Open in new window / Try shogun cloud
--- Log opened Thu Oct 13 00:00:39 2011
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]00:07
-!- alesis-novik [] has quit [Quit: Leaving]01:45
CIA-3shogun: Soeren Sonnenburg  * r1d5d860 / src/shogun/kernel/Kernel.cpp : Fix a bug in cache_multiple_kernel_row that gets triggered when support vector regression is performed with svr light and multiple threads. -
shogun-buildbotbuild #280 of python_static is complete: Failure [failed compile]  Build details are at  blamelist: sonne@debian.org11:19
-!- alesis-novik [] has joined #shogun14:01
-!- in3xes [~in3xes@] has joined #shogun18:20
-!- blackburn [~blackburn@] has joined #shogun18:27
-!- in3xes [~in3xes@] has quit [Ping timeout: 248 seconds]18:33
blackburnuh rip dennis ritchie18:34
-!- blackburn [~blackburn@] has quit [Ping timeout: 248 seconds]20:20
-!- blackburn [~blackburn@] has joined #shogun20:24
blackburnsonney2k: I've got an idea of very fun task for next gsoc or just to one of us20:33
blackburnthere could be some task to store visualization information to make easier to visualize something20:34
@sonney2kvisualization information?20:35
blackburnyeah something related to training process we could plot20:36
blackburnsonney2k: I've got a problem with new 'converter' folder, parallel and io are not visible for some reason20:37
blackburnany ideas?20:37
@sonney2kblackburn, did you derive from SGObject?20:40
blackburnsonney2k: yes, I guess the problem is I forgot to implement get_name20:40
blackburnsonney2k: where should I add new modular .i files?20:44
@sonney2kin modular/20:44
blackburnsonney2k: no, modshogun.i, somewhere else?20:45
blackburnsonney2k: where did you declared python modules for workaround like shogun.Preprocessor, shogun.Classifier?21:01
@sonney2kblackburn, that is in Makefile.template21:01
@sonney2kblackburn, we have to get memory requirements for compiling the interfaces below 1.5 G21:05
@sonney2kbetter ~1G21:05
@sonney2knone of the debs built21:05
-!- alesis-novik [] has quit [Quit: Leaving]21:27
blackburnsonney2k: have you asked anyone from swig team how to reduce size?21:41
blackburnmay be they have some tricks for that21:43
@sonney2kI don't think so...21:51
-!- sonne|work [~sonnenbu@] has quit [Ping timeout: 240 seconds]22:34
-!- sonne|work [~sonnenbu@] has joined #shogun22:49
blackburnsonney2k: around?23:32
CIA-3shogun: Soeren Sonnenburg  * r74bf297 / src/shogun/kernel/WeightedDegreePositionStringKernel.cpp : fix lhs / rhs mixup in free_feature_vector (how did this ever work?!) -
CIA-3shogun: Soeren Sonnenburg  * ra6e49bb / src/shogun/classifier/svm/WDSVMOcas.h : fix broken free_feature_vector call in wdsvmocas -
CIA-3shogun: Soeren Sonnenburg  * r12a5173 / (2 files): introduce more efficient cleanup_feature_vectors function -
CIA-3shogun: Soeren Sonnenburg  * r18a93b9 / src/shogun/features/DotFeatures.cpp : always set dot feature property on init -
CIA-3shogun: Soeren Sonnenburg  * r6356f26 / (3 files): fix static octave examples -
@sonney2kblackburn, busy watching though23:35
blackburnsonney2k: I need an idea how to stole feature matrix from simple features23:36
@sonney2kI don't think it is currently possible23:41
blackburnsonney2k: I would add leave_feature_matrix(), ok?23:43
@sonney2kwhat exactly should this do?23:43
@sonney2kget the ptr of the fm?23:43
blackburnsonney2k: yes, and make it not deleted when features will be23:43
@sonney2kand then replace the fm ptr with yours?23:44
blackburnI just want features to lose this pointer23:44
blackburnand when it is going to be deleted features won't free this matrix23:44
blackburncause it will be used by another features23:45
blackburnI could do this with copying but it is pretty bad way in case of large-scale applications23:45
@sonney2kso add a function that gets you the ptr and size23:45
@sonney2kand internally drops/resets preprocs/ptr/sizes23:46
@sonney2kSGMatrix<ST> get_feature_matrix_and_clear() ?23:46
@sonney2kor *_and_reset() ?23:46
blackburnwhy not leave_feature_matrix()?23:46
@sonney2kleave doesn't tell anything about that you steal it23:47
@sonney2kwell how about steal_feature_matrix()23:47
blackburnsonney2k: I have almost done ->converters things23:48
@sonney2kblackburn, very good23:48
blackburnand already moved all the examples23:48
blackburnand tested23:48
@sonney2kI think I fixed all octave examples now that octave runs again under debian unstable23:48
blackburnfinal step - dimensionreductionpreprocessors proxy23:49
shogun-buildbotbuild #281 of python_static is complete: Success [build successful]  Build details are at
@sonney2kI have high hopes taht the buildbot goes through now23:50
@sonney2kha :)23:50
blackburnsonney2k: we are doing nice progress last 1.5 months :)23:51
blackburnI'm running out of power with improving my dim reduction 'toolkit' though23:51
@sonney2kone of the most important things is to keep git master compilable/ tests running23:51
blackburnI would be very disappointed being rejected with paper23:51
@sonney2kand the debian package23:51
@sonney2kblackburn, dont' expect that the paper will be accepted immediately23:52
blackburnsonney2k: month?23:52
@sonney2kthere will be at least one revision23:52
@sonney2kmonths + that23:52
blackburnI see23:53
blackburnwell at least after programming I can focus on paper23:53
blackburnI feel much happier with converters concept23:53
blackburnwithout any changes you can now embed features that doesn't fit in memory23:54
blackburnsonney2k: are kernels support this?23:54
shogun-buildbotbuild #288 of octave_static is complete: Success [build successful]  Build details are at
@sonney2kstrike :)23:54
@sonney2kI don't understand what you are saying...23:55
blackburnsonney2k: can I init kernel with FileFeatures or so?23:55
@sonney2ksimplefeatures don't have to be in memory...23:56
@sonney2kso partially yes23:56
blackburnI mean that most of methods are now working with any Features supporting distance/kernel23:56
@sonney2kblackburn, btw the only way I see to get compilation below 1.5GB is to remove a few %template's23:56
@sonney2kso I guess I should make them configurable23:57
@sonney2kand drop as much as possible...23:57
@sonney2kI see. but don't you need access to the full kernel/distance matrix?23:58
@sonney2kor do you only access element pairs?23:58
blackburnunfortunately I need23:58
@sonney2kwell then you shouldn't care if features fit in memory23:58
@sonney2kthese will not be the bottleneck :)23:59
blackburnbefore it wasn't any possible23:59
blackburncause it was working only with SimpleFeatures23:59
@sonney2kthe buildbot is now working on the modular interfaces again23:59
@sonney2kblackburn, I see23:59
--- Log closed Fri Oct 14 00:00:09 2011