Open in new window / Try shogun cloud
--- Log opened Tue May 22 00:00:05 2012
--- Day changed Tue May 22 2012
-!- cronor [] has quit [Ping timeout: 252 seconds]00:00
-!- cronor [] has joined #shogun00:26
-!- karlnapf [] has joined #shogun00:44
-!- karlnapf [] has left #shogun []00:56
-!- ckwidmer [] has quit [Remote host closed the connection]01:31
CIA-113shogun: Sergey Lisitsyn master * r17f015e / (9 files in 2 dirs): Fixes for binary evaluations -
CIA-113shogun: Sergey Lisitsyn master * r4b44437 / (9 files in 4 dirs): Renamed LT_REAL to LT_REGRESSIONw -
-!- cronor [] has quit [Quit: cronor]01:51
CIA-113shogun: Sergey Lisitsyn master * r74ec526 / src/shogun/machine/DistanceMachine.h : Removed junk methods from DistanceMachine -
blackburnhahh wrong window01:54
blackburnn4ndo: ok waiting for your commit ;)02:15
-!- blackburn [~blackburn@] has left #shogun []02:15
-!- av3ngr [av3ngr@nat/redhat/x-pmomtrbvblgrgxfj] has joined #shogun02:16
-!- av3ngr [av3ngr@nat/redhat/x-pmomtrbvblgrgxfj] has quit [Read error: Connection reset by peer]02:34
-!- abn_ [av3ngr@nat/redhat/x-xuvdsxngulszmtkw] has joined #shogun02:34
-!- vikram360 [~vikram360@] has joined #shogun02:36
-!- abn__ [av3ngr@nat/redhat/x-nwncusfehavfbppt] has joined #shogun02:41
-!- abn__ [av3ngr@nat/redhat/x-nwncusfehavfbppt] has quit [Read error: Connection reset by peer]02:42
-!- abn__ [av3ngr@nat/redhat/x-kbzvzcoylxapkpmg] has joined #shogun02:43
-!- abn_ [av3ngr@nat/redhat/x-xuvdsxngulszmtkw] has quit [Ping timeout: 244 seconds]02:45
-!- abn__ [av3ngr@nat/redhat/x-kbzvzcoylxapkpmg] has quit [Read error: Connection reset by peer]02:56
-!- av3ngr [av3ngr@nat/redhat/x-gphxnzucexbjznbd] has joined #shogun02:57
-!- abn_ [av3ngr@nat/redhat/x-ptsowgvajarwgynm] has joined #shogun03:14
-!- abn_ [av3ngr@nat/redhat/x-ptsowgvajarwgynm] has quit [Read error: Connection reset by peer]03:15
-!- abn__ [av3ngr@nat/redhat/x-wglprlbbgqmbekcb] has joined #shogun03:16
-!- av3ngr [av3ngr@nat/redhat/x-gphxnzucexbjznbd] has quit [Ping timeout: 240 seconds]03:18
-!- vikram360 [~vikram360@] has quit [Ping timeout: 246 seconds]03:19
-!- vikram360 [~vikram360@] has joined #shogun03:20
-!- abn__ [av3ngr@nat/redhat/x-wglprlbbgqmbekcb] has quit [Read error: Connection reset by peer]03:50
-!- av3ngr [av3ngr@nat/redhat/x-kdvliwlttbrjbtzi] has joined #shogun03:50
-!- vikram360 [~vikram360@] has quit [Ping timeout: 246 seconds]04:02
-!- sonney2k [] has quit [Ping timeout: 245 seconds]04:12
-!- sonney2k [] has joined #shogun04:12
-!- av3ngr [av3ngr@nat/redhat/x-kdvliwlttbrjbtzi] has quit [Read error: Connection reset by peer]04:25
-!- av3ngr [av3ngr@nat/redhat/x-inlylnpychijztpn] has joined #shogun04:25
-!- abn_ [av3ngr@nat/redhat/x-cgbaacwlhhwpwsws] has joined #shogun04:39
-!- av3ngr [av3ngr@nat/redhat/x-inlylnpychijztpn] has quit [Ping timeout: 245 seconds]04:43
-!- abn__ [av3ngr@nat/redhat/x-iuhkwbiyovdvjszt] has joined #shogun04:53
-!- vikram360 [~vikram360@] has joined #shogun04:53
-!- abn_ [av3ngr@nat/redhat/x-cgbaacwlhhwpwsws] has quit [Ping timeout: 256 seconds]04:57
-!- abn_ [av3ngr@nat/redhat/x-kkvvuspxqszxjeyk] has joined #shogun05:12
-!- abn__ [av3ngr@nat/redhat/x-iuhkwbiyovdvjszt] has quit [Ping timeout: 260 seconds]05:16
-!- abn_ [av3ngr@nat/redhat/x-kkvvuspxqszxjeyk] has quit [Read error: Connection reset by peer]05:47
-!- abn_ [av3ngr@nat/redhat/x-bdztxorawqjwtbjr] has joined #shogun05:47
-!- abn_ [av3ngr@nat/redhat/x-bdztxorawqjwtbjr] has quit [Read error: Connection reset by peer]06:11
-!- abn_ [av3ngr@nat/redhat/x-chtrdmvefhcomhyl] has joined #shogun06:12
-!- gsomix [~gsomix@] has quit [Ping timeout: 240 seconds]06:35
-!- abn__ [av3ngr@nat/redhat/x-apmagomsomgqogex] has joined #shogun06:43
-!- abn_ [av3ngr@nat/redhat/x-chtrdmvefhcomhyl] has quit [Ping timeout: 250 seconds]06:47
-!- vikram360 [~vikram360@] has quit [Ping timeout: 256 seconds]06:51
-!- vikram360 [~vikram360@] has joined #shogun06:57
-!- abn_ [av3ngr@nat/redhat/x-zlsbjwqpoxgnkuui] has joined #shogun06:59
-!- abn__ [av3ngr@nat/redhat/x-apmagomsomgqogex] has quit [Ping timeout: 248 seconds]07:03
-!- sonne|work [~sonnenbu@] has quit [Read error: Operation timed out]07:04
-!- av3ngr [av3ngr@nat/redhat/x-mkyrjichayeuozkq] has joined #shogun07:06
-!- sonne|work [~sonnenbu@] has joined #shogun07:06
-!- av3ngr [av3ngr@nat/redhat/x-mkyrjichayeuozkq] has quit [Read error: Connection reset by peer]07:09
-!- abn__ [av3ngr@nat/redhat/x-hicajxdxbmsvfprp] has joined #shogun07:09
-!- abn_ [av3ngr@nat/redhat/x-zlsbjwqpoxgnkuui] has quit [Ping timeout: 250 seconds]07:10
-!- av3ngr [av3ngr@nat/redhat/x-upbeghnmqgjxwjor] has joined #shogun07:24
-!- abn__ [av3ngr@nat/redhat/x-hicajxdxbmsvfprp] has quit [Ping timeout: 276 seconds]07:28
-!- av3ngr [av3ngr@nat/redhat/x-upbeghnmqgjxwjor] has quit [Read error: Connection reset by peer]07:34
-!- abn_ [av3ngr@nat/redhat/x-dkjdkzksiuznnryi] has joined #shogun07:35
-!- abn__ [av3ngr@nat/redhat/x-wzgmklxvgtfdklwi] has joined #shogun07:44
-!- abn_ [av3ngr@nat/redhat/x-dkjdkzksiuznnryi] has quit [Ping timeout: 265 seconds]07:48
-!- mode/#shogun [+o sonney2k] by ChanServ07:50
CIA-113shogun: Soeren Sonnenburg master * rc0954f1 / (49 files): fix python modular example for new label system -
-!- abn__ [av3ngr@nat/redhat/x-wzgmklxvgtfdklwi] has quit [Read error: Connection reset by peer]08:04
-!- abn__ [av3ngr@nat/redhat/x-caxmknntobyjaorc] has joined #shogun08:04
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]08:09
CIA-113shogun: Soeren Sonnenburg master * rd7d1505 / src/interfaces/modular/Machine.i : use more fine grained %rename of apply() to get converters back to work -
-!- abn__ [av3ngr@nat/redhat/x-caxmknntobyjaorc] has quit [Read error: Connection reset by peer]08:25
-!- av3ngr [av3ngr@nat/redhat/x-hbmcwvnwmfwdwwgk] has joined #shogun08:26
-!- n4nd0 [] has joined #shogun08:36
-!- av3ngr [av3ngr@nat/redhat/x-hbmcwvnwmfwdwwgk] has quit [Read error: Connection reset by peer]08:39
-!- av3ngr [av3ngr@nat/redhat/x-htrnusvtyczkfwxc] has joined #shogun08:39
-!- av3ngr [av3ngr@nat/redhat/x-htrnusvtyczkfwxc] has quit [Read error: Connection reset by peer]08:43
-!- abn_ [av3ngr@nat/redhat/x-qjjmkztgcdxgddqm] has joined #shogun08:43
-!- uricamic [~uricamic@2001:718:2:1634:d9cd:b2c5:1776:917b] has joined #shogun08:54
sonne|workn4nd0: making progress?08:56
n4nd0sonne|work: I am right now with the problem type thing blackburn told me about yesterday08:57
sonne|workn4nd0: yeah it is very critical to get this done fast08:57
-!- pluskid [~pluskid@] has joined #shogun08:57
sonne|workthis will fix lots of crashers08:57
sonne|workpluskid: new labels are in place now, lots of stuff works - maybe you try your MC stuff08:57
pluskidsonne|work: cool!08:58
n4nd0sonne|work: I understand that vowpal wabbit is used for classification, but does it support multiclass or just binary?09:02
sonne|workI think binary09:03
n4nd0sonne|work: all right, what about CWDSVMOcas, I think this is the one that can be used both for regression and binary classification09:11
sonne|workno binary only09:11
n4nd0apply returns CRegressionLabels though09:12
sonne|workfix it!09:13
sonne|workapply() should always return CLabels09:13
sonne|workand in this case CBinaryLabels down or up cast to CLabels :)09:13
n4nd0all right09:14
n4nd0well it actually returns CLabels since the prototype of the function is CLabels* apply, but internally those labels are created as CRegressionLabels*09:15
n4nd0sonne|work: I will change that to CBinaryLabels, ok?09:16
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:23
-!- abn_ [av3ngr@nat/redhat/x-qjjmkztgcdxgddqm] has quit [Quit: That's all folks!]09:24
sonne|workn4nd0: I guess there will be a couple of such things - because I used RealLabels aka RegressionLabels now for all classifier output before we switched yesterday :)09:29
n4nd0ok, I will take it into account09:29
n4nd0is it possible to ammend a past commit?09:39
n4nd0I have some changes I want to include to a commit that is not the last one I did, but three commits ago09:40
n4nd0I have tried doing git reset HEAD^^09:40
n4nd0stage the files with the changes I want to amend to the commit09:40
n4nd0git commit --amend09:40
n4nd0and later come back to the tip of the branch using git reset and the SHA109:41
sonne|workn4nd0: wiking or pluskid might know... I have never done this but I guess git reset is needed :)09:55
-!- karlnapf [] has joined #shogun09:56
pluskidn4nd0: I don't know either. I guess git stash won't work. But maybe you can use git format-patch to create patches and re-apply them later09:56
n4nd0pluskid: yeah, I think git format-patch may be useful for this, thank you!09:57
n4nd0sonne|work: I think the problem type changes are done now, tell me if something is not ok10:16
n4nd0I may have forgotten any machine10:16
sonne|workn4nd0: just run make check-examples - you will see if more things work now :)10:16
CIA-113shogun: iglesias master * r09e6617 / (src/shogun/machine/Machine.h src/shogun/multiclass/QDA.h): + problem type macro in Machine and + return PT in QDA -
CIA-113shogun: iglesias master * r29c5724 / src/shogun/classifier/svm/WDSVMOcas.cpp : * fix CRegressionLabels* to CBinaryLabels* in apply -
CIA-113shogun: iglesias master * r13377dd / (27 files in 5 dirs): + return problem type in more classifiers -
CIA-113shogun: iglesias master * rbfab371 / (8 files in 2 dirs): + return problem type in regression machines -
CIA-113shogun: Soeren Sonnenburg master * r3801ecf / (37 files in 7 dirs): Merge pull request #538 from iglesias/problem_type -
n4nd0qutie a few of them still failing10:18
n4nd0I have selected a random one right one and I found something curious10:21
n4nd0[ERROR] assertion m_labels->get_label_type() == LT_BINARY failed in file classifier/svm/LibLinear.cpp line 8210:21
n4nd0that is used from10:21
n4nd0classifier_multiclasslinearmachine in libshogun10:21
n4nd0I guess that the problem is that in that example the labels used are CMulticlassLabels10:21
n4nd0while LibLinear requires them to be LT_BINARY10:21
sonne|workn4nd0: but that is multiclasslbilinear10:23
n4nd0I didn't understand10:24
sonne|workwe have a class for multiclassliblinear and for binaryclassliblinear10:25
n4nd0do you mean that multiclassliblinear should be used for that and not LibLinear10:25
sonne|workno idea why CLibLinear::train is even called!10:25
sonne|workif this really is a MC example10:25
n4nd0this is an example I wrote10:26
n4nd0the idea was to use a binary classifier with one of the multiclass strategies10:26
sonne|workn4nd0: ahh ok then this is a bug pluskid needs to handle10:26
n4nd0I'd say that probably one must cast labels when the MulticlassMachine uses the train method of the BinaryMachine10:27
n4nd0from MulticlassLabels to BinaryLabels10:27
pluskidI'll look at this10:28
sonne|workn4nd0: I think that linear mc machine does not create binary labels10:28
sonne|workpluskid: thx10:28
-!- blackburn [~blackburn@] has joined #shogun10:30
blackburnoh wow10:30
-!- uricamic1 [~uricamic@2001:718:2:1634:5c56:5c20:5842:ed51] has joined #shogun10:30
sonne|workblackburn: morning!10:31
blackburnguten tag10:32
sonne|workblackburn: could you fix the evaluations please?10:32
blackburnehm what to fix else?10:32
blackburnI was pretty sure I fixed it :)10:32
-!- uricamic [~uricamic@2001:718:2:1634:d9cd:b2c5:1776:917b] has quit [Ping timeout: 245 seconds]10:32
n4nd0blackburn: I've just run make check-examples after doing the problem type things but several of them still fail :(10:33
sonne|workblackburn: wait evaluations use confidences now to actually do the evaluations?10:33
blackburnsonne|work: yes10:33
sonne|workfor ROC/pRC that is?10:33
blackburnsonne|work: yes10:33
blackburnconfidences of binary labels10:33
blackburnbinary labels set confidences from vectors10:33
sonne|workwell multiclass machines are broken currently10:34
sonne|workblackburn: ImportError: cannot import name AccuracyMeasure10:34
sonne|work ??10:34
sonne|workwhat happened?10:34
pluskidproblem here is brute force type casting10:34
blackburnsonne|work: ah I removed it :D10:34
blackburndid you use it?10:34
blackburnsonne|work: I thought it is pretty useless10:34
sonne|workI like accuracy :)10:34
sonne|workno keep it10:34
blackburnsonne|work: ContingencyTableEvaluation(ACCURACY)10:35
sonne|workkeep it10:35
sonne|workexplicit is good10:35
blackburnsonne|work: a bunch of classes we don't really need10:35
sonne|workI need this10:35
CIA-113shogun: Sergey Lisitsyn master * r4e4ae17 / src/shogun/evaluation/ContingencyTableEvaluation.h : Restored explicit measure classes -
blackburnsonne|work: should work now10:41
blackburnhmm multiclass is b0rken for sure10:46
CIA-113shogun: Soeren Sonnenburg master * rd59c2b2 / (4 files in 2 dirs): Label fix for missing examples -
pluskidquite broken10:47
sonne|workof course it is10:47
blackburnI am working on it10:47
sonne|workblackburn: IIRC pluskid too?10:47
pluskidwhich part are you working on?10:47
blackburnMulticlassMachine :D10:47
sonne|workblackburn: shouldn't pluskid better work on this?10:48
pluskid149: MCLabel -> BinaryLabel10:48
sonne|workblackburn: and you better check why evaluations fail?10:48
pluskidI have fixed this10:48
blackburnsonne|work: what fails?10:48
blackburnlet me commit what I've done then10:48
sonne|workblackburn: run the examples10:48
sonne|worklibshogun, python_modular ...10:49
sonne|worklots of stuff works but some still fails10:49
blackburnI've tested ROC/PRC last night10:50
sonne|workblackburn: well just go over the failing (non-mc) examples then10:53
blackburnlet me commit first10:53
CIA-113shogun: Sergey Lisitsyn master * r31c9307 / (4 files in 2 dirs): Some fixes for multiclass machine -
blackburnpluskid: ok please continue with multiclass10:55
-!- gsomix [~gsomix@] has joined #shogun10:56
gsomixhi all10:56
sonne|workpluskid, blackburn: btw everyone no need to explicitly typecast labels - better use (the safe!) CMulticlassLabels::obtain_from_genereric(lab) etc functions10:57
sonne|workit will give you a CMulticlassLabel but check if casting is safe10:57
pluskidbtw: sonne|work why you commented out the pure-virtual get_label_type in the base CLabels?10:58
sonne|workall this RegressionLabels crap that machines return is mostly due to legacy10:58
pluskidand make it return REGRESSION?10:58
sonne|workI didn't comment it out10:58
sonne|workI didn't do that10:59
CIA-113shogun: Soeren Sonnenburg master * rdf320e8 / (src/shogun/labels/DenseLabels.h src/shogun/labels/Labels.h): make get_label_type pure virtual again -
blackburnit is not me either I think11:02
pluskidsome evil ghost is in our dev team :D11:03
pluskidvery strange problem here...11:04
pluskidin n4nd0's libshogun example, a LinearMCMachine is created11:04
pluskidthen I call get_machine_problem_type of that machine in gdb11:05
pluskidit returns PT_BINARY11:05
CIA-113shogun: Sergey Lisitsyn master * rb6a6727 / examples/undocumented/python_modular/ : Fixed serialization_matrix_modular example -
CIA-113shogun: Sergey Lisitsyn master * r9eb2a83 / (3 files in 2 dirs): Fixed GaussianProcessRegression -
blackburnn4nd0: btw inline virtual makes no sense11:06
blackburnpluskid: hmmmm11:08
pluskidforgot to recompile11:08
pluskidafter updating the shared library11:08
pluskidso the vtable is confused11:08
blackburnpluskid: but wait there is no mention of proper PT_MULTICLASS in LinearMulticlassMachine11:09
pluskidpreviously no11:09
pluskidI've added this11:09
blackburnah ok11:09
pluskidand I found a lot of machines didn't register EMachineType11:09
pluskiddoes it necessary for *every* machine to register this?11:10
blackburnpluskid: machine type is rather for static interfaces11:10
n4nd0blackburn: oh really?11:17
blackburnn4nd0: what?11:17
n4nd0blackburn: I thought they were different concepts11:17
n4nd0blackburn: why inline virtual doesn't make sense11:18
blackburnabout inline virtual?11:18
blackburnhmm can you imagine how can compiler inline a function that calls by pointer? :)11:18
CIA-113shogun: Sergey Lisitsyn master * r28082be / (2 files): Fixed ConjugateIndex -
CIA-113shogun: Chiyuan Zhang master * r52d53fd / (2 files): Fixed multiclass problem-type. -
CIA-113shogun: Sergey Lisitsyn master * rf318aee / (2 files): Merge branch 'multiclass-ecoc' of -
CIA-113shogun: Evgeniy Andreev master * r61ff1e2 / (39 files): const methods in Features -
CIA-113shogun: Sergey Lisitsyn master * r7ca1c60 / (39 files): Merge branch 'const_methods' of -
CIA-113shogun: Chiyuan Zhang master * r6216379 / src/shogun/machine/MulticlassMachine.cpp : Simple fix for mkl_multiclass crasher -
CIA-113shogun: Sergey Lisitsyn master * rde154b3 / src/shogun/machine/MulticlassMachine.cpp : Merge branch 'multiclass-ecoc' of -
blackburnpluskid: larank is failing in python_modular - could you please take a look?11:35
pluskidblackburn: I'll try, but I'm not sure I can fix, coz I know nothing about how LaRank works11:38
pluskidjust back from supper :D11:38
blackburnpluskid: no need to know - something is wrong with methods I think11:38
blackburnsonne|work: why I see terminate called after throwing an instance of 'shogun::ShogunException' w/o any SG_ERROR message?11:40
CIA-113shogun: Sergey Lisitsyn master * rd204fcd / (src/shogun/multiclass/KNN.cpp src/shogun/multiclass/KNN.h): Proper multiclass labels handling in KNN -
sonne|workblackburn: some exception at illegal place maybe?12:00
pluskidblackburn: I found the problem12:04
blackburnpluskid: what is?12:04
pluskidthe apply() method returns a CBinaryLabel12:04
pluskidwhen constructing the CBinaryLabel object12:04
pluskidin the constructor, set_label is called12:04
pluskidset_label checks whether the label is valid12:04
pluskidand found there's only one class12:05
pluskidit raise SG_ERROR12:05
pluskidhow do you think this should be handled?12:05
pluskidsonne|work: what do you think? A BinaryLabel returned by apply() may not be "valid" (i.e. two class)12:06
blackburnpluskid: does it happen in larank?12:08
pluskidyes, in the example12:09
pluskidthat's why it's failing12:09
pluskidalthough it should print "Not a two class labeling - no positively labeled examples found\n"12:09
pluskiddon't know why this message is swallowed12:10
CIA-113shogun: Sergey Lisitsyn master * rfca820b / (3 files in 2 dirs): Fixed DomainAdaptation binary svm -
blackburnpluskid: but why only one class here?12:11
pluskidblackburn: because it classify all examples as negative ?12:12
blackburnbut larank is multclass?12:13
pluskidblackburn: isn't is OvO (or OvR)?12:14
blackburnah get it12:14
pluskidthen there should be binary classifier under the sense12:14
blackburnwe should allow output to be all negative12:14
blackburnor all positive.12:14
pluskidI'm not sure how to modify this12:16
pluskidsonne|work should give some decision here12:16
blackburnyeah interesting problem12:16
-!- vikram360 [~vikram360@] has quit [Ping timeout: 244 seconds]12:33
blackburnpluskid: ok may be some option that ignores validation?12:52
blackburnpluskid: how do you obtain error message?13:01
blackburnkarlnapf: hey13:03
karlnapfblackburn hi13:03
blackburnkarlnapf: do you have time to take a look on modelselection error?13:04
karlnapfyes, just finished studying for today :)13:04
karlnapfalthough I will have lunch before13:04
karlnapfbut whats the problem13:04
blackburnok, please take a look on modelselection_grid_search_kernel13:04
blackburnit fails with wrong label type13:05
karlnapfI did not really check out the new label stuff13:05
karlnapfwhat does that mean, fails with wrong label type?13:05
blackburnkarlnapf: for some strange reason it passes wrong type of labels to evaluation13:06
karlnapfthat probably then happens in the select_model method13:06
blackburnhmm wait13:07
karlnapfwhat has happened to the git?13:07
karlnapftakes ages to update?13:07
blackburnkarlnapf: oh wrong data commit13:07
blackburnwe will remove it later I hope13:08
blackburnkarlnapf: one guy updated ASP application and added data13:08
blackburnI have removed it but it went to index13:08
karlnapfoh no ;)13:08
karlnapfargh my git program cant handle that13:08
blackburnthat sounds a bit strange hmm13:09
karlnapfI think I will clone freshly13:09
-!- vikram360 [~vikram360@] has joined #shogun13:10
karlnapfblackburn, I have to go to the post office now (closes soon) and will have lunch, I will get back here afterwards and try to fix the error13:11
karlnapfbtw only three exams left :) 2 next week13:13
blackburnargh crazy13:14
karlnapfhehe :)13:15
karlnapfbioinformatics, kernel methods and ML algorithms13:15
blackburnweren't the former about the same?13:16
pluskidblackburn: I falled asleep ...13:22
pluskidblackburn: I do a backtrace of Python13:22
pluskiddump the core, and gdb on the core13:23
pluskidthe validation is in the constructor13:23
blackburnpluskid: ok and how to dump the core? :)13:23
pluskidulimit -c unlimited13:23
blackburnrun my python script?13:24
pluskidthen the core will be dumped13:24
pluskidwhen the program aborts13:24
blackburnto file?13:24
blackburnhmm cool13:24
blackburnpluskid: and how to gdb on that file?13:25
pluskidthe validation is called on set_labels, if I remove the validation there, will sonne|work becomes crazy?13:25
pluskidgdb program core13:25
pluskidhere will be gdb python core13:25
-!- sonne|work is now known as crazy_horse13:25
blackburnpluskid: ok and the last q - how to trace back?13:26
crazy_horsepluskid: it is clear that we need to allow any kind of outputs13:26
pluskidblackburn: type "bt" in gdb13:26
blackburn(gdb) bt13:26
blackburn#0  0x00007f7e095a5445 in ?? ()13:26
blackburn#1  0x00007f7e095a8bab in ?? ()13:26
blackburn#2  0x0000000000000000 in ?? ()13:26
pluskidcrazy_horse: so, shall I just remove the validation?13:27
crazy_horsethe only way I see how to handle that currently is to add some flag via which validation can be disabled13:27
pluskidblackburn: I think you should recompile your shogun with debug enabled13:27
crazy_horsepluskid: we need this validation when training svms etc13:28
crazy_horseif we get invalid labels one gets weird results13:28
pluskidcrazy_horse: an extra parameter "do_validation" with default value true?13:28
crazy_horsequestion is whether we should call is_valid upon set_labels13:28
crazy_horseor should require the call to is_valid in all machines13:29
pluskidthings become complicated13:29
pluskidno simpler solution?13:29
crazy_horseI am open for suggestions13:29
wikingblackburn: -g13:29
blackburnwiking: -g what?13:30
wikingblackburn: as a CFLAGS13:30
crazy_horseblackburn: what are you doing?13:30
blackburnokay food time13:30
wikingblackburn: otherwise u'll have troubles with gdb-ing shogun13:30
blackburncrazy_horse: trying to understand why my script fails here13:30
crazy_horsepluskid: -g is usually on13:31
crazy_horseblackburn: which script13:31
blackburncrazy_horse: one for my thesis - uses multiclassliblinear13:31
blackburnfails on apply with exception but no message13:31
crazy_horsewhy don't you use gdb on it directly?13:31
blackburnon python?13:32
blackburnhm why not13:32
crazy_horsegdb --args python <>13:32
blackburnokay I'll try13:32
pluskidcrazy_horse: how about we do not call valid_labels on constructors, but let CMachine call it before training started?13:33
crazy_horsepluskid: that is what I suggested13:33
crazy_horseI don't have a better solution...13:34
crazy_horsemaybe is valid should print a warning only13:34
crazy_horseand return a boolean13:34
pluskidprint a warning doesn't make sense for returned CLabels, either13:35
blackburnI think validation before training is better13:46
blackburnokay bug I get is exactly that one we have in larank13:48
pluskidBTW, do we need an explicit '\n' in SG_ERROR?13:50
blackburnI think so13:52
pluskidok, thanks13:55
crazy_horsepluskid: ok then please do the changes and call is_valid() everywhere before training...14:00
crazy_horsepluskid: btw there usually is some ASSERT(labels->get_label_type ...) etc just at the start of train*()14:01
pluskidcrazy_horse: isn't it simply call in CMachine::train?14:02
pluskidall subclasses redefines train_machine14:02
pluskidbut CMachine::train is always called14:02
crazy_horsepluskid: ahh indeed14:02
crazy_horseso ASSERT(labels) should be put there too14:02
crazy_horseonly label type checking is specific to machine14:02
blackburnhehe crazy_horse it is way to crazy to add it everywhere14:04
crazy_horseblackburn, pluskid - we will need it on other occassions however, like AUCKernel etc14:06
crazy_horseso better check locations where label type is checked14:06
blackburnpluskid: will you do that?14:07
blackburncrazy_horse: did you remove ignore of operator[]?14:07
pluskidhere's the pull request14:08
crazy_horseblackburn: no14:08
pluskidblackburn: do what?14:08
blackburnpluskid: "(04:06:23 PM) crazy_horse: blackburn, pluskid - we will need it on other occassions however, like AUCKernel etc"14:08
pluskidI don't understand AUCKernel, what is it?14:09
blackburna kernel :)14:09
pluskida kernel that needs a label?14:09
CIA-113shogun: Evgeniy Andreev master * r90bc6bd / (src/shogun/base/DynArray.h src/shogun/lib/DynamicArray.h): fixes warnings and potential bugs -
CIA-113shogun: Sergey Lisitsyn master * re27e8f2 / (src/shogun/base/DynArray.h src/shogun/lib/DynamicArray.h): Merge branch 'Arrays' of -
CIA-113shogun: Sergey Lisitsyn master * rc3b256b / (14 files in 2 dirs): Merge branch 'multiclass-ecoc' of -
CIA-113shogun: Chiyuan Zhang master * r4d35f91 / (5 files): An interface for checking label type -
CIA-113shogun: Chiyuan Zhang master * r91de8f9 / (10 files in 2 dirs): CLabels::is_valid should only be called at proper situation. -
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]14:10
blackburnpluskid: yeah..14:10
pluskidI'll look at it14:11
crazy_horsepluskid: there are other such things ...14:12
pluskidcrazy_horse: please give me a list14:13
pluskidor maybe you can add them? just one line14:13
crazy_horsepluskid: I don't know which ones they are by heart14:13
crazy_horsegrep for labels->get_label_type14:13
crazy_horsepluskid: there simply are too many14:14
pluskidthat many?14:14
crazy_horsepluskid: well evaluation for sure14:14
pluskidwhy evaluation have to ensure (e.g.) a binary label has both +1 and -1?14:16
blackburnROC would fail I think14:17
blackburnunwanted /0 I think14:17
pluskidisn't ROC calculated on accuracy and recall?14:18
pluskidI think having all +1 labels is OK unless I predict them all wrong14:18
blackburnROC calculated on TPR and FPR14:20
blackburnif everything in truth labels is +1 - you would have divide by zero here14:21
blackburnpluskid: elt me fix evals14:23
pluskidblackburn: thanks14:23
-!- blackburn [~blackburn@] has quit [Remote host closed the connection]14:33
-!- blackburn [~blackburn@] has joined #shogun14:36
karlnapftsts cloning and updating takes > 10 mins now14:40
crazy_horsekarlnapf: we will take care of that but for now please help fix the issues :D14:40
blackburnargh gnome shell died14:40
CIA-113shogun: Sergey Lisitsyn master * rc196b13 / (3 files): Added ensure_valid to ROC, PRC and ContingencyTableEvaluation -
karlnapfcrazy_horse, yes youre right :), but I currently cannot even compile14:40
crazy_horsekarlnapf: why that?14:41
karlnapfbecause my code is still updating14:41
blackburnyou have bad connection there in UK :D14:41
blackburnhave someone broke that cable across strait of dover?14:42
karlnapfI know :)14:42
crazy_horseblackburn: you should be pretty quiet here you merged this!14:42
karlnapfUK internet completely sucks14:42
karlnapf100kb/s during this time of the day14:42
karlnapfand that costs you 45 punds a month14:42
blackburncrazy_horse: at least I have OK connection :D14:42
crazy_horsethese mad russians :D14:43
crazy_horsepluskid: look at the indention14:51
crazy_horsesomething is again getting wrong14:51
crazy_horseotherwise it is ready to merge14:52
pluskidcrazy_horse: oops, I'll fix this14:52
crazy_horsepluskid: other patches from today have similar problem :/14:53
pluskidit seems I accidentally configured expandtab in vim14:53
-!- wiking [] has joined #shogun14:54
-!- wiking [] has quit [Changing host]14:54
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun14:54
-!- n4nd0 [] has quit [Ping timeout: 245 seconds]14:55
crazy_horsepluskid: but ts=4 right? so we could replace 4 spaces -> 1 tab everywhere?14:57
pluskidhmm, maybe I can do a sed in the whole codebase14:58
crazy_horsepluskid: maybe we first fix this problem then do that later14:59
crazy_horseisn't there some source code beautifier that could do this anyway?14:59
crazy_horsemaybe not a good idea14:59
crazy_horsefor third party stuff we would then have to do a full diff /compare15:00
crazy_horseit is hard enough already15:00
pluskidcode beautifier might be complicated to config to obey shogun convention15:00
crazy_horsepluskid: and we don't wnat to touch external code15:02
crazy_horseso no15:02
crazy_horsepluskid: just fix this one here and lets get labels to stable for now15:03
blackburncrazy_horse: it seems you liked new nickname?15:03
crazy_horseI cannot change it back15:04
-!- crazy_horse is now known as sonne|work15:04
-!- blackburn is now known as crazy_horse15:04
sonne|workpluskid: still weird
sonne|workpluskid ^last one15:05
-!- sonne|work is now known as crazy_russian15:05
-!- crazy_horse is now known as BODKA15:05
-!- crazy_russian is now known as sonne|work15:07
-!- sonne|work is now known as blackburn15:07
-!- BODKA is now known as sonne|work15:07
blackburnsonne|work:  can you look at my pull request?15:08
sonne|workblackburn: no! I have 3 works and 9 kids!15:08
pluskidBODKA: finally this one should be correct15:08
sonne|workjobs :D15:08
CIA-113shogun: Chiyuan Zhang master * rd3b4e9c / (3 files): Ensure related kernels call labels->ensure_valid -
CIA-113shogun: Chiyuan Zhang master * r8d10a3a / (3 files): Fix indentation convention -
CIA-113shogun: Chiyuan Zhang master * r6fb43da / src/shogun/kernel/ScatterKernelNormalizer.h : fixed more indent conventions -
CIA-113shogun: Soeren Sonnenburg master * r1cc1e70 / (3 files): Merge pull request #545 from pluskid/multiclass-ecoc -
blackburnpluskid: let sonne|work merge this :D15:09
blackburnohh he did already15:09
sonne|workblackburn: did you make any progress on SLEP?15:10
karlnapfblackburn, how does this new labels stuff work15:10
karlnapfwhere do I have to pay attention?15:10
sonne|workclustering now fail15:12
sonne|workbecause of check in train()15:12
* pluskid is crazy trying to send some fax via the internet15:14
sonne|workkarlnapf: okay in new concept outputs of say svm are stored in confidences of labels15:14
sonne|workand actual labels are +1 or -115:14
karlnapfsonne|work yes just read a bit through it15:14
karlnapfthe modsel example fails because train_locked has no distinction between label types yet15:14
karlnapfI will add that15:15
sonne|workyes add please15:15
karlnapfthere also is a new memory leak if training is not performed locked15:15
karlnapfsonne|work, how was the policy for SG_REF again when a new object is returned?15:22
* blackburn thinks sonne|work is so wise15:22
sonne|workkarlnapf: no need to SG_REF new object IIRC15:23
blackburnsonne|work: greetings to russia!15:23
karlnapfdo we have any more cases than regression and binary yet?15:23
blackburnerr from :D15:23
sonne|workblackburn: I do not act like an idiot! you are a bad actor :)15:24
karlnapfapply_multiclass is not in kernelmachine so far15:24
sonne|workkarlnapf: multiclassmachine15:24
karlnapfah sure, of course15:24
sonne|workand kernelmulticlassmachine15:24
blackburnsonne|work: yeah I am stupid15:25
CIA-113shogun: Sergey Lisitsyn master * r4ead6f0 / (5 files in 2 dirs): Added virtual method indicating whether machine requires labels for training, -
-!- vikram360 [~vikram360@] has quit [Ping timeout: 260 seconds]15:38
-!- n4nd0 [] has joined #shogun15:39
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]15:50
CIA-113shogun: Sergey Lisitsyn master * r32ce04d / (2 files): Fixes for OnlineLinearMachine -
-!- vikram360 [~vikram360@] has joined #shogun16:04
CIA-113shogun: Sergey Lisitsyn master * r3f690ca / (2 files): Fixes for PluginEstimate -
-!- uricamic1 [~uricamic@2001:718:2:1634:5c56:5c20:5842:ed51] has quit [Quit: Leaving.]16:08
pluskidimport shogun.Features.MulticlassLabels16:10
pluskidcrashes python...16:10
sonne|workblackburn: you are pretty active today!16:13
blackburnthank you!16:13
-!- karlnapf [] has quit [Ping timeout: 276 seconds]16:14
-!- karlnapf [] has joined #shogun16:16
pluskidno longer crash with the latest code16:16
CIA-113shogun: Sergey Lisitsyn master * r4d58627 / (2 files): Fixes for WDSVMOcas -
-!- n4nd0 [] has quit [Quit: leaving]16:16
* blackburn strikes again16:17
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun16:18
karlnapfthe semantics of the apply code have changed16:19
karlnapfthat causes the errors in the locked examples16:19
karlnapfbefore, apply() directly called apply on all indices16:19
karlnapfnow its different16:20
-!- n4nd0 [] has joined #shogun16:23
n4nd0my blog is oficially created, yay!!16:23
pluskidn4nd0: show us the url16:24
n4nd0there is nothing yet :O16:25
pluskidoops, The connection was reset16:27
pluskidbehind the wall16:27
pluskidthis blog is created specially for shogun?16:27
karlnapfn4nd0, cool :)16:29
n4nd0pluskid: yeah16:30
n4nd0pluskid: I have no intention to blog about more stuff this far16:30
n4nd0karlnapf: thank you :)16:31
karlnapfI also got one:16:31
blackburnsonne|work: can you add them to planet shogun?16:31
karlnapfblackburn, there is an error in contingency table evaluation16:31
karlnapfpassing the same labels twice results in non 1.0 accuracy16:32
karlnapfto bee seen in the kernel_machine_train_locked example16:32
karlnapfwhich I will push now16:32
blackburnkarlnapf: uh that must be due to the new labels ...16:33
karlnapfyeh, probably :D16:33
blackburnkarlnapf: could you paste a reproducible example?16:33
-!- gsomix [~gsomix@] has quit [Ping timeout: 252 seconds]16:33
n4nd0see you later guys16:35
-!- n4nd0 [] has quit [Quit: leaving]16:35
-!- karlnapf1 [] has joined #shogun16:37
karlnapf1uuh London internet rocks16:37
-!- karlnapf [] has quit [Ping timeout: 276 seconds]16:38
CIA-113shogun: Heiko Strathmann master * r2a49637 / (6 files in 2 dirs): Merge pull request #547 from karlnapf/master -
CIA-113shogun: Heiko Strathmann master * r3d93199 / examples/undocumented/libshogun/modelselection_grid_search_kernel.cpp : extended example to use locked and unlocked x-validation -
karlnapf1blackburn, see output of kernel_machine_train_locked16:39
karlnapf1displays labels, output and accuracy16:39
CIA-113shogun: Heiko Strathmann master * r95f980f / (4 files): added label distinctions for apply_locked -
CIA-113shogun: Heiko Strathmann master * r18894b6 / src/shogun/machine/KernelMachine.cpp : removed unnecessary check -
CIA-113shogun: Heiko Strathmann master * r516773c / examples/undocumented/libshogun/kernel_machine_train_locked.cpp : some more output messages revealed an error in CEvaluation -
blackburnkarlnapf1: ok will have a look16:40
karlnapf1I think we should write more test cases16:40
karlnapf1defined input /output16:40
blackburnkarlnapf1: we have that but not activated16:40
blackburnand I guess now everything is b0rken16:40
karlnapf1I mean for simple stuff16:40
karlnapf1like single methods16:40
karlnapf1but ok16:41
karlnapf1later :)16:41
blackburnkarlnapf1: each python modular example is a test case...16:41
karlnapf1I know16:41
blackburnand they cover all simple examples like creating feature objects etc16:41
karlnapf1but I think we should do this in a more systematics way16:41
blackburnlater yes16:41
karlnapf1so that errors are easy to find16:41
blackburnmakes sense16:41
karlnapf1otherwise it always takes 20 mins to find a bug which would be immediately found by a test16:42
blackburndocumentation sprint & test sprint :)16:42
-!- blackburn is now known as crazy_horse16:48
-!- karlnapf [] has joined #shogun16:50
-!- sonne|work is now known as blackburn16:50
-!- karlnapf1 [] has quit [Ping timeout: 276 seconds]16:51
blackburnkarlnapf: cool homepage16:52
karlnapfblackburn, thanks :)16:52
karlnapfIll pass it on16:52
blackburnkarlnapf: so what is wrong with contingency table/16:53
karlnapfhave you run the example?16:53
blackburnwhich one?16:53
blackburnit was soeren you know16:53
blackburnpy or cpp?16:53
CIA-113shogun: Chiyuan Zhang master * r85330b4 / (2 files in 2 dirs): Make LLBDecoder non-abstract -
CIA-113shogun: Chiyuan Zhang master * rcc662e7 / examples/undocumented/python_modular/ : Now the equivalence of OVR and ECOCOVR+LLBDecoding validated -
blackburnkarlnapf: I didn't know you born in 1984 :D16:55
blackburnyou are a big boy comparing to me hehe16:55
-!- karlnapf1 [] has joined #shogun16:55
karlnapf1sorry blackburn, internet connection :(16:55
karlnapf1did you write anything?16:55
blackburn(06:55:06 PM) blackburn: karlnapf: I didn't know you born in 1984 :D16:55
blackburn(06:55:26 PM) blackburn: you are a big boy comparing to me hehe16:55
karlnapf1well :)16:57
blackburnok so back to contingency16:58
karlnapf128 so far :)16:58
blackburn21 here16:58
-!- karlnapf [] has quit [Ping timeout: 276 seconds]16:58
karlnapf1I know :)16:58
@sonney2khey add the magic 7 :)16:58
@sonney2kand you know my age :D16:58
@sonney2kfrom diapers to retirement16:58
karlnapf1soeren is 35?16:59
@sonney2kkarlnapf1, yeah you may call me daddy16:59
karlnapf1oh, just read the last lines :)16:59
karlnapf1hehe, you in fact are one17:00
blackburnheh I think we need a 14 yrs old developer17:00
karlnapf1lol magic 717:00
@sonney2k42 would also not be bad17:00
-!- crazy_horse was kicked from #shogun by sonney2k [crazy_horse]17:00
blackburnheh yeah that would bake you young again17:00
karlnapf1who is crazy horse?17:01
blackburnkarlnapf1: it was crazy_soeren17:01
karlnapf1oh, two much code refactoring today? :D17:01
blackburnyeah that hurts your health sometimes17:02
karlnapf1hehe :)17:02
blackburnkarlnapf1: so what is the reproducing example?17:02
blackburni checked contingency - see nothing bad here17:02
karlnapf1I already wrote 5 times17:02
karlnapf1but always my connection died before17:02
blackburnno I get it17:02
karlnapf1have a look at the output17:03
blackburnbut I can't see anything realted to evaluations17:03
blackburnterminate called after throwing an instance of 'shogun::ShogunException'17:03
blackburn[ERROR] CKernelMachine::apply(CFeatures*) cannot be called when data_lock was called before. Call data_unlock to allow.Aborted (core dumped)17:03
blackburnthat's what I see17:03
karlnapf1fetch my update17:03
blackburnok let me recompile17:03
karlnapf1I fixed a bug before that happened17:03
blackburnit seems I forgot that17:03
karlnapf1(you added this locked check in apply() although that wasnt necessary)17:04
karlnapf1I removed it again17:04
karlnapf1blackburn, I need a coffee break, will be back in 10 mins17:05
blackburnok sure17:05
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun17:11
blackburnkarlnapf1: ok I tried to reproduce - multiple calls have equal accuracies..17:13
blackburnah labels are equal17:15
blackburnbut accuracy is not17:15
blackburnI see17:15
@sonney2kblackburn, I am fixing it17:18
blackburnsonney2k: ok17:18
@sonney2kwe need to use labels17:18
@sonney2knot confidence here17:19
blackburncrazy russian bridge17:19
blackburnsonney2k: but why? it should work too - are confidences set wrong or what?17:19
CIA-113shogun: Soeren Sonnenburg master * rd27ca42 / src/shogun/evaluation/ContingencyTableEvaluation.cpp : use get_label() in contingencytableeval -
@sonney2kblackburn, because they are hardly == 117:20
@sonney2kblackburn, actually >=1 ?17:21
@sonney2kwe should put == 1 there17:21
blackburnin case of labels yes17:21
blackburnin case of confidences >=0 I guess17:22
@sonney2kblackburn, what this does is just count TPs, FPs, ...17:22
@sonney2kfor the fixed threshold17:22
@sonney2kso we can just use labels17:22
@sonney2kusing confidences only make sense if we do sign(x + thresh)17:22
@sonney2kor so17:22
CIA-113shogun: Soeren Sonnenburg master * r33a03c0 / src/shogun/evaluation/ContingencyTableEvaluation.cpp : check for equality -
@sonney2kblackburn, did you know that java throws a divide by zero exception if you do 1/0 but not 1.0/0.0 ?17:23
blackburnsonney2k: no but it looks obvious17:24
blackburn1.0/0.0 = NaN17:24
@sonney2kor ?17:24
blackburnyes inf17:25
blackburnhmm what is nan?17:25
blackburn0.0/0.0 I guess17:25
@sonney2kkarlnapf1, so you know how to fix the modsel stuff?17:25
-!- karlnapf [] has joined #shogun17:25
@sonney2kkarlnapf, so you know how to fix the modsel stuff?17:25
@sonney2kman you have some krappy KONNEXION17:26
karlnapfI dont get it: I am in the capital of europe but the internet SUCKS17:26
blackburnhaha abramovich has bought all the bandwidth in uk I guess17:26
blackburnchelsea is not enough17:26
karlnapfannoying: I am working on a cluster compute remotely ....17:26
karlnapfsonney2k, the modsel example works again17:26
karlnapfI added the label distinction17:27
@sonney2kthe libshogun ones still crash17:27
karlnapfin kernel machine17:27
karlnapflet me check17:27
-!- karlnapf1 [] has quit [Ping timeout: 276 seconds]17:27
blackburnno it works17:27
blackburnsonney2k: works17:27
@sonney2keven modelselection_grid_search_linear ?17:29
@sonney2kit crashes here17:29
@sonney2kparameters available for model selection for LibLinear:17:30
@sonney2kC1 (C Cost constant 1.): float6417:30
@sonney2kC2 (C Cost constant 2.): float6417:30
@sonney2kroot with17:30
@sonney2kC1 with values: vector=[0.25,0.5,1,2,4]17:30
@sonney2kC2 with values: vector=[0.25,0.5,1,2,4]17:30
@sonney2k[ERROR] Not a two class labeling - no negatively labeled examples found17:30
@sonney2kterminate called after throwing an instance of 'shogun::ShogunException'17:30
karlnapfsonney2k, yes I will have a look at these17:30
@sonney2kkarlnapf, does the accuracy stuff work again?17:31
@sonney2kI mean with the patch I just submitted?17:31
karlnapfsonney2k, yes17:31
@sonney2kok good17:32
karlnapfsonney2k, could we add somethign to error messages to know from which line in which file these come from?17:32
@sonney2kpluskid, can you have a look at ?17:32
@sonney2kif you are not sleeping already17:32
@sonney2kkarlnapf, just enable17:33
blackburnsonney2k: no I am fixing it17:33
blackburnalready get the error17:33
karlnapfsonney2k, nice!17:33
karlnapfsonney2k, is this a compile switch or a runtime flag?17:34
@sonney2kpluskid, ok then better go to bed before I find more bugs :)17:35
@sonney2kkarlnapf, runtime flag17:35
@sonney2kblackburn, libsvmoneclass doesn't get any labels now ? but what is this
blackburnsonney2k: could you please ignore [] and =?17:36
blackburnsonney2k: I'll check17:36
@sonney2khmmhh TypeError: Cannot create new instances of type 'ECOCSimpleDecoder'17:37
@sonney2k^ that is still sth for pluskid17:37
pluskidtoo bad I didn't go to bed just now :p17:39
CIA-113shogun: Sergey Lisitsyn master * r2741f89 / (2 files): Added problem type to clustering -
@sonney2kpluskid, apologies to keep you awake17:39
pluskidI'll spend 15min on the ECOC bug, if I fail, I'll kill it tomorrow17:39
-!- vikram360 [~vikram360@] has quit [Ping timeout: 250 seconds]17:40
@sonney2kalright got to leave train17:41
@sonney2kbtw 30 C here17:41
pluskidso hot17:41
@sonney2kand in train even hotter17:41
pluskidno air conditioner ?17:41
pluskidtoooooooo strange, you have WIFI on train but have no air conditioner....17:42
blackburnokay the only libshogun examples those are failing are modelselection ones17:48
CIA-113shogun: Sergey Lisitsyn master * r75cc249 / src/shogun/classifier/svm/LibSVMOneClass.h : Made LibSVMOneClass do not require labels -
-!- karlnapf1 [] has joined #shogun17:57
-!- karlnapf [] has quit [Ping timeout: 276 seconds]17:59
-!- pluskid [~pluskid@] has quit [Quit: Leaving]18:02
-!- karlnapf [] has joined #shogun18:03
-!- karlnapf1 [] has quit [Ping timeout: 260 seconds]18:03
puffin444Hey I have a question18:06
-!- karlnapf1 [] has joined #shogun18:06
puffin444It looks like the Google office event limits presentations to 5-8 min, 4 slides.18:07
-!- karlnapf [] has quit [Ping timeout: 260 seconds]18:07
puffin444How does one present GPs to a general audience in 5 mins?18:07
-!- vikram360 [~vikram360@] has joined #shogun18:08
-!- karlnapf [] has joined #shogun18:09
-!- karlnapf1 [] has quit [Ping timeout: 245 seconds]18:10
-!- karlnapf1 [] has joined #shogun18:12
-!- karlnapf [] has quit [Ping timeout: 245 seconds]18:13
-!- karlnapf [] has joined #shogun18:16
-!- karlnapf2 [] has joined #shogun18:17
karlnapf2why do you enforce that +1 and -1 exists in binary labels? what if machine only returns +1 in apply?18:17
karlnapf2this causes the error in x-val18:17
blackburnpuffin444: oops18:18
puffin444blackburn: oops?18:18
blackburnpuffin444: 5 mins? :)18:18
-!- karlnapf1 [] has quit [Ping timeout: 260 seconds]18:18
blackburngood question18:18
puffin444Yep. I have 5 mins to explain what Shogun is, why its cool, and and advanced machine learning algorithm :)18:19
-!- karlnapf [] has quit [Ping timeout: 245 seconds]18:20
blackburnpuffin444: I am afraid you would need to explain what means 'gaussian', 'process' and how can you apply it :)18:21
-!- karlnapf [] has joined #shogun18:21
-!- karlnapf2 [] has quit [Ping timeout: 245 seconds]18:21
puffin444blackburn: I guess so.18:21
blackburnas for shogun18:22
blackburnsay it is large scale (sonney2k applied it to whole human genome once)18:22
CIA-113shogun: Heiko Strathmann master * r653fa34 / src/shogun/evaluation/ContingencyTableEvaluation.cpp : Merge pull request #548 from karlnapf/master -
CIA-113shogun: Heiko Strathmann master * r0cb67a7 / src/shogun/evaluation/ContingencyTableEvaluation.cpp : bugfix: removed valid check since this case has to be allowed -
blackburnhave a lot of interfaces18:23
blackburnand under heavy and fast development18:23
puffin444sonney2k applied to the whole human genome! :o18:24
blackburnpuffin444: yes iirc he trained svm on some subset and applied to the whole genome18:24
puffin444Do you some of the details? I would definitely want to bring that up.18:25
puffin444blackburn: Also, didn't you mention that you applied Shogun to traffic lights?18:25
blackburnpuffin444: hmm yes I am applying it to recognize traffic signs (not lights)18:25
puffin444blackburn: Traffic signs. Got it.18:26
blackburnpuffin444: wiking is applying it to recognize medical images18:27
puffin444blackburn: Like fMRI?18:27
-!- karlnapf [] has quit [Ping timeout: 245 seconds]18:28
blackburnyes it is one of modalities18:28
-!- karlnapf [] has joined #shogun18:29
-!- karlnapf [] has quit [Ping timeout: 245 seconds]18:43
-!- karlnapf [] has joined #shogun18:46
@sonney2kpuffin444, shogun is used to do genome annotations, predictions have been made to detect genes (part of official annotation of C.elegans is mGene that uses shogun for the genomic signals)18:47
@sonney2kI used it to predict the solubility of molecules in water18:47
@sonney2kand to predict starts of the gene on the human genome18:47
@sonney2kpuffin444, we used the coffin principle to predict gender based on face images18:48
puffin444sonney2k: What classification task was used for 6-7 billion test cases in the human genome? Was that to detect genes?18:48
@sonney2kkarlnapf, the check should only be there when really needed18:49
@sonney2kpuffin444, starts of the gene18:49
puffin444It's just that I only have so much space on my maximum 4 slides18:49
puffin444sonney2k: ok18:49
@sonney2kbasically for each nucleotide scanning the human genome in both directions18:49
@sonney2k~7billion nt18:49
@sonney2kbut I also learned a splice site detector (something inside of a gene) and I trained on ~50 million sequences of length 141 in a ~10 million dimensional feature space18:51
-!- karlnapf1 [] has joined #shogun18:51
blackburnhmm I get 0.0 accuracy constantly18:51
@sonney2kblackburn, you could be better :)18:52
blackburnsome bug there..18:52
-!- karlnapf [] has quit [Ping timeout: 245 seconds]18:52
-!- karlnapf1 [] has quit [Ping timeout: 245 seconds]18:55
-!- karlnapf [] has joined #shogun18:58
-!- alexberlin [82955843@gateway/web/freenode/ip.] has joined #shogun18:58
@sonney2khi alexberlin18:58
karlnapfsonney2k, blackburn and the rest: I give up for today, this internet connection drives me mad, hope its better tomorrow19:01
karlnapfbut the modsel examples should run now19:01
karlnapfthere was this check19:01
karlnapfassertion that labels of both +1 and -1 appear in label instance in the evaluation19:01
karlnapfbut this doesnt always hold19:01
karlnapfbye :)19:02
-!- karlnapf [] has left #shogun []19:02
CIA-113shogun: Sergey Lisitsyn master * refb79fe / src/shogun/multiclass/MulticlassOCAS.h : Changed C type of multiclass OCAS to proper float64_t -
-!- gsomix [~gsomix@] has joined #shogun19:14
blackburnsonney2k: here?19:27
blackburnI do not understand why do you call obtain from generic in these macroses19:28
@sonney2kblackburn, yes19:28
blackburnsonney2k: e.g. apply_binary  returns BinaryLabels19:29
blackburnwhy to cast it19:29
@sonney2kno need to indeed19:30
blackburnsonney2k: could you please remove unneeded apply macroses too?19:30
CIA-113shogun: Soeren Sonnenburg master * r2786e87 / src/interfaces/modular/Machine.i : remove unneeded casts -
blackburnfor conjugateindex, wdocas, etc19:30
@sonney2kblackburn, why are they not needed?19:31
blackburnsonney2k: hmm no apply(CFeatures* data) here anymore19:31
@sonney2kblackburn, that is why we need to introduce one19:32
@sonney2kblackburn, are we mostly good with the transition now btw?19:34
blackburnsonney2k: yes19:34
blackburnsonney2k: I was confused with some case here19:35
blackburnapply() returned some useless shit with 0.0 accuracy19:35
blackburnand apply_multiclass() is correct19:35
blackburnI use MulticlassLibLinear19:35
blackburnI do not understand why it is so19:35
@sonney2kblackburn, please compare label/confidence outputs19:37
shogun-buildbotbuild #236 of nightly_none is complete: Failure [failed compile]  Build details are at
-!- shogun-buildbot [] has quit [Quit: buildmaster reconfigured: bot disconnecting]19:38
-!- shogun-buildbot [] has joined #shogun19:38
@sonney2kblackburn, I am asking because we should try to strip shogun src19:42
@sonney2kfrom this massive 100mb file19:42
blackburnsonney2k: ah yes19:45
blackburnsonney2k: do you know how to do that? I found some ways to do that19:45
blackburnapply() = [ 13.  13.  13. ...,  23.  23.  23.]19:46
blackburnapply_multiclass() = [ 16.   1.  38. ...,   6.   7.  10.]19:46
blackburnconfidences are []19:46
blackburnhow can that be damn..19:47
@sonney2kblackburn, please valgrind19:53
@sonney2kI will be away for some time now - blackburn if you have time please try to fix static interfaces...19:56
blackburnI wanted to get my thesis 0.05% complete tonight19:57
puffin444Going to lunch. Be back in a bit.20:13
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Quit: Page closed]20:14
-!- vikram360 [~vikram360@] has quit [Ping timeout: 276 seconds]20:19
CIA-113shogun: Sergey Lisitsyn master * r40e4ea3 / src/interfaces/modular/Machine.i : Splitted methods in apply macro in Machine.i. -
blackburnhmm that didn't help20:47
@sonney2kblackburn, can we revert it then?20:57
blackburnsonney2k: problem is still here20:57
@sonney2kso yes20:57
@sonney2kdid you valgrind?20:57
blackburnsonney2k: apply(feats) applies for train features20:57
blackburnno not memory problem20:58
@sonney2kblackburn, no valgrind errors?20:58
blackburnsonney2k: it applies to train features20:58
blackburnnot to given ones20:58
blackburnsonney2k: can you try to fix it?20:59
CIA-113shogun: Sergey Lisitsyn master * r99f81b6 / src/interfaces/modular/Machine.i : Revert "Splitted methods in apply macro in Machine.i." -
@sonney2kblackburn, can you try mcsvm.apply(testfeat)20:59
@sonney2kI have an idea...20:59
blackburnclassifier = MulticlassLibLinear(C,feats_train,labels)20:59
blackburnout = classifier.apply(feats_test).get_labels()20:59
blackburnprint out20:59
blackburnout = classifier.apply_multiclass(feats_test).get_labels()20:59
blackburnprint out20:59
blackburnsonney2k: ^20:59
blackburndifferent out20:59
@sonney2kblackburn, I guess it is because of the apply(CFeatuers=NULL)21:01
blackburnyes I tried to approach it but that was wrong guess21:01
@sonney2kblackburn, anyway can we now first fix the 100MB repository21:01
@sonney2kwhat was the filename again?21:01
blackburnjust remove *.gz2.pickle21:01
blackburnand *.gz221:01
@sonney2kI will create a fresh clone21:02
@sonney2kwill take some time21:02
blackburnsonney2k: will it break all forks/21:03
@sonney2kwe have to announce this on the mailing list21:03
@sonney2kblackburn, *.bz2 you mean?21:06
blackburnsonney2k: eb07b040ce533393043d9b3241e79a62e8f390ba21:06
@sonney2kok so *.bz2*21:07
* sonney2k 50% cloned21:08
@sonney2kfull checkout is 159M21:10
@sonney2klets remove the files21:10
blackburnsonney2k: 154M../../.git21:10
blackburnyes just like .git21:10
@sonney2kblackburn, do you want to remove sth else?21:11
@sonney2know would be a good time :D21:11
blackburnsonney2k: hmm no idea21:11
blackburnwere there any other bad commits?21:11
@sonney2kbut who cares - we suck bad w/ git at least we did when we started21:12
@sonney2kno way to hide that21:12
blackburnI do not care about lame commits at all21:13
@sonney2kme neither, I am very proud of the F... OFF SOEREN commit21:13
@sonney2kand other username not set ones :D21:13
blackburnhm there was a use21:13
blackburnI liked very much21:14
blackburnsonney2k: that's ok21:16
blackburnif you didn't gc prune21:16
@sonney2kI did21:16
blackburnsonney2k: how did you use filter-branch?21:17
@sonney2kgit filter-branch --index-filter 'git rm --cached --ignore-unmatch *.bz2*'21:18
@sonney2kit mentioned rm of all the files btw21:18
@sonney2kblackburn, let me push it to shogun-toolbox/test21:19
@sonney2kdammed connection died21:22
@sonney2k...120k/s now21:23
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun21:23
@sonney2kblackburn, it's there21:24
@sonney2kplease check it our21:25
blackburn50% 10mb21:26
blackburn43 mb21:26
blackburnsonney2k: .git is 23M21:27
@sonney2kok great21:29
@sonney2kso shall I force push it?21:29
@sonney2kbasically all history sha's since january will then be different21:29
@sonney2kand everyone will have to rebase21:30
@sonney2kblackburn, ??21:32
blackburnsonney2k: I don't mind21:32
blackburnit is much better than having 150M .git21:32
@sonney2kblackburn, done21:36
@sonney2kplease pull again from that one21:36
@sonney2kdon't overwrite the other yet21:36
blackburnsonney2k: from test?21:37
blackburnblackburn@blackburn-laptop:~/OSS/test/test$ git pull21:38
blackburnfatal: not found: did you run git update-server-info on the server?21:38
@sonney2kblackburn, not test21:38
@sonney2kthe real one21:38
CIA-113shogun: Sergey Lisitsyn master * r91829c9 / src/interfaces/modular/Machine.i : Revert "Splitted methods in apply macro in Machine.i." (+1449 more commits...) -
@sonney2ktest is RIP21:38
@sonney2kbtw look at what it does
@sonney2kbtw are branches now dead too?21:39
@sonney2kseems like21:39
blackburnsonney2k: .git is 158M here now :)21:41
@sonney2kblackburn, on a fresh clone?21:41
blackburnold one21:41
blackburnok doing fresh21:41
@sonney2kdon't erase the old one though21:41
@sonney2kwe still have backups at google / shogun-toolbox21:43
@sonney2kI guess I need to clone with --mirror to get the branches21:43
@sonney2kdoing that now21:46
-!- hoijui [] has joined #shogun21:49
@sonney2kat least we don't have to delete the whole thing21:49
@sonney2kmirror gets the branches21:56
@sonney2krunning filterbranch again21:56
@sonney2kI guess I have to push --mirror or push --all21:57
@sonney2kblackburn, btw we started with 22M21:59
@sonney2kwhen we switched to git21:59
blackburnwell 30M now not much larger22:00
@sonney2kyeah git rocks22:03
@sonney2kconsidering how much changed22:03
-!- alexberlin [82955843@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]22:07
* gsomix writes poisson brackets, live-civita symbols and thinks about him...22:13
shogun-buildbotbuild #971 of libshogun is complete: Failure [failed git]  Build details are at  blamelist: blackburn91@gmail.com22:18
@sonney2kgsomix, want to be a guinea pig?22:25
shogun-buildbotbuild #972 of libshogun is complete: Success [build successful]  Build details are at
@sonney2kblackburn, ok done22:25
@sonney2kI think we should exercise once how the rebase could work22:26
@sonney2kgsomix, you have a pull request open on git hub22:26
blackburnsonney2k: hm let me try to rebase my master22:26
blackburnin fork22:26
@sonney2kgsomix, can you please make a copy of your shogun git22:26
@sonney2kthen in this copy do a git pull --rebase ?22:26
@sonney2kblackburn, yeah please do22:26
gsomixsonney2k, copy? ok.22:31
blackburnsonney2k: loong rebase22:31
@sonney2kblackburn, check the size of the repository afterwards...22:31
@sonney2knot that we are again >100MB22:32
blackburnsonney2k: hmm my fork is not clean I am afraid22:32
gsomixsonney2k, do you mean local copy?22:32
@sonney2kblackburn, ?22:32
@sonney2kgsomix, yes22:32
blackburnsonney2k: that big commit was in my fork I think22:32
@sonney2kblackburn, I see22:32
@sonney2kgsomix, stop it then22:32
@sonney2kblackburn, I guess it has to do with the (remote) branches22:32
blackburnok rebase in progress still lets wait22:33
gsomixsonney2k, hm, ok.22:33
@sonney2kblackburn, I have no idea then how to rescue branches22:34
blackburnsonney2k: ok it failed to rebase22:34
blackburnwhich branches?22:34
blackburnwhy we want these branches but not tags?22:35
@sonney2kblackburn, tags too of course22:35
blackburnsonney2k: but what's wrong with tags?22:36
@sonney2kwe need the branches and tags22:36
blackburnI understand - what is broken?22:36
@sonney2kblackburn, I don't know how to run filterbranch and have the tags/branches22:37
blackburnsonney2k: it seems you haave to specify all branches in filter branch22:38
@sonney2kblackburn, how22:38
blackburnLets you rewrite git revision history by rewriting the branches mentioned in the <rev-list options>, applying custom filters on each revision. Those filters can modify each tree (e.g. removing a file or running a perl rewrite on all files) or information about each commit. Otherwise, all information (including original commit times or merge information) will be preserved.22:38
@sonney2k-- --all22:39
blackburnsonney2k: but our tags and branches have no .gz2 I think?22:43
@sonney2kvipins commit is from january22:45
@sonney2kno idea22:45
blackburnsonney2k: last release was in december I think22:51
blackburnor I am sure22:51
@sonney2kI had to do git filter-branch -f --index-filter 'git rm --cached --ignore-unmatch *.bz2*' --tag-name-filter cat -- --all22:59
@sonney2kblackburn, please clone again22:59
CIA-113shogun: Soeren Sonnenburg shogun_1.1 * rf000737 / src/shogun/lib/versionstring.h : Tagging shogun_1.1.0 release -
-!- hoijui [] has quit [Quit: Leaving]23:01
blackburnsonney2k: ok what to check?23:01
blackburn.git is 30M still23:01
@sonney2kblackburn, clone23:01
@sonney2kvery good23:02
@sonney2kgit branch -a23:02
@sonney2kand then try a rebase...23:02
blackburnblackburn@blackburn-laptop:~/OSS/shogun_new/shogun/src$ git branch -a23:02
blackburn* master23:02
blackburn  remotes/origin/HEAD -> origin/master23:02
blackburn  remotes/origin/boost_serialization23:02
blackburn  remotes/origin/c5023:02
blackburn  remotes/origin/galaxy23:02
blackburn  remotes/origin/master23:02
blackburn  remotes/origin/multiboost23:02
blackburn  remotes/origin/shogun_1.023:02
blackburn  remotes/origin/shogun_1.123:02
blackburn  remotes/origin/structure23:02
blackburnrebase of??23:02
blackburnsonney2k: ehm what to rebase?23:03
@sonney2kblackburn, what you did before23:03
@sonney2kI don't have anything to rebase23:03
blackburnsonney2k: fork?23:03
@sonney2kblackburn, or we ask gsomix again23:03
blackburnhmm ok let me try to rebase fork23:04
blackburnthat takes a while23:04
@sonney2kblackburn, so?23:09
@sonney2kgsomix, can you try the same?23:09
@sonney2kgsomix, do a copy of your git repo23:09
blackburnsonney2k: yeah failed to merge again23:09
@sonney2k(local copy)23:09
@sonney2kblackburn, what did you do?23:09
blackburnsonney2k: tried to rebase my fork23:10
@sonney2kblackburn, command line?23:10
blackburngit fetch upstream/master23:10
blackburngit rebase upstream/master23:10
@sonney2kblackburn, wait but do you have an additional commit?23:11
blackburnsonney2k: no it should be clean in master ..23:12
blackburnI guess it would be easier to re-create fork23:13
gsomixsonney2k, missed. =___= should I do something?23:14
@sonney2kgsomix, yes please do23:15
@sonney2k1) create a copy of your git (local copy)23:15
@sonney2kthen do a git pull --rebase on the copy23:15
-!- n4nd0 [] has joined #shogun23:16
gsomixsonney2k, ok. without fetch? I updated the repository in the afternoon.23:16
@sonney2kgsomix, all obsolete now23:16
@sonney2kbut please *keep a copy*23:16
@sonney2kgsomix, either with git fetch and then git rebase or git pull --rebase23:17
gsomixsonney2k, >> Current branch master is up to date.23:19
@sonney2kgsomix, ok23:19
@sonney2kgsomix, and if you try this on some of your pull request feature branches?23:19
blackburnsonney2k: still..23:20
@sonney2kblackburn, still what?23:21
blackburnFailed to merge in the changes.23:21
blackburnPatch failed at 0001 Introduced converters module, moved DR preprocessors to it, fixed examples, updated other classes according to changes23:21
@sonney2kblackburn, lets see if it works for gsomix...23:21
@sonney2kgsomix, any news?23:21
@sonney2kblackburn, it did work here too btw23:22
gsomixsonney2k, rebase and push?23:22
blackburnsonney2k: on your fork?23:22
@sonney2kgsomix, update your pull request23:22
@sonney2kbut you need to rebase before23:22
@sonney2kon your feature branch23:22
@sonney2kblackburn, do you remember how we added the shogun-toolbox and google code remotes23:23
@sonney2kwith --mirror=push I mean?23:23
blackburnsonney2k: I do not remember any --mirror there.. what for?23:24
@sonney2kblackburn, to push the branches23:24
@sonney2kor is this not necessary?23:24
gsomixsonney2k, sec23:24
@sonney2kplease do now23:24
@sonney2kthis is urgent23:24
blackburnhmm I do not remember sorry - I think git push should do something similar23:24
@sonney2kmaybe we used push --mirror at some point23:25
gsomixsonney2k, >> Everything up-to-date23:26
@sonney2kgsomix, on gsomix:titled_features ?23:27
gsomixlooks like that "pull --rebase" didn't rebase? I'm confused.23:27
@sonney2kwhich branch are you on?23:27
gsomixsonney2k, titled_features23:28
@sonney2kand what does git log show?23:28
gsomixsonney2k, my last commit. :(23:29
@sonney2kgsomix, and befoere that?23:29
gsomixcommit df320e82b334cd42043d239431454ec4243f370c23:29
gsomixAuthor: Soeren Sonnenburg <>23:29
gsomixDate:   Tue May 22 11:00:26 2012 +020023:29
gsomix    make get_label_type pure virtual again23:29
@sonney2kgsomix, ok then you didn't pull --rebase with master23:30
gsomixsonney2k, re-do?23:31
blackburngit pull --rebase upstream master23:32
blackburnI guess23:32
gsomixusually I do "git rebase upstream/master"...23:32
@sonney2kgsomix, but then you would need to fetch first23:33
@sonney2kso fetch first and then do it this way23:33
gsomixsonney2k, in progress.23:35
gsomixslow. I think I can solve couple of problems in optics23:38
@sonney2kI just did the same on sth23:41
@sonney2kand it just took 1 sec23:41
CIA-113shogun: Sergey Lisitsyn master * rf575c12 / src/interfaces/modular/Machine.i : Revert "Splitted methods in apply macro in Machine.i." -
CIA-113shogun: Soeren Sonnenburg master * rd126d99 / src/interfaces/modular/Machine.i : only use %rename for non-type aware interfaces -
CIA-113shogun: Sergey Lisitsyn master * r5b12cea / src/interfaces/modular/Machine.i : Splitted methods in apply macro in Machine.i. -
@sonney2kwtf do I have the revert splitted stuff 2 times?!23:43
@sonney2kgsomix, any news?23:45
@sonney2kblackburn, gsomix ok then. no idea how to rescue that situation23:47
blackburnsonney2k: recreate all forks23:47
@sonney2kI would say fresh checkout and git-format patch to the rescue23:47
@sonney2kblackburn, shall I push --force again with the last 3 commits removed?23:48
@sonney2kit is a bit weird that we have them twice23:48
@sonney2kblackburn, I am too tired for all that23:51
@sonney2kplease do and send an email to the ML23:51
blackburnhmm strange commit arrangement23:56
--- Log closed Wed May 23 00:00:41 2012