
@sonney2kHeikoS, they are worth more than the unit tests13:22
@HeikoSsonney2k: currently yes, in a year, I would say no13:22
@sonney2kHeikoS, currently unit tests work but lots and lots of integration tests are broken13:22
@sonney2kand I mean really not working13:22
@HeikoSthey dont test the correct thing13:22
@sonney2kHeikoS, what is the correct thing then?13:23
@HeikoSto assert correct results13:23
@sonney2kif not checking that we have corerct results13:23
@HeikoSthe integration tests include that, but so much more useless stuff13:23
@HeikoSso they kind of block development13:23
@HeikoSsince you always have to update them13:23
@HeikoSI would prefer if all integration tests were replaced by simple unit tests and then the examples would be changed to be more illustrative13:24
@HeikoSthis way we have the same testing power, easier development and examples that are useful rather than being abused for testing13:24
@sonney2kHeikoS, what?13:24
-!- az_de [82954e22@gateway/web/freenode/ip.] has joined #shogun13:24
@HeikoSsonney2k: some work involved but thats what I would find ideal13:24
@sonney2kHeikoS, with the same argument unit tests block development13:25
@sonney2kHeikoS, and examples13:25
@HeikoSsonney2k: no13:25
@sonney2kyou always need to update them13:25
@HeikoSsonney2k: if you change a class, say add a member the unit tests dont break13:25
@HeikoSsonney2k: they just assert correct mathematical results13:25
@HeikoSwhich are invariant of the implementation (more or less)13:25
@HeikoSbut all class changes dont affect them13:25
@sonney2kHeikoS, same with integration tests13:26
@HeikoSsonney2k: no, the tests break if you add a member13:26
@HeikoStake for example labels13:26
@sonney2konly if you add sth to the serialization framework13:26
@sonney2kand then it is actually good13:26
@HeikoSsonney2k: yeah13:26
@wikingsonney2k: wtf is the purpose of
@sonney2kbecause you want that tested too13:26
@HeikoSsonney2k: thats why we dont do that, we have all these warnings: add parameters13:26
@HeikoSand parameters are not added since they break the tests13:26
@HeikoSwhich is annoying13:26
@sonney2kHeikoS, but that is bullshit13:27
@HeikoSand doesnt help detecting errors at all13:27
@sonney2kwe can just check if everything is still fine13:27
@sonney2kthen run <>13:27
@sonney2kand commit the new data13:27
@HeikoSsonney2k: most developers (including me) have problems updating them13:27
@HeikoSand it is very annoying13:27
@HeikoSsee my point is just: if we replace them, we have the same test power13:28
@HeikoSbut minus all the hassle13:28
@sonney2kwell maybe giving you a script doing that would help13:28
@sonney2kit is just 3 lines of git commit -a / push / commit push13:28
@HeikoSsonney2k: but what do we gain from the serialisation stays fixed test?13:28
@wikingsonney2k HeikoS we can create a pre-commit hook ;)13:28
@sonney2kHeikoS, we gain that the *whole* pipeline is tested13:28
@sonney2kHeikoS, unit tests don't cover bigger stuff13:29
@HeikoSsonney2k: why is that helpful?13:29
@HeikoSfor example?13:29
@sonney2kHeikoS, because several small units may work13:29
@sonney2kbut their integration might not be well tested13:29
@HeikoSfor example?13:29
@sonney2kHeikoS, everyone in industry is doing that13:29
@HeikoSmost integration test do not test bigger stuff13:29
@sonney2kHeikoS, what for example?13:29
@HeikoSexample of such a situation where the big thing cannot be tested by unit tests?13:30
@HeikoSsonney2k:  I would rather test that all steps of the pipeline work correctly than testing the final byte-result13:30
@HeikoSbecause then you know what broke if something breaks13:31
@HeikoSalso in shogun we dont have large parts of framework parts interacting with each other13:31
@HeikoSrather multiple small algorithms13:31
-!- foulwall` [~user@] has joined #shogun13:31
@sonney2kHeikoS, sure that is what unit tests are for13:31
@HeikoSsonney2k: I agree with you that integration tests are useful if you have such big frameworks, but we dont, look at our integration tests, they can all be replaced units13:32
@HeikoSsonney2k: for example all these kernels13:32
@sonney2kHeikoS, yeah but why would one do that? You need to write an example per algorithm anyways13:33
@sonney2kand you get a test *for free*13:33
@sonney2kand it is very readable to users and a test13:33
@HeikoSsonney2k: I think they are causing more trouble than they are useful13:33
@HeikoS(given that they are replaced by units)13:33
-!- foulwall [~user@] has quit [Ping timeout: 268 seconds]13:34
@HeikoSsonney2k: also they dont tell you what is wrong when they fail, just THAT its wrong13:34
@sonney2kso far I have seen them only to detect bugs we would not have detected13:34
@sonney2kHeikoS, yeah but you can then just run -d and you get a shell where you can investigate the objects that are different13:34
@HeikoSsonney2k: its so much easier to get a failing assertion13:35
@sonney2kHeikoS, I guess you didn't notice that GaussianProcessRegression was no longer working since your last commit did you?13:35
@sonney2kHeikoS, the integration test failed and so I could have a look and fix it13:35
@HeikoSwhat failed?13:35
@HeikoSthe python example or the test?13:35
@sonney2kGaussianProcessRegression was no longer available13:35
@sonney2kin modular interfaces13:35
@HeikoSso the example failed13:35
@HeikoSnot the test13:35
@sonney2kthe integration test13:36
@sonney2kthe example works13:36
@HeikoSrunning the examples is good13:36
@HeikoSwhy does it work if the class is not available?13:36
@sonney2kif eigen3 is not available the example will give an OK13:36
@HeikoSand if it is available?13:36
@HeikoSif eigen is not available the integration test cannot be ran13:37
@sonney2kshould fail but didn't check13:37
@sonney2kHeikoS, you can look up the failure yourself in travis13:37
@HeikoSthese type of api-interface problems can be detected by running the examples13:37
@HeikoSno need for comparing serialised objects for that13:37
@sonney2kin addition it is also a good test for serializtion13:38
@HeikoSsonney2k: yep agree, but that will be handled by automatic unit tests soon in a way better way (all classes etc)13:38
@HeikoSlike clone and equals13:38
thoralfHeikoS: What happens if someone add members to the serialization (i.e. the parameters i guess?)13:38
@HeikoSthoralf: the integration test breaks13:39
thoralfHeikoS: No, technically.13:39
@HeikoSit is been written to the file upon save, so gives a different binary object on disc13:39
thoralfAnd the binary objects will be compared?13:40
@HeikoSthoralf: we want to replace this by an equals instead13:40
thoralfIs there a text-representation to compare with diff?13:40
@HeikoSthoralf: yes13:40
@sonney2kHeikoS, so back to my original question: we can drop the parametermap* class right?13:40
thoralfSo why don't do this?13:40
@HeikoSsonney2k: migration will stop working immediately13:41
@sonney2kthoralf, we do this13:41
@HeikoSsonney2k: So I would rather discuss this a bit more13:41
@HeikoSsonney2k:  see my git issues on dropping migration13:41
thoralfsonney2k: HeikoS just told you're comparing the binary.13:41
@sonney2kthoralf, no on travis you see a ascii diff of what has changed13:41
@sonney2kso you could see that a variable is added/removed/changed13:41
@sonney2kalso when sth crashes13:42
@sonney2kyou will get a gdb backtrace13:42
@HeikoSsonney2k: I would only drop migration if integration is changed, because otherwise its too painful to change things13:42
@HeikoSsonney2k: but I dont know, I would like to discuss this with others also before actually dropping things13:42
@HeikoSsonney2k: so rather not drop now but after GSoC or so13:42
@sonney2kHeikoS, but didn't you just say that migration is not used?13:43
* sonney2k gets confused13:43
@HeikoSsonney2k: I think so, but not sure, its deep down in the parameter framework so dropping classes can be dangerous13:43
@sonney2kI mean yes integration tests fail as soon as some variable name is changed/added/removed/meaning is changed/13:44
@HeikoSThere are some add calls on the migration maps done by some classes13:44
thoralfHeikoS: Is there a way to recover if a parameter was added?  Lets say: "load serialization; apply default parameters; save serialization"?13:44
@HeikoSthoralf: thats what migration does13:44
@sonney2kthoralf, yes that would work13:44
@sonney2kthoralf, also the other way around if some parameter is missing it will just not be loaded13:44
@HeikoSthoralf: but we would need a person working full time on that to be able to maintain it, its hard stuff to do13:44
thoralfHeikoS: Okay, missed that.  Too much text in this thread now. ;)13:44
@HeikoSok back to NIPS :)13:45
@sonney2kHeikoS, thoralf migration rather does the conversion - so you could *reanme* a variable and it could still work13:45
@sonney2kor you could change its meaning13:45
@sonney2kand add some code to load older revisions13:45
@HeikoSsonney2k: or type or even more things13:45
@sonney2kHeikoS, I need a short term fix then13:46
@HeikoSsonney2k: I think its too ambitios13:46
@HeikoSsonney2k: whats the problem?13:46
@HeikoSsonney2k: the parameter stuff does not to be mapped with swig btw13:46
@sonney2kParameterMapElement is in a DynArray13:46
@sonney2kand since it has virtual methods -> malloc/realloc -> kaboom13:46
@sonney2kso either drop these methods or make it a derivative of CSGObject13:47
@sonney2kand use DynamicObjectArray13:47
@HeikoSsonney2k: why did that pop up now?13:47
@HeikoSand not before?13:47
@sonney2kno idea13:47
@HeikoSwhat fails?13:48
@sonney2kor maybe I am misinterpreting the error13:48
shogun-buildbotbuild #1131 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Roman Votyakov <>13:48
@HeikoSsonney2k:  I did not change anythjing in there13:48
@sonney2kHeikoS, iglesias posted it here
thoralfsonney2k: Now I'm confused.  According to HeikoS it's a lot of work, but you're suggesting it can be done automatically.  Yes/no/maybe? ;)13:49
thoralfi.e. 4 lines of shell code13:50
@sonney2kthoralf, what?13:50
@sonney2kthoralf, what do you mean?13:50
thoralfsonney2k: The migration of serialized examples.13:50
@HeikoSsonney2k: what code caused this?13:50
@HeikoSsonney2k: I am pretty sure this can be solved without dropping the class13:50
@sonney2kthoralf, errm I am still not following13:51
@sonney2kHeikoS, running the *structur*py examples13:51
@HeikoSsonney2k: since when is this?13:52
@sonney2kthoralf, let me explain a bit more13:52
@HeikoSwhat patch broke them?13:52
@sonney2kHeikoS, no idea. I did dig to several bugs until this one popped out13:52
@sonney2kthoralf, we have examples/*/python_modular13:52
@sonney2kthoralf, these are written in a way that these are functions returning some stuff13:53
@HeikoSsonney2k: this is probably some add call to the migration map, but it worked before, so should not be solved this way13:53
@sonney2kthoralf, so a function can be called with certain parameters and the returned result is pickled to disk by a script to create reference data. if among this stuff are shogun objects these are serialized with shogun's framework13:54
@sonney2kthoralf, now when we run integration tests with the example is run again and results are compared with the serialized stuff13:55
@sonney2kHeikoS, which way?13:55
thoralfsonney2k: Yes, and HeikoS point is that it's hard to maintain the serialized stuff.  Right?13:55
@sonney2kthoralf, well one just has to run <> and then git commit in shogun-data & push and git add data in shogun and push13:56
@HeikoSsonney2k: dropping parameter map, too risky for now, and I dont have time to go into it and fix/drop things properly currently, so I propose we do this after GSoC in a clean way and fix this bug in anther way, for exampling by finding our what caused it13:56
@HeikoSthoralf: migration is hard to maintain, one needs full time people for that13:56
@sonney2kHeikoS, not sure what you are referring to - watching that stuff is changed in a compatible way or code?13:58
@sonney2kor both?13:58
@HeikoSsonney2k: fixing the bug without dropping parameter map13:58
@HeikoSdrop migration after gsoc13:58
@HeikoS(since dropping it might break things)13:58
@HeikoS(and currently too  overloaded to deal with that)13:59
@sonney2kyeah no that was I mean wrt what you said to thoralf13:59
@HeikoSsonney2k: ah sorry14:00
@HeikoSwhat? :)14:00
@HeikoSwhats the question?14:00
@sonney2kHeikoS, thatmigration is hard to maintain, one needs full time people for that14:01
@sonney2kHeikoS, so for watching that stuff is changed in a compatible way or code?14:01
@sonney2k<sonney2k> or both?14:01
@HeikoSsonney2k: yeah, its beyond our manpower, if we had some full time developers, we could do that, like a company14:01
@HeikoSsonney2k: no the idea of loading older file formats14:01
@HeikoSaka migration14:01
@sonney2kI just don't understand what requires the manpower here14:01
thoralfHeikoS: But which part requires the man power?14:02
@sonney2kyeah that is my question14:02
@wikingfucking hell14:02
@HeikoSre-writing the parameter framework in such way that migration is easy ;)14:02
@sonney2kwiking, :D14:02
@HeikoSwiking: haha :)14:02
@wikingthis conversation that you have14:02
@wikinghere for the last 30 mins14:02
@wikingis fucking pointless14:02
@wikingthere are far more urgent things14:02
@wikingand there are real concrete problems14:02
@sonney2kwiking, I only wanted to fix a bug14:02
@HeikoSactually, yes14:02
@wikingwould be great if you could concentrate on that14:02
thoralfwiking: Yeah. ;)14:02
@wikingsonney2k: get it... but still this is too mcuh14:03
@wikingsonney2k: for a bugfix14:03
@HeikoSsonney2k: so pls fix it without dropping things for now, I will take care of that later14:03
@sonney2kHeikoS, yes yes yes!14:03
@wikinghow the fuck can happen this14:03
@HeikoSsonney2k: since it worked before, that must be possible14:03
@sonney2kHeikoS, yes yes yes ...sir!14:03
@wikingUsers/wiking/shogun/examples/undocumented/java_modular/ set_feature_matrix(org.shogun.RealMatrix) in org.shogun.RealFeatures cannot be applied to (org.jblas.DoubleMatrix) feats_train.set_feature_matrix(traindata_real);14:03
@wikingdid i fuckup something in the modular generation?14:03
@sonney2kwiking, I recall seeing things like this (2 years ago)14:05
@sonney2kso no clue14:05
@wikingiglesiasg: can u test some of the examples of python_modular with the cmake generated python_modular?14:05
@wikingyou'll have to do that by hand, as the make tests is not working yet14:05
@wikingiglesiasg: if there's any errors just paste it into the common cmake bug14:06
@sonney2kfoulwall`, did you manage to put the data into a .zip?14:06
@iglesiasgwiking: so cmake for python modular, make and then in src make check-tests?14:06
@sonney2kwiking, btw are you only working on the Cmakefiel.txt?14:07
@wikingiglesiasg: nonono as said that will not work14:07
@wikingsonney2k: yes indeed14:07
@sonney2kwiking, I mean if so you should just do that in develop14:07
@sonney2kyou cannot break a build until we switch14:07
@wikingsonney2k: need a rebase still in that branch14:07
@iglesiasgwiking: well I don't get why not, I can sudo make install and the make check-tests should use the python that is intalled in the system, or??14:07
@wikingiglesiasg: cmake with python_modular14:07
@wikingiglesiasg: dude....14:08
@wikingiglesiasg: make check-tests is based on ./configure shit14:08
@iglesiasgwiking: all right :)14:08
@wikingso just fucking try to run a python_modular example14:08
@sonney2kiglesiasg, which example fails again?14:09
@iglesiasgwiking: lol much coffee this morning? :P14:09
@sonney2kiglesiasg, AdReNALiN!!14:09
@wikingand make it sure that it try to use the python_modular that is generated/installed by cmake ;)14:09
@wikingiglesiasg: yeah + no food14:09
@wikingiglesiasg: i'm on this shit since 5am14:09
@wikingi want to finish14:09
@wikingsince it's really near14:09
@sonney2kwiking, do it!14:09
@iglesiasgsonney2k: the valgrind trace above was for structure_plif_hmsvm_bmrm.py14:09
@sonney2kwiking, and merge it dammit14:09
@wikingi have to go away now for 1 hour14:09
@wikingiglesiasg: if u have some output let me know via the bug14:10
@wikingiglesiasg: dont want to browse the irclog14:10
@wikingiglesiasg: just paste it to github14:10
@iglesiasgwiking: will do14:10
@wikingthjnx heaps14:11
@wikingsonney2k: your c++11 checker in ./configure only supports gcc14:12
@wikingsonney2k: but no worries i've fucking nailed it in cmake14:12
@sonney2kwiking, no it should also work with clang14:13
@sonney2kand any other comp14:13
@wikingsonney2k: no it wont14:13
@wikingsonney2k: as -stdlib=libc++ is missing from the compiler flags14:13
@wikingand w/o that it'll die with clang14:13
@wikingwont find <atomic>14:14
@sonney2kwiking, errm but it is unsing cxx_check so that should be in there14:14
@wikinglisitsyn: your spinlock checker with cmake fails on travis.... cannot find it...14:15
@sonney2kiglesiasg, I don't that error14:16
@iglesiasgsonney2k: what?14:16
@sonney2kiglesiasg, I just ran valgrind python structure_plif_hmsvm_bmrm.py14:17
@iglesiasgsonney2k: mmm that's funny14:18
@sonney2kI see some leaks but these are rather due to pytnon not calling exit_shogun()14:18
@iglesiasgsonney2k: I run that yesterday night with the last commit message being "define M_PI if not available"14:19
@sonney2kiglesiasg, weird... same version here14:19
@sonney2kvan51, any updates or still asleep?14:20
@sonney2kor beaching :)14:20
van51sonney2k: hey I was waiting for you guys to finish :)14:20
@iglesiasgsonney2k: what about structure_discrete_*?14:20
van51sonney2k: I have sent a PR for normalization in the converter14:20
van51sonney2k: I think it's fine14:20
van51sonney2k: even on the webspam they give very close results14:21
@sonney2kvan51, did it give the same results?14:21
van51sonney2k: w/o normalization y14:21
van51sonney2k: but the converter normalizes every dimension of the vector14:21
van51sonney2k: while the dense dot normalizes the end result14:21
van51sonney2k: so there must be rounding errors or something14:21
@sonney2kvan51, yeah sure but it should be the same up to flaoting point precision issues14:22
@sonney2kvan51, well you could just compare things - get the feature vector from dotfeatures and the converted one14:22
@sonney2kthey must be exactly the same14:22
van51sonney2k: I have added a test case doing that14:22
van51sonney2k: but by using the dot product14:23
@sonney2kiglesiasg, no leaks or anything14:23
@sonney2kiglesiasg, with that script...14:23
@sonney2kvan51, dot is not a good way for comparison though there will be minimimal pertubations14:24
@sonney2kas in 1e-14 or so14:24
@sonney2kvan51, did you compute the linear kernel for that? what was the deviation?14:24
@sonney2kiglesiasg, maybe you should check again then14:25
@iglesiasgsonney2k: yeah14:26
@iglesiasgsonney2k: I will make an issue in case they keep on appearing14:27
@sonney2kiglesiasg, and assign it to yourself :D14:28
@sonney2kohh nice strong winds - sea I am coming14:30
van51sonney2k: they gave the same results14:31
van51sonney2k: I'm trying to reach my remote machine to copy the results but I can't :/14:31
van51sonney2k: also by tonight I will have a PR for "quadratic" support in the hashedDoc class14:32
van51sonney2k: what I'm doing actually is a k-skip ngram strategy14:32
-!- van51 [] has quit [Read error: Connection reset by peer]15:29
-!- van51 [] has joined #shogun15:31
-!- az_de [82954e22@gateway/web/freenode/ip.] has quit []15:39
-!- van51 [] has quit [Remote host closed the connection]15:46
-!- shogun-notifier- [] has quit [Quit: transmission timeout]15:50
-!- iglesiasg [] has quit [Quit: Leaving]15:54
-!- van51 [] has joined #shogun15:59
-!- van51 [] has quit [Quit: Leaving.]16:07
-!- pickle27 [] has joined #shogun16:18
-!- iglesiasg [~iglesias@2001:6b0:1:1da0:8c2b:7874:f378:ae37] has joined #shogun16:34
-!- mode/#shogun [+o iglesiasg] by ChanServ16:34
@iglesiasgsonney2k, around?16:34
@iglesiasgsonney2k, the valgrind errors still remain. Just created an issue, let me know about it16:42
-!- van51 [] has joined #shogun16:53
@wikingiglesiasg: any luck?16:59
@iglesiasgwiking, yes, I run a few examples without any trouble16:59
@wikingiglesiasg: cool thnx then i can add that to the tests straight away16:59
@wikingiglesiasg: can u tell me what's the output on your machine of spinlock check with cmake17:26
@iglesiasgwiking, yes, sure17:26
@iglesiasgwiking, is it enough with the entry that can be read with ccmake?17:26
@wikingjust run a cmake ..17:27
@wikingit should be there17:28
@wikingregardless of your cmake state17:28
@iglesiasgwiking, all right, give me a second. I am with the uni laptop right now and didn't have hear a copy of the repo to play with cmake17:28
@iglesiasgwiking, -- Spinlock support found17:29
@iglesiasgI guess that is not very useful17:30
@iglesiasgor maybe it is, you tell me :)17:30
@wikingthat's what i needed17:30
@iglesiasgwiking, I am going to try octave here too. Let's see what happens17:30
@HeikoSvotjakovr: I think evaluate_log_probabilities should be renamed to something that contains "predictive distribution"17:30
@wikingiglesiasg: cool17:31
@HeikoSvotjakovr: since it is a bit misleading17:31
votjakovrHeikoS: yep, i agree17:32
votjakovrHeikoS: btw is it ok for example:
@HeikoSvotjakovr: very nice! for what is that?17:34
@HeikoSgotta go, will be back in 1517:34
votjakovrHeikoS: this is python example for laplace logit/probit classifiers17:35
@iglesiasgwiking, we have a new hint: octave 3.2 here was correctly detected17:35
@wikingiglesiasg: ok that's something17:38
@wikingbut now try to build it17:38
@wikingi had problems there17:38
@iglesiasgwiking, already 25% done17:38
@iglesiasgwiking, is there an option to explicitly desactivate doxygen generation?17:39
-!- van51 [] has quit [Quit: Leaving.]17:44
@iglesiasgwiking, make worked fine, make install afterwards too but octave_modular examples fail17:44
@iglesiasgwiking, init_shogun undefined17:44
@iglesiasgwiking, I noticed this when running make install17:44
@iglesiasgto me it looks weird that liboctave appears inside a lua dir, but who knows17:45
pickle27iglesiasg: lisitsyn my PR is up to date with the style changes17:50
@wikingiglesiasg: woah17:50
@wikingiglesiasg: ok make install is not working well17:50
@wikingiglesiasg: we need there somekind of a magice17:50
@wikingto figure out the octave install dir17:50
@iglesiasgwiking, so the lua thing is not normal17:51
@wikingof course not17:51
@wikingbut i didn't have the patients for that17:51
@wikingso that still need to be fixed17:51
@wikingin the cmake17:51
@iglesiasgwiking, all right17:51
@iglesiasgpickle27, nice, let's check that everything is fine with travis and then we can merge17:52
@iglesiasgpickle27, so you solved finally the bugs you mentioned in the original implementation?17:52
@iglesiasggtg, see you later17:53
-!- iglesiasg [~iglesias@2001:6b0:1:1da0:8c2b:7874:f378:ae37] has quit [Quit: Ex-Chat]17:53
@wikingheheh great bug comment "yes Jedi is fixed now"18:03
lisitsynsonney2k: gsomix don't want money :D18:05
lisitsynohh hushell you too, please submit your gsoc thing18:05
lisitsynyou all18:05
lisitsynwiking: what is jedi?18:05
@HeikoSvotjakovr: very nice, could you make that a non-graphical one too?18:06
@HeikoSin fact, I think you dont need to show all the inference methods at once, but just one and then offer replacement lines to switch18:06
@HeikoSthis way, this can also be extended with EP18:07
@wikinglisitsyn: ica realted afaik18:13
votjakovrHeikoS: ok, btw what do you think about plotting predictive distribution? i mean, we can choose the line, say y = 0, and plot mean and 95% confidence intervals on predictive distribution18:19
lisitsynwiking: AH :D18:21
lisitsynI thought some software but now I recall18:21
@HeikoSvotjakovr: but didnt you plot the predictive distribution in the example?18:21
lisitsynfunny because I am ehm co-mentoring that project18:21
votjakovrHeikoS: i did it:) i mean we can plot confidence18:22
@HeikoSvotjakovr: ah I see, I think the way you did it is fine, just include a colorbar and mark the decision surface :)18:23
votjakovrHeikoS: oh, ok. i'd just like to use mean and variance18:29
votjakovrHeikoS: there is p(y=1) on the plot18:30
@HeikoSvotjakovr: ok18:30
@HeikoSvotjakovr: remember that these examples should be simple to demonstrate api usage. No fancy plotting involved18:30
@HeikoSvotjakovr: the cool stuff should go into your ipython notebook example, and there you should do as many cool things as possible :18:31
votjakovrHeikoS: ok18:32
@HeikoSvotjakovr: you know, just to find out how things work (the current examples)18:33
@HeikoSvotjakovr: and the fancy ones: to show off with your cool framework :) to explain how things work, to compare etc etc18:33
@HeikoSbut fancy examples later, now some simple and then EP :)18:33
votjakovrHeikoS: ok, btw didn't you see my "warnings" PR?18:34
votjakovrHeikoS: i've fixed some of them18:35
@HeikoSvotjakovr: I did, sorry, busy day. I will have a look soon18:35
@HeikoSThanks a lot for that!18:35
votjakovrHeikoS: alright :)18:36
-!- thoralf [] has quit [Quit: Konversation terminated!]18:51
-!- shogun-notifier- [] has joined #shogun18:54
shogun-notifier-shogun: Viktor Gal :feature/CMake * 68044ee / / (29 files):
shogun-notifier-shogun: Add C++11 cmake detection scripts18:54
shogun-notifier-shogun: Fix lua modular installation directory18:54
shogun-notifier-shogun: Viktor Gal :feature/CMake * 0ec5151 / .travis.yml:
shogun-notifier-shogun: Travis: use gcc for octave_modular compilation18:54
lisitsynwiking: you're doing great stuff with cmake18:56
@wikingmmm soon soon18:56
@wikingi though this fucking tests should go faster18:56
@wikingi'm having problem reseting the compiler flags at one point in cmake :S18:57
hushelllisitsyn: submitted :)19:13
hushelllisitsyn: Thanks for reminder19:14
-!- travis-ci [] has joined #shogun19:51
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []19:51
-!- foulwall` [~user@] has left #shogun ["ERC Version 5.3 (IRC client for Emacs)"]19:59
-!- van51 [] has joined #shogun20:06
-!- gsomix [~gsomix@] has joined #shogun20:09
gsomixgood evening20:10
gsomixjust returned from big city20:10
gsomix*just have20:10
-!- lisitsyn [] has quit [Quit: Leaving.]20:10
-!- foulwall [~user@] has joined #shogun20:41
-!- lisitsyn [~lisitsyn@] has joined #shogun20:43
foulwallsonney2k , I solved javascript zip extracting, but got some little bug on display the pic, I'm trying to solve it20:43
shogun-notifier-shogun: Roman Votyakov :develop * 1559b33 / examples/undocumented/python_modular/graphical/
shogun-notifier-shogun: add very basic python graphical example of GP binary classification20:55
shogun-notifier-shogun: Heiko Strathmann :develop * ba3603f / examples/undocumented/python_modular/graphical/
shogun-notifier-shogun: Merge pull request #1340 from votjakovr/feature/gp_binary_classification20:55
shogun-notifier-shogun: Added very basic python graphical example of GP binary classification20:55
@HeikoSvotjakovr: hey, you might like this here:
@HeikoSvotjakovr: I use that for my research and it is massively based on your work ! :)21:05
@HeikoSgoing home now, bye!21:05
-!- HeikoS [] has quit [Quit: Leaving.]21:05
@wikinglisitsyn: wazzzaaaaaaaaaaaaa21:07
lisitsynwiking: hey!21:07
lisitsynwiking: any help with cmake? I have some power left - just got home21:08
@wikinglisitsyn: so here i am21:09
@wikinglisitsyn: libshogun examples21:09
@wikinglisitsyn: trying to add all the .cpps as a new add_executable21:09
@wikingwhich would all be good21:09
@wikingapart from the fact that i cannot clear the compiler flags21:09
lisitsynwiking: ahh yeah21:09
lisitsyncould be a problem21:10
shogun-buildbotbuild #1132 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Roman Votyakov <>21:10
@wikingwhich means it's trying to compile the stuff like this21:10
@wikingand obviously this is not what should happen21:10
lisitsynwiking: what about that?21:10
@wikinglisitsyn: yeah tried that21:12
@wikingit appends to the list21:12
@wikingAdditional flags to use when compiling this target's sources.21:13
@wikingThe COMPILE_FLAGS property sets additional compiler flags used to build sources within the target.  Use COMPILE_DEFINITIONS to pass additional preprocessor definitions.21:13
@wikingsee 'additional'21:13
-!- van51 [] has quit [Quit: Leaving.]21:13
lisitsynwiking: yes yes got it21:13
@wikinglisitsyn: so of course there's a way around21:14
lisitsynwiking: how?21:14
@wikingbut that requires a minor refactoring of this fucking thing21:14
@wikinglisitsyn: well create a custom cmake variable and store all the compiler flags and definitions in that21:15
lisitsynwiking: and add them each time?21:15
lisitsynwould be ok I think21:15
@wikingso for the target that requires all those shit21:15
@wikingthen do the set_target_properties21:16
@wikingotherwise dont21:16
@wikingwhich would be 'all the current targets apart the libshogun-examples target'21:16
@wikingi hope u feel the irony21:16
@wikingi.e. for one fucking target we have to redo now everything :)21:16
lisitsynwiking: hah yes21:17
@wikingapart from this if u wanna help21:17
pickle27wiking: lisitsyn is it neccessary to use cmake for the examples?21:17
@wikingpickle27: well it would be good21:17
lisitsynpickle27: what else?21:17
@wikingpickle27: since then we can package them as well21:17
pickle27I seem to recall opencv uses cmake for the lib and a reg makefile for the examples21:17
@wikingpickle27: fuck ocv :D21:18
@wikinglisitsyn: so what i was saying21:18
lisitsynwiking: yes?21:18
@wikinglisitsyn: can u find out how we can add to ctest non-binary tests21:18
@wikinglisitsyn: e.g. a python test or a java test?21:18
@wikingif u know what i mean21:18
@wikingbecause until now i was doing21:19
lisitsynwiking: yeah I'll check21:19
@wikingadd_executable(targetname srcs)21:19
@wikingadd_test(<whatevertestname targetname)21:19
@wikingbut i dont know if for example we can do this21:19
lisitsynwiking: yes that's the common way21:19
@wikingadd_custom_target(OUTPUT? whatever COMMAND PYTHON_EXECUTABLE ...)21:20
@wikingand then21:20
@wikingadd_test(<wtf> whatever)21:20
lisitsynokay I'll try to do that locally21:21
@wikinglisitsyn: nono21:21
@wikinghere's your friend21:22
@wikinglisitsyn: can u test the more advanced add_test macro?21:22
@wikingadd_test(NAME <name> [CONFIGURATIONS [Debug|Release|...]] [WORKING_DIRECTORY dir] COMMAND <command> [arg1 [arg2 ...]])21:22
@wikingthis should do it21:22
@wikingsee COMMAND there21:22
lisitsynyeah that's what I wanted to do21:22
@wikingbut would be great if u could test it21:22
@wikinglet me know how's that working out21:22
@wikingin the meanwhile i'll do the fucking refactoring :)21:23
@wikingas i was googling about this for a long time now (i.e. 3 hours) w/o success -- reseting compiler flags21:23
@sonney2kfoulwall, why javascript zip extracting? why not on the server side serve a file inside a zip?21:31
votjakovrsonney2k: hi! please have a look at my "silence warnings" PR21:38
@sonney2kvotjakovr, did already21:39
@sonney2kvotjakovr, is this for clang warnings?21:40
votjakovrsonney2k: yep21:41
shogun-notifier-shogun: Kevin :develop * b9771de / / (5 files):
shogun-notifier-shogun: added QDiag and Jedi AJD methods21:41
shogun-notifier-shogun: Kevin :develop * 42b7d5b / tests/unit/mathematics/ajd/
shogun-notifier-shogun: added JADiagOrth unit test21:41
shogun-notifier-shogun: Kevin :develop * 4625e4e / / (9 files):
shogun-notifier-shogun: fixed Jedi, added Jedi unit test and some cleanup to all the ajd headers21:41
shogun-notifier-shogun: Kevin :develop * e33f051 / src/shogun/ (7 files):
shogun-notifier-shogun: coding style fixes in AJD21:41
shogun-notifier-shogun: Kevin :develop * b6b17e6 / src/shogun/mathematics/ajd/ (6 files):
shogun-notifier-shogun: ajd double -> float64_t21:41
shogun-notifier-shogun: Kevin :develop * 380249f / tests/unit/mathematics/ajd/ (5 files):
shogun-notifier-shogun: coding style changes in ajd unit tests21:41
shogun-notifier-shogun: Kevin :develop * ee75cd5 / src/shogun/mathematics/ajd/FFDiag.cpp,src/shogun/mathematics/ajd/Jedi.cpp:
shogun-notifier-shogun: ajd updated to cpp style loops21:41
shogun-notifier-shogun: Sergey Lisitsyn :develop * fc13467 / / (18 files):
shogun-notifier-shogun: Merge pull request #1326 from pickle27/develop21:41
shogun-notifier-shogun: added QDiag and Jedi AJD methods21:41
shogun-buildbotbuild #1133 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at  blamelist: Heiko Strathmann <>21:44
-!- zxtx [] has quit [Ping timeout: 245 seconds]21:45
shogun-buildbotbuild #1134 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at  blamelist: Sergey Lisitsyn <>, Kevin <>21:57
shogun-notifier-shogun: Roman Votyakov :develop * d864ec6 / src/shogun/ (7 files):
shogun-notifier-shogun: silence some warnings22:03
shogun-notifier-shogun: Soeren Sonnenburg :develop * 67344b1 / src/shogun/ (7 files):
shogun-notifier-shogun: Merge pull request #1336 from votjakovr/develop22:03
shogun-notifier-shogun: Fix various errors and warnings detected by clang.22:04
@sonney2kvotjakovr, excellent patch!22:04
@sonney2kvotjakovr, more of that kind please :)22:04
votjakovrsonney2k: thank you :) ok, i'll keep going22:05
-!- lisitsyn1 [~lisitsyn@] has joined #shogun22:08
-!- lisitsyn [~lisitsyn@] has quit [Read error: Connection reset by peer]22:08
-!- travis-ci [] has joined #shogun22:12
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 []22:12
@wikinglisitsyn1: ok i think i'm almost there ;)22:13
@wikinggood that i wrote modular interface handling modular in cmake ;P22:13
lisitsyn1wiking: well it works when you just write the command here22:13
lisitsyn1I just tried22:14
lisitsyn1wiking: though we'd have to use sth like $PYTHON22:15
lisitsyn1not just python blabla.py22:15
lisitsyn1I guess22:15
@wikingtyeye of course22:15
@wikingbut that's handled by FindPythonInterperter22:16
@wikingthat defines ${PYTHON_EXECUTABLE}22:16
@wikingso that's all cool22:16
@wikingsame with java and csharp...22:16
@wikingas well as ruby and the others22:16
lisitsyn1add_test(NAME shit COMMAND "${PYTHON_EXECUTABLE} ${TEST_SCRIPT}")22:16
lisitsyn1essentially that's it I think22:17
lisitsyn1may be some paths stuff22:17
@wikingyeye that's alright ;)22:17
lisitsyn1wiking: can I help with anything else here then?22:17
@wikingi'm in the last stage of this refactoring22:17
@wikingyes it worked! \o/22:17
@wikingok this is cooool22:17
@wikingcmake is really amazing ;)22:17
@wikinglisitsyn1: have i told you that i've created a .dmg osx package22:18
@wikingof shogun22:18
lisitsyn1wiking: yeah with cpack22:18
@wikinga native osx package :)22:18
@wikingand it was just out of box22:18
lisitsyn1I am lame with osx22:18
@wikingno worries22:18
@wikingit's just that after this22:18
lisitsyn1how native it is?22:18
lisitsyn1like deb for debian?22:18
@wikingthe package is created by the native package maker of osx22:18
lisitsyn1so you just download it22:19
-!- lisitsyn1 is now known as lisitsyn22:19
@wikingdouble click22:19
lisitsyndouble click or whatever22:19
@wikingand tada22:19
lisitsynand it is here22:19
@wikingyes yes22:19
lisitsynokay it would make Heiko happy22:19
@wikingcd /Users/wiking/shogun/build/examples/undocumented/libshogun && /usr/bin/c++    -g   -o CMakeFiles/basic_minimal.dir/basic_minimal.cpp.o -c /Users/wiking/shogun/examples/undocumented/libshogun/basic_minimal.cpp22:19
@wikingso now i have this ;P22:19
@wikingah shit i have to fix unit-test flags22:20
@wikingjust a sec22:20
@wikingok that's not generated yet with all the compiler flags22:20
@wikinglisitsyn: yeah i mean after this we can setup our own repos for rpm, deb22:20
@wikingand just fucking upload there the nightlies22:21
lisitsynwiking: is there exe?22:21
@wikinglisitsyn: ?22:21
@wikingyou mean if we could use MSVC?22:21
lisitsynwiking: some setup.exe for poor windows users?22:21
@wikinglisitsyn: yes yes22:21
lisitsynI feel sorry people use windows but .exe stuff would be cool still22:22
@wikinglisitsyn: check the screenshots22:23
@wikingi've already defined the components22:24
@wikingso that's done ;P22:24
@wikingmake package creates libshogun libshogun_headers and <whatever>_modular_interface packages22:24
@wikingor of course a monolitic tar.gz/bz222:25
@wikingcouldnt test it yet on a deb machine22:25
@wikingas i dont have access to it now22:25
@wikinganyhow we should have sooooonish the make tests ;P22:25
@wikinglisitsyn: ah if u wanna play around: find some cool cmake scripts on github and elsewhere to detect compiler flags22:26
@wikingyou know for optimizations22:26
@wikinglike is sse2 available et22:26
lisitsynwiking: okay I am on it22:26
lisitsynwiking: well22:26
@wikingfor sure there's heaps22:26
lisitsynI would just let compiler decide22:27
@wikingand i've even saw a cpuid.c based cmake script22:27
lisitsyncompilers are ok already22:27
lisitsynthey know what cpu it is22:27
lisitsynwiking: I think we should drop that thing at all22:27
@wikingso go with the22:27
@wiking-march=native ?22:27
-!- iglesiasg [] has joined #shogun22:27
-!- mode/#shogun [+o iglesiasg] by ChanServ22:27
lisitsynwiking: yeah I'd do that22:27
@iglesiasghello hello22:27
lisitsynwiking: I'll check what native skips22:28
@wikingmmm i wonder what detection i still miss then from ./configure22:28
@wikingok cool i'll finish now the unit testing22:28
lisitsynsonney2k: do you think we need these CPU deals?22:28
lisitsyniglesiasg: hey22:28
@wikingand then go back to make tests22:28
@wikinglisitsyn: ah there's one task... octave_modular22:28
@wikinglisitsyn: where the fuck the created modular interface binary should be installed22:29
@wikingand how we can find that out from cmake :P22:29
lisitsynoh okay22:29
lisitsynlet me check that22:29
-!- votjakovr [] has left #shogun ["Fallen asleep!"]22:31
@wikinglisitsyn: what do u think about having gmock/gtest in cmake with
lisitsynwiking: let me recall how I did that in tapkee22:43
@wikingalthough i should somehow enable the case where one has already the gmock/gtest localhost... so one does not really need to have internet connection for having unit testing22:45
lisitsynwiking: I wasn't very satisfied with gtest finder btw22:46
lisitsynI found a few finders and they all were a bit weird22:47
@wikinglisitsyn: ah yeah they are shit22:47
@wikingi didnt even bothered22:47
@wikinglisitsyn: but as u can see with ExternalProject we can just tell cmake to download the src from
lisitsynwiking: yes22:50
lisitsynwiking: what about trying some default location or user defined dir22:51
lisitsynelse download22:51
@wikingyeah something like that22:53
@sonney2kwiking, lisitsyn yeah -march=native but IIRC we needed -msse2 -msse3 -msse or so too - better check22:54
@wikingsonney2k: what about -fvectorize-tree?22:54
lisitsynwiking: sonney2k: we should revise flags not included to O322:54
@sonney2kand also warning flags22:55
@sonney2ksome of these don't exist on some compilers22:55
@wikingsonney2k: although that is enabled in clang in -O322:56
shogun-buildbotbuild #1135 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at  blamelist: Soeren Sonnenburg <>, Roman Votyakov <>22:56
@wikinglisitsyn: gcc -v -Q -O3 -march=native --help=target22:58
@wikinglisitsyn: all the sse instructions are enabled that are available for my cpu22:59
lisitsynwiking: well all SSEs and AVX22:59
lisitsynand mtune is correct22:59
lisitsynnot all SSEs though23:00
lisitsynwhat is SSE523:00
lisitsynis it supported by any processor?23:00
lisitsynah whatever23:00
lisitsynwiking: anyway I see most of the things are here already23:00
lisitsynso not sure it makes a lot of sense23:00
lisitsynto tune stuff23:00
@wikinglisitsyn: ok let's go with "-march=native"23:01
@wikingand if somebody has a better idea23:01
@wikinghe can contribute23:01
@wikinglisitsyn sonney2k do we want that in DEBUG mode as well?23:01
@wikingor let's go on with the DISABLE_OPTIMIZATION option and then we won't have that23:01
@sonney2kwiking, we also need support for generic cpu flags as in -O2 and no further tuning23:02
@sonney2kwiking, I would always keep -g23:02
lisitsynwiking: I guess that's done through custom targets23:02
lisitsynlike release debug23:02
@wikinglisitsyn: no not really ;)23:02
@wikinglisitsyn: one has to do that23:02
@wikinglisitsyn: realase by default won't have -g i think23:03
lisitsynwiking: we can change it23:03
@wikingand realase will have -O(something)23:03
@wikinglisitsyn: yep indeed23:03
@wikingbut this "-march=native" should be possible to disable23:03
@wikingso i'll add a cmake variable23:03
@wikingor something like that23:03
@sonney2kwiking, please make it have -g always23:03
@wikingsonney2k: o23:04
@sonney2kif someone wants to strip of symbols he can do it later with strip23:04
@sonney2kiglesiasg, I figured that doing23:05
@sonney2k    model = TwoStateModel.simulate_data(num_examples, example_length, num_features, num_noise_features)23:05
@sonney2k    sosvm = DualLibQPBMSOSVM(model, model.get_labels(), 5000.0)23:05
@sonney2kis enough to trigger the issue23:05
@iglesiasgsonney2k: oh, I see23:06
@iglesiasgsonney2k: did you see if it also appears in libshogun interface?23:06
@sonney2kiglesiasg, no didn't23:06
@wikingis this broken23:07
@wikingseems so23:07
-!- pickle27 [] has quit [Quit: Leaving]23:09
@sonney2kiglesiasg, any ideas how this is possible?23:18
@wikingsonney2k lisitsyn would u mind if test would be built by make command?23:19
lisitsynwiking: why?23:19
@wikingone one could disable example build23:19
@wikingcoz it's tricky ;)23:20
@wikingalthough i can add that trick...23:20
@iglesiasgsonney2k: not really, this was actually the party I studied and fixed something last week23:21
@iglesiasgsonney2k: it seems I have to check it further23:21
lisitsynwiking: crazy!23:21
@sonney2kiglesiasg, I mean doing23:22
@wikinglisitsyn: indeed23:22
@sonney2k    CTwoStateModel* tsm = new CTwoStateModel();23:22
@sonney2k    CHMSVMModel* model = tsm->simulate_data(100,250,10,2);23:22
@sonney2k    CStructuredLabels* labels = model->get_labels();23:22
@sonney2k    CDualLibQPBMSOSVM* sosvm = new CDualLibQPBMSOSVM(model, labels, 5000.0);23:22
@sonney2k    model = TwoStateModel.simulate_data(num_examples, example_length, num_features, num_noise_features)23:22
@sonney2k    sosvm = DualLibQPBMSOSVM(model, model.get_labels(), 5000.0)23:22
@sonney2kshould not make a difference23:22
@iglesiasgsonney2k: aah ok that the issue doesn't appear in C++23:23
@iglesiasgsonney2k: I read as you didn't check that23:23
@sonney2kiglesiasg, sure I did23:23
@sonney2kiglesiasg, but it does not appear23:23
@sonney2kbut just python23:23
@sonney2kbut why???23:24
@iglesiasgit shouldn't matter indeed23:24
@iglesiasgsonney2k:  do you find weird the model.get_labels() thing?23:24
@iglesiasginstead of doing something like23:24
@iglesiasgLabels labels = model.get_labels()23:24
@iglesiasgDualLibQPBMSOSVM(model, labels, 5000.0)23:25
@iglesiasgarggh, labels = model.get_labels()23:25
@wikinglisitsyn: so it's either the trick or the default stuff... i think on some level having the example built by default will somehow make people aware if there was really something going wrong with the modifications23:27
lisitsynwiking: yeah23:29
lisitsynmay be it makes sense to build everything23:29
@wikinglisitsyn: i mean if u disalbe BUILD_EXAMPLES variable in cmake then one can turn it off ;)23:32
@wikingbut i'll leave that option turned ON by default23:32
-!- zxtx [] has joined #shogun23:35
@wikingok i don't know how to make make test more verbose23:36
lisitsynwiking: what verbosity?23:40
@wikingok no worries i've figured it out23:44
@wikingwoah output looks cool23:45
lisitsynwiking: like ctest -VV?23:45
@wikingsolution for more verbose log:
lisitsynwiking: ctest can also send reports and etc23:47
lisitsynno idea if we can use it23:48
@wikinglisitsyn: yeah i saw23:48
@wikingwell at least we have the option for it23:48
@wikingnone the less23:48
@wikingit's working23:48
@wikinglisitsyn: do u know how to group ctest?23:48
lisitsynwiking: group?23:48
@wikingso that i have like libshogun-example and then under that i have the various tests23:48
lisitsynno I don't know how this is done23:49
@wikingnevermind it's good like this23:49
@wikingi'll add all the example runs with ctest23:49
@wikingas well as integration test23:50
-!- lisitsyn [~lisitsyn@] has left #shogun []23:51
--- Log closed Thu Aug 01 00:00:56 2013