Open in new window / Try shogun cloud
--- Log opened Sun May 06 00:00:37 2012
CIA-113shogun: Soeren Sonnenburg master * rc896d91 / src/shogun/lib/GCArray.h :00:12
CIA-113shogun: fix uninited memory reads in gcarray00:12
CIA-113shogun: SG_MALLOC -> SG_CALLOC does the job -
CIA-113shogun: Soeren Sonnenburg master * rbed68da / src/shogun/base/Parameter.cpp : fix typo in comment -
CIA-113shogun: Soeren Sonnenburg master * r9fb6130 / src/shogun/lib/SGVector.h : free old ref before copying -
-!- emrecelikten [~emrecelik@] has quit [Quit: Leaving.]01:32
-!- vikram360 [~vikram360@] has quit [Ping timeout: 240 seconds]04:21
-!- vikram360 [~vikram360@] has joined #shogun04:22
-!- pluskid [~pluskid@] has joined #shogun04:29
-!- pluskid [~pluskid@] has quit [Quit: Leaving]06:15
-!- blackburn [~qdrgsm@] has joined #shogun07:27
-!- Marty28 [] has joined #shogun07:37
-!- Marty28 [] has quit [Quit: Colloquy for iPad -]07:48
-!- gsomix [~gsomix@] has joined #shogun08:54
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 245 seconds]09:57
gsomixsonney2k, around?10:03
-!- gsomix [~gsomix@] has quit [Ping timeout: 248 seconds]10:13
-!- vikram360 [~vikram360@] has quit [Ping timeout: 245 seconds]11:22
-!- gsomix [~gsomix@] has joined #shogun11:24
-!- vikram360 [~vikram360@] has joined #shogun11:35
gsomixsonney2k, oh, we cannot use hashset for MemoryBlocks.11:49
gsomixbecause we add MemoryBlock(ptr, size) and remove MemoryBlock(prt) - different hashes.11:51
gsomixI will replace CSet to CMap<void* ptr, MemoryBlock>. ok?11:53
-!- n4nd0 [] has joined #shogun12:03
-!- gsomix [~gsomix@] has quit [Ping timeout: 276 seconds]12:11
-!- gsomix [~gsomix@] has joined #shogun12:12
-!- vikram360 [~vikram360@] has quit [Ping timeout: 255 seconds]12:36
-!- vikram360 [~vikram360@] has joined #shogun12:46
-!- pluskid [] has joined #shogun13:07
-!- n4nd0 [] has quit [Ping timeout: 250 seconds]13:12
-!- n4nd0 [] has joined #shogun13:12
-!- n4nd0 [] has quit [Ping timeout: 260 seconds]13:21
-!- n4nd0 [] has joined #shogun13:21
-!- n4nd0 [] has quit [Ping timeout: 276 seconds]13:47
-!- blackburn [~qdrgsm@] has joined #shogun14:05
-!- vikram360 [~vikram360@] has quit [Ping timeout: 244 seconds]14:07
-!- vikram360 [~vikram360@] has joined #shogun14:19
blackburnpluskid: hey15:37
blackburnit looks like recent commits caused this error15:37
blackburnit appeared on my machine too15:37
pluskidblackburn: I think I already fix this error locally15:37
blackburnpluskid: what caused it?15:37
pluskidhmm, let me think...15:38
pluskidI see15:38
pluskidvirtual void add_machine_subset(SGVector<index_t> subset)15:38
pluskidthis method in CMulticlassMachine15:38
pluskidbecomes virtual void add_machine_subset(const SGVector<index_t> subset)15:38
pluskidin CLinearMulticlassMachine15:38
pluskidone more const15:39
pluskidthen SWIG confused15:39
-!- gsomix [~gsomix@] has quit [Quit: Ex-Chat]15:39
blackburnsure it is different methods15:39
pluskidbut C++ example runs without problem15:39
blackburnpluskid: const should be everywhere here I guess15:39
blackburndo you think so?15:39
blackburnah init_machines shouldn't be15:40
pluskidblackburn: but I don't think they are different in C++ here15:40
pluskidI mean copy and const copy15:40
blackburnpluskid: hmm anyway to be fixed15:40
pluskidfixing some other errors15:40
pluskidsegfault, etc15:40
blackburnpluskid: shall I wait for yor pr or fix it?15:41
pluskidblackburn: btw: I also fixed the error you mentioned before, ECOC producing 0 in binary labels15:41
blackburnoh that's nice15:41
blackburnI forgot to send an example sorry15:41
pluskidI'll push my current fixes15:41
blackburnok I'll merge it then15:41
pluskidthe remaining segfaults seems not easy to find in a moment15:41
blackburnbtw did you recover?15:44
pluskidnot yet, but much better15:44
blackburnI had some illness yesterday and now I'm going to fix a few bugs15:44
pluskidblackburn: get well15:44
blackburnyou too :)15:45
pluskidI'm afraid I have to go to the hospital again tomorrow :p15:45
pluskidblackburn: here comes the pull request15:46
pluskidI'll continue to hunt the other segfault15:46
pluskidblackburn: do you know how to tell where's the fault when the python example simply say segfault and print nothing?15:47
CIA-113shogun: Chiyuan Zhang master * r9a18a64 / (7 files in 4 dirs): Fix typeerror in SWIG python modular -
CIA-113shogun: Chiyuan Zhang master * r786359f / (6 files in 2 dirs): Fix ECOC memory error with new SGVector. -
CIA-113shogun: Chiyuan Zhang master * r485f94a / (4 files in 3 dirs): Fix ECOC Random coding bug. -
blackburnpluskid: valgrind15:48
blackburnpluskid: a lot of python fixes but segfaults are in the end so you can check it usually15:48
blackburnpython errors*15:48
-!- n4nd0 [] has joined #shogun16:02
blackburnoh that's becoming serious in moscow16:08
pluskidwhat's that?16:09
pluskidblackburn: please another small pull request16:10
blackburnpluskid: police is trying to block rally16:10
CIA-113shogun: Chiyuan Zhang master * r555d971 / src/shogun/multiclass/MulticlassOneVsOneStrategy.cpp : Fix memory error related to new SGVector API. -
CIA-113shogun: Chiyuan Zhang master * r394b8a5 / src/shogun/multiclass/ecoc/ECOCRandomSparseEncoder.h : Increase default maxiter of random sparse coding to avoid failure. -
pluskidtotally not understand Russian16:15
blackburnpluskid: you don't have to understand they are fighting :D16:16
pluskidthis kind of event will never be live casted in China :p16:17
blackburnpluskid: yeah16:17
blackburnpluskid: tomorrow putin is going to become a president officially - that's why people rally16:19
pluskidblackburn: Vladimir Putin ?16:20
blackburnyes that piece of shit16:21
pluskidwasn't he already the president?16:21
blackburnyes that bloody bastard was :D16:21
blackburnI hope he will die really soon16:21
pluskidhow long can your president keep his seat?16:21
pluskidPutin has been there since 2000 according to wikipedia16:21
blackburnpluskid: he get one fool (medvedev) to his seat instead of him for 4 years16:22
blackburnand now he becomes a president again16:22
blackburnfor third time16:22
blackburnbut now for 6 years16:22
pluskidinteresting to see the left president comes back again, haha, this will never happen in China16:23
blackburnpluskid: yeah you have one for a long time :D16:24
-!- n4nd0 [] has quit [Ping timeout: 276 seconds]17:23
-!- wiking [] has joined #shogun17:56
-!- wiking [] has quit [Changing host]17:56
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun17:56
-!- pluskid [] has quit [Quit: Leaving]17:57
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 245 seconds]19:51
-!- blackburn [~qdrgsm@] has joined #shogun19:51
-!- n4nd0 [] has joined #shogun19:59
-!- n4nd0 [] has quit [Quit: Ex-Chat]20:08
CIA-113shogun: Soeren Sonnenburg master * r7754d8c / src/shogun/multiclass/MulticlassOneVsOneStrategy.cpp : revert change - previous (simpler) solution was correct -
CIA-113shogun: Soeren Sonnenburg master * rcf38631 / src/shogun/multiclass/MulticlassOCAS.cpp : fix crasher in mc ocas -
CIA-113shogun: Soeren Sonnenburg master * r63cce74 / src/shogun/features/Labels.cpp : remove forgotten debug code -
blackburnsonney2k: hey20:24
@sonney2kblackburn, hey20:26
@sonney2kI wonder why now some pythonmodular examples fail20:26
@sonney2kI mean some in converters that were working before...20:26
blackburnsonney2k: one solution could be to merge my libedrt stuff20:27
blackburnmakes no use of sgvectors actually20:27
@sonney2khmmh even libshogun examples fail now20:29
@sonney2kthey all worked yesterday?!20:29
blackburnno idea20:30
blackburnsonney2k: again20:38
blackburnsonney2k: let me focus on libedrt - this code is going to be obsolete anyway20:39
-!- gsomix [~gsomix@] has joined #shogun21:09
gsomixblackburn, h21:10
blackburnsonney2k: arround?21:10
-!- vikram360 [~vikram360@] has quit [Ping timeout: 276 seconds]22:22
-!- vikram360 [~vikram360@] has joined #shogun22:23
@sonney2kblackburn, you cannot use SGMatrix in LINRECONSTRUCTION_THREAD_PARAM22:32
@sonney2kor any of the objects on stack22:32
blackburnsonney2k: yeah I know22:32
@sonney2kSG_MALLOC won't work for that22:32
@sonney2kyou have to use a ptr or explicit fields22:33
blackburnsonney2k: sg_malloc for what?22:33
blackburnsonney2k: anyway it is obsolete22:34
@sonney2kbut we have to fix this first22:35
@sonney2kit is not a good idea to merge some new version at this stage22:35
@sonney2kotherwise we risk to not getting back to a stable state22:35
blackburnsonney2k: right, but this could be useless..22:36
blackburnsonney2k: but why SG_MALLOC does not fit there?22:37
blackburnfor threads I mean22:37
@sonney2kit never fits22:38
@sonney2kwhen objects need to be allocated22:38
blackburnsonney2k: do you suggest new?22:39
@sonney2kone needs new []() for that22:39
@sonney2kI suggest to not use objects22:39
@sonney2kin these structs22:39
@sonney2k(ptrs are ok though)22:39
blackburnsonney2k: I am changing it to pointers22:39
blackburnsonney2k: do you have birch trees in berlin?22:45
CIA-113shogun: Soeren Sonnenburg master * r7d802f1 / src/shogun/machine/KernelMachine.cpp :22:48
CIA-113shogun: don't use Objects in structs allocated via SG_MALLOC22:48
CIA-113shogun: - convert SGVector to ptr*, len -
@sonney2kgsomix, please explain22:49
gsomixsonney2k, a?22:49
blackburnsomeone has a bad mood :D22:50
@sonney2kgsomix, a?22:55
@sonney2kyou asked sth above ...22:55
-!- vikram360 [~vikram360@] has quit [Ping timeout: 248 seconds]22:56
blackburnsonney2k: battlesystem operational22:56
-!- vikram360 [~vikram360@] has joined #shogun22:57
gsomixsonney2k, ah. we add MemoryBlock(ptr, size) in sg_mallocs.22:57
gsomixand remove MemoryBlock(ptr).22:57
CIA-113shogun: Sergey Lisitsyn master * r3bc5f3e / (3 files): Fixed crashers in converters -
blackburnwas trivial22:58
gsomixMemoryBlock(ptr, size) and MemoryBlock(ptr) have different hashes.22:58
@sonney2kah ok22:59
@sonney2kgsomix, so did you fix it?23:00
gsomixsonney2k, in progress.23:00
blackburnsonney2k: one multiclass example and modelselection are b0rken23:01
blackburnalmost ok23:01
blackburnI will take care of multiclass crasher23:01
@sonney2kblackburn, if things are back to normal we can continue with SGMatrix and friends23:02
@sonney2kshouldn't be as intrusive (not so many things use it...)23:02
blackburnI'll try to fix multiclass today23:03
blackburntomorrow I'm out - exam on 8th23:03
CIA-113shogun: Soeren Sonnenburg master * r8264e5c / (src/shogun/features/Subset.cpp src/shogun/features/Subset.h): no longer copy subset index -
@sonney2kblackburn, k23:05
@sonney2kmultilass seems not so easy ...23:05
@sonney2ksomething wrong with subsets23:05
gsomixsonney2k, btw, gf said now that you looks very nice :) just funny fact23:08
blackburnsonney2k: it can be the same issue in evaluation, modelselection and multiclass23:09
blackburnI bet it is23:09
@sonney2kgsomix, haha - probably because picture was taken when I was your age23:15
@sonney2kwell +7 I guess :D23:15
@sonney2kblackburn, yeah23:15
@sonney2keasier for heiko to fix this23:15
blackburnsonney2k: no in ICML video you are a little younger than now :)23:16
@sonney2kstill not twenty like gsomix23:16
@sonney2kalright I need sleep23:16
blackburnsee you23:17
gsomixsonney2k, we watched videolectures.23:17
gsomixgood night :)23:17
--- Log closed Mon May 07 00:00:37 2012