Open in new window / Try shogun cloud
--- Log opened Thu Apr 19 00:00:19 2012
blackburnn4nd0: and yours?00:00
n4nd0blackburn: databases, microwave engineering and wireless communications (dsp for wireless communications)00:01
n4nd0what do you study on dsp?00:01
blackburnn4nd0: FFT, random signals processing00:03
n4nd0I think that is kind of cool :)00:03
blackburnthat's always better than optics00:04
blackburn(actually everything is better than optics)00:04
blackburnto eat some dirt for example is way much better than do some optics I think :D00:04
blackburnn4nd0: ok see you00:17
n4nd0blackburn: good night!00:17
blackburngood night00:17
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has quit [Quit: Page closed]00:17
-!- blackburn [~qdrgsm@] has left #shogun []00:17
-!- emrecelikten [~emre@] has joined #shogun00:29
-!- av3ngr [av3ngr@nat/redhat/x-jczscjkexunpkdrx] has joined #shogun01:39
-!- PhilTillet [] has quit [Ping timeout: 260 seconds]01:51
-!- emrecelikten1 [~emre@] has joined #shogun02:10
-!- emrecelikten [~emre@] has quit [Ping timeout: 276 seconds]02:11
-!- emrecelikten1 [~emre@] has quit [Quit: Leaving.]02:18
-!- PhilTillet [~Philippe@] has joined #shogun02:20
-!- PhilTillet [~Philippe@] has quit [Read error: Connection reset by peer]02:37
-!- vikram360 [~vikram360@] has quit [Ping timeout: 252 seconds]03:01
-!- n4nd0 [] has joined #shogun07:35
CIA-64shogun: Sergey Lisitsyn master * rad2707e / (35 files in 4 dirs): Moved some libraries to separate folder -
CIA-64shogun: Soeren Sonnenburg master * r48b0dd1 / (35 files in 4 dirs):07:57
CIA-64shogun: Merge pull request #454 from lisitsyn/external_libs07:57
CIA-64shogun: move a number of classes not to be used from the outside to lib/external -
shogun-buildbotbuild #738 of libshogun is complete: Failure [failed compile]  Build details are at  blamelist: blackburn91@gmail.com08:02
CIA-64shogun: Soeren Sonnenburg master * r12a09b4 / (2 files in 2 dirs): move graphical regression_lars example to graphical -
CIA-64shogun: Soeren Sonnenburg master * r677d721 / (2 files): fix wrong include paths for libsvm/liblinear -
@sonney2kblackburn - hey look the new version of liblinear arrived
n4nd0sonney2k: goood morning!08:12
n4nd0I saw a bunch of new warning caused by cover tree stuff, mainly methods, args, etc not documented08:12
n4nd0I will try to fix them asap08:12
@sonney2kn4nd0, thanks08:13
-!- wiking [] has joined #shogun09:12
-!- wiking [] has quit [Changing host]09:12
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:12
CIA-64shogun: Soeren Sonnenburg master * rc58a034 / src/shogun/classifier/svm/LibLinear.cpp : fix tron path in liblinear -
CIA-64shogun: Soeren Sonnenburg master * r519d74a / (4 files in 4 dirs): fix various documentations and thus doxygen warnings -
shogun-buildbotbuild #742 of libshogun is complete: Success [build successful]  Build details are at
wikingi have a problem with the HEAD09:43
wiking../shogun/lib/JLCoverTree.h:11:10: fatal error: 'values.h' file not found09:43
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]09:49
-!- n4nd0 [] has quit [Quit: leaving]09:57
-!- xiangwang [~chatzilla@] has joined #shogun10:02
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has joined #shogun10:09
sonne|workwiking - this file does not exist - run git clean -dfx to clean up the mess :)10:14
shogun-buildbotbuild #497 of python_modular is complete: Success [build successful]  Build details are at
-!- PhilTillet [~android@] has joined #shogun10:36
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun10:56
blackburnsonne|work: wow why did you merge it?10:57
blackburn(I planned to complete it)10:58
blackburngsomix: n4nd0: any of you around?10:58
sonne|workblackburn: I thought you did10:59
sonne|workat least there was no indication that you didn't10:59
blackburnsonne|work: ok sorry I'll mention that next time :)10:59
blackburnsome stuff to go still10:59
blackburnssl.cpp etc11:00
sonne|workblackburn: btw have you seen liblinear :)11:00
sonne|workit has regression now11:00
blackburnsonne|work: yes checking right now11:00
blackburnsonne|work: do you think it is so useful?11:00
sonne|workI plan to merge it today11:00
sonne|workyes of course11:00
sonne|worksvr is pretty good11:00
blackburnsonne|work: will you?11:01
sonne|workit should be easy11:01
blackburnI mean I can11:01
blackburnif it is needed :)11:01
sonne|workafter all SVR is reduced to svm prolbme11:01
n4nd0blackburn: yeah I am here11:01
sonne|workyou have exams :D11:01
sonne|workso no time11:01
sonne|workbut I have a train ride :)11:01
blackburnsonne|work: yes I got 5 today for my database project11:01
sonne|workgood luck then11:02
blackburnsonne|work: damned optics tomorrow11:02
blackburnn4nd0: do you need more tasks?11:03
blackburnor some exam holidays? ;)11:03
n4nd0blackburn: mmm it sounds appealing but I have still stuff to finish in shogun and exams coming next week11:03
n4nd0blackburn: but tell me in any case, it is probably interesting11:04
blackburnn4nd0: I have nothing in mind yet but if it is needed I'll come up with something ;)11:04
blackburnsonne|work: can you guide me a little how to patch libocas to solve that11:05
blackburn||w||^2 + w0 w problem11:05
sonne|workblackburn: and n4nd0 should do this cool benchmarking problem :)11:07
sonne|workblackburn: ask vojtech directly11:07
blackburnsonne|work: covertree / qsort?11:08
blackburnsonne|work: ok I will11:08
sonne|workblackburn: faster way to compute distance11:08
blackburnsonne|work: upper bounded?11:08
blackburnbtw I have seen harshit was talking about huge dataset yesterday11:09
n4nd0he wants to try knn there11:09
blackburnit was 4000 vectors?11:09
n4nd0I understood that dimension of the vectors was 400011:09
blackburnghm what a huge dataset ;)11:09
sonne|workblackburn: no by doing ||x-y||^2 = ||x||^2 -2xy + ||y||^211:10
-!- av3ngr [av3ngr@nat/redhat/x-jczscjkexunpkdrx] has quit [Quit: That's all folks!]11:10
blackburnsonne|work: precompute xy?11:10
sonne|workx^Tx can be precomputed11:10
n4nd0the tests I am doing around every day with qsort / covertree are already bigger than 4000 vectors ;)11:10
sonne|workonly x^T y needs to be done11:10
sonne|workblackburn: that is what we use in gaussian kernel btw11:10
blackburnsonne|work: yes I know11:10
sonne|workquite a bit faster there I would assume11:10
blackburnsonne|work: hmmm if we assume vectors as normalized11:11
blackburnwe can use only dot there11:11
blackburnsounds like a good option to use for me11:11
n4nd0sonne|work, blackburn so that decomposition should make it faster than using cblas function for the norm?11:12
blackburnn4nd0: hmmmm11:13
-!- PhilTillet [~android@] has quit [Read error: Connection reset by peer]11:13
sonne|workn4nd0: a lot - because you save even computing x-y11:14
blackburnno need to compute square11:15
sonne|worksquare is not the problem - it is done once at the end11:16
sonne|worksqrt I meant11:16
sonne|workforget what I said11:17
n4nd0ok ...11:17
blackburnsonne|work: sqrt is not but (x-y)(x-y) ;)11:22
-!- PhilTillet [~android@] has joined #shogun11:22
-!- wiking [] has joined #shogun11:22
-!- wiking [] has quit [Changing host]11:22
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun11:22
sonne|workwiking: see the very above11:23
wikingok i have a fix11:23
sonne|workgit clean -dfx11:23
wikingmmm i've just got connected11:23
sonne|work(it will erase everything not in git from shogun)11:23
sonne|workwiking: so no need for a fix :D11:24
wikingjlcovertree is not part of shogun?11:24
wikingotherwise it won't compile on a BSD-like system11:25
wikingas it does not have <values.h>11:25
sonne|workwiking: I see11:26
sonne|workwhat is values used for at all?11:26
sonne|workI mean why do we need it?11:26
sonne|workn4nd0: any idea?11:26
wikinghehehe no idea :)11:26
wikingi've just sent you now the pull request11:26
sonne|workdoes it compile if you remove it?11:26
n4nd0sonne|work: I am checking11:26
blackburnwhat is values?11:27
blackburnsome crazy consts11:28
n4nd0it seems that the file values.h is nowhere11:28
blackburnmystery file11:29
wikingblackburn: yo11:29
blackburnwiking: hi11:29
wikingblackburn: what was so funny yesterday?11:29
blackburnwiking: your git hotline11:30
n4nd0I don't really know why my compilation didn't fail locally11:30
wikingblackburn: well you it's a bit funny ;)11:30
sonne|workn4nd0: linux11:30
sonne|workI think we should modify the code to work w/o vlaues11:30
wikingn4nd0: as said you have linux11:30
wikingsonne|work: or move values include under some base/ header11:30
blackburnHello, Git consulting, Viktor is on line11:31
wikinghahahah yeah11:31
wikingand to see how fucken nice i am11:31
wikingor unbiased11:31
wikingi'm even supporting my competition11:31
blackburnolympics spirit lol11:31
sonne|workwiking: I don't think anyone here wants your project :D11:31
blackburnsonne|work: huh! pull request is here11:31
wikingsonne|work: yes yes they want!11:32
wikingok i have a lil libqp fix as well11:32
wikinglet me add the qp branch to my repo11:32
sonne|workthese bastards11:32
wikingoh these mother!@#11:34
blackburnsonne|work: I also thought nobody wants my project11:34
wikingso i've added libqp_gsmo
wikingsince it's included in libqp.h11:34
wikingbut the problem now that i cannot cherry-pick the pull request11:35
blackburnwhere did you get that?11:35
wikingwhich what?11:35
wikingit's from libqp11:35
sonne|workwiking: it seems to compile here w/o values.h11:35
wikingbut it has never been added to shogun11:36
sonne|workwiking: can you try if you need it at all under osx?11:36
wikingsonne|work: ok just a sec11:36
wikingn4nd0: so where did u get this code?11:37
wikingor why did you do the include <values.h> ?11:37
blackburncan somebody send me an sms if chris arrives? :D11:38
n4nd0wiking: from there
wikingah ok11:38
wikingi wonder if it uses really any of those macros11:39
n4nd0I don't even know what is defined in values.h11:39
wikingmm i think it compiles for me as well... although i'm still waiting for finishing it11:40
wikingok yeah now it's already doing doxygen11:40
wikingso it was ok11:40
sonne|workwiking: ok then11:40
wikingsonne|work: yep compiles11:40
wikingsonne|work: can u remove then the header11:40
wikingand i'll merge my repo?11:40
sonne|workletst remove the header11:40
CIA-64shogun: Soeren Sonnenburg master * r000a882 / src/shogun/lib/JLCoverTree.h : remove values.h include -
wikingok cool11:41
wikingi'll try to merge now11:41
wikingok here's my pull request for libqp11:46
wikingdon't worry about the 3 commits... it only adds the file itself.. i had to do a merge in order to not to fuck up totally my repo11:46
shogun-buildbotbuild #744 of libshogun is complete: Failure [failed compile]  Build details are at  blamelist: sonne@debian.org11:50
sonne|workwiking: looks like you forgot the rebase
wikingso fuck11:51
wikingisn't maxfloat defined somewhere in shogun already?11:51
sonne|workgrep MAXFLOAT /usr/include/math.h11:52
wikingnot on my system11:52
wikingno ok it'll work as well here11:53
sonne|work# ifdef __USE_XOPEN11:53
sonne|work/* X/Open wants another strange constant.  */11:53
sonne|work#  define MAXFLOAT  3.40282347e+38F11:53
sonne|work# endif11:53
sonne|workhere hmmhh11:53
wikingit's just in another header that it's included in math.h11:53
sonne|workyeah but math is already included in that tree?!11:54
-!- al___ [4eecc9c5@gateway/web/freenode/ip.] has joined #shogun11:58
al___hi there11:58
al___compilation issue:11:58
al___In file included from classifier/KNN.cpp:18:0: ../shogun/lib/JLCoverTree.h: In function 'node<P> batch_create(shogun::v_array<T>)': ../shogun/lib/JLCoverTree.h:277:52: error: 'MAXFLOAT' was not declared in this scope ../shogun/lib/JLCoverTree.h: In function 'void batch_nearest_neighbor(const node<P>&, const node<P>&, shogun::v_array<shogun::v_array<T> >&)': ../shogun/lib/JLCoverTree.h:764:22: error: 'MAXFLOAT' was not 11:58
al___with lastest git code11:58
al___thanks for helping11:59
blackburnhahah guys that is really fast bugreport11:59
n4nd0so it looks like values.h was used for something :S12:00
al___:p indeed12:00
blackburnn4nd0: just replace with CMath::ALMOST_INFTY or so?12:00
al___will try it right now12:00
n4nd0blackburn: I don't have access right now to the code12:00
n4nd0blackburn: I am not in my workstation12:01
blackburnI see12:01
blackburnal___: yeah you may try as well :)12:01
al___../shogun/lib/JLCoverTree.h line 277 764 766 remplace MAXFLOAT with CMath::ALMOST_INFTY do the work12:05
n4nd0al___: you can do a patch with that there ;)12:05
al___It will be a pleasure, can you tell me how to do it ? :p12:06
n4nd0you can issue a pull request from github12:08
blackburnor create a patch and send it12:09
blackburnsonne|work: can you apply a patch?12:09
blackburnhowever PR is more preferrable I think12:10
al___I m little bit nex to github, I can read the open PR but how can I create a new one ?12:19
wikingsonne|work: you should have just taken my patch :>>12:19
blackburnal___: fork shogun - clone your fork - do some changes locally - push to your remote on github - create pull request using github interface12:21
wikingfuuuucking hell my shogun fork is fsck-ed12:26
al___blackburn: I got a RSA permission denied when cloning my fork shogun12:32
n4nd0al___: are you following any tuto in github?12:32
blackburntry to set up your git according to that link ^12:33
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection]12:39
-!- wiking [] has joined #shogun12:39
-!- wiking [] has quit [Changing host]12:39
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun12:39
-!- xiangwang [~chatzilla@] has quit [Ping timeout: 240 seconds]12:41
wikingok i think i have the solution12:46
wikingjust somebody help me to rebase my giiit!12:46
n4nd0wiking: are you kidding?12:48
wikingyep no12:48
n4nd0I don't have the source with me right now, sorry :(12:48
wikingn4nd0: no worries12:49
wikingi think we should use FLT_MAX12:49
wikingsince it's iso c standard12:49
wikingnot like MAXFLOAT12:49
wikingthat is xsi12:49
wikinganybody here on debian or has access to debian12:49
wikingthat can test this for me?12:49
wikingalthough according to the repo FLT_MAX has been used in other sources and since those compiles...12:50
blackburnwiking: why not CMath's constant?12:53
wikingbecause it's not really the flt_max12:53
blackburnwhy not to change CMath's stuff to real max?12:54
wikingwell it's not the same12:54
blackburnactually I am not used to work with e+38 distances and hardly can imagine that :)12:55
wikingafaik it's there because to check float overflow12:56
blackburnhmm I thought it is kind of maximal distance12:56
-!- xiangwang [~chatzilla@] has joined #shogun12:58
blackburndrum roll chris is arriving!12:58
wikingok anyhow FLT_MAX will do the work12:58
-!- ckwidmer [] has joined #shogun12:59
wikingblackburn: pull request is there13:01
blackburnwiking: I am not able to merge it right now anyway ;)13:01
blackburnwill do in a hour once I get home13:02
-!- xiangwang [~chatzilla@] has quit [Ping timeout: 260 seconds]13:09
CIA-64shogun: Soeren Sonnenburg master * r2132e4e / src/shogun/lib/JLCoverTree.h : change MAXFLOAT -> FLT_MAX -
sonne|workwiking: it is soo easy to make you happy :D13:11
wikingshould have taken my pull request13:11
wikingbecause this way now we have this13:11
sonne|workwiking: sry didn't see this13:11
wikingok i'll try now a rebase :DDD13:12
wikingit'll be horrible for sure13:12
wikingok wasn't that bad as i thought it would be13:12
sonne|workwiking: I guess it is about time to play with git reset --hard :D13:14
shogun-buildbotbuild #745 of libshogun is complete: Success [build successful]  Build details are at
wikingsonne|work: yeps13:16
sonne|workwiking about libqp - shouldn't you rather incorporate / take the whole beast at once?13:16
wikingsonne|work: ?13:16
sonne|workwhole lipqp?13:16
wikingyou  mean when i already have the wrappers?13:16
wikingi mean otherwise the whole libqp is there13:16
wikingi've checked the latest code of libqp13:16
wikingthe _splx in shogun was up to date13:16
wikingand the header as well13:16
sonne|workahh so the rest is up to date?13:17
sonne|workthen ok13:17
wikinglet me just fix again my repo13:17
sonne|workyeah one request only please13:17
wikingi'm just trying to fix it now13:17
sonne|workso git reset --hard should do and a pull with a rebase and push --force13:17
wikingcheck now the link13:18
wikingit's only 1 commit now13:18
CIA-64shogun: Viktor Gal master * rfc8fa5e / src/shogun/lib/external/libqp_gsmo.cpp :13:20
CIA-64shogun: Add libqp_gsmo implementation from libqp13:20
CIA-64shogun: As libqp.h contains the definition for libqp_gsmo_solver13:20
CIA-64shogun: the implementation of it should be included as well in shogun -
CIA-64shogun: Soeren Sonnenburg master * r582e0cc / src/shogun/lib/external/libqp_gsmo.cpp :13:20
CIA-64shogun: Merge pull request #470 from vigsterkr/qp13:20
CIA-64shogun: add libqp_gsmo_solver -
sonne|workthere we go13:20
sonne|worknow we should get pr_loqo.cpp (in shogun already) into shape to match libqp's interface13:21
-!- xiangwang [~chatzilla@] has joined #shogun13:21
wikingloll good that you are telling me this13:23
sonne|workand actually while we are at it also CTron13:23
blackburnsonne|work: I was like O_O when I saw predictor-corrector stuff in pr_loqo13:23
n4nd0wiking: so is it libq integration in shogun better now?13:26
wikingn4nd0: not even near to it13:26
n4nd0wiking: lot of stuff left to do still?13:26
wikingyeah well we need a descent c++ wrapper for all the qp solvers13:27
-!- xiangwang [~chatzilla@] has quit [Ping timeout: 252 seconds]13:27
sonne|workn4nd0: currently code got just copied in13:28
sonne|workno mor13:28
n4nd0I see13:29
-!- xiangwang [~chatzilla@] has joined #shogun13:30
wikingsonne|work: so by the way... is there any other qp solvers in shogun?13:30
wikingand shouldn't we move qp solvers out of classifiers?13:31
sonne|workwiking: that is what I said :)13:32
wikinghahahah where? :)13:32
sonne|workwell pr_loqo and tron13:33
sonne|workbut IMHO these should go into lipqp13:33
wikingin what sense13:33
wikingi mean libqp is just as is13:33
wikinglet's separate libqp and libqp wrapping13:34
wikingi suppose you are talking about now libqp wrapper functions13:34
-!- xiangwang [~chatzilla@] has quit [Ping timeout: 248 seconds]13:36
-!- ckwidmer [] has quit [Remote host closed the connection]13:41
blackburndrum roll again13:42
blackburnchris is leaving13:42
blackburnand I am going too as well13:42
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]13:42
-!- al___ [4eecc9c5@gateway/web/freenode/ip.] has quit [Quit: Page closed]13:48
-!- xiangwang [~chatzilla@] has joined #shogun13:54
-!- xiangwang [~chatzilla@] has quit [Ping timeout: 276 seconds]14:01
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has quit [Quit: Page closed]14:03
sonne|workwiking: no I am talking about making it libqp_prloqo.cpp14:04
sonne|workand getting it core libqp14:04
sonne|workand on top of it the qp interface14:05
wikingso you wanna put prloqo into libqp?14:05
wikingor make it as part of it14:05
wikingah ok14:05
sonne|workand tron14:05
wikingtron as well14:05
wikingcan we move them to maybe14:05
wikinglets say src/shogun/qp14:05
wikingor something like that?14:05
wikingas it's not an external lib14:07
sonne|workwiking: I thought so too but now that blackburn started this lib/external stuff I think these should be in there ... maybe external/libqp14:07
wikingmmm but they are not part of libqp14:07
wikingso i mean when let's say14:07
wikingvojtech updates the libqp14:07
sonne|workno but they should become part of it14:07
wikingit's pretty easy to do an upgrade14:07
sonne|worklet me talk to vojech14:07
sonne|workabout this14:07
wikingtalk with him decide and then i'll move it14:07
wikingbut afaik we'll still need a src/shogun/qp14:07
wikingto put there the wrapper classes14:08
sonne|workmaybe better shogun/optimizer14:08
sonne|workand have the qp stuff in there14:08
sonne|workand some linear stuff etc too14:08
wikingthat sounds good14:08
wikingok i'll then make an optimizer dir and put there the abstract wrapper as a starter14:08
wikingand then we'll see what to do with pr_loqo and tron14:08
sonne|workhave a look at cplex.h too maybe remotely kind of this14:09
wikingbut first things first i want to finish my simple cccp implementation for latent14:10
wikingtoday i've done already too much branch switching till now :)14:10
-!- blackburn [~qdrgsm@] has joined #shogun14:14
-!- xiangwang [~chatzilla@] has joined #shogun14:17
sonne|workwiking: as you wish :)14:19
-!- PhilTillet [~android@] has quit [Ping timeout: 240 seconds]14:21
-!- xiangwang [~chatzilla@] has quit [Ping timeout: 252 seconds]14:25
wikingsonne|work: let me know when you've concluded the business with vojtech ;)14:26
-!- gsomix [~gsomix@] has quit [Ping timeout: 260 seconds]14:37
-!- pluskid [~chatzilla@] has joined #shogun14:51
sonne|workhey pluskid14:51
sonne|workwelcome back14:51
pluskidjust back from the trip14:51
pluskidI saw your feedbacks14:51
sonne|workI guessed so14:51
pluskidwill fix those soon (tomorrow)14:52
pluskidtoo tired to work tonight14:52
sonne|workno worries14:52
sonne|workif none of you applied at a different org we already now know who will be in14:52
pluskidbtw, do you think I should keep the code for reading and writing multiclass strategy in MulticlassSVM?14:52
pluskidI only applied shogun14:53
pluskidso am I in? :p14:53
pluskidwon't tell me this moment? haha, bad boy14:53
sonne|workthat only means you are selected by us then you will be in :)14:53
sonne|workI am not allowed to14:53
sonne|workbut I don't think it is too hard to guess14:54
pluskidbtw, I'm just curious, sonne|work , do you hate STL? :p14:54
sonne|workanyways multiclass svm -> maybe multiclass kernel machine?14:54
sonne|workpluskid: yes but don't ask my why14:54
sonne|workblackburn asked me that so often that I forgot14:54
sonne|workexcept that it looks ugly (syntax wise)14:55
wikingwhen is the list going to be released?14:55
pluskidOK, that's why they decided to sent you a gift of STL book, haha14:55
wikingonly next tuesday?14:55
sonne|workgoogle will announce on 23rd14:55
pluskidI can also feel your hate toward STL, too :)14:55
wikingSTL for sonne|work \o/14:55
wikingbut honestly one should consider valarray :)14:55
sonne|workwiking: we have SGVector :P14:56
pluskidI think the read/write of multiclass-strategy should be put in a read/write function in MulticlassMachine14:56
wikingsonne|work: yes but within sgvector14:56
pluskidbut it seems current there's no such function in MulticlassMachine14:56
wikinginstead of having float*14:56
sonne|workpluskid: multiclasskernelmachine14:56
wikinga valarray would allow a lot of automatic optimization...14:56
sonne|workthese functions are not used anyways I would guess14:56
sonne|workmaybe some used them long time ago for multiclass svms14:56
pluskidthen I'll not worry too much about that14:57
pluskidI'd try to keep it similar to the original code14:57
sonne|workbut nowadays the save_serializable / load_serializable stuff is much more powerful/usefull14:57
pluskidso, this two manually crafted function might be removed some day14:58
sonne|workI think all load/save() functions should go14:58
pluskidwiking: what is a valarray?14:58
pluskidsonne|work: at this moment?14:58
wikingpluskid: the forgotten child of STL14:58
sonne|workwiking: why the massive overhead?14:59
wikingit's really a very very small overhead14:59
wikingcompared to other STL containers14:59
wikingas this is really not a real STL container14:59
sonne|workwe simply want to be fast for low-dimensional data too15:00
wikingand recent g++ implementation use a done a lot of optimization with that15:00
wikingi.e. using SIMD and multiple thread out of box15:00
sonne|workand having more than size / refcount as overhead is imho not acceptable15:00
wikingsince valarray is a really simple class15:00
pluskidoh, it seems what a valarray is what we wish to do with our SGVector?15:00
sonne|workpluskid: yeah the a+b, sin(a) etc operations15:01
wikingsonne|work: check out the definition of valarray template: /usr/include/c++/4.2.1/valarray15:01
wiking   private:15:02
wiking      size_t _M_size;15:02
wiking      _Tp* __restrict__ _M_data;15:02
wiking      15:02
wikingthis is all the overhead ;)15:02
pluskidbtw, I seriously think we need some basic matrix operations support after doing the LARS work, Blas/LaPack is a bit difficult to use, especially for simple tasks like dot product of vectors, we need some wrappers15:02
pluskidwiking: maybe we need to do some standard benchmarks someday to persuade sonne|work that STL (at least valarray) is not that evil15:04
wikingpluskid: hehehe there's already such available... just google it15:04
pluskidwiking: about valarray or STL?15:04
pluskidI googled "valarray benchmark", and the first result it "why is valarray so slow", haha15:05
pluskidcool, seems especially optimized for "small" (10^3) sized vector15:10
wikingi mean in this sense maybe having GSL will solve all the problems ;)15:10
pluskidhaha, another huge dependency that I guess sonne|work also hates :p15:11
pluskidbe back in a while15:12
wikingbut gsl is just about over optimizing all kind of vector/matrix operations15:12
-!- pluskid_ [~chatzilla@] has joined #shogun15:25
-!- pluskid [~chatzilla@] has quit [Read error: Connection reset by peer]15:25
-!- pluskid_ is now known as pluskid15:25
wikinganybody else might need a CMath:: function that not only returns the max value of a vector but the position/index of that max value as well?15:26
blackburnwiking: isn't it available already?15:27
wikingmmm is it? i was looking for ::max15:27
wikingonly found  static inline T max(T* vec, int32_t len)15:27
wikingtill now15:27
wikingthere's no such15:28
wikingor at least i couldn't find till now in Math.h15:29
blackburnline 900 or so15:29
sonne|workstatic inline int32_t arg_max(T * vec, int32_t inc, int32_t len, T * maxv_ptr = NULL)15:29
wikinggit grep arg_max15:30
wiking0 resuls15:30
sonne|workno objections for having a simpler one though15:30
wikingis it maybe in a recent commit?15:30
blackburnby fernando I guess15:30
sonne|workQDA commit15:30
wikingsince i haven't rebased my latent branch for a hiwhile15:30
wikingah ok15:30
blackburnshogun/classifier/QDA.cpp:              out->set_label(i, CMath::arg_max(norm2.vector+i, num_vecs, m_num_classes));15:30
blackburnshogun/mathematics/Math.h:              /// return arg_max(vec)15:30
blackburnshogun/mathematics/Math.h:                      static inline int32_t arg_max(T * vec, int32_t inc, int32_t len, T * maxv_ptr = NULL)15:30
wikingso that much be the reason15:31
wikingok i'll rebase then15:31
-!- gsomix [~gsomix@] has joined #shogun15:33
blackburnsonne|work: chris is in tuebingen/berlin up to may so exclude the usa for kickoff15:34
sonne|workexcept if we have a student there15:34
blackburnforget what I said then heh15:42
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 246 seconds]15:51
-!- n4nd0 [] has joined #shogun15:53
-!- pluskid [~chatzilla@] has quit [Ping timeout: 246 seconds]16:17
-!- pluskid [~chatzilla@] has joined #shogun16:22
-!- gsomix [~gsomix@] has quit [Ping timeout: 260 seconds]16:44
-!- Marty28 [~marty@] has joined #shogun16:47
-!- gsomix [~gsomix@] has joined #shogun16:59
-!- Marty28 [~marty@] has quit [Quit: Colloquy for iPad -]17:07
-!- n4nd0 [] has quit [Read error: Operation timed out]17:35
-!- PhilTillet [] has joined #shogun17:47
-!- pluskid [~chatzilla@] has quit [Ping timeout: 264 seconds]18:13
-!- n4nd0 [] has joined #shogun18:44
-!- wiking_ [] has joined #shogun18:58
-!- wiking_ [] has quit [Changing host]18:58
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun18:58
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 248 seconds]19:01
-!- wiking_ is now known as wiking19:01
* sonney2k alright19:07
* sonney2k lets kick as and chew bubblegum19:08
* sonney2k goes berserk and updates liblinear19:08
wikingi'm gonna use ocas as a solver for now in latent svm19:08
wikingthat's what i've decided :)19:08
wikingok off to train19:18
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]19:19
-!- PhilTillet [] has quit [Ping timeout: 264 seconds]19:29
gsomixit's seems that `sublime text 2' become my favourite editor19:38
gsomixI still not able to use vim...19:39
@sonney2kgsomix, try vimtutor19:39
gsomixsonney2k, I tried :)19:40
CIA-64shogun: Soeren Sonnenburg master * r9815db6 / src/shogun/lib/external/libqp_gsmo.cpp : fix compile warning in libqp_gsmo -
CIA-64shogun: Soeren Sonnenburg master * r723ccd1 / (2 files in 2 dirs): fix doxygen documentation related errors -
n4nd0gsomix: maybe you are an emacs person ;)20:03
gsomixn4nd0, huh20:04
* sonney2k thinks that gsomix still needs his fingers20:11
-!- PhilTillet [] has joined #shogun20:16
gsomixn4nd0, which exams in this year?20:26
n4nd0gsomix: so I'll do next week databases and microwave engineering20:27
n4nd0gsomix: wireless communications in a couple of weeks20:27
gsomix>> databases | theoretical course?20:28
n4nd0not really20:28
n4nd0the first semester was rather theoretical20:28
n4nd0now it is half of it sql20:28
n4nd0and other half security and normlization20:28
n4nd0gsomix: what about you?20:29
gsomixn4nd0, math analysis, complex analysis, philosophy and optics20:30
gsomixin june, i think.20:30
gsomixn4nd0, I have horrible course about databases... FoxPro 9 =____=20:31
gsomix*in last year20:31
n4nd0FoxPro 9, no idea20:31
n4nd0I have never taken a course itself in complex analysis, but I have done stuff with transforms20:32
n4nd0and last year I used it quite a bit in a course in electromagnetism20:32
n4nd0I think complex analysis looks nice20:32
n4nd0do you like it?20:32
gsomixn4nd0, yep. but i'm missing a lot of lectures about it.20:34
gsomix>> in a course in electromagnetism | aha, impedances and other20:35
-!- puffin444 [230bf329@gateway/web/freenode/ip.] has joined #shogun20:35
n4nd0in this one impedances didn't appear that much20:36
n4nd0it was about electromagnetic propagation in plasma media20:36
n4nd0with tensors and stuff like that20:37
n4nd0hey puffin44420:37
puffin444Have I walked in the middle of a physics conversation? :)20:37
gsomixn4nd0, I see20:37
n4nd0haha just asking about our forthcoming exams20:37
puffin444Oh are you students of physics?20:37
n4nd0telecommunications + CS here :)20:38
-!- PhilTillet [] has quit [Ping timeout: 252 seconds]20:38
gsomixpuffin444, I'm undergraduate student in physics.20:39
gsomixpuffin444, ah, hello :)20:39
puffin444Oh I see. I'm CS+Math+Econ20:39
-!- mohataher [3e8758c4@gateway/web/freenode/ip.] has joined #shogun20:41
gsomixwhat is english equivalent of exam without assessment?20:42
n4nd0what do you mean?20:42
puffin444An ungraded exam?20:43
@sonney2kpuh, liblinear 1.9 has *lots* of changes20:43
@sonney2ktoo many for any train ride20:43
n4nd0sonney2k: I think you need the Siberian train for that :)20:45
gsomixn4nd0, hmm, exams with only binary assessment (pass / fail)20:45
gsomixproof that you attended a classes.20:47
gsomixok, never mind :)20:47
-!- PhilTillet [~Philippe@] has joined #shogun20:47
@sonney2kn4nd0, heh I think some train to frankfurt would suffice20:48
@sonney2kbut still I didn't know how much (subtle) changes they did even to old liblinear20:48
@sonney2kthis + our modifications don't make live so easy20:49
puffin444How does liblinear fit in with shogun?20:49
PhilTillethelo everybody :)20:49
gsomixPhilTillet, hi20:51
puffin444PhilTillet, hey20:52
-!- mohataher [3e8758c4@gateway/web/freenode/ip.] has quit [Quit: Page closed]20:52
PhilTilletwhat's up guys?20:52
-!- mohataher [3e8758c4@gateway/web/freenode/ip.] has joined #shogun20:53
-!- mohataher [3e8758c4@gateway/web/freenode/ip.] has quit [Client Quit]20:55
PhilTilletI have powned my keyboard???21:01
PhilTilletI have somewhat broken the "j" key21:01
puffin444What happened? You can't press the "j" key?21:02
PhilTilletI can :p but it fell off my keyboard, and it is broken so I can't put it back :/21:06
puffin444That's quite annoying. That happened to my touchpad mouse buttons.21:08
-!- mohataher [3e8758c4@gateway/web/freenode/ip.] has joined #shogun21:08
PhilTilletI think i'll try to carefully glue it21:10
-!- PhilTillet [~Philippe@] has quit [Quit: Leaving]21:14
-!- PhilTillet [~Philippe@] has joined #shogun21:14
PhilTilletsonney2k, I still can't think of a nice way to put my opencl patch in another file21:14
@sonney2kn4nd0, uh! I just got an email from JL - he is no longer at yahoo!21:20
n4nd0sonney2k: oh, is he already at another company?21:21
n4nd0John Langford21:21
@sonney2kpuffin444, inventor of isomap21:22
@sonney2kand author of the fastest covertree on earth21:22
n4nd0and the cover tree ;)21:22
@sonney2kand vowpal vabbit :)21:22
n4nd0we are still not done with ours :P21:22
gsomixlegend :)21:22
puffin444Oh isomap? The manifold learning algorithm?21:23
n4nd0puffin444: yeah, dimensionality reduction21:23
@sonney2kpuffin444, this guy
n4nd0I think he has also been with Thrun with particle filters21:24
puffin444Ok I knew that name sounded familiar21:24
puffin444Sebastian Thrun?21:24
puffin444Oh wow!21:24
n4nd0they are both authors of this paper
@sonney2khe is a big shot and a cool guy (as you can guess from the picture :)21:26
n4nd0I think it is pretty awesome that he was mentor of a project in shogun last year21:27
n4nd0sonney2k: no project suited or not enough time this year?21:28
puffin444Oh wow he has also been involved in captcha21:28
@sonney2kn4nd0, also I didn't have enough time this year21:30
puffin444Looking at his website, it looks like he has done a lot of cool stuff. It must have been exciting to have him as a mentor.21:30
@sonney2kand am having21:30
@sonney2kunfortunatly shashwat didn't continue development21:32
n4nd0sonney2k: what do you think about particle filters? could it be a place in shogun for it?21:32
@sonney2kI would have hoped he does some more on vw / shogun21:32
n4nd0I'm just curious about it21:33
PhilTilletwah, paper with Thrun!21:33
@sonney2kI don't know anything about it21:33
@sonney2khmmhh I know JL better than Thrun (would even say he is more famous)21:33
n4nd0oh do you think so?21:33
@sonney2kbut I guess due to this stanford lecture thrun is more known21:33
n4nd0in the environment I mode Thrun is like a rock star :D21:33
@sonney2kand of course for winning the darpa thing21:33
n4nd0I move*21:34
PhilTilletyes :p that's how I knew him21:34
PhilTilletthat ai-class made him a rockstar lol21:34
n4nd0for the people in my department he has been a rockstar since fast slam21:34
@sonney2kfast slam?21:34
n4nd0they have told me about a lecture he gave here at KTH about it21:34
n4nd0about fast slam21:35
n4nd0it is a method used for localization in robotics21:35
n4nd0SLAM stands for simultaneous localization and mapping21:35
n4nd0a robot doesn't know where it is and goes around building a map and getting an idea where it is at the same time21:36
n4nd0sonney2k: about particle filters, are you more familiar with the condensation algorithm?21:37
@sonney2kok then about rockstars21:37
@sonney2kdoes anyone know gary bradski here?21:37
@sonney2kn4nd0, not at all21:37
n4nd0Willow Garage, OpenCV21:38
puffin444Doen't ring a bell21:38
@sonney2kyeah ok21:38
n4nd0sonney2k: I checked the videolecture with he and you ;)21:38
@sonney2kleon bottou?21:38
n4nd0new for me21:38
@sonney2kyann lecun?21:38
@sonney2kalex smola?21:39
@sonney2kbernhard schoelkopf?21:39
@sonney2kklaus-robert mueller :D21:39
n4nd0I don't know that many21:39
@sonney2kn4nd0, ok then last one21:39
mohataherI have a question, when to use SGVector and when to use SGMatrix ? (when having one column only of course), I ask coz I was advised in a pull-request to use SGMatrix instead of SGVector coz it's lightweight. Any explanation ?21:40
@sonney2kvladimir vapnik?21:40
n4nd0sonney2k: SVMs!!21:40
PhilTilletsonney2k,  yes for Yann Lecun ! he's french after all :p21:40
@sonney2kn4nd0, ok then you should know smola / schoelkopf too21:40
@sonney2kand actually klaus :)21:40
PhilTilletlol gsomix21:41
@sonney2kgsomix, hehe21:41
puffin444Yann Lecun sounds familiar21:41
n4nd0lately I see Joachims everywhere21:41
@sonney2kn4nd0, where?21:41
n4nd0lot of papers and work related to SO21:42
@sonney2kold ones then or?21:42
@sonney2kJL goes to microsoft...21:42
n4nd0mmm I think 5 years old at most, are those considered old ones?21:43
@sonney2kI guess they are in  better shape than yahoo21:43
n4nd0oh microsoft21:43
@sonney2kyes old21:44
n4nd0he is the author too of SVM struct21:44
-!- genix [~gsomix@] has joined #shogun21:45
@sonney2kn4nd0, yeah that is all old stuff21:45
@sonney2khe did svmlight in the 90's21:46
n4nd0SVM struct appears dated 2008 on the webpage21:46
@sonney2klast update21:46
n4nd0I see21:46
-!- gsomix [~gsomix@] has quit [Ping timeout: 252 seconds]21:47
@sonney2kI guess I am old known all these in person21:47
@sonney2kanyways they know me too :D21:47
@sonney2kand now ... time to sleep21:48
n4nd0sonney2k: you are also a rock star :)21:48
n4nd0good night21:48
-!- fman [b890a72d@gateway/web/freenode/ip.] has joined #shogun21:48
n4nd0sonney2k: btw, particle filters, are least nice to see in action21:48
-!- n4nd0 [] has quit [Quit: leaving]21:53
-!- wiking [] has joined #shogun22:04
-!- wiking [] has quit [Changing host]22:04
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun22:04
-!- puffin444 [230bf329@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:19
-!- fman [b890a72d@gateway/web/freenode/ip.] has left #shogun []22:45
-!- genix [~gsomix@] has quit [Ping timeout: 245 seconds]23:31
-!- genix [~gsomix@] has joined #shogun23:33
-!- genix is now known as gsomix23:33
gsomixwiking, yo23:33
-!- mohataher [3e8758c4@gateway/web/freenode/ip.] has quit [Quit: Page closed]23:35
--- Log closed Fri Apr 20 00:00:19 2012