Open in new window / Try shogun cloud
--- Log opened Sat May 05 00:00:37 2012
-!- n4nd0 [02893bbe@gateway/web/freenode/ip.] has quit [Quit: Page closed]00:07
-!- cronor_ [] has joined #shogun00:52
-!- cronor [] has quit [Read error: Connection reset by peer]00:52
-!- cronor_ is now known as cronor00:52
-!- cronor [] has quit [Client Quit]00:57
-!- PhilTillet [] has joined #shogun01:15
-!- PhilTillet [] has quit [Remote host closed the connection]01:19
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]03:07
-!- wiking [] has joined #shogun03:13
-!- wiking [] has quit [Changing host]03:13
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun03:13
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit]03:15
@sonney2kblackburn I hope you will see a big one from me tonight too... I changed sooo much SGVector related...06:04
-!- gsomix [~gsomix@] has quit [Quit: Ex-Chat]08:11
-!- gsomix [~gsomix@] has joined #shogun08:11
-!- gsomix [~gsomix@] has quit [Remote host closed the connection]08:21
-!- vikram360 [~vikram360@] has quit [Ping timeout: 244 seconds]08:47
-!- vikram360 [~vikram360@] has joined #shogun09:39
-!- blackburn [~qdrgsm@] has joined #shogun09:55
-!- wiking [] has joined #shogun10:00
-!- wiking [] has quit [Changing host]10:00
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:00
blackburnwiking: hey10:15
wikingyes yes10:16
-!- vikram360 [~vikram360@] has quit [Ping timeout: 252 seconds]10:39
blackburnwiking: could you send me clef data somehow?11:04
wikingblackburn: ah yeah11:04
wikingi have to redownload as they changed something the other day11:05
blackburnwiking: they require some shity papers to sign :D11:05
wikingbut i'll make it available for you11:05
wikingyeye no worries11:05
blackburnI guess you did it already11:05
blackburndo you use dropbox?11:05
wikingmmm my drop is full :)))11:06
wikingand i really don't want to buy11:06
wikingi'll put it on the web for ya11:06
-!- vikram360 [~vikram360@] has joined #shogun11:06
blackburnwiking: ok thanks11:06
blackburnwiking: bad I can't expand your dropbox with shared folder :(11:07
wikingeheheh noup11:07
blackburnwould be nice hehe11:08
wikingblackburn: i've made some space on it11:09
wikingwhat's your db account ?11:10
blackburnwiking: how large is data btw?11:10
wikingit's around 65mb11:10
wikingthe training set11:10
wikingyep it's just 1000 pics11:10
wikingblackburn: invitation sent... but it's still uploading...11:12
blackburnwiking: thanks!11:12
wikingcan i ask maybe actually a favour11:12
wikingHOG the pictures?11:12
wikingi mean generate hog features of it?11:13
blackburnwiking: do you mean my favor11:16
blackburnlet me see pictures first ;)11:16
wikingu don't want to :p11:20
wikingsome of them are really scary :)11:21
blackburnwiking: damn they are really varying11:22
blackburnwiking: I don't think HOGs can help out there..11:23
wikingyeah i was guessing so11:23
wikingbut it's good to have it11:23
wikingas well11:24
blackburnsome classes are really really crazy11:24
blackburnwiking: did you use MKL on top of that??11:25
wikingi used BoW11:25
wikingwith sift11:25
blackburnyeah BoW is the only thing that could work here11:26
blackburnwiking: and it was something like of 90% accuracy?11:26
wikingnot with this11:26
wikinglast year11:26
wikingthis year this is fucking crazy11:26
blackburnoh shit I opened DVOR class11:26
wikingnot interested :)11:27
blackburnoh damn11:27
wikingtold you11:28
blackburnI will not sleep anymore11:28
blackburnDVDM is nice too..11:28
blackburnwiking: I guess this competition is for ones who can to not vomit whole day long seeing this pictures11:29
blackburnwiking: text is to be extracted somehow there..11:29
blackburnwiking: fmri is much more clean and not vomitting :D11:30
CIA-113shogun: Sergey Lisitsyn master * re6525c7 / src/shogun/machine/KernelMachine.cpp : Workaround for kernel machine segfault -
-!- blackburn [~qdrgsm@] has quit [Read error: Operation timed out]13:00
-!- vikram360 [~vikram360@] has quit [Ping timeout: 252 seconds]16:56
-!- vikram360 [~vikram360@] has joined #shogun16:57
-!- karlnapf [] has joined #shogun17:38
karlnapfanybody here?17:41
karlnapfmy current git doesnt compile, some hash -set related problems17:44
karlnapfsonney2k, do you have a clue where that might come from17:47
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun17:48
karlnapfblackburn, hi17:51
karlnapfjust cloned current git, does not compile17:51
blackburnkarlnapf: hey17:51
blackburnkarlnapf: really ?17:51
karlnapfmy options are17:51
blackburnfeel free to fix ;)17:52
karlnapf--enable-debug --disable-optimization --enable-trace ...17:52
karlnapfjust asking whether you knew about that17:52
blackburncompiles here hmm17:52
karlnapfwhich switches are you using?17:52
blackburnkarlnapf: opt one only17:53
blackburnlet me check with yours17:53
blackburnkarlnapf: python_modular?17:53
karlnapfbut fails before the module17:53
karlnapfworks without any switches17:54
blackburnkarlnapf: uh right17:55
blackburntrace mallocs is broken17:55
karlnapfis the problem obvious to you? if yes I wont get into it :)  ...17:57
karlnapfif no, I will fix it17:57
blackburnkarlnapf: not obvious but I'll try to take care17:57
karlnapfok then :)17:58
karlnapfWill fix my own bugs instead then :)17:58
blackburnkarlnapf: strange bug actually17:59
karlnapfblackburn, I was wondering whether we should add an assert that zero blocks are allocated in the exit_shogun() method17:59
karlnapfbecause some examples contain memleaks17:59
karlnapfbut they are hidden17:59
blackburnkarlnapf: hmmm I don't mind it17:59
karlnapfif we would enforce not leaking if trace-mallocs is activated17:59
karlnapfthat would automatically find all of them in the libshogun examples18:00
blackburnkarlnapf: make check-valgrind does the job btw18:00
blackburnin libshogun examples18:00
karlnapfoh, really?18:00
karlnapfok did not know that18:00
karlnapf :)18:01
blackburnkarlnapf: but not trace mallocs18:01
blackburnruns valgrind on each example18:01
karlnapfbut valgrind is better18:01
karlnapfalthough slower :)18:01
blackburnand outputs error count18:01
blackburnI have added it some day18:01
karlnapfblackburn, how does the ref-counting for vectors work?18:02
karlnapfdo I call ref method18:02
karlnapfor do I use SG_REF?18:02
blackburnkarlnapf: hmmm actually it relies on magic now18:03
karlnapfI dont have to do stuff?18:03
blackburnkarlnapf: copy constructor and constructor increases ref18:03
karlnapfand unref?18:03
blackburndestructor unrefs18:03
karlnapfnice thing18:03
karlnapfwill play around with that18:03
blackburnkarlnapf: e.g. if you have function that has SGVector as parameter18:04
blackburnfunc (SGVector vec)18:04
karlnapfyes get it18:04
blackburnnot const &18:04
karlnapfwhat does const& do?18:04
blackburnit calls copy constructor and increases ref for local scope18:04
karlnapfoh its a reference18:04
blackburnnot "const SGVector& vec"18:04
blackburnkarlnapf: sounds pretty cool actually18:05
karlnapfyes indeed18:06
karlnapfWill try to incorparate that into the subset stuff18:06
karlnapfcurrently, everything is copied18:06
karlnapfthat sucks18:06
blackburnkarlnapf: ohoh18:07
karlnapfwhats the matter?18:08
blackburnkarlnapf: I think we will die in that hell.. :D18:08
blackburnkarlnapf: painful transition18:08
karlnapfwith all the vectors?18:08
karlnapfI think its less bad than last year :)18:09
karlnapfalthough we could have done that properly from the beginning :)18:09
karlnapfbut thats how it works ;)18:09
CIA-113shogun: Heiko Strathmann master * r04e2696 / src/shogun/mathematics/Math.h : added copyright and a permute_vector method -
CIA-113shogun: Heiko Strathmann master * r49cccd0 / (7 files): initial sketch for new statistical testing framework -
CIA-113shogun: Heiko Strathmann master * r20e72f4 / src/shogun/statistics/LinearTimeMMD.cpp : new SGVector system changes -
CIA-113shogun: Heiko Strathmann master * rf432068 / (8 files in 2 dirs):18:10
CIA-113shogun: Merge pull request #494 from karlnapf/master18:10
CIA-113shogun: Suggestion for a statistical testing framework -
blackburnkarlnapf: that bug is really really strange..18:12
blackburnwould be strange if it caused by inline18:13
-!- gsomix [~gsomix@] has joined #shogun18:45
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:49
gsomixhi all18:49
gsomixkarlnapf, hey, what's wrong with hashset?18:53
gsomixkarlnapf, ah, okay. =___=18:56
karlnapfgsomix ,hi :)18:58
karlnapfyou got it :)18:58
karlnapfgsomix, are all the memory errors in make test caused by the SGVector transition?19:10
gsomixkarlnapf, I don't know really. I'm checking now for my bugs.19:14
@sonney2kkarlnapf, git clean -dfx19:21
karlnapfsonney2k, thanks I keep forgetting that19:21
@sonney2kgsomix, nothing wrong with it19:21
karlnapfsonney2k, about this SGVector<char> thing19:21
@sonney2kpeople always forget to do it when things change19:21
karlnapfI will replace it via a non-sgvector alternative19:22
karlnapfotherwise I will have to dig into generich classes which is messy19:22
gsomixsonney2k, huh? I found some bugs.19:22
@sonney2kgsomix, which?19:22
@sonney2kkarlnapf, can you be more concrete?19:23
@sonney2kvoid ptr and length or what?19:23
karlnapfsonney2k, yes19:23
karlnapfand actually19:23
karlnapfnot registering parameters19:23
karlnapffor these parameter trees its not necessary anyway19:23
@sonney2kwell ok19:24
gsomixsonney2k, wait a minute, please19:24
karlnapfperhaps override the serialize method19:24
* sonney2k continues with some intrusive changes all over the place19:24
karlnapfsonney2k, even fater git clean -dfx I get all these errors19:41
karlnapfI made make distclean and the git clean19:42
karlnapfjust valgrinded, its SGVector related, will check whether configure switches change something19:44
@sonney2kkarlnapf, which error do you get?19:46
karlnapfsonney2k, let me check it a bit more, give me a moment19:47
karlnapfcurrently this spooky one:19:47
karlnapfvex amd64->IR: unhandled instruction bytes: 0xC5 0xF9 0x57 0xC0 0xE8 0x6319:47
karlnapf64 bit stuff perhaps?19:47
CIA-113shogun: Soeren Sonnenburg master * r1f419bb / examples/undocumented/libshogun/statistics.cpp : slightly simplify example -
CIA-113shogun: Soeren Sonnenburg master * reaf5c09 / (17 files in 8 dirs):19:49
CIA-113shogun: continued sgvector transition19:49
CIA-113shogun: - convert linear machines to use SGVector w instead of double ptr19:49
CIA-113shogun: - remove various unneedef SG_FREEs -
@sonney2kkarlnapf, when you run things with valgrind always use --enable-debug --disable-optimization19:49
@sonney2kkarlnapf, but yes you will still see several crashers due to sgvector19:50
karlnapfoh yes of course thats this error19:50
karlnapfbut this wasnt the one I meant19:51
karlnapfah ok19:51
karlnapfalright then19:51
@sonney2kkarlnapf, and it will get worse19:52
@sonney2konce sgvector is done19:52
@sonney2ksgmatrix will follow19:52
@sonney2ksgsparse amtrix19:52
@sonney2kbut one by one19:53
karlnapfmh, I should wait with any changes then19:53
@sonney2kand sgvector is most used so it is not that bad19:53
karlnapfalmost half of the tests are failing here19:53
@sonney2kwhich tests?19:53
karlnapfin a freshly cloned repo19:53
@sonney2konly one fails here19:53
@sonney2kand I am currently fixing it19:53
karlnapfclassifier_mklmulticlass, classifier_qda, classifier_multiclasslinearmachine, and more19:55
karlnapfgotta go now, will be back in 30 mins19:55
gsomixsonney2k, segfaults with --enable-trace-mallocs. in hashset.19:56
@sonney2kgsomix, heh19:57
@sonney2kcould be endlessly looping19:57
gsomixsonney2k, what do you mean?20:04
gsomixah, I got it20:05
@sonney2kgsomix, you have to check how it was done before to avoid this20:07
@sonney2kkarlnapf, in CKernelMachine - will indices work already?20:13
@sonney2kand what is params.indices=SGVector<index_t>(0) in apply() supposed to do20:13
karlnapfsonney2k, let me have a look20:14
karlnapfwhich line which file?20:15
karlnapfk got it20:15
@sonney2k(it is disabled right now)20:15
karlnapfinitialise vector with 020:16
karlnapfthe 0 probably sneaked in20:16
karlnapfremove it20:16
karlnapfbut the vector has to be NULL20:16
karlnapfsorry not 0, but NULL20:16
karlnapfso SGVector<index_t>() will do it20:16
@sonney2kkarlnapf, so removing it will work as good20:17
@sonney2k(default constructor does NULL already)20:17
karlnapfyes true20:17
karlnapfsonney2k, any clue why I could have all these memory errors?20:18
karlnapfand you dont?20:18
CIA-113shogun: Soeren Sonnenburg master * r052e3c3 / src/shogun/classifier/mkl/MKLMulticlass.cpp : fix mkl multiclass sgvector bug -
CIA-113shogun: Soeren Sonnenburg master * r39d9382 / src/shogun/machine/KernelMachine.cpp : remove unneeded initialization in KernelMachine -
@sonney2kkarlnapf, old shogun?20:19
@sonney2kkarlnapf, btw I have one more question20:19
@sonney2kI am getting a new error in training error: 1.00542220:21
@sonney2kcross_validation estimate:20:21
@sonney2k[2449.555420,2462.500676] with alpha=0.050000, mean=2456.02804820:21
@sonney2k[ERROR] assertion result.mean<2 failed in file evaluation_cross_validation_regression.cpp line 10320:21
@sonney2kbut no memory errors or anything20:22
@sonney2kvalgrind errors I mean20:22
karlnapfthese values sound like uninitialised memory20:22
karlnapfthe assertion is there to ensure that the value is small20:22
karlnapfso valgrind errors?20:22
@sonney2kkarlnapf, so that value should be around what you get for training error?20:23
karlnapfI mean the estimate is obviously wrong20:23
karlnapfno, but the data is linear seperable,20:24
karlnapflet me have a look at the example20:24
karlnapfoh no, not lin. sep.20:24
@sonney2kin this transition I am doing a few intrusive changes that can change anything20:25
@sonney2kone is that get_labels() etc never copies20:25
@sonney2kand same with CAlphabet20:25
@sonney2k(we used to always copy there)20:25
karlnapfthat might be a problem for cross-validation ...20:26
@sonney2kso if this is assumed somehwere then -> kaboom20:26
karlnapfperhaps something with the subsets20:26
@sonney2kbut you call get_labels_copy20:26
karlnapfbecause get_labels with a subset HAS to copy20:26
karlnapfthat is probably the reason for the large x-val error and to mem-error: wrong labels20:27
@sonney2kalso get_int_labels will copy20:30
karlnapfoh there is at least one bug20:31
karlnapfoh wait20:32
karlnapfI might be wrong20:32
karlnapfno sorry20:33
karlnapfwrong alert20:33
karlnapfdid the error occur when you changed that get_labels thing? or before?20:34
@sonney2kno idea20:34
@sonney2kI guess after change20:35
@sonney2kbut still I this is more tricky20:35
@sonney2kI guess I will continue to try to fix the errors I see in the more elaborate python_modular exmaples20:35
@sonney2kmaybe it is a side effect20:35
karlnapfyes, you could remove the assertion for now20:36
@sonney2kwell not remove - I will just work on python_modular and revisit it later20:36
karlnapfI guess its something with the splitting strategy producing wrong labels20:36
karlnapfI will try to complete the ModelSelectionparameters thing now20:37
gsomixsonney2k, fixed20:43
CIA-113shogun: Soeren Sonnenburg master * r3ceca27 / (5 files in 2 dirs): polish get_labels_copy() and drop load/save from CLabels -
@sonney2kgsomix, where?20:44
gsomixsonney2k, in CSet.20:45
gsomixjust added use_sg_mallocs flag.20:45
gsomixI'm testing now... with --enable-trace-mallocs now. =___=20:46
karlnapfsonney2k, gsomix, there are some memory leaks in the model-selection stuff20:46
karlnapfmodelselection_grid_search_kernel.cpp for example20:46
karlnapfmmh, kind of hard to fix as long as there is no trace mallocs, Ill check in later when this works and try to find out whats going on20:48
karlnapfCannot fix the other thing because I dont know whether I cause mem-leaks20:48
-!- blackburn [~qdrgsm@] has joined #shogun20:49
karlnapfis the trace-mallocs working again?20:49
@sonney2kkarlnapf, gsomix found the issue - so very soon20:49
blackburnsorry I get a little bit ill20:49
gsomixblackburn, grustno20:50
@sonney2kblackburn, working too hard...20:50
blackburnsonney2k: hardly work rather hehe20:50
gsomixblackburn, I also got ill.20:51
blackburnok according to logs gsomix you will take care of it?20:51
@sonney2kbasically all multiclass stuff crashes20:51
@sonney2kgsomix - once this is fixed - please remove all the load() and save() crap ... seems to be in many many classes not only CMachine derived ones...20:51
blackburnsonney2k: I know the reason probably20:52
@sonney2kwow - quite many python examples work already20:52
blackburnsonney2k: issue is mainly related with linearmachine's w20:52
gsomixsonney2k, okay. Added to my todo list. :)20:52
@sonney2kwill be back in ~20 min20:52
gsomixkarlnapf, ping20:58
karlnapfgsomix, hi20:59
karlnapfgot it?20:59
gsomixkarlnapf, yep. can you review and merge it?20:59
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]20:59
blackburnwow I never expected21:00
karlnapfwill do21:00
karlnapfgsomix, does it compile with trace-mallocs without adding more errors in tests?21:01
blackburnkarlnapf: set is used only in memory blocks counting ;)21:02
gsomixkarlnapf, yep, it seems.21:02
karlnapfpatch looks nice, I dont know what you talked about with soeren about this default constructor, if you agreed on that I iwll merge21:03
gsomixkarlnapf, just a second21:09
gsomixkarlnapf, done21:20
karlnapfgsomix, did you yet push?21:20
gsomixkarlnapf, yep21:22
CIA-113shogun: Evgeniy Andreev master * r3fc0b13 / (5 files in 3 dirs): fixes for using non-shogun mallocs/callocs -
CIA-113shogun: Evgeniy Andreev master * r1786947 / (src/shogun/lib/memory.cpp src/shogun/lib/memory.h): default constructor is not needed -
CIA-113shogun: Heiko Strathmann master * r64d63bd / (4 files in 3 dirs):21:22
CIA-113shogun: Merge pull request #498 from gsomix/CSet21:22
CIA-113shogun: Fixes in CSet -
gsomixkarlnapf, check it. I need a feedback. :)21:22
karlnapfcompiling ...21:23
blackburngsomix: what are your ongoing tasks?21:23
gsomixblackburn, hmm, Arrays -> DynArray.21:23
gsomixmap in titled features21:23
gsomixand load/save deletions21:24
blackburnok no need for other tasks then21:24
-!- vikram360 [~vikram360@] has quit [Read error: Connection timed out]21:27
-!- vikram360 [~vikram360@] has joined #shogun21:27
blackburn==17052== Invalid read of size 821:28
blackburn==17052==    at 0x4C2A327: memcpy (mc_replace_strmem.c:635)21:28
blackburn==17052==    by 0x53088AF: shogun::CSubset::CSubset(shogun::SGVector<int>) (Subset.cpp:29)21:28
blackburn==17052==    by 0x52DC6C5: shogun::CSubsetStack::add_subset(shogun::SGVector<int>) (SubsetStack.cpp:105)21:28
blackburn==17052==    by 0x52959FD: shogun::CLabels::add_subset(shogun::SGVector<int>) (Labels.cpp:257)21:28
blackburnsomething crazy is going here heh21:28
karlnapfgsomix, ehm lots of errors with trace-mallocs21:28
karlnapfin the libshogun tests21:28
karlnapfwithout trace-mallocs runs ok21:29
gsomixkarlnapf, what tests?21:29
karlnapfclassifier_libsvm is the first one21:31
karlnapfgaussian_naive_bayes does not terminate21:31
gsomixkarlnapf, okay, thanks.21:32
blackburnfor brave souls who want to solve multiclass machines - the error is in copy constructor of linear machine21:32
karlnapfgsomix, do the tests run for you?21:33
blackburnit is being double freed for some reason21:33
blackburnah and also subsets are b0rken21:33
karlnapfblackburn, yes these are probably because of SGVector stuff21:33
karlnapfI will review that once the other stuff works21:33
blackburnkarlnapf: exactly because of it :)21:34
blackburnI just told what I have found out while checking valgrind21:34
blackburnhave to rest now - my cold is too annoying to fix code right now21:34
karlnapfhah blackburn, same here21:35
blackburnkarlnapf: are you cold too?21:35
karlnapfsince almost two weeks now, annoying21:35
blackburnah I hope you get rid of it soon21:36
blackburnhowever after I saw DVOR category of medical CLEF I can't complain anymore21:36
-!- vikram360 [~vikram360@] has quit [Read error: Operation timed out]21:37
-!- vikram360 [~vikram360@] has joined #shogun21:37
karlnapfehm what?21:37
blackburnkarlnapf: in the medical CLEF benchmark there are a lot of nice photos21:38
blackburnI lost my sleep today probably seeing it :D21:38
karlnapfI guess I dont wanna see that21:39
blackburnkarlnapf: yeah that's why I didn't send it to you ;)21:39
blackburnok see you tomorrow21:39
karlnapfk blackburn, take care21:39
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]21:40
karlnapfgsomix, tell me when do sent another path ;)21:41
karlnapfwill do some cooking now21:41
gsomixkarlnapf, okay.21:41
gsomixit seems, I need asleep a little.21:41
-!- wiking [] has joined #shogun21:41
-!- wiking [] has quit [Changing host]21:41
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun21:41
karlnapfgsomix, dont forget to take a break from time to time ;)21:42
gsomixgood night, guys21:49
-!- Marty28 [] has joined #shogun21:50
-!- gsomix [~gsomix@] has quit [Ping timeout: 276 seconds]21:54
@sonney2kI guess it is best pluskid has a look at multiclass errors21:58
-!- Marty28 [] has quit [Quit: Colloquy for iPad -]22:00
-!- karlnapf [] has quit [Ping timeout: 272 seconds]22:03
CIA-113shogun: Soeren Sonnenburg master * re8c646b / src/shogun/clustering/Hierarchical.cpp : fix crasher due to double free in hierarchical -
CIA-113shogun: Soeren Sonnenburg master * r6b5dcbc / src/shogun/regression/KernelRidgeRegression.h : move enum to upper class level -
-!- emrecelikten [~emrecelik@] has joined #shogun22:08
CIA-113shogun: Soeren Sonnenburg master * r21b11be / src/shogun/distributions/LinearHMM.cpp : fix double free in linearhmm -
CIA-113shogun: Soeren Sonnenburg master * rd847327 / src/shogun/kernel/AUCKernel.cpp : fix double free in auc kernel -
CIA-113shogun: Soeren Sonnenburg master * rb27af96 / (3 files in 2 dirs): fix double free errors in mkl / combined kernel / wds kernel -
CIA-113shogun: Soeren Sonnenburg master * ra576554 / (5 files in 2 dirs):23:24
CIA-113shogun: re-introduce load/save to labels23:24
CIA-113shogun: it does make sense to have these functions to be able to load/save23:24
CIA-113shogun: labels in standard file formats -
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]23:34
CIA-113shogun: Soeren Sonnenburg master * rea70748 / src/shogun/transfer/domain_adaptation/DomainAdaptationSVM.cpp : fix double free in da svm -
--- Log closed Sun May 06 00:00:37 2012