Open in new window / Try shogun cloud
--- Log opened Sun Jul 01 00:00:17 2012
-!- blackburn [~blackburn@] has quit [Ping timeout: 248 seconds]00:37
-!- blackburn [~blackburn@] has joined #shogun00:47
CIA-18shogun: Sergey Lisitsyn master * r4bc9001 / src/configure : Added Eigen3 configure check -
CIA-18shogun: Sergey Lisitsyn master * rc86b318 / src/shogun/io/AsciiFile.cpp : Removed warning-causing empty backslashed line in AsciiFile -
CIA-18shogun: Sergey Lisitsyn master * r6375108 / (3 files in 2 dirs): Fixed least squares multitask L1/Lq regressor and made it fit with SLEPs analogous method -
CIA-18shogun: Sergey Lisitsyn master * rf408929 / (3 files in 2 dirs): Merge branch 'slep' of git:// -
-!- blackburn [~blackburn@] has quit [Ping timeout: 246 seconds]02:36
-!- blackburn [~blackburn@] has joined #shogun02:41
CIA-18shogun: Sergey Lisitsyn master * r5643f9a / src/shogun/transfer/multitask/TaskGroup.h : Added a few doc lines for task group -
CIA-18shogun: Sergey Lisitsyn master * rdf424fa / (10 files in 3 dirs): Added tree regularized logistic regression and fixed least squares regressors support -
CIA-18shogun: Sergey Lisitsyn master * r102ecb0 / (6 files in 3 dirs): Added PegasosSVM -
CIA-18shogun: Sergey Lisitsyn master * rc586669 / (13 files in 6 dirs): Moved liblinear to optimization -
-!- blackburn [~blackburn@] has quit [Ping timeout: 250 seconds]06:47
shogun-buildbotbuild #14 of deb1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Sergey Lisitsyn <>07:10
-!- n4nd0 [] has joined #shogun09:32
-!- pluskid [~pluskid@] has joined #shogun09:49
-!- blackburn [~blackburn@] has joined #shogun13:21
-!- blackburn1 [~blackburn@] has joined #shogun13:30
-!- blackburn1 [~blackburn@] has quit [Quit: Leaving.]13:38
-!- n4nd0 [] has quit [Ping timeout: 248 seconds]13:41
@sonney2kblackburn, you broke the build...13:57
@sonney2k...instead of fixing it13:57
blackburnnahh easy13:59
@sonney2kblackburn, please fix it then... we really need to get the buildbot back to green14:01
blackburnsonney2k: but we have pegasos now :D14:01
CIA-18shogun: Sergey Lisitsyn master * r88ff286 / (3 files in 3 dirs): Updated headers due to liblinear migration -
blackburnshould work now14:03
shogun-buildbotbuild #15 of deb1 - libshogun is complete: Success [build successful]  Build details are at
CIA-18shogun: Sergey Lisitsyn master * r6857283 / (2 files in 2 dirs): Fixed things causing compilation warnings -
blackburnsonney2k: hmm heiko stuff fails14:10
blackburnthe only thing that fails I guess14:11
blackburnI disable it for now14:14
CIA-18shogun: Sergey Lisitsyn master * r7ef9212 / (2 files): Disabled statistics tests and updated MT LR test -
-!- blackburn [~blackburn@] has left #shogun []14:57
-!- n4nd0 [] has joined #shogun16:19
n4nd0sonney2k: around?16:24
-!- pluskid [~pluskid@] has quit [Quit: Leaving]16:46
-!- n4nd0 [] has quit [Quit: leaving]17:29
-!- heiko1 [] has joined #shogun19:08
@sonney2kheiko1, seen what blackburn said?19:09
@sonney2k<blackburn> -sonney2k: hmm heiko stuff fails19:09
@sonney2k<blackburn> -the only thing that fails I guess19:09
@sonney2k<blackburn> -I disable it for now19:09
@sonney2k<CIA-18> -shogun: Sergey Lisitsyn master * r7ef9212 / (2 files): Disabled statistics tests and updated MT LR test -
heiko1oh no!19:09
heiko1Ill fix it19:10
@sonney2khmmhh actually sparse stuff is still broken for lots of static interfaces...19:11
@sonney2kblackburn - why did you add pegasos? it is the same as SGD but slow19:12
heiko1blackburn, sonney2k, with an updated git version, both statistics examples dont fail here19:19
heiko1what happened?19:19
@sonney2kheiko1, did blackburn not just deactivate them in the patch avove?19:33
heiko1yes he did19:33
heiko1I think I know19:34
heiko1they might fail sometimes 1/100 times19:35
heiko1Ill deactivate the assertions19:35
heiko1actually stupid of me leaving them in there19:35
-!- heiko1 [] has left #shogun []19:36
-!- blackburn [~blackburn@] has joined #shogun19:38
blackburnsonney2k: I became bored and checked our main page19:40
blackburnso we had no pegasos and I thought why not to add it :D19:40
blackburnI'll fix sparse don't worry - I just forgot to do it for octave_*19:40
@sonney2kblackburn, so you converted it to use dotfeatures?19:42
blackburnsonney2k: pegasos? yeah I think so19:42
@sonney2kand is it the original algorihtm?19:42
@sonney2kI meant what is the source?19:43
blackburna little modified though19:43
blackburnfrom shai shalev19:43
@sonney2khmmhh heiko is gone again19:43
blackburnsonney2k: I even didn't really test and compare it19:43
@sonney2kblackburn, then file a bug for that19:44
@sonney2kotherwise we will forget that this doesn't work :D19:44
blackburnsonney2k: I have no idea whether it works19:44
blackburnmay be it does19:44
blackburnsonney2k:  insomnia makes some crazy things with me19:44
@sonney2kseems so :)19:46
@sonney2kbetter fix the sparse bugs for static/modular interfaces...19:46
blackburnsonney2k: will do19:47
@sonney2kI would expect there are more in java/ etc to fix19:47
@sonney2kbut I guess the bug is always the same19:47
blackburnsonney2k: do we have sparse in java?19:47
@sonney2kif someone converts dense -> sparse :D19:48
blackburnsonney2k: then it is not affected19:48
blackburnsonney2k: basically bug is related with MALLOC19:48
blackburnsonney2k: we allocate some crap and operator= tries to free it19:48
blackburnso what I did is just added inplace new19:48
@sonney2khmmhh wouldn't it be better to have just some simple struct for that internally and for I/O have some overhead with refcounts and copying?19:50
blackburnsonney2k: I've never known about inplace new actually and pluskid helped with it19:50
@sonney2kI don't like the new[] crap19:50
blackburnsonney2k: not new[]19:50
blackburnnew (&matrix[i]) SGSparseVector<float64_t>(); that is19:50
blackburnsonney2k: when you allocate memory for sparse vectors you should initialize it somehow19:50
blackburnthat thing does this job19:51
@sonney2klooks horrible19:51
@sonney2khow many people can parse this expression?19:51
blackburnsonney2k: there is no other way to have array of SGReferencedData19:51
@sonney2kblackburn, that is why I mean we better don't19:52
blackburnsonney2k: what do you suggest then?19:52
@sonney2kso we could just use some sparse vector struct as we had internally19:52
blackburnI'd like to easyfix it now rather than changing internals again19:52
blackburnstruct like?19:52
@sonney2klike we had19:53
@sonney2kand then on top of it have the SGReferencedData derived sparse vector class that copies though19:53
blackburnI am not so unhappy with it - you only should do it once you allocate memory19:53
@sonney2kblackburn, but don't we have the overhead now?19:54
blackburnsonney2k: no overhead at all19:54
@sonney2kwait - so sparse features uses sparse matrix19:54
@sonney2kand sparse matrix doesn't use sparse vector?19:54
blackburnsonney2k: sparse features make no use of sparse matrix19:54
blackburnonly input19:55
blackburnit contains array of sparse vectors19:55
blackburnso each time you allocate new array you have to initialize refs19:55
@sonney2kblackburn, well it should make use of sparse matrix19:55
blackburnsonney2k: it will be the same - but in sparse matrix then19:56
@sonney2klike CDenseFeatures19:56
blackburnsonney2k: you *have* to new each sparse vector of array19:57
blackburnbecause operator= will corrupt it in other case19:57
@sonney2kblackburn, well you don't if we don't use SGReferecedData right?19:57
@sonney2kso now we add the overhead to each sparse vector19:58
blackburnoverhead like?19:58
blackburnreference count?19:58
blackburnsonney2k: I thought we discissed it is ok19:59
@sonney2kI don't remember...20:00
@sonney2kit is 12 byte / vector20:00
@sonney2klets consider 100 Mio examples20:01
@sonney2kso just 1.2G20:01
@sonney2klooks OK I gues20:01
blackburnsonney2k: but much easier to work with20:02
@sonney2kblackburn, yeah except for expressions like (&sparse_matrix[i])->~SGSparseVector<T>(); and the one above :/20:05
blackburnsonney2k: hah well I can live with it - it appears only once20:05
blackburnwe have some uglier code20:05
@sonney2kblackburn, where ?20:06
@sonney2kwell ok if it just appears once20:07
@sonney2kblackburn, so then convert the sparsefeatures also to use sgsparsematrix20:07
@sonney2kthen it is ok20:07
blackburnsonney2k: pthreads are unparseable20:07
@sonney2kyeah but that is easily fixable20:07
@sonney2kif we treat sparse features like this20:07
@sonney2kwe should do the same with strings later20:08
blackburnyeah but I'd like to not convert sparse features just now20:08
@sonney2kblackburn, btw I really wonder if we shouldn't have something like that instead of the dense feature matrix20:09
CIA-18shogun: Sergey Lisitsyn master * r6637f20 / (2 files in 2 dirs): Fixed sparse matrix handling in octave interfaces -
@sonney2kI mean also just list of vectors20:09
@sonney2kone can permute vectors nicely...20:09
blackburnwell it makes sense for me actually20:10
@sonney2kon the other hand blas/lapack stuff needs such fortran style matrice20:10
blackburnsonney2k: well we rarely work on full matrices20:20
blackburnsonney2k: do we work on matrices at all?20:20
@sonney2kblackburn, don't GMM / GPs do?20:23
@sonney2kblackburn, I guess not an issue now though20:23
@sonney2kmuch more important is get things back to stable20:24
@sonney2kthen sparsefeatures -> sparsematrix, GMMs, buildbots, tests running lalal20:24
blackburnsonney2k: lets see if tests will go now20:25
@sonney2kexamples you mean :)20:25
@sonney2ktests are all disabled unfortunately :/20:26
blackburnnearly no difference20:26
@sonney2kI would wish20:27
@sonney2klast time we said that too :D20:27
@sonney2kand we were fixing things for a few weeks20:27
blackburnif it compiles - nice20:27
blackburnif it does not produce segfaults - awesome :D20:28
blackburnsonney2k: what is treesize?20:30
blackburn573276 KiB??20:30
blackburn573 mb?20:30
@sonney2kthe compiled stuff including data20:31
blackburnpretty much20:32
blackburnthat didn't help20:47
blackburnsonney2k: any idea what can be wrong with kernel_lik?21:23
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun21:44
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Quit: Page closed]23:01
CIA-18shogun: Sergey Lisitsyn master * rf482b5d / examples/undocumented/matlab_and_octave/kernel_lik.m : Temporary commented out kernel_lik.m example -
--- Log closed Mon Jul 02 00:00:17 2012