Open in new window / Try shogun cloud
--- Log opened Sat Feb 16 00:00:14 2013
--- Day changed Sat Feb 16 2013
n4nd0sonney2k: where is that address from?00:00
@sonney2kn4nd0, you need to get the full trace00:08
n4nd0let me try again00:08
@sonney2kEpisode.IV / A.NEW.HOPE00:08
n4nd0nothing here00:09
@sonney2kstar wars!00:09
n4nd0but I have seen something like through telnet some time00:09
@sonney2kthe guy who did it00:10
n4nd0sonney2k: not around any more I guess?03:08
-!- n4nd0 [] has quit [Quit: leaving]03:10
-!- cavalier- [~cavalier-@] has quit []03:44
-!- FSCV_ [~FSCV@] has quit [Quit: Leaving]05:15
-!- lambday [0e8b6149@gateway/web/freenode/ip.] has joined #shogun07:54
lambdayhello, this is soumyajit again.. I have a few queries regarding the "fast food" idea... I was unable to see the video but I went through the paper that talks about introducing Taylor features to approximate the kernel...08:08
lambdayfirstly, they apply the method for Gaussian kernels but says "a similar approach could also work for other kernels which depend on distances or inner products between feature vectors, e.g. the sigmoid kernel."... does that mean all CDotKernels in shogun?08:09
lambdaysecondly, how do you plan to implement this? say, for Gaussian kernel, extending it to another class that uses this method? and do it for all the kernels that the method applies to?08:10
lambdaythirdly, where can I look for some paper reference of the "fast food" talk?08:23
-!- sumit [73f91219@gateway/web/freenode/ip.] has joined #shogun08:34
blackburnlambday: it is only a draft and more references will come once mentor of that idea describe it more09:25
lambdayblackburn: okay.. thanks :)09:26
-!- sumit [73f91219@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:26
-!- lambday [0e8b6149@gateway/web/freenode/ip.] has left #shogun []09:34
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has joined #shogun10:05
-!- pari [6eeaec2a@gateway/web/freenode/ip.] has joined #shogun11:21
-!- flxb [] has joined #shogun11:41
wikingi'm wondeirng about this fast food11:55
wikingas there are other approaches as well11:55
wikinglike from vedaldi11:55
wikingHKM ++11:56
wikingso to say ;)11:56
parihi all, I have just finished installing shogun. But i am not sure if everything is installed correctly. Is there any way to check it?11:57
wikingpari well try to run an example? :)11:57
wikingblackburn: ping11:57
pariok, thanks.. let me see11:59
parii couldn't get the example running i am getting the following error: basic_minimal.cpp:(.text+0x29): undefined reference to `shogun::init_shogun(void (*)(_IO_FILE*, char const*), void (*)(_IO_FILE*, char const*), void (*)(_IO_FILE*, char const*), void (*)(bool&, bool&))' basic_minimal.cpp:(.text+0x2e): undefined reference to `shogun::exit_shogun()' collect2: ld returned 1 exit status12:32
pariwhat could be possibly wrong?12:32
wikingi reckon the path where the shogun library is installed (if u did make install) is not in the library path12:51
-!- pari [6eeaec2a@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]12:54
-!- flxb [] has quit [Ping timeout: 244 seconds]12:56
-!- n4nd0 [] has joined #shogun13:18
-!- n4nd0 [] has quit [Quit: leaving]13:37
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]13:37
-!- n4nd0 [] has joined #shogun15:28
n4nd0sonney2k: around?15:40
n4nd0sonney2k: proof of concept for the website
n4nd0I have isolated there the addition of the new bar15:44
n4nd0let me know if you were thinking of something like that15:44
blackburnwiking: yeah fast food is not the only way16:16
-!- flxb [] has joined #shogun16:26
-!- n4nd0 [] has quit [Quit: leaving]16:33
-!- flxb [] has quit [Ping timeout: 252 seconds]16:45
-!- heiko1 [] has joined #shogun16:48
@sonney2kheiko1, hey there - I have a question about serialization again20:13
@sonney2kwhen an array of SGVectors is de-serialized20:13
@sonney2kthen the serialization code will call SG_MALLOC(SGVector<type>, len) right?20:14
@sonney2kwiking, I just did make valgrind in tests/unit20:14
@sonney2kguess what - ocas examples have uninited memory reads20:14
@sonney2kso no wonder we see weird stuff20:14
blackburnsonney2k: is that mcocas?20:32
wikingsonney2k: lol?20:33
blackburnsonney2k: so libocas? I tested mc one before IIRC20:59
-!- shogun-notifier- [] has joined #shogun21:06
shogun-notifier-shogun: Sergey Lisitsyn :master * 749d966 / .gitignore:
shogun-notifier-shogun: Ignore gcda file extension21:06
shogun-notifier-shogun: Sergey Lisitsyn :master * c4d485e / src/shogun/lib/tapkee/tapkee_shogun.cpp:
shogun-notifier-shogun: Fixed totally wrong tapkee integration feature access callback21:06
-!- in3xes [~in3xes@] has joined #shogun21:21
shogun-notifier-shogun: Sergey Lisitsyn :master * 4091d89 / src/Makefile.template:
shogun-notifier-shogun: Added gcda for clear target21:30
shogun-notifier-shogun: Sergey Lisitsyn :master * 9860c73 / src/shogun/converter/LaplacianEigenmaps.cpp:
shogun-notifier-shogun: Removed unnecessary and memory-wise wrong distance initialization in laplacian eigenmaps21:30
@sonney2kwiking, please run make valgrind in unit tests there are also errors in LatentModel21:48
shogun-notifier-shogun: Soeren Sonnenburg :master * 99a8236 / src/shogun/.project:
shogun-notifier-shogun: add eclipse project file21:54
shogun-notifier-shogun: Soeren Sonnenburg :master * 2b60dbe / / (9 files):
shogun-notifier-shogun: Fix serialization of Sparse features21:54
shogun-notifier-shogun: - remove inplace (de-)constructors for SGSparse21:54
shogun-notifier-shogun: - add malloc support for SGSparseVector<T>21:54
shogun-notifier-shogun: - enable unit test for sparse feature serialization21:54
heiko1sonney2k, eclipse project file?21:55
@sonney2kheiko1, yeah21:55
@sonney2kwith own shogun formatter :D21:56
heiko1sonney2k, array of SGVectors, mmmh cannot really remember got to check the code21:56
@sonney2kheiko1, I managed to fix this21:56
@sonney2kheiko1, I did some magic to SG_MALLOC21:56
heiko1sonney2k, that is cool, so I can just always press ctrl-shift-f and formatting will be fine?21:56
heiko1sonney2k, cool!21:56
@sonney2kwhenever you do SG_MALLOC(SGVector<T>, len) then it will call new SGVector<T>[len]();21:57
@sonney2kthis will call the inplace constructor automagically21:57
@sonney2kand SG_FREE() will know what to do too21:57
heiko1I see21:57
@sonney2kso this will e.g.make dynarray of SGVector work etc21:57
heiko1btw thanks for fixing this base-class problem21:58
@sonney2kslightly stupid to have so many sg_malloc functions now but this fixes things at least in a nice way for the user21:58
@sonney2kunderneath it is hacky though21:58
heiko1yeah, agree, this is a good solution21:58
@sonney2kbut I don't see a way around this if we don't want to switch to new[] / new()21:59
@sonney2kwhich is bad because we then can no longer interact with C malloc'd memory directly21:59
heiko1nono, agreed on this21:59
@sonney2k(no more realloc / direclty using python matrix memory etc)21:59
@sonney2kheiko1, so while we are at it - serialization I mean - what would it cost to support SGMatrixList / SGNDArray?22:00
blackburnsonney2k: so you like working in eclipse now?22:00
heiko1sonney2k, you mean how much effort that would be?22:00
heiko1Can't really tell, these things tend to be a bit unpredictable22:01
heiko1For what do we need those?22:01
heiko1anything in particular?22:01
@sonney2ke.g. serialization of QDA22:01
@sonney2kbut yes in general I would like to have serialization support for all SG* datatypes22:02
heiko1yes true22:02
heiko1dont know, at some point just has to take the pain :)22:02
heiko1sonney2k, btw I should this week finish my MMD stuff, so I can spend some time on doing things22:02
heiko1but whats top priority in your opinion?22:02
@sonney2kheiko1, getting shogun 2.1 out22:03
heiko1there is also this svm C thing22:03
heiko1yeh regarding that22:03
@sonney2kI don't know if we should add new features...22:03
@sonney2kmaybe better fix warnings / bugs etc22:03
heiko1I will focus on that then22:04
heiko1also a bit documentation22:04
heiko1there are also still some bugs in the streaming features22:04
heiko1and unit tests would also be good22:04
@sonney2kheiko1, if you find bugs you should put them to github issues22:05
heiko1so far I did all these "tests" in the examples22:05
heiko1sonney2k, I know, the ones I mentioned are in there in fact22:05
@sonney2kthen we can write a unit test for that and fix the bug22:05
heiko1I am really happy about the unit tests22:05
@sonney2kheiko1, so what would be needed for SGNDArray & serialization?22:05
@sonney2kI mean I think we need a new type or?22:06
@sonney2kheiko1, btw eclipse is dog slow... but it is very easy to jump around in tons of code / do refactoring22:06
heiko1sonney2k, I dont know currently, would have to look in the code and think about it, its some time since I last read code for this22:06
heiko1sonney2k, eclipse is very good in my opinion, for exactly those things, searching classes, refactoring, all this blabla that makes life easier22:07
heiko1I also like the debugger22:07
heiko1I was doing lots of emacs before22:07
heiko1but eclipse is just more comfortable22:07
heiko1and less RMI also :)22:07
@sonney2kheiko1, well I am using it at work so why not give it a try for C++ /' shogun22:09
heiko1sonney2k, I dont use it for compiling though22:09
@sonney2kbut it is too slow :/22:09
heiko1I mean22:09
heiko1its not that bad22:09
@sonney2kI would be interested in trying shogun & java22:10
heiko1slower than emacs - of course22:10
@sonney2kit is here22:10
@sonney2kbut my machine is aging...22:10
heiko1what takes long?22:10
heiko1ah ok, yes22:10
@sonney2kclick ... wait ... click ... wait22:10
heiko1I remember on my old machine, it was a bit annoying22:10
heiko1but you can disable a few things22:10
heiko1then it gets faster22:10
@sonney2kheiko1, I think doing ndarray is actually easy22:10
@sonney2kone needs to write one vector with the dimension sizes22:11
@sonney2kthen all that is needed is to write out the whole data22:11
@sonney2kin the same way like any vector/matrix22:11
heiko1how is the data stored in the ndarray?22:11
@sonney2kjust a ptr to data22:11
@sonney2kin memory22:11
@sonney2klike for sgmatrix22:11
heiko1so all in a row22:12
heiko1I was wondering whether one would jiust have multiple arrays at different positions or like this22:12
@sonney2kno why should one22:15
@sonney2kthat is what sgmatrix list is for22:15
@sonney2kit can handle e.g. images of different size22:15
@sonney2kheiko1, I just had a look. I think all that is needed is to refactor these new_cont_begin stuff22:22
@sonney2kto get a SGVector as argument with the dims22:22
heiko1sonney2k, with this structure it should be quite easy to do the nda,22:22
@sonney2kif dims==1/222:22
@sonney2kthen it does the same like before22:22
@sonney2kthat's it22:22
heiko1the migration stuff is a bit more tricky22:22
@sonney2kbut there are none22:23
heiko1cannot remember the methods names but there are some which also do the case distinctions over the types22:23
heiko1let me check22:23
heiko1void TParameter::copy_data22:24
heiko1well they should not be too hard to extend22:24
heiko1since already working22:24
blackburna few memory errors left!22:27
@sonney2kheiko1, looks easy indeed22:28
@sonney2kthe biggest issue is to do the refactoring of new_cont_begin but that's it22:28
@sonney2kblackburn, context?22:28
heiko1sonney2k, maybe you find some inconsistencies on the way :)22:29
blackburnsonney2k: I mean in make check-valgirnd22:29
@sonney2kblackburn, in libshogun examples you mean?22:29
@sonney2kblackburn, go go :)22:30
blackburnI fixed a great bug with tapkee22:30
@sonney2knot sure if there is anything like a 'great' bug :)22:30
@sonney2khorrible maybe?22:30
blackburnwell all linear methods weren't working22:30
blackburnthat's just great22:31
@sonney2kgreat == cool22:31
@sonney2kdoesn't mean big22:32
@sonney2kyeah sarcasm would work for great bug :D22:32
blackburnthanks for pointing this out!22:33
blackburnokay one more fix22:33
@sonney2kheiko1, blackburn btw if n4nd0 pops up - I like the draft and want it for the main site!22:33
@sonney2kI have to repair my bike now22:34
blackburnI don't mind and don't understand how to put it on22:34
heiko1I can only see the menu22:34
blackburnsonney2k: that's *great* time to do that22:34
blackburnfebruary night :D22:34
@sonney2kheiko1, yeah it is a mockup22:34
shogun-notifier-shogun: Sergey Lisitsyn :master * 538ee43 / src/shogun/lib/tapkee/tapkee_methods.hpp:
shogun-notifier-shogun: Commented out unused projecting functions to avoid unhandled memory22:35
blackburnoh another great bug22:40
shogun-notifier-shogun: Sergey Lisitsyn :master * 6cbe6de / src/shogun/converter/LinearLocalTangentSpaceAlignment.cpp:
shogun-notifier-shogun: Fixed lost nullspace shift in LLTSA22:49
* sonney2k watches a youtube howto do bicycle gearing22:49
blackburnsonney2k: gearing like?22:49
@sonney2klike switching gears22:50
@sonney2kfrom bevel to bevel22:50
blackburnsonney2k: which part do you switch?22:50
blackburnuh! that appears to be hard to speak about these mechanisms in english hah22:51
@sonney2kalright time to get hands dirty22:51
-!- flxb [] has joined #shogun22:52
shogun-buildbot_build #216 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Soeren Sonnenburg <>22:52
@sonney2kwish me luck22:54
-!- in3xes [~in3xes@] has quit [Ping timeout: 255 seconds]22:58
shogun-notifier-shogun: Heiko Strathmann :master * afc7499 / tests/unit/features/ (2 files):
shogun-notifier-shogun: renamed test23:15
shogun-notifier-shogun: Heiko Strathmann :master * 4a6b977 / tests/unit/statistics/ (5 files):
shogun-notifier-shogun: unit tests for kernel selection23:15
shogun-notifier-shogun: Heiko Strathmann :master * 6e4af9c / tests/unit/ (7 files):
shogun-notifier-shogun: Merge branch 'mmd_kernel_selection'23:15
shogun-notifier-shogun: Heiko Strathmann :master * 0ab4a45 / tests/unit/ (7 files):
shogun-notifier-shogun: Merge pull request #879 from karlnapf/master23:15
shogun-notifier-shogun: unit tests for kernel selection23:15
@sonney2kblackburn, I didn't know that my bike's gear can be switched that smoothly!23:43
* sonney2k is happy23:44
blackburnsonney2k: so you mean gears like say 1/2/3 and 1/2/3/4/5/6/7?23:44
blackburnso what did you do with it?23:44
@sonney2kjust fine tune it23:44
blackburnah yeah that's a good thing23:44
@sonney2kI could only use 3/7 gears23:45
@sonney2k3 of 723:45
@sonney2kand these were stuttering23:45
@sonney2kI mean switching between them was horrible23:45
blackburnhuh that's not normal23:45
@sonney2know it is totally smooth23:45
blackburnwhen I ride bike I switch them pretty often I think23:45
blackburnsonney2k: do you ride bike already?23:47
blackburnas I can see it is +2 in berlin23:47
@sonney2kmakes a lot of sense to keep the pace/energy requirements consistent and just switching gears to adjust23:47
@sonney2kblackburn, every day I do!23:48
blackburnsonney2k: well in general I prefer to keep them at something normal (3/3 for me)23:49
blackburnbut when road gets smooth (it happens 30% of time in russia :D)..23:50
blackburnsonney2k: isn't that a little cold to ride a bike now? :)23:52
--- Log closed Sun Feb 17 00:00:39 2013