Open in new window / Try shogun cloud
--- Log opened Thu Aug 16 00:00:17 2012
gsomixgood book about 3 generations of programmers: MIT (1950-1960), hardware engineers, game developers and... Stallman.00:11
gsomixsonney2k, ta-dam00:17
gsomixaww, author is Levy00:20
gsomixis it what we wanted?00:27
gsomixok, I need clean  up my code and sleep00:34
-!- gsomix [~gsomix@] has quit [Ping timeout: 252 seconds]00:42
-!- zxtx [] has quit [Ping timeout: 272 seconds]01:16
-!- zxtx [] has joined #shogun02:02
shogun-buildbotbuild #56 of nightly_none is complete: Failure [failed git]  Build details are at
shogun-buildbotbuild #65 of nightly_default is complete: Failure [failed git]  Build details are at
shogun-buildbotbuild #52 of nightly_all is complete: Failure [failed git]  Build details are at
-!- zxtx [] has quit [Ping timeout: 246 seconds]03:30
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun04:25
puffin444hey wiking04:25
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]04:50
-!- shogun-buildbot_ [] has joined #shogun05:26
-!- shogun-buildbot [] has quit [Ping timeout: 268 seconds]05:27
-!- shelhamer [] has joined #shogun08:54
-!- uricamic [~uricamic@2001:718:2:1634:876:e01f:ab62:a2a] has joined #shogun08:54
-!- shelhamer [] has quit [Client Quit]08:54
-!- gsomix [~gsomix@] has joined #shogun08:56
-!- heiko1 [] has joined #shogun09:59
-!- heiko1 [] has left #shogun []10:01
gsomixgood morning10:22
-!- heiko1 [] has joined #shogun10:27
-!- heiko1 [] has left #shogun []10:27
-!- gsomix_ [~gsomix@] has joined #shogun11:12
-!- gsomix [~gsomix@] has quit [Ping timeout: 244 seconds]11:15
-!- blackburn [~blackburn@] has joined #shogun11:41
-!- nicococo [] has joined #shogun11:52
-!- nicococo [] has left #shogun []12:02
@sonney2kgsomix_, so you make the zero copy optional right?12:28
@sonney2kbecause one still has to use frombuffer() ?12:29
gsomix_sonney2k, yep. I did not figure out, how overload constructor, so there is frombuffer method.12:32
CIA-21shogun: Soeren Sonnenburg master * r63c8091 / (data doc/tutorial): require current data / doc -
gsomix_sonney2k, btw, maybe do you know, how raise exceptions in python c/api?12:33
@sonney2kgsomix_, IIRC you have to return some empty python object (and incref it ... IIRC PyNone or so?)12:34
@sonney2kand set the error message12:35
@sonney2kgsomix_, the static python interface should have that code12:35
@sonney2kgsomix_, instead of frombuffer you could just modify the typemaps12:35
shogun-buildbot_build #348 of deb1 - libshogun is complete: Success [build successful]  Build details are at
@sonney2kgsomix_, or is this even possible? I mean do you need access to the underlying SGObject?12:36
gsomix_sonney2k, no, I don't.12:37
wikingmeeting :D12:37
gsomix_>> you could just modify the typemaps || and constructor. I want something like in numpy - DenseFeatures(array, copy=True/False).12:38
shogun-buildbot_build #287 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at  blamelist: Soeren Sonnenburg <>12:38
@sonney2kgsomix_, yeah that would be cool ...12:39
blackburnI fixeed that12:40
@sonney2kblackburn, no idea - looks like recent commits broke even the standard git submodule update process12:41
-!- heiko [] has joined #shogun12:42
wikingok warnning fix:
CIA-21shogun: Viktor Gal master * r1f29074 / src/shogun/machine/LinearLatentMachine.cpp : Remove apply() implementation from LinearLatentMachine -
CIA-21shogun: Sergey Lisitsyn master * rbf3f2dd / src/shogun/machine/LinearLatentMachine.cpp : Merge pull request #720 from vigsterkr/latent -
CIA-21shogun: Sergey Lisitsyn master * rad15be9 / (5 files in 5 dirs): Fixed KRR examples -
-!- vojtech [9320543b@gateway/web/freenode/ip.] has joined #shogun12:44
blackburnuricamic: vojtech: is it possible to extend ocas or/and bmrm to use mixed norms like L1/Lq?12:45
vojtechblackburn: it would not be easy.12:46
blackburnin both cases?12:46
vojtechboth translate the original hard problem to a simpler reduced problem which is QP task in the case of L2 norm12:47
vojtechif you would use L1 norm it would lead to LP12:47
CIA-21shogun: Sergey Lisitsyn master * ree71cb3 / examples/undocumented/libshogun/test.cpp : Removed test example introduced by mistake -
blackburnI see12:47
vojtechit would not be stable12:48
vojtechit would require additional stabilization term12:48
vojtechnot easy12:48
wikingheiko: yo12:48
blackburnI thought of two applications - multitask and group lasso-like thing12:48
blackburnthat's why I asked12:48
heikowiking hi12:48
wikingheiko: how's the mocking going?12:49
blackburnvojtech: what is use case of bmrm with proximal term?12:49
heikowiking, the gmock installation is still not working, but I worked on something else the last days (cross-validation output stuff and labels)12:49
heikoDo you have a linux system where you can run the tests?12:49
vojtechblackburn, for example structured output SVM12:49
wikingheiko: cool, u mean you are having problem with the configure script?12:49
wikingheiko: yes indeed.. then i'll try to fix it for you12:49
wikingheiko: i've already installed gmock there from .deb12:50
heikothat would be great12:50
heikowhich version?12:50
wikingso i see the problem with ./configure12:50
blackburnvojtech: I mean, what can we do with that w_t term in objective?12:50
heikosince the current one in debian/ubuntu doesnt have a config script12:50
heikoto get the paths12:50
wikingheiko: yeah neither this one... i mean i'm using latest LTS of ubuntu12:50
heikoI added the possibility to add a custom one to shogus confgure script12:50
wikingafaik htat's the latest ubuntu release ;P12:50
heikobut still I have problems12:50
wikingheiko: i see that12:50
heikosame here, latest ubuntu12:50
wikingheiko: i'll try to steal from somewhere the configure script for gmock ;P12:51
wikingheiko: i guess checkheader check lib will work as well12:51
vojtechblackburn, the proxy term is introduce to avoid zig-zag behaviour at early stages of convergence. I don't know what you can do with it ...12:51
heikowiking, ok12:51
vojtechblackburn, may be I didn't understand the question12:51
blackburnvojtech: so it is just a helper thing?12:51
vojtechyes, to speedup convergence12:52
blackburnvojtech: well there are some domain adaptation formulations that pull w to some w_t12:52
vojtechI know12:52
wikingheiko: a bit scary stuff though.... is that for some reason the gmock package does not have gtest bundled into the package12:52
wikingheiko: so i'll have to add a gtest test as well12:53
blackburnvojtech:  is that possible with that proximal term?12:53
vojtechblackburn, the implemented method is generic, i.e. you can add any convex term, e.g. additional proxy trerm12:53
blackburnvojtech: I see, thanks12:54
heikowiking, ok , strange12:54
heikowiking, I got loads of tests ready, so looking forward to this working :)12:54
wikingheiko: aaaah ok it does ...12:54
wikingheiko: it's just a dep package: Depends: libgtest-dev12:55
heikoI got that installed12:57
-!- cheng [] has joined #shogun12:59
-!- nicococo [] has joined #shogun13:00
-!- alexlovesdata [] has joined #shogun13:02
@sonney2khmmhh what happened to the NY folks?13:04
@sonney2kblackburn, is chris away?13:05
blackburnsonney2k: I don't know13:05
blackburnwho are NY folks?13:05
@sonney2kwho else is missing - n4nd0 (who said that he cannot make it)13:05
blackburnI know only one13:06
@sonney2kblackburn, chris, puffin13:06
blackburnpuffin is pretty far away from NY :)13:06
@sonney2ktimezone wise - same13:06
blackburnyou mean timezone wise13:06
@sonney2kpluskid also cannot make it13:07
@sonney2kso I guess we can start then13:07
@sonney2kAlright everyone - welcome to our final celebration.13:07
@sonney2kLast weeks lots of PRs have been merged and I am very happy to see how much you have contributed (and hopefully will be contributing in the future).13:07
@sonney2kNow please everyone help getting shogun stable such that we can release version 2.0 with your code merged! So focus now is bugfixes (get the warnings down to ZERO), tests, documentation, tutorial.13:07
@sonney2kThat said - only heiko sent me some figures for the new website. Excuses anyone?13:07
@sonney2klook heiko even shows his angry face13:08
@sonney2kuricamic, wiking, blackburn  c'mon!13:09
-!- pluskid [72f6b3c3@gateway/web/freenode/ip.] has joined #shogun13:09
@sonney2kpluskid, and you too :D13:09
wikingpluskid: you should hide! :)13:09
blackburnsonney2k: ich spreche englisch nicht13:09
wikingpluskid: sonney2k is angry :>13:09
@sonney2kwiking, no heiko is13:09
pluskidwow, so few people?13:09
* heiko raging13:09
chengpluskid: Hi, was worried that I had to actually do some work. ;-)13:10
pluskidcheng: haha :p13:10
@sonney2kso pluskid, uricamic, wiking, blackburn  - please send pictures created with the aid of shogun *TODAY*13:10
pluskidsonney2k: sorry for absent during the last several days, but what kind of pictures do you need?13:11
@sonney2kgsomix_ is somehow an exception - protocol buffers are not really hmmh screenshots13:11
blackburndas fasse ich nicht13:11
heikopluskid, just something that shows how your methods work13:11
@sonney2kpluskid, some nice ones for the website13:11
heiko(which is easy for GP)13:11
vojtechsonney2k, could you remind us what kind of pictures. E.g. what picture should be send for a generic SVM solver ?13:11
pluskidsonney2k, heiko OK13:12
heikoyou can do all these cool heatmap plot with posterior probabilities13:12
chengpluskid: Once you get the sign dataset working, I guess there will be nice pictures.13:12
pluskidcheng: that is impossible for the *TODAY* deadline13:12
@sonney2kvojtech, well some illustrative application - something where you can show what a method does / how it does it in a figure13:12
@sonney2ksomething *fancy*13:12
@sonney2kI don't mind if you don't do it today but create a more fancy one a little later13:12
@sonney2kif you tell me when I get it13:13
pluskidsonney2k: are there any examples?13:13
blackburncheng: pluskid: ah right sorry I forgot to prepare it13:13
chengpluskid: I know. Didn't mean to pressure you.13:13
@sonney2kcheng, that is my job anyways ;)13:13
shogun-buildbot_build #359 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Soeren Sonnenburg <>13:14
@sonney2kvojtech, e.g. for dimension reduction methods some embedding. for svm the usual figure good / vs bad with kernels (we have that already btw)13:14
heikodoorbell ..13:14
blackburncowbell ..13:14
@sonney2kusually that is some small python script so you can aswell look for examples in examples/undocumented/<if>/graphical13:15
@sonney2kand add yours there13:15
vojtechsonney2k, ok we will create something13:15
@sonney2kand then just save the figure13:15
@sonney2kanyone still has questions to that?13:15
@sonney2kif not...13:15
@sonney2klets continue13:16
@sonney2kNow for some formal stuff: According to the timeline there are two upcoming todo items:13:16
@sonney2k1. Mentors/Students - please submit your final evaluations by August 23rd latest (24th is the hard deadline from google).13:16
@sonney2k2. Students submit your code samples to google (blackburn will send around a script on the mailinglist to do this) after August 31.13:16
@sonney2kso 1) is very important not to miss for mentors/students13:16
@sonney2kotherwise no money for the student13:16
heikoone more thing about the plots: please use the latex integration script in the graphical folder in python_modular, it ensures that all plots integrate well with the tutorial13:17
@sonney2kso students push you mentors!13:17
@sonney2kit will be the same effort as last time btw13:17
@sonney2kheiko, well I just need .png's or /jpgs for now - I think we can do the rest for the tutorial later13:18
gsomix_sonney2k, *push*13:18
heikosonney2k, I know, still, its just one line import13:18
@sonney2kgsomix_, no worries13:18
heikoThen they all have same fonts etc13:18
@sonney2kthat reminds me, heiko, pluskid, blackburn have been writing some nice tutorial / documentation for their parts13:18
@sonney2kyou can find the latex source under shogun/doc/tutorial13:19
chenglike +113:19
@sonney2kplease in the final weeks contribute to that tutorial13:19
pluskidits in submodular btw13:19
@sonney2kvojtech, uricamic - you could probably just copy formulas description over from your .tex source of your papers13:20
@sonney2knicococo, and you too13:20
nicococosi claro13:20
gsomix_btw, if somebody wants to check shogun on python3 - welcome. :)13:20
@sonney2kgsomix_, I did only once but weeks ago - it worked nicely - maybe we need a buildbot test...13:21
@sonney2kstudents: regarding 2) this is an easy task with the script blackburn will send around - so don't worry too much about it13:21
blackburn67 warnings left13:21
@sonney2kplease help us getting these 67 -> 013:22
uricamicI have few warnings still there, I am waiting for the wiking PR to merge and I will fix them13:22
blackburnuricamic: are you ok to merge it?13:22
@sonney2kas a reminder: that's the url where you can see the status
uricamicblackburn: yes13:22
blackburnokay that's what I was waiting for13:23
CIA-21shogun: Viktor Gal master * ra72e3cf / (25 files in 4 dirs):13:23
CIA-21shogun: Refactor RiskFunction in structure methods This code refactoring basically adds13:23
CIA-21shogun: RiskFunction functionality into StructuredModel. This way we can remove13:23
CIA-21shogun: RiskFunction and all the related objects and have only one user defined class13:23
CIA-21shogun: (StructuredModel) for SO methods. -
CIA-21shogun: Sergey Lisitsyn master * r3ce4fd2 / (25 files in 4 dirs): Merge pull request #719 from vigsterkr/so -
uricamicI see, I didn't know that :)13:23
uricamicok, thx13:23
blackburnyeah I kept it in secret13:23
blackburn(forgot to ask)13:23
@sonney2kNow please everyone tell us what you liked about doing shogun - GSoC, what you did hate. Any kind of honest criticism / feedback is very welcome such that we can learn what we should improve. I very much prefer unfiltered direct feedback (like fuck of Soeren!!!) in contrast to not letting us know you did hate us :D13:23
wikingwooohoooo BIGPATCH :)13:23
@sonney2kwho wants to start?13:23
pluskidsonney2k: so this time complain instead of weekly summary?13:24
@sonney2kpluskid, is this a complaint?13:24
@sonney2kplease complain :D13:25
pluskidI would say great org!13:25
pluskidbut shall I complain about SG_REF/SG_UNREF?13:25
@sonney2k(I don't mind others joining pluskids complaints)13:25
blackburnsonney2k: tarball generator is in the mailing list13:25
pluskidI always forget about when should a SG_REF be put :p13:25
@sonney2kpluskid, sure complain about what you hate13:25
@sonney2kpluskid, that is why we need you post-gsoc13:26
@sonney2kotherwise this mechanism will stay until the nd of days13:26
pluskidoh, I'm starting to regret for saying that :>>>13:26
chengpluskid: Is there stuff from an organisation viewpoint of GSoC that we could make better?13:26
blackburnpluskid: we know you are going to do some research so please take a little effort if required to use shogun13:26
blackburnit is worth the thing I am sure13:27
@sonney2kat least it paid off here13:27
pluskidI guess sending some small gifts to the students will be nice, like bages or T-shirts with SHOGUN on it13:27
pluskidblackburn, sonney2k: sure~~13:27
blackburnshogun get me paid with a lot of useful contacts and skills one can miss when developing they own solution13:27
@sonney2kpluskid, we need some shogun design for the t-shirt ...13:28
pluskidbtw, I saw someone comparing their algorithm with the MKL in shogun at KDD13:28
@sonney2kpluskid, yeah shogun is known for MKL :(13:28
@sonney2kor :)13:28
pluskidsonney2k: why :( ? :D13:28
blackburnI'd like to have a t-shirt with
@sonney2kpluskid, any other feedback?13:29
pluskidbtw, I'm finished with my complain13:29
@sonney2kvojtech, uricamic - mind to continue?13:29
@sonney2kblackburn, then print it yourself13:29
blackburnsonney2k: no I believe it should be official13:29
@sonney2kblackburn, but I usually don't wear pants o_O13:30
blackburnsonney2k: only t-shirt?13:30
gsomix_blackburn, \m/13:30
blackburnbad boy13:30
vojtechsonney2k, Michal should say what he likes/dislikes as he did the work. I was just a scientific support :)13:30
@sonney2kvojtech, well you can complain too :D13:31
@sonney2kwiking / alexlovesdata you too please start to complain13:31
uricamicwell, I quite enjoyed GSoC, so I don't have much complains13:31
blackburnsonney2k: is that a nice pic? :D13:32
uricamicmaybe just our little misunderstanding that resulted in little divergence in so framework :)13:32
@sonney2kuricamic, I hope that this is resolved by now :)13:32
vojtechsonney2k, This year everything went quite smoothly. w.r.t organization I new it from the previous year and w.r.t coding Michal works independently = minimal load for me :)13:32
uricamicthere is only one thing left I need to discuss with n4ndo13:33
blackburnvojtech: sonney2k removed your last gsoc project in the middle of summer :D13:33
@sonney2kwe never had such strong interdependence like in wikings/n4nd0s/uricamics task13:33
blackburnthanks to me :D I restored it with stl vector instead13:33
@sonney2kbut it is back now13:33
uricamiche made some change in the last PR that results in shogun exception when BMRM used for trainig13:33
wikingall's great but why cannot we use STD! :D13:34
uricamicbut I guess, it will be easy to fix that too13:34
wikingespecially STL :D13:34
vojtechblackburn: which project ? The EM algorithm from the last year ?13:34
blackburnvojtech: yeah13:34
@sonney2kwiking, you sound a bit like pluskid - anyway some exceptions are OK but only in .cpp files (though I still don't like it)13:35
vojtechvery good, I know people who use it in their work ...13:35
wikingsonney2k: nah it was just a joke... i mean we can live w/o it13:35
@sonney2knicococo, heiko, blackburn anything you want to complain about?13:35
wikinggot used to SGVector and SGMatrix13:35
nicococon4nd0 is not here i can't complain for him. for me it was pure joy ;)13:35
blackburnsonney2k: I am so integrated to shogun - I can complain only about myself I guess13:36
heikonot here, I wanted the unit-tests, which we now have. Apart from that, everything fine! :) pretty much enjoyed it13:36
@sonney2kblackburn, shall I complain about you then?13:36
blackburnsonney2k: yeah feel free :D13:36
@sonney2kblackburn, you always break the build :D13:36
@sonney2kis that a hobby?13:36
blackburnno I don't, I always repair the build13:37
wikingheiko: dont lie you dont have it! :D13:37
wikingheiko: but will soon ;P13:37
heikowiking, I HOPE SO!!!! :D13:37
wikingah yeah13:37
wikingas a complaint maybe13:37
wikinga serious one13:37
wikingwhy we no switch to cmake completely?13:37
@sonney2kOK so does anyone have an opinion in which direction shogun should be developed in the future?13:38
uricamicwiking: yep I would appreciate that too :)13:38
wikingcmake it :)13:38
uricamic+1 :)13:38
@sonney2kwiking, that sounds like a longer discussion we should rather do offline13:38
wikinghehe ok i'm just saying that now with all the ./configure script hacks i had lately13:38
wikingi know that half of them are not portable13:39
chengHow about making sure that shogun goes into the standard distributions?13:39
wikingi.e. woulnd't work on some arch/distrib combination13:39
@sonney2kwiking, huge effort in any case13:39
wikingbut yeah i know that it does take some time13:39
wikingbtw: another good thing would be if some GUI guy would come around13:39
chengI mean updating the other repositories with more recent versions.13:39
wikingand make a nice gui for shogun :)13:40
blackburngui? why?13:40
heikowiking, gui?13:40
@sonney2kcheng, yeah I tried to do shogun packages for debian/ ubuntu but badly maintained them (due to other errors in clang on powerpc / python packaging stuff)...13:40
heikopretty much work, huge13:40
heikoand also, why13:40
blackburnno idea how gui looks like :D13:40
@sonney2kIsn't python the guy gui?13:40
wikingwhy not? :) i mean i know so many people who wouldn't touch anything that's command line (and they all hack weka because of that)13:40
shogun-buildbot_build #360 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Viktor Gal <>13:40
wikingyeeey i broke it!13:41
@sonney2kit was all broken anywyas13:41
blackburnit was13:41
wikingah ok it wasn't me13:41
chengHow about instead of a gui, a html interface?13:41
@sonney2kbut no one fixed it...13:41
wikingcheng: that's gui as well :)13:41
blackburnI know the cause13:41
wikingcheng: i mean yeah a nice html fronted would be great as well13:41
gsomix_only command line, only hardcore :)13:41
chengsonney2k: for many people, installing from source sounds too scary.13:42
heikosonney2k, yes, I double that13:42
blackburnfrom my experience shogun is not enough and you do a cross-library thing13:42
blackburnno idea how to gui that13:42
heikoI know quite a feq scientists who never compiled anything, so it would be great to have at least debian/ubuntu biulds for shogun13:42
blackburnyeah auto .deb13:43
@sonney2kdjango / shogun ... possible but danger is that we get lots of newbie users13:43
@sonney2kheiko, blackburn, cheng pre-built stuff is on the todo then13:43
@sonney2kanyone has some new ML domains / algorithms in mind?13:44
blackburnsonney2k: ahh we need to emphasize that static is not recommended unless you know what you do13:44
chengwiking: I suggest html instead of gui because building cross platform windows is tough.13:44
blackburncheng: we do not support windows natively anyway13:44
chengsonney2k: deep networks13:44
wikingcheng: second that13:45
@sonney2kcheng, find us some mentor ...13:45
@sonney2kcheng, ask Yann at NIPS :D13:45
chengblackburn: I mean like OSX (sorry, overloaded the word windows(13:45
chengblackburn: Qt and Tk are tricky on osx.13:45
blackburnand I do not want to develop gui13:45
-!- gsomix_ is now known as gsomix13:46
blackburncheng: sonney2k: I agree we have to have deep learning here13:46
@sonney2kbut seriously, cheng, nicococo, vojtech ask at ML conferences if you find sth interesting if these authors wouldn't mind to mentor next year13:46
blackburnhowever it is a kind of magic13:46
@sonney2knone of us here has any clue about deep learning13:46
@sonney2k(except from literature)13:46
wikinguse Theano13:46
nicococoi can ask gregoire montavon (sitting next to me) he is well known for his deep networks...13:46
chengsonney2k: I will do some salesman stuff.13:46
wikingpretty heavy weight shit :)13:46
@sonney2kso we need an expert13:47
@sonney2kotherwise better not touch this field13:47
blackburnwe wanted to implement LMNN13:47
blackburnno idea if it is hot still13:47
@sonney2kcheng, well there is yann's lush, torch5 or 7 , theano ...13:47
@sonney2kI would rather steal code than doing everything from scratch though13:48
chengThe other popular area is sparse learning, compressed sensing.13:48
blackburnyeah compressed sensing is really hot13:48
chengI just saw David Donoho being a superhero at an MRI conference.13:48
blackburni should contact igor carron and ask about mentoring something13:48
@sonney2kok if you still have ideas bring them on any time but lets continue for now13:49
@sonney2kStudents/Mentors as usual please give us an update - what is left to do / or do you still intend to do before the planned shogun release (September 1) and maybe later. Did you manage to implement everything you put on your schedule / are you satisfied with the work you did or what do you want to improve?13:49
chengQuite amazing for a mathematician to be an invited speaker  at an application conference.13:49
@sonney2kpluskid, you wanted to start with that I remember :D13:49
pluskidbefore the end of GSoC13:50
pluskidI was mainly cleanning up the code, and writing the tutor13:50
pluskidI wish to complete the multiclass part of the tutor13:50
@sonney2kpluskid, I know you want to do SG_REF/UNREF in the future (you all are witnesses)13:50
pluskidsonney2k: hope I have time for that13:50
@sonney2kvojtech, cheng, btw can you contribute some latex for the tutorial? you also have written quite a bit of papers that could fill a few sections...13:51
chengsonney2k: Sure13:51
@sonney2kpluskid, us too.13:51
@sonney2kgsomix, mind to continue?13:51
blackburnpluskid: cheng: let us discuss road sign data just after meeting?13:52
vojtechsonney2k, yes we can write it13:52
pluskidblackburn: ok!13:52
chengblackburn: great13:52
gsomixsonney2k, mmm?13:52
gsomixI lost the thread of the discussion :(13:53
wikinggsomix: complain :)))13:53
blackburngsomix: what is left to do?13:53
blackburnwhat do you want to do and what you are happy and unhappy with?13:54
@sonney2kgsomix, you can also say what you did if you want :)13:54
@sonney2kwiking, alexlovesdata - maybe you next then?13:55
gsomixI just want to finish protocols/directors/typemaps for needed classes. And I'm happy with it.13:55
@sonney2kgsomix, to the point as usual :)13:55
wikingsonney2k: i have complained already :D13:56
wikingsonney2k: and gave suggestion about what woudl be gread in long run for shogun13:56
@sonney2kwiking, yes but what is left todo?13:56
wikingsonney2k: ah ok13:56
@sonney2kwhat will you do before release?13:56
wikingsonney2k: left to do: give you nice pictures, add a part into tutorial and now after the last 2 PRs of yesterday merge latentmodel with structuredmodel resulting in latent so svm13:57
wikingthey'll be done in reverse order :)13:57
@sonney2ksome nice examples that create figures would be optimal13:58
@sonney2kvojtech, uricamic - please continue13:58
wikingmaybe i push then tutorial the last task13:58
alexlovesdataye,we had also an idea like that13:58
alexlovesdatato create an example that yields some pics of detection finds13:58
uricamicok, todo: produce examples that creates some pictures13:59
uricamic+ contribute to the turorial13:59
@sonney2kuricamic, code is ready otherwise right?13:59
@sonney2kuricamic, btw didn't you want to do some benchmark / comparison?13:59
uricamicthe learning algorithms are ok13:59
blackburnuricamic: we need to fix multiclass14:00
blackburnit is not being initialized when using risk14:00
wikingblackburn: ?14:00
blackburnwiking: python example fails14:00
wikingok that might be my bad as well14:00
blackburnit was there before14:00
uricamicyep, I guess the problem is in the last PR from n4ndo, now when someone uses BMRM for training it then results in exception14:00
wikingah ok14:00
@sonney2kuricamic, ^14:01
blackburnuricamic: in argmax number of classes is set14:01
blackburnyou don't call argmax so number of classes is not set14:01
vojtechsonney2k, we plan to submit a paper about the method. This will require doing benchmarks.14:01
blackburnthat's the only reason14:01
uricamicyep, but the current multiclass risk function does not use argmax at all14:01
blackburnuricamic: that needs to be handled somehow14:01
@sonney2kvojtech, would be nice to have this figure / comparison in the tutorial I think14:02
uricamicI know, I wanted to discuss this with n4ndo, but haven't catch him yet14:02
vojtechsonney2k, we have some comparison - figures - already. We will put it to the tutorial14:02
uricamicalso there have been some problem with MulticlassSOLabels that valgrind complained about14:02
@sonney2kheiko, maybe you continue?14:02
@sonney2kvojtech, ok14:03
heikoyeah, ok14:03
heikoI finished everything, that is all two-sample and independence tests14:03
@sonney2kuricamic, just fix this then or talk to us after the meeting :)14:03
heikorecently did minor updates and optimisations14:03
uricamicsonney2k: sure14:03
* sonney2k worldmodel complete14:03
heikoso my project is ready from what was is the proposal14:03
heikothere are lots of things that I still like to integrate, but that will happen after shogun2.014:04
@sonney2kheiko, so what would you want to do before (if you have time...)14:04
heikothis week, I want to get grips with the unit tests14:04
heikoI wrote a lot of them, they are currently in the examples14:04
heikoalso, currently I am working on a more sophisticated cross-validation intermediate result thing14:05
heikobased on blackburns ModelSelectionOutput14:05
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has joined #shogun14:05
heiko(they guy from cambridge wanted this and I remember other poeple asking14:05
yoohi all14:05
heikoI also integrated this john plat sigmoid fitting for the labels, that has to be integrated. and there is an extension for multiclass that I wrote and want to add to shogun14:05
heikoapart from the bugfixing7warnings etc14:06
heikoone thing:14:06
@sonney2kthat is a lot already14:06
@sonney2k(if you intend to sleep)14:06
shogun-buildbot_build #361 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>14:06
heikofrom next week, I will have to take a step back since my Master's thesis deadline is on 7th Sept14:06
heikoso a few hours less per week for shogun14:06
@sonney2kI was wondering already...14:06
heikoIts already in a pretty good state14:07
heikobut writing things up takes time14:07
@sonney2kbetter focus on your thesis then14:07
@sonney2kwe don't want you to fail with that!14:07
heikoand yes, sleeping is important for me :D14:07
blackburnI am wondering what should I do in my master thesis :D14:07
blackburnI am totally lost in ML now14:07
chengblackburn: in computer science?14:07
heikoblackburn, you can jsut ask people to host you14:07
blackburncheng: yeah I want to do ML but don't know what to do at all14:07
blackburnhost me?14:08
heikoyes your thesis14:08
heikoYou can ask other people that at you uni14:08
blackburnI am unsure I understand what do you mean14:08
@sonney2kI guess blackburn you should take over14:08
blackburnI am the biggest expert in ML here14:08
shogun-buildbot_build #288 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at  blamelist: Sergey Lisitsyn <>, Viktor Gal <>14:08
gsomixsad, but true :(14:08
@sonney2kright after vladimir V.14:09
heikoHe's in London :)14:09
chengHey, if you contribute to shogun, you are by definition an ML expert.14:09
blackburnin my university rare professor know what SVM is14:09
blackburnmay be 3-4 of they do know14:09
heikoblackburn, thats why I suggested to ask other people :)14:09
blackburnheiko: other people where/14:09
heikoe.g. London/Berlin14:10
blackburnwell that's something optional for they14:10
heikoThen you got one supervisor locally and the other one somewhere else14:10
heikoanyway, lets talk about that later :)14:10
blackburnyeah ok14:10
@sonney2kblackburn, if you want to go to berlin, or anywhere it would probably not be a problem14:10
@sonney2kjust choose the city / group...14:11
@sonney2ktask etc14:11
blackburnsonney2k: I am not able to leave country yet14:11
@sonney2kblackburn, but please continue with what you have done14:11
@sonney2kwant to do :D14:11
blackburnI will just be arrested and happily go to army lol14:11
chengML for military. Very bad.14:12
blackburncheng: that's mandatory here but not when I am studying here14:12
@sonney2kseems like blackburn is shy today14:12
blackburnsonney2k: shy?14:12
blackburnwhy shy?14:12
blackburnprobably I interrupted the process :D14:13
@sonney2kblackburn, anything you want to do before shogun 2.0 / sth still on todo?14:13
@sonney2kwhat you want to improve in your (other) code?14:13
blackburnyes, I am going to extend liblinear with MTL graph setting for regression problems14:13
blackburnthen I wanted to unify its API14:13
blackburnand extend existing SLEP solver with overlapping group lasso14:14
@sonney2kvojtech, before I forget - labels in multiclass ocas are 0...<nr_classes-1> ?14:14
blackburnmay be some examples to do too14:14
@sonney2kblackburn, pictures :D14:14
blackburnI got no idea about pictures14:15
chengcrazy idea: would shogun work on a phone or tablet?14:15
@sonney2kblackburn, well discuss with chris - you can probably show how a decision surface is changed when multi/dual task stuff is used14:16
blackburncheng: probably that's possible, but what to do with it?14:16
@sonney2kcheng, sure it runs on gunnars iphone for years14:16
chengyou can go around classifying stuff with the phone camera.14:16
@sonney2k(we even tweaked configure for that)14:16
blackburnoh well it is rather opencv example14:16
alexlovesdata@cheng: cloudi nterface for shogun :P14:16
-!- emrecelikten [] has joined #shogun14:16
blackburnbut actually I know how to implement that14:16
vojtechsonney2k, labels in libocas: 1,2,3...nY14:16
wikingcheng: it works with raspebrry pi14:17
wikingcheng: *raspberry14:17
wikingcheng: alas it should be possible to use it on iphone14:17
@sonney2kwiking, hmmh then I guess we need to label[i]+1 ...14:17
wikingsonney2k: but that way i get really all the labels shifted :)14:17
@sonney2kok then does anyone have any final comments / questions?14:17
chengI think it has been a great GSoC.14:18
wikingsonney2k: the labels are shifted by +114:18
chengVery lively group of students this year.14:18
pluskidhow time flies14:18
@sonney2kFinally, once GSoC is over real live will usually require much more time - so IRC might be less helpful than postings on the mailinglist. I think it would be nice if we authors would post using some official email address. So everyone here who intends to stick around should get this forward.14:19
@sonney2kSo final words - thanks all for your great efforts. It has been fun (at least for me and I hope for you too). We hope you will stick around - please keep in touch in one way or another...14:20
blackburnfor ones who are interested in some careers it is useful for sure14:20
blackburnI got contacted by google twice because of shogun :D14:20
chengAlso, spread the word about shogun to your colleagues. We want more users and also more potential developers.14:21
* pluskid is hesitating to update his resume14:21
wikingblackburn: :>14:21
@sonney2kplease stay subscribed to the mailinglist - you will notice that at some point people start to use your code :D14:22
@sonney2kbut usually only when they get stock14:22
@sonney2kanyways, meeting is over from my side!14:22
@sonney2kyou can stay here and keep on partying14:22
heikonice one :)14:22
alexlovesdatathanks for the great organization effort Soeren!14:23
@sonney2kblackburn has some vodka for all of you ready14:23
@sonney2kvirtual only though14:23
@sonney2kbut hopefully next year we can organise some shogun mini conf14:23
@sonney2k(in berlin...)14:23
emreceliktenApparently I came in just in time14:23
@sonney2kheiko, that reminds me - any news on the verein issue - otherwise we loose 50% of the money...14:23
gsomixsonney2k, thanks14:24
@sonney2kthat would just mean less travel support...14:24
heikosonney2k, argh, sorry, I will check that out now14:24
@sonney2k(next year)14:24
gsomixmay the Forc^W Carol Smith be with you14:24
heikosonney2k, I think a meeting would be very cool!14:24
wikingheiko: i think i've got it :)14:24
@sonney2kpluskid, find out if you have vacations at MIT or if there is a conference in germany you need to attend!14:25
heikowiking, great, send something14:25
wikingheiko: i'll just double-check if it works smoothly on the ubuntu14:25
wikingheiko: as now i've tested on my machine... it works w/o the gmock-config script14:25
pluskidsonney2k: :D I know I can get free lunch from you14:25
alexlovesdatahey blackburn: a world fusion approach for your thesis :
blackburnpluskid: cheng: mind to join some channel to discuss? it could be some flood like to discuss here14:26
@sonney2kalright - I have to get kids from kindergarden but just ignore me - keep on partying14:26
pluskidblackburn: cool, how to set up a channel?14:26
gsomixtime to work14:26
blackburnjust join #shogun214:26
blackburnalexlovesdata: what to do with it?14:26
alexlovesdataread it :)14:27
alexlovesdatait is sufficiently unusual to be YOUR thesis14:27
alexlovesdatain the discriminative part you can do everything14:27
blackburnalexlovesdata: you mean just copy that and treat it as my thesis? :D14:27
heikoI also gotta go, wiking, I will check back on the testing later this evening14:28
blackburnsounds like russian science14:28
alexlovesdataaehmmm, we are NOT in social sciences, blackburn!14:28
-!- heiko [] has left #shogun []14:29
blackburnalexlovesdata: I am unsure what does it mean14:30
alexlovesdatain principle they interpret bayes rule as an optimization problem14:33
alexlovesdatathen they can add an additional penalty on the space of probabilistic models14:33
alexlovesdatathey are doing something like probabilistic mixture of local svms14:33
alexlovesdatatheir idea is fun14:34
alexlovesdataa mixture of both worlds ;)14:34
wikingwtf :)))))14:34
alexlovesdataI need to read it in detail ,too14:34
wikingwhy no ?!14:34
wikingon ubuntu14:34
alexlovesdatawould that be the next GSoC 2013 project? :D14:34
blackburnalexlovesdata: okay I will check that14:34
wikingthere's libgtest-dev with header files but no actual libgtest14:35
alexlovesdatablackburn: are you satill pursuing ILSRVC14:35
blackburnalexlovesdata: wiking does14:35
blackburnalexlovesdata: I am waiting for thing I can help with14:35
wikingbut now i'm truing this shitfuck:)))14:35
wikingwhy no gtest library on ubuntu14:35
blackburnalexlovesdata: I have ~0 clue about that kind of recognition :)14:36
wikingsomething is wrong with ubuntu :)14:39
-!- alexlovesdata [] has left #shogun []14:40
-!- nicococo [] has left #shogun []14:40
wikingahhaha fuckers14:41
wikingsomebody fucked up ubuntu release14:41
wikingthey left out libgtest014:41
wikingnatty is the last release that still has it14:41
wikingafter that they only have libgtest-dev14:41
wikingthat only contains the headers :D14:41
wikingwtf :D14:41
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]14:49
wikingblackburn noooo14:49
wikingwho's gonna merge noooow! :)14:49
-!- cheng [] has quit [Quit: Leaving.]14:51
-!- pluskid [72f6b3c3@gateway/web/freenode/ip.] has quit []14:52
-!- pluskid [72f6b3c3@gateway/web/freenode/ip.] has joined #shogun14:53
gsomixsonney2k, around?15:03
shogun-buildbot_build #362 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>, Viktor Gal <>15:04
shogun-buildbot_build #289 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at  blamelist: Sergey Lisitsyn <>, Viktor Gal <>15:06
pluskid:( completely no idea what nice figure to create15:09
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has quit [Quit: Page closed]15:28
-!- pluskid [72f6b3c3@gateway/web/freenode/ip.] has quit [Quit: Page closed]16:03
-!- vojtech [9320543b@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:15
-!- zxtx [~zv@] has joined #shogun17:27
-!- uricamic [~uricamic@2001:718:2:1634:876:e01f:ab62:a2a] has quit [Quit: Leaving.]17:37
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has joined #shogun17:54
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has quit [Client Quit]17:54
@sonney2kgsomix, yes?18:00
gsomixsonney2k, nevermind. :)18:00
CIA-21shogun: Viktor Gal master * rae2ea5c / src/configure : Detect gmock framework without gmock-config -
CIA-21shogun: Heiko Strathmann master * reecb090 / src/configure : Merge pull request #721 from vigsterkr/utest -
wikingi hope heiko will manage to get this work18:25
CIA-21shogun: Michal Uricar master * r4f930ca / (12 files in 2 dirs): BMRM training exception fixed + several warning fixes -
CIA-21shogun: Heiko Strathmann master * rf28f0ee / (12 files in 2 dirs): Merge pull request #723 from uricamic/BM_SOL_EXAMPLE -
-!- blackburn [~blackburn@] has joined #shogun18:39
wikingblackburn: got 10 mins for testing?18:53
blackburnwiking: yeah18:53
blackburntesting of what?18:53
wikingblackburn: u have the latest ubuntu right?18:53
wikingok just a sec to cehck out something18:54
wikingso can u download compile& install this18:54
wikingas u won't have package of it18:54
blackburnokay let me try18:55
wikingah before18:55
wikingdo this18:55
wikingsudo apt-get install google-mock18:55
wikingand once ready check out the HEAD of shogun/master18:56
wikingand try to run ./configure script18:56
wikingand give --libs=<where u've installed libgtest>18:56
wikingand paste here what's the output of ./configure script for google c++ mocking framework18:57
wikingshould work ;)18:57
shogun-buildbot_build #363 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Viktor Gal <>18:57
wikingi don't know what's with serialization18:58
blackburnecho "'make install' is dangerous and not supported18:58
wikingbut bmrm should work in the next build18:58
wikingdue to uricamic's fixes18:58
shogun-buildbot_build #290 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at  blamelist: Heiko Strathmann <>, Michal Uricar <>, Viktor Gal <>19:00
wikingblackburn: ?!19:00
wikingblackburn: not supported? :))) you cannot install it under /usr/local/ ?19:00
blackburnwiking: so how can I install gtest to some default dir19:01
wikinggood question19:01
wikingi'm trying on some other machine now :)19:02
wikingwell this is rather a huge hack19:03
wikingbut fuck it i mean i don't know what were actually ubuntu people thinking to supply a -dev package of a library w/o the actual library19:03
wikingsudo cp lib/.libs/* /usr/local/lib19:04
wikingsudo cp lib/.libs/* /usr/loca/lib19:04
wikingthat should do the trick19:04
wikingand then: in shogun/src19:05
wiking./configure --libs=/usr/local/lib19:05
blackburn 346 /usr/lib/gcc/x86_64-linux-gnu/4.6/../../../../lib/ undefined reference to `pthread_key_create'19:10
blackburn 347 /usr/lib/gcc/x86_64-linux-gnu/4.6/../../../../lib/ undefined reference to `pthread_getspecific'19:10
blackburn 348 /usr/lib/gcc/x86_64-linux-gnu/4.6/../../../../lib/ undefined reference to `pthread_key_delete'19:10
blackburn 349 /usr/lib/gcc/x86_64-linux-gnu/4.6/../../../../lib/ undefined reference to `pthread_setspecific'19:10
blackburnwiking: should I compile it w/o thread support or so?19:10
wikingooooh motherfucker :))19:10
wikingcan u paste me the part of configure.log about gmock19:11
blackburnwiking: ^ that's it19:11
wikingthe whole19:12
wikingwith the cc line an eerything19:12
wikingdo u still have the ./configure-229-20517.cpp file?19:13
wikingok just a test19:14
wikingin ./configure script19:14
wikingGMOCK_LDADD="-lgmock -lgtest"19:14
wikingreplace it with19:14
wikingGMOCK_LDADD="-lgmock -lgtest -lpthread"19:14
wikingbut i think you'll have to add still one thing19:15
blackburnadd to?19:15
wikingGMOCK_LDADD="-lgmock -lgtest -lpthread -D_THREAD_SAFE"19:15
blackburnwell it is detected now19:15
wikingso do it like that19:16
wikingtry to compile the whole libshogun19:16
wikingor just type: make unit-tests19:16
wikingshould compile libshogun19:16
wikingand then run the unit tests19:16
wiking../shogun/lib/versionstring.h:3:27: note: expanded from macro 'VERSION_REVISION'19:17
wiking#define VERSION_REVISION 0x19:17
wikingwtf :D19:17
wikingor still compiling :)19:20
CIA-21shogun: Viktor Gal master * r6a96d7e / src/configure : Fix GMOCK_LDADD flags -
CIA-21shogun: Sergey Lisitsyn master * rbfd1965 / src/configure : Merge pull request #724 from vigsterkr/latent -
wikinglet'shope it works now :))19:24
wikingok new PR for fixing some doxygen warnings19:28
CIA-21shogun: Viktor Gal master * rb7037d3 / (3 files): Fix doxygen comments in structure -
CIA-21shogun: Sergey Lisitsyn master * r271aeed / (3 files): Merge pull request #725 from vigsterkr/so -
wikingno unit tests yet? :)19:29
blackburnI forgot to19:30
blackburncompile only libshogun19:30
blackburnlast interface to go19:31
shogun-buildbot_build #364 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Heiko Strathmann <>, Michal Uricar <>19:35
wikingagain? :)19:35
blackburnwiking: I was thinking about static analysis of code19:51
wikingblackburn: so?19:51
blackburnhave an opinion on that?19:51
wikingor what's happening :)19:51
blackburnI mean we could use static analysis to track some bugs19:51
blackburntreehydra would work19:52
blackburnor sparse19:52
blackburndo you have any experience in that?19:52
wikingmm not really19:53
wikinghow's your unit testing?19:53
blackburnwill try in a minute19:53
wikingthere's one coming with clang19:54
wikingpretty cool19:54
wikingi've tried it couple of times19:54
blackburnSGVectorTest.misc failed19:56
wikingoh cool why :)19:57
blackburnlib/ Failure19:58
blackburnValue of: b[i]+1.3*b[i]19:58
blackburn  Actual: -1931.7419:58
blackburnExpected: d[i]19:58
blackburnWhich is: -1931.7419:58
blackburnlib/ Failure19:58
blackburnValue of: b[i]+1.3*b[i]19:58
blackburn  Actual: -819.94719:58
blackburnExpected: d[i]19:58
blackburnWhich is: -819.94719:58
wikingmmm that is funny19:58
wikingsince they seemed to be the same value19:58
blackburnprecision issue?19:58
wikingall the others passed?19:58
wikingcan u run it again?19:59
wikingwhat what what20:00
blackburnworked once20:00
blackburncomputing probability20:00
wikingok then there's some precisiou problem20:00
wikingi'm just running that checker stuff on shogun20:01
wikingit has some funny reports ;)20:01
blackburnwiking: clang one?20:01
wikingi'll put the report somewhere20:01
blackburnyeah interesting to see20:01
wikingjust let it finish20:01
wikingcompilation is much slower of course20:01
wiking../shogun/mathematics/Math.h:372:19: warning: Array access (from variable 'xy') results in a null pointer dereference area += 0.5*(xy[2*i]-xy[2*(i-1)])*(xy[2*i+1]+xy[2*(i-1)+1]);20:02
blackburnI know the method probably20:02
blackburncan't agree with analyzer20:03
wikingheheh funny ones20:06
wikingkernel/Kernel.cpp:530:34: warning: Division by zero20:06
wiking                (int32_t)(kernel_cache.buffsize/kernel_cache.activenum);20:06
wiking                          ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~20:06
wikingkernel/Kernel.cpp:862:27: warning: Array access (via field 'vector') results in a null pointer dereference combined_kernel_weight = weights.vector[0] ;20:06
shogun-buildbot_build #365 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>, Viktor Gal <>20:11
blackburnokay this 530 one is not a bug20:11
blackburnbut good it points it20:11
wikingblackburn: is there a check for kernel_cache.activenum != 0?20:12
shogun-buildbot_build #291 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at  blamelist: Sergey Lisitsyn <>, Viktor Gal <>20:13
blackburnwiking: it just should not happend but no, no check20:13
wikingblackburn: well then at least an ASSERT()20:14
wikingif it should not happen20:14
blackburnthat's a good thing I should say20:14
-!- Marina_ [5ce7df29@gateway/web/freenode/ip.] has joined #shogun20:14
wikingi have the report20:14
blackburnlet me see full report of that thing20:14
wikinglet me check how i can share it20:14
blackburnI will put a few asserts20:14
Marina_hello, can anyone help me by the arts application? i get the following error when by running
Marina_Usage: [options] seq.fa error: incorrect number of arguments20:15
blackburnMarina_: how do you execute it?20:16
blackburnexact command line string I mean20:16
Marina_python arts.py20:17
blackburnyou should provide some sequence files for that20:17
blackburnwiking: cool thanks20:18
wikingblackburn: have fun :) should i post it on mailinglist?20:18
blackburnMarina_: e.g. ./ data/ARTS.dat.bz220:18
blackburnoh it fails in current build :D20:19
Marina_ah ok, i try20:19
Marina_hist[255]=4763 get_num_bits_in_histogram()=8 > get_num_bits()=2 [ERROR] ALPHABET too small to contain all symbols in histogram Traceback (most recent call last):   File "", line 64, in <module>     preds = arts.predict(seq)   File "/home/mvidovic/POIM/ARTS/", line 199, in predict     f = self.sensors[i].get_test_features(chunk, self.window)   File "/home/mvidovic/POIM/ARTS/", line 104, in get_te20:21
Marina_i got this error20:21
Marina_can you help me?20:21
blackburnwill try to20:21
wikingblackburn: heheh nice memory leak reports :)20:22
wikingblackburn: there's on in slep20:22
wikingmmm i guess it would be a good idea to have this automatically generated :)20:24
wikingmmm should we set it up somewhere ?:)20:27
wikingi mean of course it depends as well which static analyser we want to use20:28
wikingi used this one because i had it at hand20:28
blackburnMarina_: okay I recall I tried to fix that but do not remember the root cause20:28
blackburnI will check now and try to fix20:28
Marina_thank you!!20:31
Marina_the arts programm will show me the important transcription starts, or?20:32
blackburnMarina_: I'd like to help you if I knew anything about transcription starts or anything like that :D20:33
blackburnit detects something and it fails on creating string features20:33
blackburnthat's what I know20:33
Marina_thats super - thats enough ;-)20:36
wikingblackburn: so nullpointer dereferencing fixes are coming up from u ?:)20:40
blackburnyes everything from me20:40
wikingblackburn: i'll fix the so part20:40
blackburnah ok20:40
wikingthe memleaks20:40
wikingblackburn: what should be with the Undefined allocation of 0 bytes (CERT MEM04-C; CWE-131) errors? :)20:47
blackburnwiking: no idea what it is20:47
wikingwell that it can happen that u call malloc with 0 bytes20:48
blackburnhmm just assert it then20:48
wikingok i think i'll push this mem leak fix20:53
wikingbut i need uricamic to comment on it20:53
wikingso don't merge it20:53
blackburnMarina_: hah okay it was my bad20:54
blackburnit works ok - I just said wrong thing20:54
blackburnMarina_: you need to put a fasta file after ./arts.py20:54
Marina_which should i use?20:55
blackburnwell the one you want to detect (something) in20:56
blackburnafaik by default it uses trained model20:56
blackburnfrom that file20:56
Marina_but thats the file we used -- data/ARTS.dat.bz220:57
blackburnno, it is a model and it is loaded by default20:57
blackburnyou need to specify fasta file containing sequencies you want to work with20:57
blackburnit works with single sequence20:58
blackburnokay updated20:58
Marina_is there any example for such a file -- how the input have to be`?20:59
blackburnMarina_: for example you need to detect what it detects in some sequence20:59
blackburnjust sequences20:59
blackburnin line20:59
blackburnsequence* sorry21:00
blackburnso just create some file file.fa and put some sequence21:00
blackburnand ./ file.fa will do something with trained model21:00
Marina_i try --- thanks!!21:00
blackburnI should learn about bioinf more :D21:01
blackburnhowever it is kind of legacy for us now21:01
wikingthis is really heavy weaight21:02
wikingi think i've found the bug why the multiclass results were not so good :DDD21:03
wikingcheck that link21:03
wikingand you'll see21:03
wikingSO multiclass i mean21:03
wikingblackburn: got it? :)21:04
blackburnif score > max_score21:05
blackburnscore = max_score21:05
blackburnit provides nice output21:05
blackburnwe must have it on buildbot21:05
wikingblackburn: and then on the end21:06
wikingmax_score is used21:06
blackburnyeah let me fix it21:06
wikingi'll fix it :)21:06
wikingi'm in structure :D21:06
blackburnah ok21:06
wikingthis shit is good21:06
wikingwe need this shit running automatically21:06
Marina_mhh, i have put 2 sequences into testfile.fa :  ACCTTTTTGAAAA    and TTTTATACCATAT    and there comes the error21:07
wikingn4nd0 where are you!?21:07
Marina_loading model file......................done number of chunks for contig: 1 processing chunk #0 [ERROR] assertion max_string_length>=window_size || (single_string && length_of_single_string>=window_size) failed in file features/StringFeatures.cpp line 1197 Traceback (most recent call last):   File "", line 64, in <module>     preds = arts.predict(seq)   File "/home/mvidovic/POIM/ARTS/", line 199, in predict    21:07
blackburnMarina_: it should be bigger21:07
blackburnMarina_: your string has lenght less than window size21:08
Marina_oh :-)21:08
blackburnlet me find the window size21:08
wikingblackburn: i've requested uricamic to comment on that PR... let's hope he agrees21:08
wikingblackburn: i wonder if ASSERT will remove these nullpointer dereferencing errors21:13
blackburnwell it should21:13
blackburnwiking: one bad thing about require macro - it has to be in { } with if21:18
blackburnin other case it will be expanded to if if21:18
blackburnif () REQUIRE() else ... will produce if () if () else21:19
wikingman this bug is crazy  :))21:21
wikingthis with max_score :)21:21
wikingi'll have to see the difrerence21:22
blackburnMarina_: oops sorry21:23
blackburnformat is21:23
blackburn> anything21:24
blackburnand then sequence21:24
blackburnso fasta with single file21:24
Marina_? i dont understand21:24
Marina_can you spot the right window_size?21:25
blackburnit is 14021:25
blackburnwith default model21:25
Marina_ah thanks21:25
Marina_and what do you mean with :21:25
Marina_anything [21:23] <blackburn> and then sequence21:25
wikingshit with the fucking mosek :DDD21:26
blackburn> 0 label=121:26
blackburnMarina_: ^21:26
blackburnbut sequence should have length greater than 14021:26
blackburnwill not work with tweets :D21:26
blackburnwiking: great idea to tweet something like that ^, right? :D21:29
blackburnwiking: ACTGCCATTACGCA!21:30
Marina_now i get an assertion error:  loading model file......................done Traceback (most recent call last):   File "", line 64, in <module>     preds = arts.predict(seq)   File "/home/mvidovic/POIM/ARTS/", line 180, in predict     assert(num_chunks > 0) AssertionError21:30
blackburnMarina_: okay that's a different thing now :D21:30
wikingneed a new license21:32
wikingSO-SVM: 90.3021:36
wikingMC:     50.0021:36
Marina_whats wrong now? what can i do?21:36
wikingi think n4nd0 will be happy21:36
wikinglet's see with the old code21:36
blackburnMarina_: nothing yet21:37
wikingcomeoooooon COMPILE BITCH!21:38
-!- gsomix_ [~gsomix@] has joined #shogun21:38
blackburnMarina_: I am checking what can I do :)21:38
Marina_thanks blackburn21:39
wikingSO-SVM: 36.2021:41
wikingMC:     46.8021:41
wikingn4nd0 owes me a beer21:41
-!- gsomix [~gsomix@] has quit [Ping timeout: 272 seconds]21:42
wikingi fucked it up21:42
wikingwith the PRs21:42
wikingok i'm closing 72621:42
blackburnwhat did you break?21:45
wikingnothing.. just added the 2 separate commits into one pr21:45
wikingfixing it21:46
blackburnMarina_: okay it worked with sequence of length ~2500 and small patch21:49
wikingblackburn: so great idea for the static analyzer ;)21:51
CIA-21shogun: Viktor Gal master * r2bd9905 / src/shogun/structure/MulticlassModel.cpp : Fix argmax function in MulticlassModel -
CIA-21shogun: Sergey Lisitsyn master * r7eb8b72 / src/shogun/structure/MulticlassModel.cpp : Merge pull request #728 from vigsterkr/latent -
Marina_blackburn: what do you mean by small patch?21:51
blackburnMarina_: I will commit it in a sec21:51
blackburnMarina_: here it goes, please update git21:52
CIA-21shogun: Sergey Lisitsyn master * r4c4d431 / applications/arts/ : Added probably missed kernel setting in arts -
-!- puffin444 [6c646990@gateway/web/freenode/ip.] has joined #shogun21:55
puffin444hey blackburn, hows it going?21:56
blackburnpuffin444: hey puffin44421:56
blackburnwhoa that was recursive21:57
blackburnpuffin444: I wanted to ask a few thing21:57
puffin444lol how was the meeting?21:57
puffin444oh sure21:57
blackburnwell we were having vodka and dance a little21:58
blackburnI completely missed your gradient evaluation stuff21:58
blackburnis that the thing that learns a gp on top of results?21:58
blackburnor is that model selection for gps?21:58
puffin444what do you need concerning gtadient evaluation?21:59
blackburnwhat is the method21:59
puffin444they are the same21:59
Marina_blackburn: the assertion error is still there22:00
blackburncan that work for svm?22:00
blackburnMarina_: what is length of your sequence?22:00
puffin444assertion error?22:00
blackburnthat's pretty strange, okay22:01
blackburnMarina_: how your input file look like? can you paste it via
blackburnpuffin444: so can that work with general machine?22:02
blackburnor is that the thing to select parameters for gp regression machine?22:02
puffin444the gradient framework should be applicable to anything with a differentiable function class22:03
blackburnany example of that?22:03
Marina_ahhhh it works22:03
puffin444if you encounter problems with it, please let me know22:03
blackburnnice to hear22:03
blackburnpuffin444: can I learn svm's C with that?22:04
puffin444I have not tried it with something other than a gp22:04
Marina_thank you very much blackburn22:04
blackburnthere is no gradient for sure, but what are other examples of differentiable evaluations?22:04
blackburnMarina_: no problem, let me know if there are more problems with it22:05
puffin444but it should work in theory as long as it can work with the interface22:05
blackburnpuffin444: okay other issue22:05
puffin444no examples yet except for inference method class22:05
blackburnthere was a class with eigen members22:06
blackburndid you fix that already?22:06
puffin444yes that should be fitcinference method22:07
-!- n4nd0 [] has joined #shogun22:07
blackburnn4nd0: he22:07
blackburnpuffin444: can we avoid it somehow?22:07
n4nd0blackburn: hey!22:08
n4nd0wiking: thank you for the bug22:08
wikingn4nd0: heavy shit :D22:08
puffin444errors? warnings? I found no errors on my machine, but there are warnings that appear to come from an incorrect config on my box22:08
n4nd0BTW there is another bug I found this morning22:08
n4nd0wiking: :D22:08
wikingn4nd0: but now it performs really good :))))22:08
puffin444no more eigen?22:08
wikingn4nd0: but never would have found it w/o blackburn's idea to do a static analysis22:08
blackburnpuffin444: I mean if we have eigen in headers it has to be in included path22:08
n4nd0wiking, blackburn: then I own you both a beer :)22:09
n4nd0blackburn: it can be vodka too :P22:09
blackburnpuffin444:  if it is in .cpp - it is not a dependency22:09
blackburnsee what I mean?22:09
puffin444yes. dont I use just the .h file from shogun though? do I directly reference eigen anywhere?22:10
blackburnpuffin444: I am not sure I got what you mean22:10
puffin444oh sry you mean in the cpp files22:10
blackburnpuffin444: easiest solution I have in mind is to use Map from eigen22:11
blackburndo you know how to use it?22:11
puffin444not sure22:11
blackburnpuffin444: for example Map<VectorXd> somevector(pointer); will produce something that acts just like VectorXd22:12
puffin444so should I just reference the cmap header if I want to use eigen?22:13
puffin444oh ok22:13
blackburnpuffin444: this way you can map SGMatrices to eigen ones22:13
puffin444oh I see so that I can just use one set instead of having both eigen and shogun matrices22:14
blackburnpuffin444: we have to avoid eigen members because in other way we have to add eigen include path everywhere - in tests/examples and even add a dependency in .deb package22:14
blackburnyeah just map it with Map<>22:15
puffin444I go ahead and work on this. by the way I have a tuorial underway22:15
blackburnpuffin444: here comes the example -
puffin444sorry if I was absent the last few days. I had to pack up and I am currently on route to Pittsburgh22:16
blackburnI see22:16
puffin444where should I put the tutorial under shogun-tutorial?22:18
-!- Marina_ [5ce7df29@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:18
blackburnpuffin444:  well did you check available structure?22:19
CIA-21shogun: Evgeniy Andreev master * rd3f9d75 / (6 files in 3 dirs): added design of zero-copy methods -
CIA-21shogun: Sergey Lisitsyn master * rc6792c3 / (6 files in 3 dirs): Merge pull request #722 from gsomix/buffer_protocol -
puffin444yes I think it should go under algorithms but  not sure22:19
blackburnyeah put it to algorithms22:20
blackburnanyway changing structure is so much easier task than writing a book :D22:20
wikingn4nd0: check on the report that i've sent to the mailing list.. i think there won't be any more such big bugs, but would be nice to eliminate those reports before the 2.0 release22:21
puffin444oh one last thing- any issues with the Laplacian PR?22:21
shogun-buildbot_build #366 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Viktor Gal <>22:22
blackburnpuffin444: I've waited for other commits but if it can go - please rebase and I will merge22:22
n4nd0wiking: I have read this static analysis, although I have not gone through the report. I'll do that22:22
n4nd0good idea btw!22:22
wikingn4nd0: yeah i think w/o this we would be debugging MulticlassModel for a while to find out why is it so bad :)22:23
puffin444I will do asap. I am currently on route to Pittsburgh so it may take some time before I can rebase22:23
n4nd0wiking: hehe yes22:24
blackburnpuffin444: no reason to hurry :)22:24
shogun-buildbot_build #292 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at  blamelist: Sergey Lisitsyn <>, Viktor Gal <>22:24
blackburnpuffin444: did they put your massive notebook into a special car?22:24
blackburnoh gosh22:24
puffin444lol no, but the gas mileage is pretty bad :)22:25
-!- zxtx [~zv@] has quit [Ping timeout: 240 seconds]22:25
blackburnare you driving a car just now? :D22:25
blackburnthis can be dangerous unless you have well-trained structured output SVM22:26
puffin444goodness no lol thankfully my father is helping me move22:26
blackburnsend him a hello from soviet russia :D22:27
puffin444I will :).22:27
CIA-21shogun: Sergey Lisitsyn master * r11b38ab / examples/undocumented/cmdline_static/ : Fixed KRR example -
puffin444ok Blackburn I have to go. I hope to come back and work on these things asap.22:29
blackburnpuffin444: no need for asap :)22:30
-!- puffin444 [6c646990@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:30
CIA-21shogun: Sergey Lisitsyn master * rc727079 / data : Updated data submodule -
wikinggit submodule update                                                                                                                  [wiking@welitron:~/shogun/]22:33
wikingfatal: reference is not a tree: 874abd8a2269380c6e15c4bdadda697a8b81c98722:33
wikingUnable to checkout '874abd8a2269380c6e15c4bdadda697a8b81c987' in submodule path 'data'22:33
blackburnwiking: are your data up to date?22:34
wikingblackburn: that's what i'm trying to do22:34
wikingupdate it22:34
blackburnwhat if you go to the folder and update?22:35
wikingit's not working like that22:35
wikingit's not an actuall branch22:35
wikingit's just a reference to a given commit22:35
wikingi mean git submodule should not work like that22:35
blackburnI don't know then22:35
wikingit's not as if you have a given branch checked out into a subdir...22:35
wikingwho wrote the mosek detection in mosek?22:37
wikingi mean in configure script?22:37
blackburnI did IIRC22:37
blackburnfeel free to improve :D22:37
n4nd0wiking: anyway, let me ask you - what example do you run when you test the MulticlassModel?22:37
wikingblackburn: did it work for you?22:38
wikingn4nd0: yours22:38
wikingn4nd0: so_multiclass22:38
blackburnwould it be a surprise if I said I didn't test it?22:38
wikingblackburn: no :)22:38
wikingblackburn: as it doesn't work for me22:38
wikingi knew how to patch for myself22:38
wikingn4nd0: yeps22:39
n4nd0wiking: I find it strange that you don't find a seg fault (actually it says glibc detected) when running it22:39
wikingn4nd0: lol?!22:39
wikingn4nd0: it ran smoothly for me22:39
n4nd0wiking: aham22:39
n4nd0there is a bug in any case :)22:39
wikingn4nd0: although it at shitload amount of memory22:39
n4nd0wiking: what?22:40
n4nd0oh sorry, ok22:40
wikingso it used like 1.2G ram22:40
wikingor so22:40
n4nd0yes, I am trying to fix now the mem leaks22:40
n4nd0there are some still here22:40
wikingheheh go for it22:40
wikingblackburn: so which static analyzer then?22:40
blackburnwiking: no idea, I didn't try others22:41
n4nd0wiking: but you can check the bug I am telling you, Mosek.cpp:24122:41
wikingn4nd0: lemme see22:41
CIA-21shogun: Sergey Lisitsyn openmp * r74a774c / : Merge remote-tracking branch 'origin/openmp' into openmp (+530 more commits...) -
n4nd0wiking: when A.num_rows = 0 (the constraints matrix A is empty) that is not valid22:41
n4nd0ptr[A.num_rows-1] ...22:42
n4nd0I am taking care of this so don't bother fixing it right now22:42
wikingnot touching anything :)22:42
wikingnot even the lousy mosek detection in ./configure script22:43
n4nd0sorry about that :(22:44
wikingnw :)22:44
n4nd0didn't do anything for auto-detect mosek with configure22:44
blackburnno mosek no cry22:45
shogun-buildbot_build #367 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>22:49
shogun-buildbot_build #356 of deb1 - libshogun is complete: Failure [failed git]  Build details are at  blamelist: Evgeniy Andreev <>, Sergey Lisitsyn <>22:51
blackburnI do not understand what happens with data22:52
-!- gsomix_ is now known as gsomix22:53
wikingblackburn: it's ok now23:02
wikingi mean the data...23:03
-!- nonoelph12345 [] has joined #shogun23:20
n4nd0leaks out in so_multiclass :)23:23
CIA-21shogun: Sergey Lisitsyn master * r89a4836 / (2 files in 2 dirs): Fixed serialization issue -
shogun-buildbot_build #357 of deb1 - libshogun is complete: Success [build successful]  Build details are at
-!- n4nd0 [] has quit [Ping timeout: 246 seconds]23:31
-!- n4nd0 [] has joined #shogun23:33
CIA-21shogun: Evgeniy Andreev master * ra9e8ca1 / src/interfaces/python_modular/DenseFeatures_protocols.i : removed code that requires directors -
CIA-21shogun: Sergey Lisitsyn master * r40f88f7 / src/interfaces/python_modular/DenseFeatures_protocols.i : Merge pull request #729 from gsomix/buffer_protocol -
shogun-buildbot_build #293 of deb2 - static_interfaces is complete: Success [build successful]  Build details are at
-!- n4nd0 [] has quit [Client Quit]23:37
shogun-buildbot_build #368 of deb3 - modular_interfaces is complete: Failure [failed compile python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>23:40
shogun-buildbot_build #369 of deb3 - modular_interfaces is complete: Failure [failed compile ruby_modular]  Build details are at  blamelist: Evgeniy Andreev <>, Sergey Lisitsyn <>23:54
--- Log closed Fri Aug 17 00:00:17 2012