Open in new window / Try shogun cloud
--- Log opened Sat Nov 05 00:00:13 2011
blackburn1damn I can't solve it :(00:42
blackburn1oh.. it is a lapack bug01:10
-!- blackburn1 [~blackburn@188.122.237.76] has quit [Quit: Leaving.]01:21
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun06:44
-!- in3xes_ [~in3xes@180.149.49.227] has joined #shogun06:58
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 245 seconds]07:01
-!- in3xes_ is now known as in3xes07:12
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 252 seconds]08:20
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun08:31
-!- in3xes [~in3xes@180.149.49.227] has quit [Remote host closed the connection]09:06
-!- blackburn [~blackburn@109.226.81.30] has joined #shogun10:06
-!- blackburn [~blackburn@109.226.81.30] has quit [Read error: Connection reset by peer]10:14
-!- blackburn [~blackburn@188.122.224.207] has joined #shogun10:14
-!- blackburn [~blackburn@188.122.224.207] has quit [Read error: Connection reset by peer]10:36
-!- blackburn [~blackburn@85.114.187.132] has joined #shogun10:37
-!- blackburn [~blackburn@85.114.187.132] has quit [Read error: Connection reset by peer]10:47
-!- blackburn [~blackburn@188.122.228.165] has joined #shogun10:56
-!- blackburn1 [~blackburn@31.28.36.115] has joined #shogun11:12
-!- blackburn [~blackburn@188.122.228.165] has quit [Ping timeout: 258 seconds]11:14
-!- blackburn [~blackburn@85.114.185.74] has joined #shogun11:21
-!- blackburn1 [~blackburn@31.28.36.115] has quit [Ping timeout: 260 seconds]11:23
blackburnsonney2k: around?11:39
blackburnsonney2k: my issue with thread unsafety was related to thread-unsafe DLAMCH routine, which determines machine eps11:54
blackburnit is fucking awful - the whole LAPACK is thread-safe but DLAMCH11:55
-!- blackburn1 [~blackburn@109.226.124.12] has joined #shogun12:17
-!- blackburn [~blackburn@85.114.185.74] has quit [Read error: Connection reset by peer]12:18
CIA-37shogun: Sergey Lisitsyn master * r4b6f385 / (3 files in 2 dirs): Removed float lapack wrappers - http://git.io/N57b8A16:40
CIA-37shogun: Sergey Lisitsyn master * rdafece0 / src/shogun/converter/LocalTangentSpaceAlignment.cpp : Optimized LTSA - http://git.io/sWLt1w16:42
CIA-37shogun: Sergey Lisitsyn master * r5280282 / src/shogun/converter/KernelLocalTangentSpaceAlignment.cpp : Optimized KLTSA - http://git.io/6YBvFw16:42
CIA-37shogun: Sergey Lisitsyn master * r59c976a / examples/undocumented/python_modular/converter_localitypreservingprojections_modular.py : Added LPP python modular example - http://git.io/c4MKsw16:44
shogun-buildbotbuild #312 of octave_static is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/31216:48
@sonney2kblackburn1, it is a bit surprising - thought that lapack is 100% thread safe...19:16
@sonney2ka bit weird for a lib of this kind19:16
blackburn1sonney2k: 3.3 is so, but 3.2.1's dlamch is not19:16
@sonney2kblackburn1, well then add a test for that version and force unicore if <3.319:17
blackburn1sonney2k: how can i test lapack version in runtime?19:17
@sonney2kblackburn1, compile time - not runtime19:17
blackburn1ehh okay19:17
@sonney2kblackburn1, but maybe the even have a api function for checking the version?19:18
blackburn1sonney2k: I don't know it yet19:18
@sonney2kno idea - check the source :)19:18
@sonney2kblackburn1, have you read Madeleines question on the mailinglist?19:26
blackburn1sonney2k: yes, but I'm afraid I can't help she19:29
blackburn1sonney2k: don't you think alessandro is a bit offensive? :D19:30
@sonney2kblackburn1, no why?19:32
@sonney2khe is right19:32
@sonney2kif it doesn't work w/o jblas then it doesn't make sense to just print a warning19:32
@sonney2kand one should fail instead19:32
@sonney2kblackburn1, do we have MeanAbsoluteError?19:40
@sonney2kin evaluation I mean?19:40
blackburn1sonney2k: mean squared error19:42
blackburn1sonney2k: I'm about his writing style :)19:42
@sonney2kbut that cannot do absoute right?19:43
@sonney2kI just started writing it...19:43
blackburn1sonney2k: what is the difference?19:43
@sonney2kone is ||.||^219:43
@sonney2kone ||.||19:43
@sonney2k| |19:44
blackburn1sonney2k: does it have some 'fundamental' difference?19:44
@sonney2kfundamental?19:44
@sonney2kit will punish differences close to zero much more19:45
blackburn1eh nevermind, bad word19:45
blackburn1ah I see19:45
blackburn1sonney2k: you could meansquarederror -> meanerror19:45
blackburn1and add options to square or not19:45
blackburn1sonney2k: about madeleine question - I really do not understand much about her problem, it is related to things you are better with than me19:47
@sonney2kblackburn1, ok implemented that mean absolute error thing19:48
blackburn1http://www.netlib.org/lapack/lapack-3.3.0.html19:49
blackburn1sonney2k: Thread safe xLAMCH: SLAMCH and DLAMCH were the only two routines not thread-safe in LAPACK-3.2. This is fixed: all routines in LAPACK-3.3 are now thread-safe.19:49
blackburn1it is really awful19:49
blackburn1sonney2k: could you please describe how to link shogun with my own lapack instead of atlas' one&19:50
blackburn1?19:50
blackburn1I spent too much time on it, decided to ask you instead19:50
@sonney2kyour own lapack?19:51
blackburn1hah19:51
@sonney2kjust set the library / include paths on configure?19:51
blackburn1again bad wording19:51
blackburn1sonney2k: yes a did a try19:51
blackburn1but it links with atlas' one :(19:52
blackburn1I did*19:52
blackburn1I'll make a try again19:52
blackburn1sonney2k: today I've tested kernel ltsa on 84x112 pictures19:53
blackburn1it was 80s faster than original LTSA with the same results19:54
blackburn180x*19:54
blackburn1in fact I started writing a paper, what do you think, does this kind of algo extension worths to be published in any journal?19:55
blackburn1I'm a total newbie in science affairs hah19:56
@sonney2kblackburn1, if the extension is not trivial and ltsa is some established algo - sure20:00
blackburn1sonney2k: what is the trivial extension?20:00
CIA-37shogun: Soeren Sonnenburg master * r857311c / (4 files in 2 dirs): introduce MeanAbsoluteError - http://git.io/H8jXVw20:00
@sonney2kblackburn1, something obvious for experts20:00
@sonney2kblackburn1, btw look at the copyright line in the above ^ commit20:01
blackburn1sonney2k: can I consider you as expert? hehe20:01
blackburn1why me?20:01
@sonney2kblackburn1, no - I don't know anything about LTSA20:01
@sonney2kblackburn1, I used your MSE code20:02
blackburn1sonney2k: do you know anything about SVD and kernel matrix?20:02
@sonney2kkernel matrix? no clue what that could be20:02
blackburn1matrix of pairwise kernels?20:02
blackburn1k11 k12 k13 ...20:02
blackburn1sonney2k: in the ltsa algorithm there are right singular vectors of centered feature matrix [X-[mean feature vector]] are being computed20:04
blackburn1I proved that it is the same to find eigenvectors of centered gram matrix20:05
blackburn1=> faster, makes possible to use kernels20:06
blackburn1sonney2k: it is obvious for me, but I don't know if it is for anyone else20:07
blackburn1sonney2k: do not leave me alone when I'm ask stupid Qs :D20:20
-!- heiko [~heiko@host86-160-249-224.range86-160.btcentralplus.com] has joined #shogun22:23
heikosonney2k, just read the mails regarding the modsel problem, will try to figure our what is going on now.22:24
blackburn1howdy!22:24
blackburn1saturday night fixing model selection show22:25
heikohi blackburn22:27
heikohehe, well just read the mail22:27
heikoand i just got a minute22:28
heikovery busy the rest of the time22:28
heikoeverything ok in russia? :)22:28
blackburn1heiko: sure, we are vodka-enough22:29
heikohehe :)22:29
heikowhat about your studies, everything going well?22:29
blackburn1heiko: oh studies are going ok, though I have a little spare time because I started working at netcracker22:30
heikonetcracker?22:30
blackburn1heiko: yeah, http://netcracker.com/en/22:31
heikoah ok, new job22:32
heikocool22:32
heikois it nice?22:32
blackburn1heiko: it would be better to say first job :)22:32
blackburn1not very nice but 2 pros: a) I'm paid b) I work part-time :)22:33
heikook :)22:33
blackburn1heiko: what about you? what do you do?22:34
heikoI am started with a master in machine learning in london22:34
heikohttp://www.csml.ucl.ac.uk/courses/msc_ml/22:34
heikoreally cool22:34
heikobut consumes all of my time22:34
blackburn1heiko: so sad you can't do shoguning :)22:36
heikoyes, indeed22:36
heikobut thigns will change22:36
heikobut for the moment it is like it is22:36
heikothink i will have some free time around january22:36
heikoand then in march22:36
heikoand then in the summer also22:36
blackburn1I guess we will never have much free time as we had in summer22:36
heikoyes i think so22:38
heikobtw current head does not compile here22:38
blackburn1heiko: error?22:38
heikoIn file included from base/class_list.cpp:66:0:22:39
heiko./kernel/MultitaskKernelMaskPairNormalizer.h: In member function 'virtual bool shogun::CMultitaskKernelMaskPairNormalizer::init(shogun::CKernel*)':22:39
heiko./kernel/MultitaskKernelMaskPairNormalizer.h:76:7: error: 'string' is not a member of 'std'22:39
heikoguess include is missing22:39
blackburn1I would personally kick chris' ass for usage of std but he is moving to NY22:39
heiko;)22:39
blackburn1that would be hard to catch him22:40
blackburn1okay I'll take a look22:40
heikoblackburn can you help me with a little git problem?22:41
blackburn1heiko:22:41
blackburn1sure22:41
heikoI still have some work in my master branch and i committed it already to my github repo because i wanted to backup it22:41
heikohowever, the code is not yet really working22:41
heikobut if i want to do a bugfix i cannot send pullrequests22:41
heikobecause all my changes are in there22:42
heikodo you know how to move this stuff to another branch or something so that i can work on head without any of my other changes involved?22:42
blackburn1you can stash your changes for now22:42
blackburn1git stash22:43
blackburn1it will keep your changes until you will git stash pop it22:43
heikodoes that even work if they are already in my github repo?22:43
blackburn1heiko: ah did you commit it?22:44
heikoyes22:44
blackburn1hmm22:44
heikoits also already pushed22:44
heikoyou know i wanted to backup itonline, kind of stupid22:45
blackburn1not a stupid22:45
heikobut i was in a hurry and i did not want to loose all of my changes22:45
heikomoving to london, plane, laptop broken or anything...you know22:45
heikobut now I want to put them away to work on them later22:45
blackburn1huh that's difficult22:46
heikoyes.22:46
heikoi thought of perhaps to the brute force way and save my repo copy here and just reset my github account22:46
heikoand then work the stuff in again later22:46
heikoi have to reconsider it anyway22:46
heikothink i will jsut do this22:47
blackburn1heiko: I'm sorry but I have no idea how to change your base in this way22:48
blackburn1I'm still lame with some git issues22:48
heikome too :)22:48
heikoreforking ...22:49
blackburn1heiko: I would know more if I was using forks and etc22:49
blackburn1but I commit right into the shogun and it is pretty simplified22:50
heikoyes, with all the forks and stuff -- thats pretty messy sometimes22:50
heikowell i just cloned it22:50
blackburn1heiko: from my experience it comes easy to have some dev branch22:51
heikoyes, true true, i normally do that22:52
blackburn1but even this becomes messy sometimes22:52
heikobut not this time :)22:52
blackburn1oh I really need to stop procrastinate and compiling lapack22:53
blackburn1and start research on road sign recognition22:53
blackburn1:D22:53
heikofor studies?22:53
blackburn1yeah, it is my bachelor's work or so22:54
blackburn1but if I go through it with high accuracy I would even get paid for that22:54
heikothat sounds cool22:55
heikoblackburn1, the file with the error is full of std::22:57
heikoi will just add an include or so22:57
blackburn1heiko: are you going to fix it? I can22:57
heikoi can do it22:57
blackburn1heiko: if you have time please do, cause it is not being reproduced on my22:57
blackburn1machine22:58
heikodone23:02
blackburn1heiko: does it work?23:06
blackburn1heiko: there are vector too, is it ok?23:06
heikoyes, but they are included in Kernel.h23:06
blackburn1ah ok23:06
CIA-37shogun: Heiko Strathmann master * r54afa01 / src/shogun/kernel/MultitaskKernelMaskPairNormalizer.h :23:07
CIA-37shogun: temporary added include to compile again23:07
CIA-37shogun: (string include was recently removed out of Kernel.h) - http://git.io/qOdR6w23:07
-!- heiko [~heiko@host86-160-249-224.range86-160.btcentralplus.com] has quit [Quit: Leaving.]23:40
-!- heiko [~heiko@host86-160-249-224.range86-160.btcentralplus.com] has joined #shogun23:51
heikoblackburn1 goodn823:51
-!- heiko [~heiko@host86-160-249-224.range86-160.btcentralplus.com] has left #shogun []23:51
blackburn1heiko see you23:52
--- Log closed Sun Nov 06 00:00:19 2011