Open in new window / Try shogun cloud
--- Log opened Fri Mar 29 00:00:17 2013
@sonney2kit is 24:00 :)00:00
@sonney2kthought my phone crashed when it displayed 00:0000:01
@sonney2kblackburn, how shall we do it about the proposal?00:04
blackburnsonney2k: do whatever you want to do with it00:04
@sonney2kshall I directly edit the proposal gdoc document?00:04
@sonney2kand you fill in the stuff tomorrow?00:04
blackburnsonney2k: if you want to do it not solely - yes00:04
@sonney2knot sure what you mean00:05
@sonney2kIt would be best to have another iteration00:05
blackburnsonney2k: if you want help from fernando or heiko side00:05
blackburnyou'd have to put it in gdocs00:05
blackburnas they can't see the real application00:05
blackburnsonney2k: sent a mail00:11
@sonney2kblackburn, I did too :)00:13
blackburnsonney2k: what for?00:14
blackburnokay whatever I am off for today00:15
n4nd0blackburn: ook, good night! and good job :)00:17
blackburnn4nd0: good night00:17
-!- zhuoran [80eefe5d@gateway/web/freenode/ip.] has joined #shogun00:32
zhuoranHi, is anyone online?00:32
n4nd0zhuoran: Yep00:32
zhuoranI am a newbie here, trying to do a program solving a graph problem00:32
n4nd0zhuoran: It sounds fun00:33
zhuoranIs there any general common graph class or structure in shogun?00:33
n4nd0zhuoran: not explicitly00:33
n4nd0but you could use SGMatrix for dense representation or also something for sparse00:34
zhuoranOr I should write my own adjacency list in program00:34
n4nd0zhuoran: as you wish :)00:35
zhuoranThat seems what I want.00:35
zhuoranLet me have a glance.00:35
n4nd0zhuoran: You are welcome00:35
n4nd0zhuoran: What are you working on btw?00:35
zhuoranIssue 87300:39
zhuoranI have already wrote my own program to solve this problem.00:40
n4nd0aham cool!00:40
zhuoranHowever, I really not familiar with SHOGUN. I am trying to merge it in.00:40
n4nd0zhuoran: Ok, just ask away stuck00:41
n4nd0if* stuck00:41
zhuoranThanks :D00:41
@sonney2kn4nd0, fighting w/ the application00:46
n4nd0sonney2k: do you think it needs that much improvement??00:46
@sonney2ksome parts yes00:46
n4nd0I thought it was ok..00:46
n4nd0mm I see00:46
n4nd0I am going to open the gdocs00:47
n4nd0sonney2k: are you modifying it there?00:47
@sonney2kfor example first part was asking about the organizaton00:47
@sonney2kn4nd0, yes00:47
@sonney2kbut we told only about shogun00:47
@sonney2knot so much about us00:47
n4nd0I see a much longer description now there00:48
@sonney2kbut it is not much more work00:48
@sonney2kyes mostly from previous year with adjustments00:48
@sonney2kwhat you are marking is what I am working on currenlty00:48
n4nd0hehe ok :)00:48
n4nd0I tend to highlight things as I read them00:49
@sonney2kwhat also will need improvemnt is 6)00:49
@sonney2kwe have to explicitly say why we choose which mentor00:49
@sonney2kdoing that will leave no room for confusion00:49
n4nd0I understand00:49
@sonney2ke.g. vojtech -> libqp expert, shogun mentor me shogun expert etc00:50
n4nd0let me write a list with the mentors in there00:50
@sonney2kyes please I will continue with 200:51
@sonney2kthe rest likely needs not much polishing00:51
n4nd0sonney2k: I just realized - Christian is not mentoring this year? Nor Alex Binder?01:01
@sonney2kno one asked them I guess?01:01
n4nd0too bad01:02
@sonney2kwell we can still add to the ideas list next week01:02
-!- FSCV [~FSCV@] has quit [Quit: Leaving]01:05
n4nd0sonney2k: there are some people I don't really know about01:13
-!- zhuoran [80eefe5d@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]01:15
@sonney2kn4nd0, yeah well01:15
@sonney2kjsut put their names there and I will write more details01:15
n4nd0there are some TODOs together with their names01:16
@sonney2kn4nd0, somehow the questions in the gdoc don't match the one in melange01:18
@sonney2kanyway please read 1 & 201:19
n4nd0shit idk how to make proper comments01:21
n4nd0got it now01:22
n4nd0you established the JMLR software track? wow01:25
@sonney2kyeah,, too01:26
@sonney2kn4nd0, please read 3-6 now too01:27
n4nd0I wrote a couple of minor comments01:29
n4nd0if you agree I can modify01:30
@sonney2karg I am tired01:31
@sonney2kso what is missing01:31
n4nd0write sth about this 4 guys01:32
@sonney2klet me start from the end01:32
@sonney2kplease read 1101:38
@sonney2kplease read 1001:41
@sonney2kn4nd0, ^01:41
n4nd0sonney2k: should we put in 10) that students are also encourage to "idle" at IRC and the communication between students and some mentors is daily there?01:43
@sonney2kn4nd0, yes please!01:44
@sonney2k[;ease do01:45
@sonney2kn4nd0, do you know what blackburn meant in 8 with  Second, we do coordinate plans and schedule before GSoC as much as possible and chances are low for busy students.?01:48
n4nd0sonney2k: since it is required to write an schedule for the application, students who will be busy during the summer with other occupations are detected??01:49
n4nd0I didn't express myself well either TBH...01:50
@sonney2kso this is about overloaded students!01:50
n4nd0I think so01:50
@sonney2kyou can read 9 and 801:55
@sonney2kn4nd0, I hope 8 is better now01:56
n4nd0yes, I think it is good01:58
n4nd07 and we are done then01:58
@sonney2kI am dying02:03
@sonney2kplease read the first paragraph of 702:03
n4nd0hehe tomorrow is holiday :)02:03
n4nd0yeah I am reading on the way02:03
n4nd0hehe hardcore machine learners? :)02:04
@sonney2kkids wake up in a few hours they won't understand why I waste time on randomly hitting keystrokes at night02:04
n4nd0in a few hours?! what time do they normally wake up?02:05
n4nd0my gf will wake me up latest at 9h saying, I am hungry, frukost! :D02:05
@sonney2kwell I have 4 hours max before being woken up02:06
n4nd0wow they really wake up early then02:06
@sonney2kn4nd0, ok done02:12
@sonney2kplease check02:12
n4nd0we are good then I think02:14
n4nd0sonney2k: are you going to put it in melange right away?02:15
@sonney2kn4nd0, let me try02:18
@sonney2kalright submitted02:30
@sonney2kbed time!!!!!02:31
n4nd0good night!02:32
-!- n4nd0 [] has quit [Quit: leaving]02:39
shogun-buildbotbuild #337 of nightly_default is complete: Failure [failed test]  Build details are at
-!- naywhayare [] has joined #shogun03:51
-!- blackburn1 [~blackburn@] has joined #shogun06:12
-!- blackburn [~blackburn@] has quit [Ping timeout: 256 seconds]06:15
-!- sumit_ [73f91219@gateway/web/freenode/ip.] has joined #shogun08:21
-!- blackburn1 [~blackburn@] has quit [Quit: Leaving.]09:02
-!- blackburn [] has joined #shogun09:38
-!- sumit_ [73f91219@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:47
-!- hoijui [] has joined #shogun10:19
@sonney2kblackburn, yes indeed :D11:29
@sonney2kblackburn, please check whether you are OK with the proposal now11:30
blackburnsonney2k: yeah why not11:31
blackburnI am ok with everything11:31
blackburnI see you filled in a lot of stuff there11:32
@sonney2kblackburn, ok12:14
blackburnsonney2k: I am thinking about removing dimension reduction thing as it is a separate developed module - would be a bad thing12:20
blackburnmoving code back and forth is an idiot thing12:21
@sonney2kremove from where?12:22
blackburnsonney2k: from the ideas list12:23
-!- n4nd0 [] has joined #shogun12:23
@sonney2kblackburn, I thought you do that in tapkee only anyways and only write wrappers to make it work in shogun?12:25
blackburnsonney2k: yes and it sounds like a bad idea for *shogun*12:26
blackburnbecause student would work on tapkee directly and can tend to avoid learning shogun12:26
@sonney2kwe did the same with some other libs and have libqp similarly in the proposal12:26
@sonney2kblackburn, well you mentor him right?12:26
blackburnyes that's why vojtech student didn't know about shogun at all12:26
@sonney2kso you can make sure he gets it right12:27
@sonney2kblackburn, yeah but he did integrate the stuff12:27
blackburnok whatever lets keep it then12:27
@sonney2kthere can be synergies12:27
blackburnyeah if I didn't develop a standalone library I wouldn't manage to learn a lot things12:28
@sonney2ki.e. the external lib profits, shogun profits from the external lib (less code directly maintained in shogun, interfaceses to all $LANG)12:28
blackburnyes that was my point too but you didn't like that before?12:28
@sonney2kblackburn, so what is missing now is to get shogun's build back to green12:29
blackburnI have no idea what is missing12:29
@sonney2kblackburn, of course not in your case12:29
@sonney2kit is like 1 shogun developer less12:29
@sonney2kthe stuff with vojtech scales as in he continues to develop libqp12:30
@sonney2ksame with libvw12:30
blackburnI see nothing different12:30
blackburnif you'd rather keep my skills low - yes12:31
@sonney2kblackburn, ohh you did remove VW ?12:31
blackburnsonney2k: yes I didn't yet write an description12:31
@sonney2kI thought you wanted to have it in the ideas list?12:31
-!- lambday [3ba2174d@gateway/web/freenode/ip.] has joined #shogun12:32
blackburnI have no plan about it yet12:32
@sonney2kblackburn, the difference is that john langford / vojtech will never work on shogun but their library12:32
blackburnmay be I'll will write it this weekend12:32
blackburnsonney2k: and?12:32
@sonney2kblackburn, you are very active w/ shogun. when develop a separate library your shogun activity decreases12:33
@sonney2kthat is the difference12:33
blackburnbut we are not striving for producing a noise but quality right?12:33
@sonney2ksure why?12:34
blackburnwhat do you mean by decreasing activity then?12:34
@sonney2kblackburn, one core shogun developer less12:35
@sonney2kin contrast to an indirect win of one shogun developer when libvw and libqp are improved12:35
@sonney2kn4nd0, good morning! I will go to bed now to get some more sleep. If you have time please fix the dead link on the front page (middle more button).12:37
n4nd0sonney2k: good afternoon :P12:37
n4nd0let me see what link is wrong12:37
blackburnsonney2k: well people lose interest that's normal12:38
-!- heiko [] has joined #shogun12:39
-!- heiko [] has quit [Client Quit]12:39
blackburnI also learned that have not the best BDFL12:39
n4nd0sonney2k: the digit recognition application shows completely wrong in my browse :S12:39
n4nd0blackburn, sonney2k : is it fine there?
blackburnn4nd0: it has kinda 2x of normal width12:40
blackburnn4nd0: do you mean that?12:40
blackburnn4nd0: oh even worse12:42
n4nd0sonney2k: I think that PR should fix it, merge please if you have a second12:48
n4nd0blackburn: do you have shogun-web rights as well? :)12:49
blackburnn4nd0: yes12:53
n4nd0I don't really know how it works for shogun-web when one merges things12:54
n4nd0if they go directly to the webpage or one should do something else12:54
n4nd0it seems it didn't change yet12:55
blackburnn4nd0: probably that doesn't change anything13:05
n4nd0il maestro must probably update something too13:06
-!- heiko [] has joined #shogun14:26
lambdayhi heiko14:41
lambdayjust saw your comments..14:41
lambdayfixing the unittest..14:41
heikolambday: these are just minor comments14:41
heikogood work otherwise14:41
lambdayheiko: thanks :)14:43
lambdayand you mean fixed matrix of 100x100?14:44
heikolambday: yes14:44
heikoand do not use log(det(M))14:44
heikobut use matlab or something to compute the number to compare to14:44
lambdayheiko: okay..14:45
heikouse the log(det(M)) for the small one14:45
heikoshould be safe14:46
heikoin facts, its better to use the number14:46
heikoto leave it :)14:46
heikono need for randomness14:46
heikofixed tests are better14:46
lambdayokay.. will be updating asap..14:47
heikoif you want more than one test, what you could do is to introduce some "difficult" matrices with a large condition number, but I think thats not really needed14:47
lambdayone more thing.. I was working on the sparse case...14:47
heikolambday: yes I saw the mail, let me read it :)14:49
lambdayheiko: okay.. so I have to manually enter 100,000 numbers, in the test? will be using matlab rand for generating one though.. :(14:49
lambdayalright :)14:49
heikocreate some matrix with structure and then use that one14:50
heikolet me think of an example ...14:51
-!- n4nd0 [] has quit [Quit: leaving]14:51
heikolambday: just do A=1:10014:54
heiko(as col vector)14:54
heikoand then A=A*A'14:54
heikoand then add some number to the diagonal14:54
lambdaygot it14:54
heikoto make prevent problems14:54
heikoyou have to add something otherwise smalles Eigenvalue will be zero14:55
heikosum(log(diag(chol(A*A'+eye(size(A,1)))))) =  6.36591954859210914:56
heikoon my machine14:56
heikowith 1:10014:56
lambdayyes, same result15:01
lambdayokay.. using that in the test then15:01
heikomaybe for the small test, use the log(det(M)) thing and the number15:02
heikolambday: btw you can assert high accuracy floating point results with EXPECT_EQ(CMath::abs(number - expected), 1E-15)15:03
heikoEXPECT_LE (less or equal)15:03
heikothats how I usually do it15:03
lambdayheiko: okay..15:06
lambdaywith floating points this really sounds the best way to check I guess.. I had to fight with it for quite a long time that day..15:07
heikoyes the google test one only checks a few digits, which is not enough for us15:07
blackburnheiko: ???15:08
blackburnheiko: EXPECT_NEAR15:08
lambdayi tried with expect double eq but that failed15:08
heikoblackburn: same thing15:09
heikobut yes, actually less to write :)15:09
heikolambday: so check EXPECT_NEAR15:09
heikoEXPECT_NEAR(val1, val2, abs_error);  the difference between val1 and val2 doesn't exceed the given absolute error15:09
lambdayheiko: alright...15:10
blackburnheiko: but if there is a bug in cmath::abs?15:10
heikosometimes one has to play with the error tolerances a bit ...15:10
heikoblackburn: well we should have a test :)15:10
blackburnheiko: and if you don't cover a bug?15:11
heikoblackburn: I think its possible to cover all cases for CMath::abs15:12
heikoin fact, for this one, I can even prove that it works correctly with some hoare calculus or whatever ;)15:12
blackburnheiko: you can't prove that15:14
heikoblackburn: I can15:14
heikoI learned that at university15:15
blackburnheiko: I don't think you can prove that code will work on any compiler/machine15:15
blackburnthat's why people stick with other tools rather than writing their own15:15
heikoblackburn, well wrt to some conditions15:15
heikoblackburn: but the same holds for gtest right?15:16
blackburnheiko: of course but it is a duty of gtest to write that15:16
blackburnheiko: it is sonney2k-like approach to rewrite all the shit and pray15:16
blackburnI don't like that15:16
blackburnI don't believe my code more than any other code15:16
heikoblackburn: so it is the duty of any compiler to check if then else right?15:17
heikofrom our perspective, it doesnt matter whether we use CMath::abs or gtest since for both to be correct, we have to assume that either compiler or gtest work correct15:17
heikobut under the assumption that compiler works correct, the abs code is fine15:17
blackburnyes but you have to stick to testing framework when checking invariants15:18
heikoblackburn: well what happens if there is a bug in abs?15:18
heikotest fails15:18
blackburnit is not about that abs thing but in general15:18
heikoso with my test I also test abs15:18
heikoblackburn: yes15:18
blackburntesting through untested15:18
blackburnit cascades bugs15:18
-!- sumit [73f91219@gateway/web/freenode/ip.] has joined #shogun15:19
blackburnheiko: btw the same principle applies to the test coverage thing15:20
heikoblackburn: what do you mean?15:20
blackburnif we use a lot of CMath etc15:20
blackburnit is marked as covered15:20
blackburngiving a false feeling of safety15:20
heikotest coverage is always a fals feeling of safety right?15:21
heikotest coverage can only be used to create feelings of non-safety15:21
heikonamely if its low15:21
blackburnheiko: yes but the more things you avoid the better that estimate is15:21
heikoblackburn: but this is also a hard problem right?15:21
blackburnof course15:21
heikoyou cannot always check everything from scratch in your tests15:22
blackburnheiko: that's why you should not use code you don't test there if possible15:23
blackburnand that's why I don't like using any of abs or etc there15:23
heikoblackburn: well I agree, but I would say for abs it is not important due to the simplicity of it15:23
heikoblackburn: but will avoid in the future15:23
blackburnfeel free heiko I just warned15:24
blackburnheiko: about the proposal - have you checked?15:27
blackburnit now contains a lot of stuff15:28
blackburnsonney2k rewrote the whole thing or so15:28
heikoblackburn: no thats why I am here in fact15:30
heikois it still in the google thingi?15:31
blackburnheiko: yes15:32
heikoblackburn: btw I read your paper draft.15:52
heikoI dont have anything to complain except for a few typos /grammar things15:52
heikowill send you a scan of my notes soon15:52
blackburnheiko: cool! thanks for reading it15:52
blackburnheiko: I owe you a beer :)15:53
heikoalso like it!15:53
heikobeer or bear? :)15:53
blackburnheiko: bear hmm I am not sure they live there15:54
heikohaha :)15:54
blackburnheiko: if you go with me a little eastern15:54
blackburnI could find you some bear15:54
heikohehe :)15:54
heikoso when is deadline?15:54
blackburndeadline for?15:55
blackburnheiko: gsoc?15:55
blackburnheiko: I don't know - is there any deadline? :)15:55
blackburnI mean it is not even submitted yet15:56
heikoblackburn: ok I see, so no stress in sending you the feedback15:56
blackburnyeah a month is ok15:56
heikomy parents are here thats why I ask , so I do this afterwards :D15:57
lambdayheiko: I made the changes, it seems that 1E-15 is not working... I've kept it 1E-1015:58
heikolambday, thats fine - go as accurate as you can15:58
blackburnheiko: yeah you should rather get to your parents :)15:59
blackburnlambday: 1e-15 is too tight15:59
heikoblackburn: I sent them to the british museum which is 5 mins from here :)15:59
heikoblackburn: I sometimes use 1E-1515:59
heikobut 10 should also be fine15:59
lambdayI'm committing then15:59
heikolambday: nice15:59
blackburnheiko: it should be ok for small numbers15:59
heikoI have to leave soon, so if you hurry, I will merge before :)16:00
blackburnheiko: but 1 != 1 with 1e-15 I think?16:00
blackburnmay be 1==1 but could be that 3!=3 :)16:01
heikoblackburn, lambday, just go maximum that gets accepted :)16:02
heikojust more than 6 that was my point16:02
lambdayheiko: okay.. :)16:02
lambdayand I need a good tutorial for git I guess :( 80% of the time I am wondering about git..16:02
lambdayI git a git add and then git commit...16:03
heikolambday: haha, I totally know that and have that too from time to time :)16:03
heikobest is ask blackburn16:03
heikolambday: thats fine --- add/commit for new files16:03
lambdayheiko: :( :( ...16:04
heikoblackburn: the application is very good I think16:04
heikoblackburn: so not changing anything - you should commit :)16:04
heikoI have to leave in 2 mins16:05
blackburnheiko: boss made the deal already I thnik16:05
lambdayblackburn: :D16:05
heikoblackburn: I see, nice then, should be no problems16:05
lambdayhow to do the git pull --rebase thing? git pull --rebase upstream master... here shall I put upstream as "" or my forked repo?16:05
heikolambday, from your master branch you do16:06
heikogit fetch upstream16:06
heikogit pull --rebase upstream master16:06
heikothe you change to your feature branch and do16:06
blackburnfetch is redundant here16:06
heikogit rebase master16:06
heikothen you go to master  again and do16:06
lambdayheiko: alright..16:06
heikogit merge feature_branch_name16:06
heikothen you do git push16:07
heikolambday: and if you have changes you have to stash or commit them first16:07
heikoand a good tip is:16:07
heikonever modify the master branch16:07
heikoonly for pushing :)16:07
heikootherwise you can create a mess easily16:07
heikohad that often16:07
heikobut might be my stupidity :)16:08
heikolambday: ok sorry, gotta go now, will check tonight16:08
heikoblackburn,lambday  bye!16:09
lambdayheiko: okay... thanks :) ciao16:09
-!- heiko [] has quit [Quit: Leaving.]16:09
-!- blackburn [] has quit [Quit: Leaving.]16:40
-!- lambday [3ba2174d@gateway/web/freenode/ip.] has quit []16:42
-!- shogun-notifier- [] has joined #shogun18:25
shogun-notifier-shogun: rahul :master * afb7f2b / src/ (4 files):
shogun-notifier-shogun: added a new kernel SSKStringKernel which implements the subsequence kernel (by lodhi et. al.). Also added to the modular interface18:25
shogun-notifier-shogun: lambday :master * c5a7cac / src/shogun/mathematics/Statistics.cpp,src/shogun/mathematics/Statistics.h:
shogun-notifier-shogun: Added log_det method in CStatistics class18:25
shogun-notifier-shogun: lambday :master * 0891f5a / src/shogun/mathematics/Statistics.h:
shogun-notifier-shogun: indentation fixed in Statistics.h18:25
shogun-notifier-shogun: lambday :master * ac917f2 / / (7 files):
shogun-notifier-shogun: Fixed the previous messy commits18:25
shogun-notifier-shogun: lambday :master * 6cf55d3 / src/shogun/mathematics/Statistics.cpp:
shogun-notifier-shogun: Unit test for log_det added. log_det updated.18:25
shogun-notifier-shogun: lambday :master * 6b28e87 / src/shogun/mathematics/Statistics.h:
shogun-notifier-shogun: Fixed doxygen info in log_det18:25
-!- n4nd0 [] has joined #shogun18:28
shogun-buildbotbuild #843 of deb2 - static_interfaces is complete: Failure [failed test python_static]  Build details are at  blamelist: lambday <>18:34
shogun-buildbotbuild #844 of deb2 - static_interfaces is complete: Failure [failed test python_static]  Build details are at  blamelist: Heiko Strathmann <>, lambday <>18:46
-!- n4nd0 [] has quit [Remote host closed the connection]18:48
-!- n4nd0 [] has joined #shogun18:48
shogun-buildbotbuild #845 of deb2 - static_interfaces is complete: Failure [failed test python_static]  Build details are at  blamelist: rahul <rahul@localhost.localdomain>18:56
wikingmmm we are still committing to master branch ? :S19:05
@sonney2kwiking, yes sure until we have the build fixed19:12
@sonney2khmmhh but not features19:12
@sonney2kohh well19:12
@sonney2kwe are good in producing features and suck in fixing bugs19:12
shogun-buildbotbuild #971 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Heiko Strathmann <>, lambday <>19:15
-!- lambday [0e8b614d@gateway/web/freenode/ip.] has joined #shogun19:28
shogun-buildbotbuild #972 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: rahul <rahul@localhost.localdomain>19:34
@sonney2klambday, I just realize that your PR did contain all the previous commits - please for next time use git commit --amend or git reset <version> and commit a single clean version19:39
@sonney2klambday, this way the buildbot will not fail for the old broken commits (it builds every commit...).19:39
lambdaysonney2k: yes.. sorry for messing it up this time..19:39
@sonney2kn4nd0, thanks for fixing the link19:40
@sonney2klambday, don't worry we are no better in all that19:40
lambdaysonney2k: next time onwards I'll be more careful before sending a PR..19:40
@sonney2klambday, btw please also set a proper user name for your git stuff19:40
@sonney2klambday, don't worry it is our mistake not to tell you upfront19:41
n4nd0is it fixed now?19:41
@sonney2kwe do / did the same stuff19:41
@sonney2kn4nd0, I didn't put it to the website yet19:41
n4nd0I tried but it goes still to the old link!19:41
lambdaysonney2k: I have been struggling with git... hope to get better asap :(19:41
@sonney2kI was thinking that maybe we should put the workshop instead of the demos to the front page19:41
n4nd0all right19:41
@sonney2klambday, we still struggle with git too so welcome to the team :)19:42
@sonney2kn4nd0, would you agree?19:42
n4nd0writing on the phone here - 5 keystrokes per minute :)19:42
@sonney2kI mean it does help a lot19:42
@sonney2kn4nd0, pretty fast you are then :D19:42
lambdaysonney2k: everyone around here is so helpful.. :) thanks for being so patient19:43
n4nd0git yeah... our eternal archienemy/best friend :D19:43
-!- sumit [73f91219@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:44
n4nd0I agree with the workshop idea, probably lot of people will visit the page soon19:45
n4nd0the more visible the better I think19:45
@sonney2kn4nd0, ok then we should do it rather now. If you have time please write up sth. I am trying to find /ask for a nice c-base figure to put on the front page19:47
n4nd0I am not at home now unfortunately19:48
n4nd0no computer around to work, maybe tonight when I get back19:48
@sonney2kahh true19:48
@sonney2kphone :D19:49
* sonney2k is asking for a pic at #c-base19:49
n4nd0what kind of pic you have in mind?19:49
n4nd0the spaceship one is fancy hehe19:50
@sonney2kthe logo on top would be cool19:50
@sonney2kor the starshop19:51
@sonney2klooks cool for sure19:51
n4nd0so you have to ask them for the permission to use it I understand19:52
@sonney2kor this one
@sonney2kI think so19:52
n4nd0I like these ones better than the spaceship19:58
n4nd0but any is fine I think19:59
lambdaysonney2k: one stupid question - why did you named this project shogun? I mean, what does "sho gun" mean?20:03
n4nd0soeren sonnenburg and gunnar raetsch I think - main original authors20:08
n4nd0sort of acronym of their names20:08
@sonney2klambday, yes indeed what n4nd0 said20:13
@sonney2kn4nd0, I will just fix the website as is - please when you have time write some text for the middle block20:14
lambdayn4nd0: hmmm.. :)20:14
@sonney2kit seems it takes some time to get permission to use a c-base logo20:14
@sonney2klambday, we had other ideas like Reason/ Raeson / but in some survey people did prefer shogun :D20:15
lambdaysonney2k: shogun sounds cool.. I put that in google translator from japanese to english and it showed "military treatment" :D20:17
n4nd0ok will do tonight probably20:17
n4nd0tomorrow afternoon I am going to travel and I will be until Tuesday eveningn completely off so I will do it before going20:18
@sonney2kn4nd0, thanks. I will fix the website now and then we still have this burning issue with tests not being all passing :/20:19
@sonney2kI could use some help for that...20:20
n4nd0any clue what is wrong?20:20
n4nd0I can spend some time tonight20:20
n4nd0sonney2k ^20:25
-!- blackburn [~blackburn@] has joined #shogun20:27
@sonney2kn4nd0, thanks20:35
@sonney2kblackburn, do you have time to help fixing the buildbot errors?20:57
blackburnsonney2k: yes20:58
wikingsonney2k: what's the bug that holds back switching to dev branch?20:58
@sonney2kwiking, I've made tests / example failures critical20:58
blackburnwiking: something is wrong20:58
@sonney2ki.e. when one fails -> boom20:58
@sonney2kbuildbot fails20:59
@sonney2kbefore that we didn't enable this stuff20:59
wikingah ok20:59
@sonney2kand because of that we saw lots of breakage recently20:59
wikingyeah i know that examples were not critical eariler20:59
@sonney2kso we didn't detect the issues20:59
wikingsonney2k: some of the code always core dumped on the fbsd bot i saw that in dmesg always21:00
@sonney2kwiking, I am still not sure I really enabled all of them21:00
blackburnsonney2k: we've got to ignore these 3 kernels21:00
blackburnand hmm21:00
@sonney2kwiking, yeah some examples generate random data21:00
blackburnthat's not possible to solve21:00
@sonney2kand so they fail only sometimes (well did fail - I've set the seed now)21:00
wikingok, where's the list of failures? :)21:01
@sonney2kwiking, buildbot21:02
wikingblackburn: u have your visa for july workshop? :)21:02
@sonney2kbut caution21:02
wikingsonney2k: just checking21:02
blackburnwiking: no21:02
@sonney2kwe only see python failures for now21:02
wikingblackburn: :( get it!21:02
blackburnwiking: yeah I should21:03
@sonney2kthe other stuff will only become visible once python failures are fixed21:03
@sonney2kwiking, and once we fixed that all I still don't know what we do with buildbots21:03
wikingblackburn: is the tsn or what the hell in tapkee now?21:04
@sonney2kI've checked config and we explicitly run the buildbot for master branch only21:04
blackburnwiking: yes t-sne is in21:04
wikingblackburn: mmm i ought to test that21:04
wikingblackburn: in shogun as well? or only tapkee for now?21:04
blackburnwiking: no I didn't manage to write a wrapper class yet21:04
wikingblackburn: so tapkee only cool i'll hae to check that tonight then21:04
blackburnwiking: if your data is dense you may use that simple app that comes with tapkee21:05
@sonney2kwiking, you know there was more important stuff to do gsoc application!21:05
wikingblackburn: yeah it is so then i can test it out of box21:05
-!- lambday [0e8b614d@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]21:05
wikingsonney2k: :S21:05
blackburnwiking: it reads ascii and outputs ascii21:05
blackburnwiking: one row = one feature21:05
wikingblackburn: cool21:05
wikingblackburn: csv?21:05
blackburnwiking: no I am lazy21:06
blackburnjust ascii21:06
wikingblackburn: space?21:06
wikingblackburn: i'll test it then soonish and let you know what did it do with my stuff21:06
blackburnwiking: ./bin/tapkee -m t-sne -i input.dat -o output.dat21:06
blackburnwiking: ./bin/tapkee -m t-sne -i input.dat -o output.dat --sne-perplexity 50 --sne-theta 0.521:07
blackburnwiking: two params21:07
wikingblackburn: where's more info about those two params21:08
wikingblackburn: and what's the output btw? clustered vectors?21:08
blackburnwiking: I have to admit I don't know the exact meaning21:08
blackburnwiking: clustered??21:08
blackburnwiking: ./bin/tapkee -m t-sne -i input.dat -o output.dat -td 2 --sne-perplexity 50 --sne-theta 0.521:08
blackburn-td is target dimension21:08
blackburn2 by default21:08
blackburn that's the paper on barnes-hut-sne21:09
wikingwell okeeey21:09
wikingi guess i'll need a bit bigger -td :P21:09
blackburnwiking: why?21:09
wikingi have a feeling21:10
wikingbut then again no idea21:10
wikingso i'll just test it21:10
blackburnwiking: 3?21:10
blackburndo you see through multidimensional spaces?21:10
* wiking haz magic eyez21:10
wikingblackburn: just for those hours when u give a fuck21:11
blackburnwiking: t-SNE is purposed for visualization (d=2,3)21:12
blackburnthat's why I am surprised21:12
blackburnwiking: furthermore barnes hut is *specialized* to support only 2 IIRC21:12
blackburnwiking: yeah you can't set it to 3 even :D21:13
wikingahhaha ok21:13
blackburnwiking: I can check if I can extend it21:13
blackburnwiking: alrightt should work with other valuesto21:14
blackburnwiking: PARAMETER(ScalarType, perplexity,        SNE_PERPLEXITY,    IN_RANGE(perplexity,0,(N-1)/3));21:14
blackburnwiking: last parameter tells you about possible range of perplexity :)21:15
blackburnwiking: and theta=0 means t-SNE while theta>0 means barnes-hut-sne21:15
-!- n4nd0 [] has quit [Remote host closed the connection]21:20
-!- shogun-notifier- [] has quit [Quit: transmission timeout]21:25
blackburnwiking: I can't say it works with td!=221:29
blackburnlets check basic t-sne21:30
blackburnwithout barnes-hut21:30
blackburnwiking: what do you think?21:32
@sonney2kblackburn, looks nice? gsoc idea?21:33
blackburnsonney2k: I think a week of work21:34
@sonney2kblackburn, I don't remember whether you responded to my comment to this: one can do this all in c++21:35
@sonney2kwith some kind of builder pattern and then it should immediately work from all *modular interfaces like that21:36
blackburnsonney2k: it is not a pure builder pattern21:36
blackburnit is a builder without need to call build()21:36
@sonney2kbut you are right thinking about it it should be very easy. would even say just 1 day of work21:36
blackburnsonney2k: not a day - we do not support list of sgobjects21:37
@sonney2ktrue woudl need a new typemap21:38
blackburnsonney2k: with proper design it could be heavily advertised21:38
@sonney2kwhat I don't understand with this syntax though is how one would do nested things21:38
@sonney2klike in your example try out gaussian_kernel with certain params and then linear kernel with other params etc21:39
blackburnsonney2k: how does it work now?21:39
@sonney2kI don't recall exactly - heiko would know21:40
blackburnsonney2k: actually I see it like that21:40
@sonney2kI had this bracket based solution that supports that21:40
@sonney2kbut it would require typemaps all over21:40
blackburnthat's all21:43
blackburnwhen you are done21:44
blackburnno tree messing at all21:44
blackburnsonney2k: ^21:44
@sonney2kblackburn, I totally swapped out how that works internally21:45
blackburnsonney2k: step by step21:45
@sonney2kmight work I am not sure21:45
@sonney2kI mean this syntax is certainly powerful enough21:46
blackburn1) gaussian_kernel.parameter("width") returns an instance of some class parameter (not parameter we have)21:46
@sonney2kI just don't know if we can easily map it to the structure we have21:46
blackburn2) it sets default parameters to that parameter21:46
blackburn3) as(ExponentialRange) pushes the parameter to the exponential state21:46
blackburn4) from(-10) sets lower bound21:47
blackburn5) to(10) sets upper bound21:47
@sonney2kblackburn, I've committed the submitted proposal to publicity & updated the website21:47
blackburnokay a few issues more21:47
blackburn1.1) each parameter("width") call marks parameter inside a class as *to-be-modelselected*21:47
blackburn3.1) as(ExponentialRange) has some default thing too21:48
blackburnsonney2k: the only flaw21:48
blackburnif we call wrong order or wrong methods21:48
blackburnit is runtime error21:48
blackburnnot a compile-time error21:48
blackburnbut that's default tradeoff21:48
blackburnI am totally sure people would get excited about that21:49
@sonney2kblackburn, I just checked modelselection_grid_search_kernel.cpp ?21:54
@sonney2kand I think your idea would work if one includes a21:54
@sonney2kCModelSelectionParameters* in SGObject21:55
@sonney2konly issue is that the complete tree has to be constructed somehow21:55
blackburnsonney2k: I do not know exactly but I see it constructing each time parameter() method is called21:55
@sonney2kModelSelectionParameters can handle nested stuff21:57
@sonney2kso I guess simplest soluton would be to have non-nested stuff inside each class21:57
blackburnsonney2k: there should be major changes in the framework to support that21:57
@sonney2kI don't think so21:57
@sonney2kthen when the whole object to be model selected is ready21:58
@sonney2kone creates the tree by going top->down21:58
@sonney2kcreating the nested CModelSelectionParameters21:58
@sonney2kI think that's all21:58
blackburnsonney2k: no it should not be top to down in final21:58
blackburnI'd vote for magic21:59
@sonney2kwhat magic?21:59
blackburneach time parameter() of some sgobject is activated21:59
blackburnthe tree is reconstructed etc21:59
blackburnor may be it is constructed only when doing model selection22:00
blackburnthat makes sense to me as well22:00
@sonney2kthat is what I meant22:00
@sonney2k(the latter idea)22:00
blackburnI see22:00
blackburnI have two major ideas now22:00
blackburnclient server and that22:00
@sonney2kbut no time to do either one22:01
@sonney2kthat is why I am saying -> gsoc idea!22:01
blackburnthat's a tough idea22:02
blackburnbasically one has to be a shogun developer to understand what to change22:03
blackburnI imagine if I would be asked to do that 2 years ago22:03
blackburnsonney2k: ahh btw yesterday was the day I joined #shogun22:04
blackburn2 years ago22:04
blackburnI feel youth is slipping away haha22:04
@sonney2khabby 2nd b'day22:08
@sonney2kblackburn, well heiko did the whole model selection framework in his first gsoc22:09
@sonney2kso what is more tough?22:09
blackburnsonney2k: his current idea22:09
@sonney2kyes indeed *lol*22:09
blackburndevelop a new machine learning paradigm22:10
blackburnstructured risk minimization was developed by vapnik in the end of 20th century. we need a new framework22:11
@sonney2kstructured output on general graphs & large scale is already sufficient22:12
blackburnsonney2k: but it is not heiko thing?22:14
@sonney2kblackburn, no but if you have some idea how to even do that - there is a workshop at this years icml about it :D22:19
blackburnI don't know about machine learning22:19
blackburnso no22:19
@sonney2kyeah and you don't drink vodka all day22:20
* sonney2k is in css hell22:20
blackburnboth are true22:20
-!- shogun-notifier- [] has joined #shogun22:21
shogun-notifier-shogun: Sergey Lisitsyn :master * 99135b4 / tests/integration/python_modular/
shogun-notifier-shogun: A splat for KRR test22:21
blackburnsonney2k: svmsgd parameters were changed you say?22:22
@sonney2kblackburn, I hardly remember22:24
@sonney2kthere was something22:24
blackburnsonney2k: trying to understand what is the difference between modular and static sgd calls22:24
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: leaving]22:26
shogun-buildbotbuild #846 of deb2 - static_interfaces is complete: Failure [failed test python_static]  Build details are at  blamelist: Sergey Lisitsyn <>22:27
blackburnalright will take a look at these examples tomorrow22:42
@sonney2kblackburn, look
@sonney2kI managed to get the border back22:53
@sonney2ksome 'trivial' css22:53
@sonney2k.ideas tr:not(:first-child) td {border:1px solid #8cacbb;}22:53
blackburnsonney2k: right border is not showed here22:54
@sonney2kblackburn, relaod!22:54
blackburngood you remind me22:54
blackburnI forgot to put benoit rostykys22:54
blackburnsonney2k: rostykus22:55
-!- n4nd0 [] has joined #shogun22:55
blackburnguy with olivier22:55
@sonney2kohh yes22:55
blackburnn4nd0: sonney2k: did you fix anything in content of that gsoc2013 html?22:55
@sonney2kwas there any bug?22:56
n4nd0blackburn: mmmm I don't remember right now22:56
@sonney2kblackburn, ohh btw you had this nice header telling students that they should select at most 3 ideas22:56
@sonney2kthat should also be in there again22:56
@sonney2kand the patch requirement22:56
blackburnsonney2k: alright22:57
blackburnn4nd0: I want to sync with repo stuff22:57
@sonney2kn4nd0, some students don't read what we write in the form...22:57
@sonney2kblackburn, what?22:57
@sonney2kwhich repo?22:57
@sonney2kahh the html22:57
blackburnsonney2k: I want to copy from publicity to the website22:57
n4nd0blackburn: yes take the version in the page then22:57
@sonney2kI see22:57
@sonney2kblackburn, plesae do the other way round22:58
@sonney2kcopy from website22:58
n4nd0blackburn: I think there was sth minor I changed22:58
@sonney2kbetter be safe22:58
n4nd0oh yeah some carriage returns22:58
@sonney2kwe have no history or so22:58
@sonney2kon the website22:58
blackburntabs are ruined22:58
n4nd0sonney2k: why?22:58
n4nd0that they don't read forms22:58
@sonney2kn4nd0, bern4rd didn't program any22:59
@sonney2kit is some effort...22:59
blackburnthat's a problem22:59
blackburnsonney2k: we have history22:59
@sonney2kblackburn, where?22:59
blackburnsonney2k: but it doesn't store old22:59
blackburnjust mentions changes22:59
blackburnohh it has eaten all spaces23:00
@sonney2kn4nd0, well you could program some simple solution easily - just add a revision to each object23:00
@sonney2kand increase that revision all the time23:00
@sonney2kand show only the largest revision23:00
n4nd0we are talking about storing history of the articles right?23:01
@sonney2kn4nd0, yes23:01
@sonney2kI think subpages/pages is not so important23:01
@sonney2kthat's more layout23:01
blackburnohh I can't stop laughing about heiko idea23:01
blackburnthere are three mentors now23:01
n4nd0"easily", I think there would be some things to handle like what if two people are modifying at the same time?23:01
n4nd0there must be for sure something done already somewhere to draw inspiration from though23:02
@sonney2kn4nd0, ahh no we don't do that23:02
shogun-buildbotbuild #973 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Sergey Lisitsyn <>23:02
@sonney2kn4nd0, we usually coordinate stuff no?23:03
blackburnsonney2k: are they using shogun at that 'criteo'?23:03
n4nd0sonney2k: yes23:03
n4nd0simple idea would be if someone is editing one article then another one cannot have the chance to do it23:03
@sonney2kblackburn, no idea23:03
@sonney2kn4nd0, yes but this is (not yet) a problem23:04
blackburnsonney2k: I think heiko need a co-mentor to increase # of mentors to 423:04
n4nd0sonney2k: btw, there is no place yet in the homepage for the workshop thing or am I missing sth?23:04
@sonney2kwhen we are 100 people then 2 might edit the website at the same time23:04
@sonney2kn4nd0, I was hoping you write something?23:04
n4nd0sonney2k: aham got it23:05
n4nd0I misunderstood23:05
n4nd0so the idea is to write a few words about the workshop, put it in the homepage together with a link to the main workshop page under events23:05
@sonney2kn4nd0, exactly23:05
n4nd0sonney2k: ok, I am on it then23:06
@sonney2kn4nd0, btw I managed to get the table border back23:07
@sonney2kin the ideas list23:07
n4nd0wow I see23:08
n4nd0sonney2k: it is looking very nice!23:08
@sonney2k`just' .ideas tr:not(:first-child) td {border:1px solid #8cacbb;}23:08
n4nd0css I guess23:08
@sonney2kanyway I have to sleep23:08
@sonney2khard w/o a css editor23:09
n4nd0sonney2k: I will mail you these short text for the homepage soon23:09
@sonney2k& nite23:09
n4nd0good night23:09
blackburnsonney2k: that css family guy pic fits the case23:09
n4nd0hehe indeed23:09
--- Log closed Sat Mar 30 00:00:18 2013