Open in new window / Try shogun cloud
--- Log opened Fri Feb 08 00:00:41 2013
-!- zxtx [] has quit [Ping timeout: 244 seconds]00:57
-!- zxtx [] has joined #shogun01:11
-!- shogun-buildbot [] has quit [Ping timeout: 246 seconds]03:32
-!- shogun-buildbot [] has joined #shogun03:33
-!- Netsplit *.net <-> *.split quits: syst3mw0rm, blackburn, sonne|work, shogun-buildbot, cavalier-, @sonney2k, zxtx, wiking04:13
-!- sonney2k [] has joined #shogun05:25
-!- ServerMode/#shogun [+o sonney2k] by zelazny.freenode.net05:25
-!- blackburn [~blackburn@] has joined #shogun05:26
-!- zxtx [] has joined #shogun05:27
-!- shogun-buildbot [] has joined #shogun05:27
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun05:27
-!- cavalier- [~cavalier-@] has joined #shogun05:27
-!- sonne|work [] has joined #shogun05:27
-!- syst3mw0rm [] has joined #shogun05:27
-!- syst3mw0rm [] has quit [Ping timeout: 246 seconds]05:31
-!- syst3mw0rm [] has joined #shogun05:33
-!- FSCV_ [~FSCV@] has quit [Ping timeout: 255 seconds]07:41
-!- FSCV [~FSCV@] has joined #shogun07:55
-!- FSCV [~FSCV@] has quit [Excess Flood]07:58
-!- FSCV [~FSCV@] has joined #shogun07:58
-!- n4nd0 [] has joined #shogun09:05
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has joined #shogun09:48
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has quit [Quit: Page closed]09:59
@sonney2kblackburn, so you don't want gsoc to happen this year?10:19
blackburnsonney2k: why so?10:19
@sonney2kblackburn, I didn't see any emails...10:20
blackburnsonney2k: I wanted to discuss with you10:20
blackburnsonney2k: wrote in jabber actually :D10:20
@sonney2kI am not at work...10:21
@sonney2kso no jabber10:21
blackburnsonney2k: I am convincing igor carron now10:22
blackburnto mentor something10:22
blackburn his blog10:22
@sonney2kblackburn, that comes later10:23
@sonney2kfirst announce our interest10:23
@sonney2klook at what I did last year...10:23
@sonney2k(email archives to the list)10:23
blackburnsonney2k: should I ask to do everything in public?10:23
@sonney2kblackburn, but yes cool10:24
@sonney2kblackburn, ?10:24
@sonney2kHEAD^^ igor10:24
blackburnwell for example should I ask possible mentors to announce their interest10:24
blackburnin public10:24
blackburnor it is ok to contact me10:24
blackburnI don't mind any10:24
@sonney2kblackburn, what I did:10:25
blackburnyeah I am searching that mail10:25
@sonney2k1) tell everyone on the mailinglist we want to participate in gsoc10:25
@sonney2k2) give them a list of old ideas10:26
@sonney2k3) mention the new ideas url that we want to update10:26
@sonney2k4) talk to everyone (phone/chat but in private) who could potentially mentor and what exactly10:26
@sonney2k5) update the ideas list10:26
@sonney2k6) apply for gsoc (write proposal)10:27
@sonney2k7) announce that we are in or not10:27
blackburnsounds like an algorithm, thanks :)10:27
blackburnsonney2k: so it seems I should take it over completely in any case?10:27
blackburnlike I should say I am responsible for all the stuff about that?10:28
blackburnokay I'll write some big mail tonight10:30
@sonney2kget stuff done man10:30
@sonney2knot like helping me to fix this bug10:31
blackburnsonney2k: ?10:31
blackburnsonney2k: well I asked that cause if I don't say anything - you will get mails10:32
blackburnwhich you'd need to redirect anyway then10:32
@sonney2kblackburn, recall the bug about failing GPs?10:35
@sonney2kblackburn, just post it publicly on the    mailinglist - people will know what to do10:35
@sonney2kand draft the gsoc2013 ideas list website10:36
blackburnsonney2k: ahh that was irony meaning I didn't help you to fix the bug ;D10:56
blackburnsonney2k: quite acid!10:57
@sonney2kblackburn, which it was...11:01
blackburnI keep not getting what you mean11:02
@sonney2kblackburn, recall that you said you will have a look at the bug?11:03
blackburnI said I tried to fix that but didn't managed11:03
blackburnsonney2k: I need admin rights in shogun-toolbox org at github11:04
@sonney2kthe compute incremental hash function was called on elements of an empty array (a check was missing...)11:05
@sonney2kblackburn, what for?11:05
blackburnsonney2k: at least to control service hooks11:06
@sonney2kwhich hook is missing?11:06
blackburnhaha okay11:06
-!- shogun-notifier- [] has joined #shogun11:23
shogun-notifier-shogun: Soeren Sonnenburg :master * 680939f / src/Makefile.template:
shogun-notifier-shogun: fix (harmless) error messages on make install11:23
shogun-notifier-shogun: better avoid them to not confuse us users or devlopers :)11:23
blackburnwiking: hello to you too11:29
wikingsonney2k: here?11:29
@sonney2kwiking, wassup god of the north?11:30
wikingsonney2k: unittesting11:31
wikingsonney2k: have u seen somewhere that stuff explained?11:31
* sonney2k wants - check11:31
wikingsonney2k: ok11:31
wikingsonney2k:  i'll try to see how we can do it11:31
wikingbut afaik it'll only work with ubuntu/debian systems11:31
wikingas i have no clue how the others are handling those11:31
wikingi mean the gtest+gmock package11:32
@sonney2kwiking, perfectly ok with me - IMHO none of the devs uses anything els11:34
wikingsonney2k: meeee11:34
wikingsonney2k:  i use os11:34
@sonney2kyeah but you can take care of yourself :D11:34
n4nd0I am using arch11:35
wikingn4nd0: hehehe ok then you can help me11:35
n4nd0but I can try to care of me too :D11:35
wikingn4nd0: can u check if you have google unit test package?11:35
wikingn4nd0: and google mock package?11:36
n4nd0wiking: I installed gtest this morning actually, I needed it for ROS - uni work11:36
wikingn4nd0: ehehhe u use ROS? cool11:36
n4nd0wiking: I have started today11:36
wikingn4nd0: ok can you send me in pastebin or somewhere what is the package content for gtest?11:37
n4nd0I am going to use it for my thesis11:37
wikingn4nd0: when r u finishing?11:37
n4nd0wiking: enough?11:37
n4nd0wiking: I have started this week :D11:37
wikingn4nd0: mmmm yeah that one11:37
n4nd0I plan to be finished by August - September11:37
wikingbut do you have that in package on arch?11:37
wikingor you installed it from src/svn?11:38
n4nd0wiking: yep, it is in arch11:38
n4nd0it can be installed with pacman11:38
n4nd0I believe it was only available through AUR not so long ago11:38
wikingcan u give me the content of package?11:39
wikingsonney2k: have u heard of this: google-glog11:39
n4nd0wiking: let me see11:39
wikingsonney2k: but imho log4cxx is more mature11:39
wikingn4nd0: thnx just put the whole file list into pastebin or something11:40
wikingn4nd0: with debian you would do dpkg -L <package name>11:40
@sonney2kwiking, if I would switch then to log4cxx but there is no urgent need to do it11:40
wikingbut i don't know for arch11:40
n4nd0I had to installed that one too before log4cxx, it seems that people from ROS the same tools11:40
wikingsonney2k: k11:40
blackburnwiking: what can you mentor?11:40
n4nd0wiking: I have a few files here under /usr/include/gtest, I am not sure which one you need11:41
n4nd0stupid question probably hehe ^11:41
wikingblackburn: mmm well i'd happily mentor either a full redesign of optimization. to finally have a good general interface for optimizers11:41
wikingblackburn: or continue working with SO or latent stuff11:41
n4nd0I'd really like to continue with SO!11:41
blackburnwiking: can you please try to describe these ideas next week11:42
n4nd0atm I am reading the tutorial that Cheng posted on the mailing list regarding SO in CV11:42
@sonney2khmmh does anyone know what special stuff happened on february 1st? we had >600 new visitors that day11:42
n4nd0sonney2k: something workshop related maybe?11:42
wikingsonney2k: i guess it's because they've realised from last year11:43
@sonney2kn4nd0, your email?11:43
wikingsonney2k: that they need to have a PR before doing anything with shogun in gsoc ;)11:43
n4nd0sonney2k: I just checked and I sent it February the 2nd, 01:12 am CET11:43
wikingsonney2k: you thaught them well ;)11:43
wikingblackburn: till next week?11:43
wikingn4nd0: ok so nothing else just /usr/include/gtest ?11:43
wikingn4nd0: do you maybe have something under /usr/src?11:44
wikingn4nd0: or maybe /usr/lib11:44
blackburnwiking: till end of next week I think11:44
@sonney2kall these new visitors came from china!11:44
blackburnoh sh11:44
wikingblackburn: okey11:44
blackburnn4nd0: we'd need a mentor for you11:44
wikingn4nd0: wantme? :)11:44
n4nd0wiking: I have stuff in /usr/src/gtest11:45
wikingi mean as a mentor :)11:45
n4nd0wiking: sure :)11:45
wikingn4nd0: ooooh COOOL11:45
blackburncoupling you with wiking would be not so good I think - we need fresh people11:45
wikingblackburn: :D11:45
wikingblackburn: fresh meat :)11:45
blackburnyeah in means of mentoring too11:45
wikingn4nd0: ok then it's the same as in debian11:45
wikingn4nd0: can u check if there's google mock package for arch?11:45
blackburnhowever if you want it so much we'd think about it11:45
n4nd0wiking: these files are under /usr/src/gtest/src11:46
n4nd0wiking:   gtest.h        gtest_main.cc11:46
wikingn4nd0: yep that's cool11:46
n4nd0wiking: I have not found it in pacman repos11:46
wikingn4nd0: aha ok!11:46
n4nd0there should be sth in AUR I guess11:46
n4nd0let me see11:46
@sonney2kwiking, n4nd0, blackburn once we have some more tests active we should really release 2.111:47
n4nd0wiking: here you go
@sonney2knot like last year11:47
-!- alexlovesdata [~binder@2001:638:806:e001:eda1:c6f9:301a:5329] has joined #shogun11:47
n4nd0I was thinking these days11:47
@sonney2klast year we failed to release before n4nd0 made us drown in patches11:47
n4nd0what about a project just focused on examples??11:47
blackburnalexlovesdata: hey there - good time to visit us11:47
@sonney2kand so we had no more time to release11:47
blackburnalexlovesdata: want student this year?11:47
wikingsonney2k: yep11:48
n4nd0I think Georg Zeller did great last year helping me, we could ask him too11:48
wikingn4nd0: thnx11:48
n4nd0wiking: do you need me to install it?11:49
@sonney2kn4nd0, then please ask him if he has anything11:49
wikingn4nd0: well i'm just checking now whether i can find out what's the content of this package... i guess from the tarball i can get that info11:49
n4nd0wiking: ok11:49
n4nd0sonney2k: ok, I am mailing him right now11:49
@sonney2kwiking, well make it work for debian / ubuntu first then one step at a time...11:51
n4nd0blackburn: what is your address, I'll CC you11:51
wikingsonney2k: yeah i'm on it now11:51
blackburnn4nd0: sergey.lisitsyn@11:51
n4nd0blackburn: ty11:52
@sonney2kdoes anyone know what is?11:53
blackburnsonney2k: ??????????????11:53
blackburnhow can you don't understand?11:53
blackburnif I press translate in my chrome it says it is microblogging11:54
@sonney2kso someone blogged about shogun there I guess11:54
@sonney2kalexlovesdata, did you ask about rooms at TU for the after workshop party?11:55
blackburnmay be that11:55
n4nd0published 2012-04-24 I think11:56
@sonney2kblackburn, rather not yeah exactly^11:56
blackburnhah yeah11:56
@sonney2kblackburn, btw if you want to know how shogun is used - check this
@sonney2kA food image recognition system with multiple kernel learning :D12:01
@sonney2kHow far can you get with a modern face recognition test set using only simple features?12:01
alexlovesdatastudent for what, blackburn?12:01
blackburnalexlovesdata: gsoc12:01
@sonney2kGeotagged photo recognition using corresponding aerial photos with multiple kernel learning12:02
blackburnsonney2k: old.. - just nothing from 201212:02
@sonney2kA surf-based spatio-temporal feature for feature-fusion-based action recognition12:02
alexlovesdata@sonney2k ... not yet ... I want to check that mean assert but now12:02
alexlovesdatathat mean assert bug now12:02
@sonney2klook there are you :) ????????????? ???????? ?????? ? ??????? ?????? ??????? ???????? ? ?????????? ??????????????? ??????????12:03
blackburnsonney2k: lol12:03
blackburnyou scary people with these words12:04
blackburnsonney2k: can you translate it?12:04
blackburnwith your mind I mean12:04
@sonney2ksome words are missing12:05
@sonney2kalexlovesdata, please put this next on the list then12:05
blackburnsonney2k: which?12:05
@sonney2kalexlovesdata, about the assertion failure12:05
@sonney2kit assumes that 0.9 accuracy on that data set is possible12:05
@sonney2kdid anyone test if this is reasonable?12:06
@sonney2kI mean what does KNN get on this dataset?12:06
alexlovesdatait has to do with traffic sign detection12:06
alexlovesdata@sonney2K: i finmd this non reasonable12:06
blackburnsonney2k: alexlovesdata: may be some student polishing mkl?12:06
@sonney2kblackburn, ????????????? ????????12:06
alexlovesdatabecause we use some data and some kernels and non-sparse MKL12:06
blackburnsonney2k: recognition, road12:06
alexlovesdataHogs for traffic sign detection12:07
blackburnalexlovesdata: bingo12:07
alexlovesdatabut i do not know opornyi vektor12:07
blackburnalexlovesdata: stupid to write about that, isn't it? :D12:07
@sonney2kalexlovesdata, support vector12:07
blackburnalexlovesdata: support vector12:07
alexlovesdataohh, thx blackburn12:07
alexlovesdata@sonney2k: I want to check the performance of that example12:08
blackburnoriginal vapnik's name :D12:08
alexlovesdataand then if it is reasonably high with good kernels remove that assertion12:08
blackburnnot the name of vapnik I mean - but the method's name12:08
alexlovesdatahehehe, I got it blackburn ... opornyi vektor ;)12:08
@sonney2kblackburn, why did you say there is nothing from 2012 in the scholar list: Predicting the outcome of renal transplantation, Forensic identification with environmental samples ...12:08
@sonney2klots of12:08
blackburnsonney2k: hm well there are some12:09
blackburntop matches are from other libraries though12:09
alexlovesdata@sonney2k: I will take care of the room question, ... for how many people and what days?12:09
alexlovesdataaboutl iterature and things using shogun: my Plos ONE paper, my CVIU paper12:10
@sonney2kalexlovesdata, I guess <20 people for the weekend sat/sunday12:10
alexlovesdata@ sonney2k thanks! I willask andrea about our room first ... why not?12:10
@sonney2kblackburn, let me guess weka and libsvm/liblinear?12:10
alexlovesdataandrea = Klaus secretary12:10
blackburnsonney2k: I mean other libraries that publish mention us12:10
blackburnlike 'toolbox for doing all the shit'12:11
blackburnand they say there is other toolbox, shogun12:11
@sonney2kblackburn, sure they have to compare against12:12
alexlovesdata@blackburn: about Gsoc 2013 ... not sure yet ... too many things now above my head12:12
-!- blackburn [~blackburn@] has quit [Read error: Connection reset by peer]12:12
alexlovesdataI will know better in march because then some things have been decided12:13
-!- blackburn [~blackburn@] has joined #shogun12:13
@sonney2kalexlovesdata, when is your defense btw?12:16
alexlovesdata@sonney2k 27.2. 15.30  Marchstr 23 room 4.0something12:18
alexlovesdatawhen I know a party location for the evening i will announce it officially12:18
@sonney2kalexlovesdata, quite some time still but good luck nevertheless12:19
alexlovesdataroom 4.03312:21
-!- n4nd0 [] has quit [Quit: leaving]12:32
-!- n4nd0 [] has joined #shogun12:34
alexlovesdataquestion: where is data in the sgonu git?12:35
alexlovesdatain the shogun git?12:35
n4nd0alexlovesdata: it is in a git submodule12:36
n4nd0alexlovesdata: do you want to get it locally?12:36
alexlovesdatayes, exactly nando12:36
n4nd0alexlovesdata: go to the root of your git repo and run12:37
n4nd0git submodule update --init12:37
alexlovesdataahh ok, ... it is cloning data12:38
n4nd0alexlovesdata: :)12:38
-!- FSCV_ [~FSCV@] has joined #shogun12:39
alexlovesdata@sonney2k: the example works fine  with me with errors > 0.9312:40
alexlovesdatabut the results are non-deterministic12:41
alexlovesdataso it can happen that sometimes it drops < 0.912:41
alexlovesdataI suggestion to lower the assertion to 0.7512:41
alexlovesdataas a fix12:41
@sonney2kalexlovesdata, ohh I see12:41
@sonney2kalexlovesdata, maybe we better initialize the random number generator12:42
alexlovesdataI will try to push a commit ... i must learn that anyway :)12:42
alexlovesdata@sonney2k ... good idea12:42
-!- FSCV [~FSCV@] has quit [Ping timeout: 260 seconds]12:42
@sonney2kalexlovesdata, I guess it is due to the way the x-validation split is chosen?12:42
alexlovesdataI will try to init the random number generator and check whether this is deterministic then12:43
alexlovesdatagood! : CMath::init_random(1); yields a reproducible result12:51
alexlovesdataabove 0.912:51
alexlovesdataI will push a whatever fix12:51
@sonney2kalexlovesdata, that will be your first pull request right? did it just take 2 years to convince you github is not your enemy ;)12:58
alexlovesdatahahaha, sonne, no ... I just needed some hours of time without haste ... nothing can be really appreciated when running too fast13:22
alexlovesdataI will chekc how to init a pull request from my remote to shogun ...13:22
-!- n4nd0 [] has quit [Ping timeout: 252 seconds]13:24
alexlovesdataI needed to understand how to update my remote/master from shogun's upstream master first13:24
blackburnalexlovesdata: you need fork13:28
blackburndo you have one?13:28
alexlovesdatanot yet I think13:36
alexlovesdataI am reading for that13:37
-!- n4nd0 [] has joined #shogun13:39
blackburnalexlovesdata: ask me in case you need any assist13:41
@sonney2kblackburn can you merge it?14:04
blackburnyeah sure14:05
blackburnI was talking to alexlovesdata about github stuff14:05
shogun-notifier-shogun: AlexBinder :master * dfe64a8 / examples/undocumented/libshogun/evaluation_cross_validation_multiclass_mkl.cpp:
shogun-notifier-shogun: initialized random number generator for reproducible outcomes in the light of an assertion ASSERT(result->mean>0.9);14:05
shogun-notifier-shogun: Sergey Lisitsyn :master * 253b5bc / examples/undocumented/libshogun/evaluation_cross_validation_multiclass_mkl.cpp:
shogun-notifier-shogun: Merge pull request #859 from AlexBinder/master14:05
shogun-notifier-shogun: init random number generator in evaluation_cross_validation_multiclass_mkl.cpp14:05
@sonney2kpending valgrind runs this should bring us back to normal.14:07
alexlovesdatado we have Kernel fisher discriminant in shogun?14:11
alexlovesdatashould be easy to implement ... ten lines in matlab when we use spectral regression14:11
blackburnif you give matlab code to me I can implement that using eigen314:12
blackburnor you can :)14:12
n4nd0or maybe it is a good patch for someone who wants to get involved in gsoc ;)14:12
blackburnn4nd0: you?14:13
n4nd0blackburn: haha why not14:13
n4nd0but I was thinking of new-comers14:13
blackburnn4nd0: if you want to do some monotone work - feel free to check for some code in .h14:13
blackburnyou know what to do with that code ;)14:14
n4nd0are we purifying the code from those things?14:14
n4nd0online inline methods to survive?14:14
blackburnI don't see much help from inline methods14:15
alexlovesdatawhat is .h ?14:15
n4nd0I did in my head only inline -> online14:15
blackburnalexlovesdata: extension, header :)14:15
alexlovesdataye ... but what has that t odo with monotone work?14:15
blackburnthe more code we have in .h the slower compilation is14:15
alexlovesdataI know .h is a C header14:15
blackburnalexlovesdata: we need to get all the code to .cpp14:15
blackburnalexlovesdata: yes apparently .hpp is a proper extension14:16
alexlovesdataI never code routines in .h14:16
alexlovesdatanot even inits ...14:16
blackburnalexlovesdata: that's gut :)14:16
blackburnI am the guilty who did that I think14:16
alexlovesdataI could also try to  fight with eigen3 myself ... but thanks you for the offer14:17
blackburnalexlovesdata: feel free to ask me about eigen3 as well - I spent a lot of nights having luv time with eigen314:17
blackburnthere are some issues when you don't understand how to do something - it takes time to google14:18
alexlovesdataye ... true14:22
alexlovesdata preprocessors lacks
wikingn4nd0: here?14:37
wikingn4nd0: what is the make/configure env for ROS? cmake?14:37
n4nd0wiking: they use cmake yes14:38
wikingn4nd0: ok14:39
wikingn4nd0: thnx14:39
n4nd0wiking: what is it you are working on?14:40
n4nd0just curious14:40
wikingn4nd0: trying to integrate into shogun to compile the gtest with the same flags as shogun is build with14:40
wikingas that's the proper way to use gtest14:40
n4nd0aham I see14:40
blackburnalexlovesdata: hah - >50% of methods are missing there14:45
-!- heiko [] has joined #shogun14:45
blackburnI am trying to replace one pointer field in kernel with shared_ptr14:46
heikosonney2k, sonne|work, blackburn around?14:46
blackburnit goes ok actually - a few changes14:46
blackburnheiko: ja14:46
heikohi :)14:46
heikoI just saw a cool talk14:46
blackburnabout love?14:46
heikoon a hybrid cholesky algorithm which works on multiple core cpus and GPU14:46
heiko :)14:46
heikoI talked the author into integrating that into shogun14:46
n4nd0hi heiko14:47
n4nd0is it CUDA?14:47
heikon4nd0, hi :)14:47
blackburnheiko: may be he should rather integrate it to eigen14:47
heikoquite interesting thing he did14:47
blackburnheiko: cholesky is needed in gps, right?14:47
heikocholesky is needed everywhere where there is the word "Gaussian"14:48
blackburneigen has pretty nice cholesky anyway14:48
blackburnare there such big matrices?14:48
heikoblackburn, the GPU is used14:48
heikothis changes things14:49
heikowith eigen, you are right, but eigen has no cuda dependecies :)14:49
blackburnheiko: how does he code gpu? it is a great matter14:49
heikocuda c14:49
blackburnheiko: eigen runs on cuda14:49
heikomaybe then this would be a better idea you are right14:49
heikobut you know what?14:49
blackburnyeah gael works on some port14:49
heikoit will be exclusive to shogun otherwise :D14:49
heiko*and* he could try to make it mature in a safe environment14:50
n4nd0SHOGUN - Machine learning and linear algebra toolbox :D14:50
heikoand then once its working fine, put it into eigen14:50
heikon4nd0, well :)14:50
blackburncuda is nvidia property I am not the biggest fan of it14:50
n4nd0blackburn: why not?14:50
heikoblackburn, agreed, but everyone has a GPU14:50
heikoso one should use them14:50
blackburnI don't14:50
heikoits like 20 times faster on 10000 matrices14:51
blackburnHD4000 is not a gpu :D14:51
heikoblackburn, doesnt it have opencl bindings?14:51
heikoit is a gpu for sure14:51
blackburnheiko: kind of useless14:52
blackburnheiko: but we need gpu desperately - can't say anything against it14:52
n4nd0isn't any graphics card a GPU?14:52
heikoblackburn, so what about the argument, that all the people were very interested in it14:52
heikon4nd0 I think yes14:52
blackburnn4nd0: well sure but compare some nvidia board in big computer14:52
heikobut cuda does only work on nvidia14:52
blackburnand hd4000 in my notebook :D14:52
n4nd0blackburn: sure14:53
-!- SCV__ [~FSCV@] has joined #shogun14:53
-!- FSCV_ [~FSCV@] has quit [Ping timeout: 255 seconds]14:53
blackburnheiko: I am not against that14:54
heikohave a look, it also depends whether the guy is willing to port it to shogun (requires at least some work)14:55
blackburnheiko: we all need new technologies to have more freedom14:55
heikoblackburn, indeed14:56
n4nd0blackburn: you made me think when you said the nvidia thing14:56
n4nd0I assumed that CUDA must be used if one wants to use GPU14:56
blackburnn4nd0: no there is opencl14:57
blackburnwhich is powerful as well14:57
n4nd0I see14:57
blackburnand *open* :)14:57
blackburnI don't like to be stallman but still14:58
n4nd0from wikipedia14:58
n4nd0"if the OpenCL implementation is correctly tweaked to suit the target architecture, it performs no worse than CUDA"14:58
n4nd0it sounds like CUDA tends to perform better :)14:58
blackburnn4nd0: yes that's natural14:59
n4nd0but limited to nvidia14:59
n4nd0yes, I continued reading and it's true14:59
blackburnyou pay for generality14:59
heikocuda is completely closed14:59
heikothe used to give away source code14:59
heikobut not anymore14:59
heikothe guy just told me14:59
heikoso this is a bad thing in general, I agree with blackburn14:59
blackburnI can't run cuda on my machine14:59
heikoblackburn, the guy also talked about opencl, seems to be not too complicated to port it from cude to opencl15:00
n4nd0how different is to program in OpenCL and CUDA?15:00
heikomaybe we can convince him15:00
blackburnquite different I think - it depends on the complexity15:00
heikon4nd0, I think its very similar, though opencl may not have all the features15:00
heikoblackburn, lol :)15:00
blackburnif it is matrix algorithm15:00
blackburnshoudl be similar I believe15:01
heikoyes, this is what I meant15:01
blackburnin case of something with kernels - it could get worse15:01
heikokernels on gpus?15:01
heikothis is crazy :)15:01
blackburnheiko: gpu kernels15:01
heikohaha :)15:01
heikoI see15:01
blackburnyou can tailor some code to run on gpu15:01
blackburnusing C15:01
n4nd0blackburn: both cuda and opencl use the concept of kernels right?15:01
blackburnI think so n4nd015:01
heikoMetropolis-Hasting-Kernel, Gaussian Kernel, Covariance Kernek, GPU-kernel, Linux-kernel, Kernel of a linear map15:02
n4nd0haha people seem to like that word15:02
blackburnheiko: epanechnikov kernel15:02
blackburnheiko: in density estimation15:02
heikothat sound too Russian for me :)15:02
heikobb in 5 mins15:02
blackburnheiko: it is15:02
n4nd0blackburn: AFAIK you need kernels to use the GPU at least in CUDA15:03
blackburnn4nd0: yes but you don't have to write complex stuff if you do something with matrices I think15:03
blackburnn4nd0: however as I see you'd need to write kernel anyway yes15:04
blackburnthere is an idea that keep me nervous15:05
n4nd0which one?15:05
blackburnI am curious if it is possible to write algorithms using some 'workers'15:05
n4nd0blackburn: tell me more :)15:06
blackburnso you want to train svm which means you run a few workers that do that15:06
blackburnthey should be kind of actors or os15:06
blackburnthat idea is so simple but I can't understand how to do that in general15:06
n4nd0workers are to be run in parallel I understand15:06
blackburnnot only parallel but also distributed15:06
blackburnproablby mahout works somehow like that15:07
n4nd0what is it that you don't understand how to make it general?15:07
blackburnwould be so cool to have something like general principle of doing that15:08
blackburnprobably I am talking about map-reduce :D15:09
n4nd0blackburn: hehe maybe15:09
blackburnI wouldn't like to write map-reduce code - it is mahout stuff15:09
n4nd0blackburn: you can use frameworks like hadoop for the map/rduce part15:10
heikobtw I advertised the workshop a bit here15:10
blackburnn4nd0: that's what mahout does15:10
heikofew people said they might come15:10
n4nd0heiko: that's cool :)15:10
n4nd0blackburn: true15:11
heikoall the kernel people are here in London, so we might already have two of them :) I will also harass the others15:11
n4nd0heiko: what kernel are we talking about now? :D15:11
heikohaha :D15:11
blackburnso, heiko, all mercer cult fellows?15:12
heikoblackburn, yes, more the RKHS guys with characteristic kernels, distribution embeddings, kernels on MDPs15:12
blackburnhare krishna hare RKHS15:12
heikoI will also invite John-Shawe-Taylor  :)15:12
heikobut he is ever busy15:12
blackburnno idea who it is15:13
heikomade them popular for ML15:13
heikoblackburn, I also have some other ideas on things to integrate into shogun15:14
heikoone being the computation of determinants of very large matrices15:14
blackburnheiko: what about this year gsoc?15:14
heikoblackburn, what do you mean?15:14
blackburnheiko: student/mentor? :)15:14
heikoblackburn, I can do a mentor this year15:14
blackburnyou probably have two choices only15:14
heikosince I have a job finally :)15:15
blackburnyeah same here15:15
heikoCould also implement stuff myself, but maybe its less stressful to tell others what to do15:15
blackburngsoc have the same monetary value for me now15:15
blackburnso I wouldn't like to left my job because of gsoc :)15:16
heikoyeah, I mean are you even a student now?15:16
blackburnheiko: I am15:16
heikoI am not, would have to become one in order to participate15:16
heikooh, ok15:16
heikohopefully we can manage the gsoc this year15:16
blackburnand will be for an year more after that15:16
heikobut I will have a bit more time from spring on15:16
blackburnI could participate twice more15:16
n4nd0heiko: but you are phd student now right?15:16
heikon4nd0, not yet15:17
heikohopefully in sept15:17
n4nd0heiko: aaaah ok15:17
heikojust had my interview last week15:17
n4nd0oh, I wish you good luck with the selection then!15:17
heikothanks :)15:17
n4nd0aiming at UCL?15:17
blackburnheiko: I will announce gsoc initiative tonight15:18
heikobut its quite competetive, the alternative is here:
heikoblackburn, initiative? like ideas etc?15:18
heikowhen do we have to apply?15:18
blackburnheiko: gsoc is not yet announced :)15:18
heikon4nd0, how are your studies going?15:18
heikoblackburn, so what will you announce?15:19
blackburnheiko: that we are going to apply15:19
heikoisn't it better to wait until it is announced?15:19
n4nd0heiko: fine, I started this week my master thesis at KTH finally15:19
blackburnheiko: sonney2k doesn't think so15:19
heikon4nd0, nice master in? CS?15:19
heikoblackburn, well ok, actually I would agree15:19
n4nd0heiko: the plan is to be done with the few things I have left for Spain by May-June and finish the master thesis by August-September15:20
heikosince we are currently slow with everything15:20
blackburnheiko: but I am checking for mentorship  already15:20
n4nd0heiko: half way between CS and EE, it is a robotics master15:20
heikon4nd0, very cool!15:20
n4nd0heiko: but I do CS stuff15:20
heikoyou have to show some cool robot videos15:20
n4nd0heiko: I have something from a course I did a couple of years ago15:20
heikon4nd0, are you using your special svms somewhere in that?15:21
n4nd0heiko: no hehe15:21
blackburnheiko: I wanted to talk about GPs :D15:21
blackburnheiko: just in a few words - why?15:21
blackburnwhat makes it special?15:21
heikon4nd0, show me, I also have one cool video of an AR-drone15:21
heikoblackburn, what?15:21
blackburnheiko: why should I choose GP and when?15:21
n4nd0heiko: this is very crappy compared to a drone hehe15:22
n4nd0but we built it from scratch15:22
blackburnn4nd0: you never showed that to me - it was a mistake I married you!15:22
heikon4nd0, cool! :)15:22
n4nd0blackburn: haha15:22
n4nd0I think I had posted that video around here before!15:22
n4nd0heiko: this is the lab where I am at KTH
heikocool stuff15:23
n4nd0I want to become a phd here too15:23
heikoblackburn, well you can use them to model complex functions15:24
heikowithin a probabilistic framework15:24
heikoso quite flexible, they are used everywhere15:24
heikoand fully Bayesian, which is sometimes very useful15:24
blackburnheiko: my gf is working with hyperspectral images (d ~80) - should I recommend to try GPs?15:25
blackburntask is classification mostly15:25
heikowhat is the goal?15:25
blackburnheiko: top accuracy I think15:25
heikoto obtain high-accuracy predictions?15:25
heikothen better use an SVM15:25
heikoalthough they are similar15:25
blackburnbut training samples are very noised15:25
heikobut if you are interested in predictive distributions, Bayesian kernel selecton or similar, use GPs15:25
blackburnheiko: so it is still kind of theoretical?15:26
heikoblackburn, no, just depends on what you want15:26
heikopredictive distributions can be very useful15:27
blackburnheiko: how can one use it?15:27
heikosince you for example have probabilities for multiclass problems15:27
heikovery useful15:27
heikobut not so much for binary decisions15:27
heikoand kernel selection also, you can have a posterior distribution over Gaussian kernel sizes in every dimension for example15:27
blackburnheiko: predictions are multiclass in that case (water, asphalt, ...)15:28
heikoblackburn, use svms, much easier if you just want to predict :)15:29
blackburnheiko: well her supervisor wants to promote some method so they use it yet15:29
heikoblackburn, probabilitites can be useful15:29
heikosince you can add a decision "dont know"15:29
blackburnyes that's clear15:29
n4nd0heiko: wow the drone is following his face?15:29
heikon4nd0, yes15:30
n4nd0heiko: that's pretty cool15:30
heikousing opencv15:30
heikothis ar drone is really cool15:30
blackburnheiko: is it you on the video with long hair? :)15:30
heikoyeh with the red trousers, thats me :)15:30
heikoabout 1,5 yrs ago15:31
blackburnlooks pink :D15:31
heikonow I look more exhausted ;)15:31
blackburnheiko: even more?15:31
blackburnon the 0006 it detected your groin heiko15:32
heikothat was a cool thing, this printer company, oce, bought us a few of those drones and sponsored us playing with them for one week - in holland15:32
n4nd0oh that's nice!15:32
blackburnheiko: did you implement that?15:33
blackburnit looks like you were kinda presenting it15:33
heikoblackburn, yes15:34
heikoused opencv to track the face15:34
heikothen generated movement commands from that15:34
blackburnheiko: typical VJ stuff?15:34
heikowhere the difficulty was to estimate the distance to the face with one camera15:34
heikosince face size is not linear to distance15:34
blackburnheiko: yes d^215:34
heikodid a power-regression15:34
heikoblackburn, no interestingly not15:35
heikoanother function15:35
blackburnhmm yes sure15:35
blackburnI am wrong15:35
heikoand with that we got a +-10cm estimate15:35
blackburnit looks like something else15:35
heikowhich can be used to move back and forth15:35
blackburnI can imagine a plot of that function15:35
alexlovesdatahey n4ndo ... you are also a computer vision guy ...15:35
heikovery simple, but looks cool15:35
blackburnwell when it comes to say 5 meters15:35
blackburn5.5 and 4.5 - no matter15:35
blackburnsize won't change at all :D15:36
heikoblackburn, yes also the camera resolution is horrible as you can see15:36
blackburnalexlovesdata: what to do if not CV15:36
heikobut our "data" (measured distance by hand) was explained with almost no error15:36
heikoguys, I gotta do some work now, see you later!15:36
n4nd0alexlovesdata: I am more into robotics localization, mapping, SLAM and that stuff15:37
n4nd0but yeah there is lot of CV used in that too15:37
n4nd0kinect sensors, kinect sensors eveywhere :)15:37
@sonney2kalexlovesdata, I just git pull'd and guess what I still have the issue plus a couple more due to blackburn's arpack removal15:51
@sonney2kand fibonaccyheap and convertree15:52
blackburnsonney2k: ?15:52
n4nd0what happens with them?15:52
blackburnsonney2k: 2 mins15:52
alexlovesdatayou still have that assertion error, sonney2k ??15:54
alexlovesdatawhat accuracy is printed before the assert fails?15:54
alexlovesdataI get a 0.94 ...15:54
shogun-notifier-shogun: Sergey Lisitsyn :master * 7d0a59b / examples/undocumented/libshogun/ (3 files):
shogun-notifier-shogun: Removed fibheap and arpack examples15:55
@sonney2kalexlovesdata, 0.7615:57
@sonney2kblackburn, btw why did you remove arpack / fibheap?15:57
blackburnsonney2k: why are not needed anymore15:58
blackburnsonney2k: I strongly don't expect we use it somewhere out of tapkee15:58
@sonney2kI see ok15:59
blackburnso the only reason they appeared in shogun was dimred15:59
blackburnfibonacci heap stayed in tapkee15:59
blackburnarpack wrapper was replaced with better one15:59
wikingok here comes a semi-working google unit tester15:59
@sonney2kalexlovesdata, did you git pull and have > 0.9 accuracy?15:59
@sonney2kalexlovesdata, look the buildbot is also not happy
wikingsonney2k: do we have a $() variable for ar, i.e. static linking16:01
n4nd0cu later people16:02
-!- n4nd0 [] has quit [Quit: leaving]16:03
@sonney2kwiking, hmmhh don't think so16:04
blackburnnot used anymore as well -  sonney2k16:05
shogun-notifier-shogun: Sergey Lisitsyn :master * 9c17a1d / examples/undocumented/libshogun/ (2 files):
shogun-notifier-shogun: Removed covertree example16:06
@sonney2kwiking, ar is hardcoded - just found it line 576 in Makefile.template16:07
@sonney2kwiking, feel free to introduce a variable for it16:07
blackburnsonney2k: would you like if we put shared_ptr from boost to our lib?16:07
@sonney2kblackburn, haha have you hear yandex is more popular than bing :D16:08
blackburnsonney2k: actually it looks like changing internal (class field) pointers to shared_ptrs is not that difficult16:08
@sonney2kblackburn, this is something we cannot do inside gsoc16:08
blackburnsonney2k: no, really? bing sucks I knew that16:09
blackburnbut I don't use yandex at all16:09
@sonney2kI am afraid this might be much more work16:09
blackburnsonney2k: it could introduce bugs - that's I am more afraid of - but I changed one field in CKernel to shared_ptr and just had to change a few casts in siblings16:10
blackburnand getter/setter16:10
-!- SCV__ [~FSCV@] has quit [Quit: Leaving]16:10
@sonney2kand I would not want to risk to get the same chaos we had last gsoc when we had all these transitions not yet done16:10
@sonney2kif we want to do such stuff then directly after gsoc16:10
@sonney2kblackburn, hope that is OK... I recall that you became quite desperate last summer when nothing worked...16:11
blackburnsonney2k: I don't remember that exactly16:11
@sonney2kblackburn, did we have chatlogs back then? I certainly remember...16:12
blackburnsonney2k: well I am desperate 90% of time ;D16:12
-!- FSCV [~FSCV@] has joined #shogun16:12
blackburnbut yes I agree it is not the time to do that16:13
-!- travis-ci [] has joined #shogun16:14
travis-ci[travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []16:14
-!- alexlovesdata [~binder@2001:638:806:e001:eda1:c6f9:301a:5329] has left #shogun []16:16
wikingmmm this is a shit16:36
wikinggtest is as a source in ubuntu16:36
wikingbut not gmock :S16:36
blackburnwiking: yes but it was known before?16:43
wikingblackburn: well yes and no.... :)16:59
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has joined #shogun17:36
wikingok i think this is more or less done17:37
wikingone will have to have google mock source as well at hand to use it17:38
wikingthe package in ubuntu is faulty17:38
shogun-notifier-shogun: vigsterkr :master * 0c0581e / src/configure,tests/unit/Makefile:
shogun-notifier-shogun: Fix google mocking framework detection17:43
shogun-notifier-shogun: Google unit testing and mocking framework has to be compiled17:43
shogun-notifier-shogun: with the same compiler flags as the code you are testing.17:43
shogun-notifier-shogun: Viktor Gal :master * 24a7529 / src/configure,tests/unit/Makefile:
shogun-notifier-shogun: Merge pull request #860 from vigsterkr/utest17:43
shogun-notifier-shogun: Fix google mocking framework detection17:43
wikingsomeone shold test is17:44
wikingshoudl test this as well17:44
wikingworks for me on ubuntu LTS 12.x17:44
-!- travis-ci [] has joined #shogun17:56
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []17:56
-!- heiko [] has quit [Quit: Leaving.]18:05
-!- heiko [] has joined #shogun18:05
wikingheiko: there you go18:34
wikingheiko: unit testing with gmock18:34
heikowiking, the unit test thing?18:34
heikowiking, you are awesome!18:34
heikoI will use that for the stuff I am currently doing, after figuring out how18:34
wikingheiko: ok18:34
heikohave some time on the weekend, nice work18:34
wikingheiko: note that you'lll need google mock source... the package in ubuntu is not good18:35
heikojust the source?18:35
wikingand of course gtest src as well18:35
wikingbut for that the ubuntu package is good18:35
heikowill try out soon18:36
heikothis will help us a lot18:36
heikoonce it works fine, we can only accept commits if they include some tests18:36
heikoso we wont have to rely on those big picture tests in the c++ examples18:36
blackburnwiking: I got may be stupid question18:37
wikingblackburn: go ahead18:38
blackburnokay we have latent svm18:38
blackburnwhat is the difference between it and sliding window detector?18:38
blackburndo I understand right it is the same at some point?18:38
blackburnit is rather a matter of training18:39
wikingwell the detection is quite similar18:47
wikingit's just rahter yeah the training18:47
blackburnwiking: have you seen that paper?
blackburnwiking: pedestrian detection at 45 fps18:48
shogun-buildbotbuild #777 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Viktor Gal <>18:55
-!- heiko [] has left #shogun []19:09
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:18
shogun-notifier-shogun: Soeren Sonnenburg :master * 159ddb0 / / (4 files):
shogun-notifier-shogun: fix miscelleaneous erros/warnings19:40
shogun-notifier-shogun: - fix example compiler error19:40
shogun-notifier-shogun: - fix doxygen warnings19:40
shogun-notifier-shogun: - fix compiler flag19:40
@sonney2kwiking, which packages does one need to install for gmock/gtest?19:43
@sonney2klibgtest-dev and google-mock?19:43
wikingsonney2k: apt-get install libgtest-dev19:45
wikingand download gmock src19:45
wikingas the one in ubuntu is broken (does not provide sources under /usr/src)19:46
wikingand then just19:46
@sonney2kit is there /usr/src/gmock/src19:46
wikingis it?19:46
@sonney2kdebian here though19:46
wikingah ok19:46
wikingubuntu does not give have that19:46
wikingbut yeah19:46
shogun-buildbotbuild #778 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at
wikingby default it treis to find it under /usr/src/gmock19:47
wikingso then it should be fine for you19:47
@sonney2kg++ -I/usr/src/gmock -c /usr/src/gmock/src/gmock-all.cc19:47
wikingdon't do that19:47
wikingjust cd tests/unit19:47
@sonney2kg++ -I/usr/src/gtest -c /usr/src/gtest/src/gtest-all.cc19:47
wikingafter ./configure19:47
@sonney2kwiking, no I mean it fails to find things on configure19:47
wikingwhat does it say19:48
wikingi mean configure19:48
wikingChecking for Google C++ Mocking Framework ... yes19:49
wikingif it's not yes then plz copy-paste here the error line19:49
blackburncheck da announcement19:51
blackburnat mailing list19:54
@sonney2kwiking, it just says20:00
@sonney2k============ Checking for Google C++ Mocking Framework ============20:01
@sonney2kResult is: no20:01
@sonney2kyou could have made it a little more verbose :-]20:01
@sonney2kwiking, can I split it up in separate gmock/ gtest tests?20:04
@sonney2kwiking, hmmhh afk? then let me just do it20:05
blackburn hah google can't suggest another theme this time20:05
@sonney2kblackburn, so they are running out of money?20:06
blackburnsonney2k: that's a great implication out of my mind20:07
blackburntoo complex for me :)20:07
blackburnmay be it feels ok for them to keep old graphics20:07
* sonney2k continous the gmock/gtest endeavor20:07
blackburnsonney2k: we should name shogun versions by bacterias20:08
blackburnsonney2k: escherichia coli20:08
@sonney2kblackburn, or virii20:08
@sonney2kbut only the very bad bad ones20:08
blackburnsonney2k: yes, may be 3.0 should be called HIV20:09
@sonney2kblackburn, when we introduce self modifying codes and template meta programming then it surely will be!20:09
blackburnsonney2k: we should take some bacteria branch with increasing complexity20:10
blackburnand then put next name each time we release :)20:10
wikingsonney2k: i've tried that20:16
wikingsonney2k: i mean splitting20:16
@sonney2kwiking, give a a few more minutes...20:17
wikingsonney2k: but one without the ohter is useless20:17
wikingsonney2k: but not in configure.log20:17
wikingsonney2k: the simple output of configure20:17
wikingsonney2k: as there's more info where it actually died20:17
wikingsonney2k: ah ok i see it now20:19
wikingi forgot to print an error20:19
shogun-notifier-shogun: Soeren Sonnenburg :master * 77862dd / src/configure:
shogun-notifier-shogun: hopefully fix gmock/gtest detection20:25
@sonney2kwiking, please try if it works ^20:25
@sonney2kwiking, I think we might want to make this more robust and actually run a cxx_check compiling something like g++ -I/usr/src/gtest -c /usr/src/gtest/src/gtest-all.cc20:27
@sonney2kblackburn, nice announcement :)20:29
wikingsonney2k: well20:30
@sonney2kblackburn, now we only need the ideas url20:30
@sonney2kwiking, ?20:30
blackburnsonney2k: I'll add one once we get some20:30
@sonney2kblackburn, just copy it from last year for now20:30
wikingsonney2k: works20:30
@sonney2krecall everyone can edit now so it is a good start20:31
@sonney2kin particular the optimization thing is still valid20:31
@sonney2kand I want someone to code examples and write d3js / django interactive code20:31
@sonney2kfor some show down20:31
@sonney2klet me ask Mr. ICA20:33
blackburnsonney2k: please cc me if that's ok20:33
@sonney2kif he would be interested in getting ICA in20:33
blackburnsonney2k: ICA is pretty small20:34
blackburneven fastICA is not that much code20:34
@sonney2kthere are tons of algorithms for that20:34
@sonney2ka. ziehe - tdsep20:35
blackburnsonney2k: haha I didn't expect you write in that lang20:40
@sonney2kblackburn, I can hardly talk in $OTHERLANG to a friend :D20:41
blackburnsonney2k: google translate helped me a little20:41
@sonney2kblackburn, lets hope AZ is in. he did some cool stuff on that matter and has some nice examples too!20:45
blackburnsonney2k: nice20:46
@sonney2kblackburn, if I wasn't afraid of multiboost I would ask them20:46
blackburnsonney2k: why you are afraid?20:46
@sonney2kmission impossible20:46
@sonney2kwe also need some small ideas to be implemented for potential candidates...20:47
@sonney2kwiking, ok nice20:47
@sonney2kwiking, how do I run a test now?20:48
blackburnsonney2k: what is impossible in multiboost?20:49
wikingsonney2k: in ./tests/unit run make20:50
-!- n4nd0 [] has joined #shogun20:50
blackburnsonney2k: oh gosh that's a list!20:54
blackburnby gunnar20:54
blackburnthat's THE LIST20:54
@sonney2knotice the peak?20:54
@sonney2kat ~201020:54
blackburnof code?20:55
@sonney2kthat is when I merged and removed their code20:55
blackburnof commits?20:55
@sonney2kblackburn, which list?20:55
@sonney2kboosting list?20:55
blackburnsonney2k: gunnar forwarded my mail20:55
@sonney2kheh :)20:56
wikingsonney2k: works?20:58
wikingsonney2k: should get some errors21:00
blackburnsonney2k: 'das Programm announced noch' is that german or you switched lang accidentally?21:03
blackburnI am just curious21:03
@sonney2kblackburn, my german is ruined :)21:05
blackburnsonney2k: I feel inspired when trying to read deutsch21:06
blackburnlooks so fancy :)21:07
blackburnnoch wissen! wir ob wir! aus berline!21:07
blackburnmachinnes lernen21:07
blackburnUn?berwachtes Lernen!21:09
blackburnoh correct is maschinelles lernen21:10
n4nd0wiking: do you think gtest/gmock should work fine here?21:17
n4nd0I can test it if so21:17
shogun-notifier-shogun: Sergey Lisitsyn :master * 55a234d / src/shogun/kernel/string/LocalAlignmentStringKernel.cpp:
shogun-notifier-shogun: Removed shameful not stable mark from Local Alignment string kernel21:23
@sonney2kn4nd0, please try21:25
n4nd0sonney2k: ok21:26
n4nd0just need to configure for it and build I guess21:27
@sonney2kblackburn, n4nd0, wiking - what does ./evaluation_cross_validation_multiclass_mkl give you when you run it?21:27
blackburnoh well let me compile latest21:27
blackburnsonney2k: was ->
@sonney2kwiking, can you please fix the 2 failing gtests ? I want to enable them in the builds!21:31
@sonney2kbuildbots taht is21:31
@sonney2kotherwise we will never have them21:31
@sonney2kblackburn, *lol*21:32
blackburnsonney2k: ->
blackburnwell anyway the best is that:21:39
blackburnthey asked - photoshop me please to not look drunk21:39
wikingn4nd0: try21:47
wikingsonney2k: btw make unit-tests in ./shogun/src should work as well21:48
blackburnsonney2k: I can't compile examples here21:49
blackburnyeah your thing with ;21:50
blackburnsonney2k: why did you do that? it is better to have ;21:50
@sonney2kwiking, there is no make unit-tests?21:51
@sonney2kblackburn, was not possible21:52
blackburnsonney2k: mean assert failed here21:52
blackburnhah well setting seed won't work21:52
@sonney2kblackburn, issue is that when you have a more complex macro that includes if (cond) { sth; }21:52
blackburnit is not machine independent21:52
n4nd0wiking: moment please21:53
blackburnsonney2k: ah you added unlikely21:54
@sonney2kblackburn, well it fails everywhere I tried21:54
@sonney2kso I wonder if someone here gets a 0.921:54
@sonney2kn4nd0, you still have a 32bit system?21:54
n4nd0sonney2k: I am using 64 OS21:55
@sonney2kblackburn, yeah and I need these extra { } brackets21:55
blackburnsonney2k: you could put it to { } block21:55
@sonney2kso only fix would be to change coding style to require21:55
@sonney2kif ()21:55
blackburnthen ; would work21:55
blackburnah it is already in21:55
@sonney2kblackburn, sure long tiem21:56
blackburnso { ... }; is not possible?21:56
@sonney2kit is but not21:56
@sonney2kif () {}; else {}21:56
@sonney2kthe else is then out of context21:56
blackburnno I mean21:56
@sonney2kbut if we had if () { {}; } else {} all good21:57
blackburnokay got it21:57
@sonney2kso that is why maybe we should change it to using {} always21:57
blackburnmacro w/o ; in the end is bad21:57
@sonney2kblackburn, log4cxx does that too...21:57
blackburnsonney2k: if we get igor we could get some factorizations/sparse stuff21:59
blackburnI don't give a fuck what they are doing but sounds magical21:59
@sonney2kblackburn, watch your language...22:01
blackburnsonney2k: are you getting too old for such language? :)22:02
blackburnanyway sonney2k not only ICA but some NMF or so would be nice22:03
blackburnit is really cool as we are now learning features not only classifiers22:03
blackburndesperately need group lasso specialist22:09
@sonney2kwiking, haha you unit tests found an error in SGVector::max - the function you introduced utilizing atlas to find the max :)22:15
@sonney2kall hail unit tests!22:19
blackburnsonney2k: does shogun pass unit tests?22:19
@sonney2kyou mean the 8 we have :D22:20
blackburnyes still22:20
@sonney2k6/8 currently22:20
@sonney2kbut I think I fixed the max issue now22:20
blackburnsonney2k: we need code coverage computation22:21
blackburnto understand how it is covered22:21
@sonney2k0.000001 % :D22:21
blackburnit is so nice for python but I don't know about C++22:21
blackburnbut if we see now it is 0.1% we are happy22:21
blackburnsonney2k: lol, gcov22:23
blackburnit is already here22:23
@sonney2kso what is it?22:23
blackburnsomething gnu to compute coverage22:23
wikingsonney2k: :D22:25
@sonney2klooks like this is more like grof22:25
@sonney2kalright 1 test fails now :)22:25
blackburnyeah I don't understand how to use it22:26
@sonney2kwiking, good boy, bad boy!22:26
wikingsonney2k: well then good that we have unit test finally :P22:26
blackburnwiking: any idea?22:26
wikingsonney2k: heheh i wonder what was happening there with the max thing :)22:26
blackburnabout coverage22:26
wikingblackburn: about.... sorry i was showering till now22:26
@sonney2kwiking, yeah we need them with programmers like you ;-)22:26
* wiking hides22:26
wikingsonney2k: but actually what happened in that max calculation?22:26
blackburnwe need to know what is tested and what is not22:26
wikingblackburn: ah that's easy22:27
blackburnwiking: how?22:27
wikingblackburn: almost nothing is tested :D22:27
blackburnwiking: I will find you22:27
wikingblackburn: and kill me? :)22:27
wikingblackburn: ok you want it per object or per object per function?22:27
blackburnI want it somehow22:27
wikingbecause the per obj is quite easy22:28
wikingas you can see for each .cc you need a _unittest.cc22:28
wikingso that's quite easy22:28
blackburnno that's not interesting22:28
wikingbut i reckon that's not so nice22:28
wikingi think we need an obj->func coverage22:29
blackburnI found that gconv can be used with google test22:29
@sonney2kwiking, the cblas_idamax compute argmax_i(abs(x[i]))22:29
wikingblackburn: mmm22:30
@sonney2kwiking, but we used it to compute max22:30
blackburnblas is pure evil22:30
@sonney2kdangerous for negative numbers...22:30
wikingsonney2k: i was totaly convinced from some doc that it's a simple max for double22:30
@sonney2kwiking, yeah but it is not22:31
blackburnIDAMAX finds the index of element having max. absolute value.22:31
blackburnwiking: never trust anything but netlib22:31
wikingblackburn: yep22:31
wikingi can see that now22:31
wikingok well it's good for abs_max :)22:31
wikingif we'd have that22:31
wikingbtw: blackburn well i guess we could use that22:31
blackburnwiking: yes, I'd like to try22:32
wikingblackburn: but i guess some simple ctags hack could do it as well no?22:32
@sonney2kinterestingly we use it in arg_max too22:32
@sonney2kbut we have no test for that that could fail22:32
* sonney2k runs away screaming22:32
blackburnsonney2k: well at least we can be sure iranian missile will fail22:32
@sonney2kblackburn, sure of that? anything south of the equator will be mapped to north of it22:33
@sonney2ksame east of  the null meridian -> west22:34
@sonney2kwest <-> east22:34
@sonney2kscary no?22:34
blackburnsonney2k: hmm22:34
blackburninteresting what it will reach then22:34
blackburnhave you all seen new iranian aircraft?22:35
@sonney2kindeed some drones attempting to land on berlins new airport ;^)22:35
wikingblackburn: the plastic mockup?22:36
blackburnwiking: !22:36
wikingno i haven't seen it :D22:36
wikingblackburn: well let's try that gconv thingy then22:37
blackburnwiking: I am trying it now22:37
blackburnwiking: sonney2k: this way we can accept any patches that increase coverage22:42
@sonney2kblackburn, so you mean tasks for gsoc applicants?22:42
@sonney2khow demotivating :/22:43
blackburnwell may be22:43
@sonney2kporting examples over from python_modualr to other $LANG would also be valuable22:44
blackburnsonney2k: isn't that demotivating?22:44
@sonney2kblackburn, at least you see that stuff runs some real thing22:45
@sonney2kand maybe even with fance gfx22:45
@sonney2kohh boy23:11
@sonney2klook svm ocas is working again...23:11
@sonney2kall due to that isamax thing23:11
wikingsonney2k: \o/23:13
wikingsonney2k: megafail by me23:13
n4nd0wiking, sonney2k : sorry, I had an issue with my python, but now it is fine :)23:13
n4nd0wiking: Checking for Google C++ Testing Framework ... unit testing framework not detected, for unit testing please install it23:13
wikingbut seems not many people use shogun as nobody complained23:13
n4nd0Checking for Google C++ Mocking Framework ... unit testing framework not detected, for unit testing please install it23:13
n4nd0let me check configure too23:13
wikingn4nd0: ok i guess it cannot find the src dir for either of those libraries23:14
wikingit tries to find it under /usr/src/{gtest,gmock} by default23:14
wikingif the src is somewhere else you have to specify it for the ./configure by --gmock=PATH --gtest=PATH23:15
shogun-notifier-shogun: Soeren Sonnenburg :master * 0017902 / / (5 files):
shogun-notifier-shogun: fix argmax/max in sgvector23:15
shogun-notifier-shogun: for float/double: isamax computes the argmax of abs(vector)23:15
shogun-notifier-shogun: not the max! introduce new max_abs / arg_max_abs functions for that and23:15
shogun-notifier-shogun: add tests for these too.23:15
n4nd0wiking: /usr/src/gtest is there23:16
wikingn4nd0: where's your gmock src?23:16
wikingif you dont have that you'll have to download it23:16
wikingn4nd0: but still i wonder why it cannot find gtest src23:17
n4nd0wiking: yeah that was weird23:17
n4nd0gmock is in...23:17
wikingn4nd0: but the src?23:18
wikingbecause those are only the headers23:18
wikingn4nd0: but still need to figure out why gtest fails for u23:19
wikingn4nd0: anything informative in the configure.log?23:20
n4nd0wiking: does gmock dir look fine?
n4nd0checking configure.log now23:20
wikingn4nd0: yeah headers looks great23:21
n4nd0wiking: nothing apart from what I pasted before, in the configure.log23:21
wikingbut you would need the src files as well:  gmock-spec-builders.cc23:21
wikingif u dont have those23:21
wikingdownload gmock src somewhere23:21
wikingand give the path for the configure script23:22
@sonney2kwiking, please install gmock on your buildslaves...23:22
wikingsonney2k: mmm letmecheck what freebsd has23:22
wikingsonney2k: i have gtest installed on ubuntu23:22
@sonney2kI hope I find the time to do it on all the rest and then we enable this thing23:22
@sonney2kbut for now23:22
wikingbut as said earlier on ubuntu23:22
wikinggmock package is fucked23:22
wikingso i cannot enable it :S23:23
blackburnwatch your language! :D23:23
@sonney2kwiking, then put the files manually somewhere23:23
blackburnah soeren is sleeping23:23
blackburnnow we can say bad words23:23
wikingsonney2k: /home/wiking/gmock-1.6.023:23
wikingsonney2k: for the ubuntu buildbot23:23
wikingsonney2k: but afaik you should not run make unit-tests there23:23
n4nd0wiking: I've got some python files in /usr/shar/gmock too23:23
wikingsonney2k: as its doing static analyzer there23:24
blackburn songs of sonney2k's youth23:24
n4nd0no clue of cc files though23:24
n4nd0wiking: ok gtest detected now23:25
wikingn4nd0: cool now i think u should download gmock for yourself23:27
wikingn4nd0: if the package doesn't have the src installed somewhere23:27
wikingi mean the c++ src23:27
n4nd0wiking: I have found the sources in a zip in AUR23:27
wikingof gmock that is23:27
n4nd0wiking: take a look at it23:29
wikingthat's what u need23:31
n4nd0wiking: complete install or just point in the configure to the .src files?23:32
wikingn4nd0: the former one... just point to the root dir of what you've got extracted from that zip23:33
wikingi mean latter23:33
wikingthe compilation will be done by shogun23:33
n4nd0wiking: let's try it out23:34
blackburnwhat do you guys think if we make shogun a browser?23:34
wikingif the ./configure finds both23:34
wikingblackburn: most def :)23:34
wikingn4nd0: just run make unit-tests23:34
wikingthat should both compile the shogun lib and try to execute the unit tests23:34
blackburnwe will get a lot of users by a matter of seconds23:34
blackburnpeople would like to see how ML library changed to browser23:35
wikingblackburn: i guess we'll compile it for all the mobile platforms as well :P23:35
n4nd0wiking: gmock no detected still23:36
wikingn4nd0: ok how do you run ./configure?23:36
n4nd0wiking: a sec I think I know what it might be23:36
n4nd0mmm nothing it didn't work out23:37
n4nd0so I have23:37
n4nd0oh shit I am idiot23:37
n4nd0I still had the /usr/include/gmock from before...23:37
wikingok just paste here your ./configure line :)23:37
n4nd0got it detected :)23:38
n4nd0./configure --disable-optimization --interfaces=python_modular --gtest=/usr/src/gtest/ --python=python2 --gmock=/usr/src/gmock23:38
wikingmm and you extracted your gmock zip to /usr/src/gmock ?23:38
n4nd0but it is working now23:39
n4nd0Checking for Google C++ Testing Framework ... yes23:39
n4nd0Checking for Google C++ Mocking Framework ... yes23:39
wikingheheh then you can just remove both --gtest=/usr/src/gtest/ and --gmock=/usr/src/gmock23:39
wikingas it's by default looking for it at that location23:39
n4nd0hehe I like to type :P23:39
wikingok well now type23:39
wikingmake unit-tests23:39
wikingand it should do the whole thing23:39
wikingand on the end see the unit test running23:40
n4nd0make unit-tests is a gmock thing?23:40
n4nd0make unit-tests23:40
n4nd0make: Nothing to be done for `unit-tests'.23:40
wikingi wonder23:40
wikingok then just do this23:40
wikingcd ../tests/unit && make23:40
n4nd0but I didn't make yet23:40
n4nd0maybe it needs to be done23:40
wikingdont worry about that23:41
wikingit should invoke automatically the compilation of shogun library23:41
n4nd0making in tests/unit ...23:41
wikingso that'll make everything that u need for unit testing23:41
wikingand then run the unit tests23:41
n4nd0wiking: successful make23:48
--- Log closed Sat Feb 09 00:00:43 2013