Open in new window / Try shogun cloud
--- Log opened Wed Jul 25 00:00:05 2012
--- Day changed Wed Jul 25 2012
yoodense, string, any ..00:00
blackburnwell in case of dense00:00
yooI have my own function then for printing string features00:00
blackburnstring is a little more complex - you iterate from 0 to features->get_num_vectors() and get and print strings inside00:01
yooyes I have that00:01
yoojust wanna to understand to way iterator works but I loosing my french in that00:01
yooby the way .. can you explain me the purpose of CMulticlassMultipleOutput ?00:03
blackburnyoo: heh00:03
blackburnto support top k outputs00:03
blackburni.e. if you have 50 classes and you want to store top 5 predictions00:04
yooI am suffering of the lack of c++ example towards to python modular ^^00:04
blackburnyoo: feel free to contribute ;)00:04
blackburnyoo: it is much easier to use shogun from python00:05
yoosure I did a little  bit with mkl and cross valid example :p00:05
yooblackburn: yes but lots of little modification have to be made on c++00:05
blackburnI am open for good ideas00:06
yooblackburn: for example in crossvalid and multiclass: cannot get the average confusion matrix after many crossvalid runs ..00:06
yooI tried to add this but thats not trivial to consider multiclasslabels for base crossvalidation object00:07
yoosee what I mean ?00:07
yoosorry its late my english is going down ..00:07
blackburnisn't it native for you?00:07
blackburnI thought it is :)00:08
yooI would like to ass more evaluation for multiclass00:08
yoo-ss +dd :p00:08
blackburnyeah we can ass some00:08
blackburnas for confusion matrix00:09
yooconfusion matrix after Xvalidation and EER for intra class could be fun )?00:09
blackburnI think one way is to extend it using msoutput00:09
blackburnwhat is EER?00:09
yooequal error rate00:09
-!- blackburn [~blackburn@] has left #shogun []00:10
yoofor each class in multiclass evaluation00:10
-!- blackburn [~blackburn@] has joined #shogun00:10
blackburnyoo: how to compute it?00:11
-!- needsch175 [] has quit [Ping timeout: 245 seconds]00:11
yoothe same way you compute ROC curve00:13
blackburnso ROC one vs rest?00:13
blackburnI wanted to add that too00:14
yooI ll be back tommorow to discuss this if you want some help00:14
yoomsoutput should be the right way00:15
blackburnit is pretty easy actually00:15
blackburnI think00:15
blackburnI'll do that a little later00:15
yooI will follow this to better understand shougunlib00:15
blackburnone problem is here though00:15
blackburnlet me check00:15
yoobtw I never succeed in passing msoutput in evaluate function without strange outputs ...00:16
yooand seg fault but maybe I mentioned it already before00:16
blackburndo you have that example still?00:16
yoowhich one ?00:16
blackburnsegfaulting one00:16
yoolet me check00:17
yooI just use this one
yooand add msoutput object in evaluate00:18
yoothink to change the inputs00:18
blackburnyoo: hmm it works here00:22
blackburnokay my bad probably00:27
blackburnyoo: will commit a fix in a few mins00:29
yoogood night then, see you tomorrow00:29
blackburnyoo: what is you are applying shogun for?00:30
yoomachine learning =)00:30
blackburnokay good night00:30
yooschool project in action recognition00:30
yooapply to face expression mostly00:30
blackburnschool like high school?00:30
yooyes I am 13 y old00:31
blackburnno way00:31
yoocoded my first neural nets at 1100:31
yoosee ya, good night00:31
blackburnhard to believe but you would become youngest shogun user ever :D00:32
yoojust joking00:32
-!- yoo [575b08cb@gateway/web/freenode/ip.] has quit [Quit: Page closed]00:32
wikingblackburn: :DDDD00:34
CIA-18shogun: Sergey Lisitsyn master * r8e7d14c / src/shogun/modelselection/ModelSelectionOutput.cpp : Added support of other than linear machines in modelselectionoutput -
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]00:41
-!- romi_ [] has quit [Quit: Leaving]01:15
-!- alexlovesdata [] has left #shogun []01:47
shogun-buildbotbuild #33 of nightly_none is complete: Failure [failed compile]  Build details are at
shogun-buildbotbuild #39 of nightly_default is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #29 of nightly_all is complete: Success [build successful]  Build details are at
-!- n4nd0 [] has joined #shogun07:00
-!- n4nd0 [] has quit [Quit: leaving]08:35
-!- uricamic [~uricamic@2001:718:2:1634:3de0:9a42:a489:6401] has joined #shogun08:46
-!- pluskid [] has joined #shogun09:06
-!- uricamic1 [] has joined #shogun09:36
-!- uricamic [~uricamic@2001:718:2:1634:3de0:9a42:a489:6401] has quit [Quit: Leaving.]09:36
-!- uricamic1 [] has quit [Quit: Leaving.]10:01
-!- uricamic [~uricamic@2001:718:2:1634:3de0:9a42:a489:6401] has joined #shogun10:04
-!- needsch [] has joined #shogun10:34
-!- bern4rd [53368e31@gateway/web/freenode/ip.] has joined #shogun10:46
-!- bern4rd [53368e31@gateway/web/freenode/ip.] has quit [Quit: Page closed]11:16
-!- n4nd0 [] has joined #shogun11:18
n4nd0someone available to help me out?11:28
koen-shogunthe question would be... with what?11:29
n4nd0koen-shogun: no idea about why this seg fault happens
pluskidn4nd0: it might be caused by some memory error, wild pointers or anything terrible... maybe run with valgrind might give some hint?11:35
n4nd0pluskid: it doesn't seg fault with valgrind11:36
n4nd0pluskid: do you think that reading valgrind's trace can give me a clue about it in any case?11:36
pluskiddoes valgrind report any memory leak or something?11:36
pluskidif valgrind says some pointers are used before initialized, or some memory are written out of bound, etc. it might be the cause of the problem11:37
n4nd0I have some Conditional jump or move depends on uninitialised value(s) and Invalid write of size ...11:38
n4nd0I am going to investigate11:38
pluskidn4nd0: good luck then, I'll be right back (for supper :))11:38
n4nd0pluskid: thank you! I don't have the habit to use valgrind to check seg faults, but to look for memory leaks11:39
-!- needsch [] has quit [Ping timeout: 250 seconds]11:51
n4nd0pluskid: problem detected and solved :)11:54
droopycool :^)11:54
n4nd0a couple of errors doing arithmetic with indices11:54
n4nd0pluskid: thank you again!11:54
pluskidn4nd0: cool!11:55
koen-shogunwhat you can do to find uninitialized memory: during compilation define -DMALLOC_PERTURB_=6911:59
koen-shogunthis will initialize all memory allocated using malloc with 69, instead of being uninitialized12:00
koen-shoguncan shake out some memory bugs quite easily12:00
n4nd0koen-shogun: thank you for that12:07
n4nd0this one was a problem related to writing out of bounds though12:07
n4nd0I will be back later, bye!12:08
-!- n4nd0 [] has quit [Read error: Operation timed out]12:12
wikingoh you are not here12:45
-!- gsomix [~gsomix@] has joined #shogun13:34
gsomixgood day13:34
gsomixsonney2k, around?13:34
wikingheiko ?13:48
wiking!seen heiko13:48
droopyheiko ( was last seen parting #shogun 18 hours, 9 minutes ago stating "{}".13:48
wikingthnx droopy13:48
-!- yoo [575b08cb@gateway/web/freenode/ip.] has joined #shogun14:09
yoohi all14:09
gsomixdroopy, mimimi.14:10
-!- needsch [] has joined #shogun15:29
-!- gsomix [~gsomix@] has quit [Read error: Operation timed out]15:33
-!- bern4rd [53368e31@gateway/web/freenode/ip.] has joined #shogun15:41
-!- heiko1 [] has joined #shogun15:54
-!- yoo [575b08cb@gateway/web/freenode/ip.] has quit [Quit: Page closed]16:06
-!- pluskid [] has quit [Quit: Leaving]16:40
-!- uricamic [~uricamic@2001:718:2:1634:3de0:9a42:a489:6401] has quit [Quit: Leaving.]16:41
CIA-18shogun: Heiko Strathmann master * rd8d1158 / (16 files in 3 dirs): changed class structure a bit to avoid redundancy and allowing easier -
CIA-18shogun: Heiko Strathmann master * rfac0f66 / examples/undocumented/python_modular/ : comment changes -
CIA-18shogun: Heiko Strathmann master * rf3b1c98 / (17 files in 4 dirs): Merge pull request #674 from karlnapf/master -
CIA-18shogun: Heiko Strathmann master * rfa25275 / examples/undocumented/python_modular/ : initial commit -
CIA-18shogun: Heiko Strathmann master * ra88feed / examples/undocumented/python_modular/ : Merge pull request #675 from karlnapf/master -
-!- bern4rd [53368e31@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:13
-!- n4nd0 [] has joined #shogun17:34
-!- n4nd0 [] has quit [Client Quit]17:34
-!- blackburn [~blackburn@] has joined #shogun17:44
-!- tiger_eye [] has joined #shogun17:55
CIA-18shogun: Heiko Strathmann master * rb427525 / (2 files): added inverse gamma cdf -
CIA-18shogun: Heiko Strathmann master * r8f57540 / examples/undocumented/libshogun/statistics.cpp : added test for inverse gamma cdf -
CIA-18shogun: Heiko Strathmann master * radc555a / (4 files): added threshold computation in a better way -
CIA-18shogun: Heiko Strathmann master * r0f8b2a9 / (7 files in 3 dirs): Merge pull request #677 from karlnapf/master -
-!- heiko1 [] has left #shogun []17:58
-!- yoo [575b08cb@gateway/web/freenode/ip.] has joined #shogun18:03
-!- yoo [575b08cb@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:10
CIA-18shogun: Heiko Strathmann master * r6ef226b / (2 files): renamed inverse_normal_distribution to inverse_normal_cdf -
CIA-18shogun: Heiko Strathmann master * ra32c81b / src/shogun/statistics/HSIC.cpp : added comments -
CIA-18shogun: Heiko Strathmann master * r83f385e / src/shogun/statistics/LinearTimeMMD.cpp : minor checks and enabling threshold computations -
CIA-18shogun: Heiko Strathmann master * r5b3d566 / (4 files in 2 dirs): Merge pull request #678 from karlnapf/master -
blackburnwiking: ping18:42
blackburnI just compared asm outputs of eigen3, loop and blas18:50
blackburntotally crazy18:50
blackburneigen3 one in particular18:50
-!- shogun-buildbot [] has quit [Ping timeout: 255 seconds]18:53
-!- shogun-buildbot [] has joined #shogun18:53
shogun-buildbotbuild #179 of deb2 - static_interfaces is complete: Failure [failed test octave_static]  Build details are at  blamelist: Heiko Strathmann <>19:38
wikingblackburn: resutl19:52
blackburnwiking: I compared on vectors of length 1 000 00019:53
blackburncomputed average or 20000 runs19:53
blackburnblas takes 2251 clock in average here19:53
blackburnbasic loop 223419:54
blackburneigen takes 193419:54
blackburnI run it a few times and results are similar so I expect eigen is really faster20:00
-!- n4nd0 [] has joined #shogun20:41
n4nd0I found this article interesting:
CIA-18shogun: Sergey Lisitsyn master * r086ee0a / (3 files in 2 dirs): Added custom evaluations support for model selection output -
shogun-buildbotbuild #180 of deb2 - static_interfaces is complete: Success [build successful]  Build details are at
-!- needsch [] has quit [Ping timeout: 244 seconds]21:58
-!- needsch175 [] has joined #shogun22:00
-!- heiko [] has joined #shogun22:01
-!- heiko [] has left #shogun []22:01
-!- zxtx [] has quit [Ping timeout: 264 seconds]22:32
wikingblackburn: why wouldn't it throw23:00
wikingit throws :)23:00
blackburnwiking: well at least it works ;)23:01
@sonney2kblackburn, did you run your blas benchmark on optimized blas?23:01
blackburnsonney2k: atlas23:01
@sonney2kor just the crap that gets installed?23:01
@sonney2kwell atlas selfcompiled?23:02
blackburnsonney2k: installed atlas23:02
@sonney2kor the standard atlas23:02
@sonney2kok then23:02
@sonney2kthen it is the same as basic loop23:02
blackburnsonney2k: let me share code23:02
@sonney2kyou need to compile blas23:02
@sonney2kyes let me try this23:02
blackburnfurthermore may be you will fix something in test that can be unfair23:03
@sonney2kwhat did you compute btw?23:03
blackburnsonney2k: dot23:03
blackburnsonney2k: first one is to power up my cpu :D :D23:05
blackburnwe have no step back actually - I already told eigen team we use they stuff :D23:10
@sonney2kblackburn, arg I recently updated atlas and don't will have to compile it again...23:19
@sonney2kthat can take ages23:19
@sonney2kblackburn, current output23:19
@sonney2k elapsed 2973.500000 clocks23:19
@sonney2kblas elapsed 3006.500000 clocks23:19
@sonney2kloop elapsed 3031.500000 clocks23:19
@sonney2keigen elapsed 3006.000000 clocks23:19
blackburnwell pretty same23:19
blackburnthey all are23:19
@sonney2kyeah no difference23:20
@sonney2kI think dot is not the best op23:20
blackburnsonney2k: I measured it because we heavy rely on it23:20
@sonney2ksth like matrix matrix mul or vector woudl be better23:20
@sonney2kblackburn, but measuring time diff with clock() is also not so good23:20
blackburnmatrix matrix is rare here in shogun23:20
blackburnsonney2k: gettimeofday?23:20
blackburnI failed to recall the best way23:21
@sonney2kahh I totally forgot about renameing Ctime functions23:21
blackburnsonney2k: after we add -I/usr/include/eigen3 I have a lot of warnings23:22
blackburnany idea why they haven't been there before?23:22
-!- n4nd0 [] has quit [Quit: leaving]23:47
--- Log closed Thu Jul 26 00:00:17 2012