Open in new window / Try shogun cloud
--- Log opened Thu Mar 01 00:00:19 2012
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun00:22
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]00:33
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun00:39
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]01:24
-!- delcoder [~ishaan@101.63.94.119] has joined #shogun02:34
-!- delcoder [~ishaan@101.63.94.119] has quit [Quit: Leaving]04:22
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:01
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]09:57
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:08
-!- bais [~bais@n136-p64.kthopen.kth.se] has joined #shogun10:57
-!- bais [~bais@n136-p64.kthopen.kth.se] has quit [Client Quit]10:58
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]10:59
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun11:16
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection]11:19
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun11:20
-!- bais [~bais@n136-p64.kthopen.kth.se] has joined #shogun11:23
baisI have just found the CustomKernel class, that should help!11:24
-!- bais [~bais@n136-p64.kthopen.kth.se] has quit [Client Quit]11:24
-!- Karl-London [57718696@gateway/web/freenode/ip.87.113.134.150] has joined #shogun11:40
Karl-LondonHi. I'm a Shogun beginner and I'm trying to run some examples in Octave (modular).11:41
Karl-LondonThe problem is, the data directory is missing. Could someone please tell me where the data directory is? Or send me the data?11:42
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun11:46
Karl-LondonCould someone please send me the data file: label_train_twoclass.dat, fm_train_real.dat and fm_test_real.dat ?11:51
Karl-LondonMy email is: ksh309@gmail.com       Many thanks for your help!11:52
n4nd0Karl-London: hey!11:59
n4nd0Karl-London: you can download them from the github repo11:59
n4nd0Karl-London: you can find the files you want here https://github.com/shogun-toolbox/shogun-data/tree/master/toy12:00
n4nd0Karl-London: but you may be interested also in getting a local clone of the repo :)12:01
Karl-LondonGot it! That's great. Thank you very much for your help!12:04
n4nd0Karl-London: no problem, you are welcome12:04
-!- Karl-London [57718696@gateway/web/freenode/ip.87.113.134.150] has quit [Quit: Page closed]12:08
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]12:53
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun13:02
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection]13:17
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun13:17
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]13:19
-!- bais [~bais@n136-p64.kthopen.kth.se] has joined #shogun13:51
baishi13:52
baisi'm having a problem using SVMs in the octave modular interface13:52
baiswhen I try executing the train() method it simply prints "sorry not yet implemented"13:55
baisThe question at this point might be stupid, but does this mean there is not way of using SVMs with octave right now?!13:55
baisshould i maybe be using the LibSVM class?14:01
baisit seems like the SVM class is just an interface to actual classes like LibSVM..14:16
-!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: Operation timed out]14:35
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun14:36
sonne|workbais: yes SVM is just the baseclass15:09
sonne|workuse any of its child-classes http://www.shogun-toolbox.org/doc/en/current/classshogun_1_1CSVM.html15:09
sonne|workwiking: have a look at the updated ideas list - vojtech's libqp has some more detailled descr. now15:15
wikingsonne|work: hey! cool i'll check i've had one email exchange with vojtech. he said that he's going to be pretty busy so i'll have to be independent but that's fine by me15:16
sonne|workhe has a great plan now I think15:18
sonne|workjust look at it (I recall that this and the latent svm where of interest to you)15:19
wikingyep15:19
wikingi think15:20
wikingalthough it's a bit much but there could be a great way to do them both15:20
wikingi mean creating nice interface for libqp within shogun15:20
wikingand use libqp for the latent svm15:20
wikingthat'd be great15:20
wikingsince now some of the latent svm 'implementations' use mosek15:21
wikinghmm that reminds me i should clean up my latest patch and submit a pull request ;P15:30
baissonne|work: yes thanks I am finally making my way through the documentation in a more ordered way :P Although, I cannot find some form of description of the various classes, like what would be the difference between using LightSVM and SVMLib, and such..15:36
sonne|workyeah that is missing15:37
sonne|work^ bais15:37
sonne|worksome more documentation is certainly needed in particular when people don't want to google 'libsvm' 'svmlight' etc15:37
sonne|workwiking: well it depends if you intend to sleep15:38
sonne|workand of course if you want to stick around after gsoc ... in the end gsoc is just 3 months15:38
wikingsonne|work: hahaha well it's part of my phd15:39
wikingsoooo15:39
wikingi mean i'm not doing this for gsoc15:39
wikingit's for myself and my phd15:39
sonne|workwell then just start the work :D15:39
wikingyep15:39
wikingi'd just like to talk about it with somebody15:40
wikingso that you guys get something in a form you'd like to see it15:40
wikingso everybody wins :P15:40
sonne|workI havent read the latentsvm papers but I can very well imagine what it does and that the resulting problem is non-convex15:41
wikingyep15:41
sonne|workbut in the end it is just an SVM :)15:41
wikingcurrently that's why i'm trying out other optimizers15:41
wikingbut yep.. essentially it's an sum :)15:42
wiking*svm15:42
sonne|workwiking: what about http://opencv.itseez.com/trunk/modules/objdetect/doc/latent_svm.html15:42
wikinglesse15:43
wikingyeah15:43
wikingi know the opencv based one15:43
wikingit's very very specific15:43
sonne|workwiking: do you want to do that for linear svms?15:47
sonne|workI mean linear latent svm15:47
sonne|work(I just had a look at the paper)15:47
wikingwell afaik the latent svm that joachim's student did was linear... he haven't made the kernel extension of it...15:48
sonne|workkernel would be too slow15:48
sonne|workso then we need hog and lbp features and then live is good :)15:48
baissonne|work: oh ok i didnt figure out they were external references to the project, i thought they were internal stuff like "KernelMachine" or such15:51
sonne|workbais: yeah we use other people's code if we can...15:52
baisok well, thanks15:53
-!- bais [~bais@n136-p64.kthopen.kth.se] has quit [Quit: leaving]15:53
sonne|workwiking: certainly quite some effort to get things to work...15:54
sonne|workcan be as hard a general structured output learning (HM-SVM)15:54
wikingnyiha15:55
wikingi haven't even knew about HM-SVM15:55
wikinglet me check what's that thingy :)15:55
wikingapart from being a hidden markov -svm :)))15:55
sonne|workit is frequently used in vision/text/bioinformatics to segment things15:55
sonne|worke.g. dna into genes/non-genes15:55
wikingmmm15:56
wikingshogun has it? :)15:56
wikingor it's only implemented in hm-svm toolbox15:57
sonne|workwell shogun has the argmax but no more16:01
sonne|workone of the projects is to get hm-svm toolbox back into shogun16:02
sonne|workwith some nice design though16:02
sonne|workcode quality is not so good in that part...16:02
sonne|work^wiking16:11
wikingmmm16:12
wikingi'll give it a look16:12
sonne|workwiking: well, choose what you are most interesting in / what does motivate you to do most :)16:33
-!- delcoder [~ishaan@115.242.61.210] has joined #shogun16:42
-!- delcoder [~ishaan@115.242.61.210] has quit [Quit: Leaving]17:14
-!- blackburn [~qdrgsm@188.168.128.87] has joined #shogun17:46
-!- blackburn [~qdrgsm@188.168.128.87] has quit [Quit: Leaving.]18:04
-!- blackburn [~qdrgsm@188.168.128.177] has joined #shogun18:10
-!- n4nd0 [82ede325@gateway/web/freenode/ip.130.237.227.37] has joined #shogun18:22
n4nd0blackburn: how is it going?18:28
blackburnn4nd0: hey, fine, what about you?18:28
n4nd0blackburn: fine too18:29
n4nd0blackburn: I am taking a look to the gsoc ideas at the webpage18:29
blackburnand what do you think about it?18:29
n4nd0I think they are really interesting in general18:30
n4nd0I find quite appealing the one of structured output18:31
n4nd0but being honest, it may be a bit out of scope for my knowledge right now18:31
n4nd0a bit too out of scope :P18:31
blackburnwell I didn't know anything about dimreduction last year when I did a lot of dimreduction :)18:31
n4nd0oh18:32
n4nd0I thought you had done sth with it before maybe in a math course or the like18:33
n4nd0you are working in multitask right?18:34
blackburnwill be working on MT, yeah18:34
n4nd0the first one about writing tutorials and documentation looks nice too18:37
n4nd0is that a priority for shogun?18:37
blackburnwell probably most interesting and important idea, yeah18:38
n4nd0yeah, that one looks good definetely18:39
n4nd0it can be a good way to learn the fundamentals so later (post-gsoc) one can continue with more advanced stuff18:40
n4nd0blackburn: anyhow, do you have any other idea do work on in the meantime?18:42
blackburnwell.. not really18:47
blackburnlet me think about it18:47
blackburnn4nd0: maybe QDA?19:01
n4nd0blackburn: it sounds almost like LDA :P19:02
n4nd0I guess they must be related to each other19:02
blackburnyes, but no code for QDA here :)19:02
n4nd0I will take a look to it then19:03
n4nd0do you know something it is applied for?19:03
blackburnnot really, but we are general-purpose ML lib19:06
blackburn:)19:06
n4nd0I just wondered :P19:07
blackburnmay be spectral clustering19:07
blackburnbtw spectral clustering is a very simple task19:11
blackburnone need to just combine ready KMeans and LaplacianEigenmaps19:11
n4nd0is it used for dimensionality reduction?19:11
blackburnspectral clusterin is used for clustering, but it uses laplacian eigenmaps19:12
blackburnsimply it is kmeans in laplacian eigenmaps domain19:13
-!- blackburn [~qdrgsm@188.168.128.177] has quit [Quit: Leaving.]19:30
-!- n4nd0 [82ede325@gateway/web/freenode/ip.130.237.227.37] has quit [Quit: Page closed]19:50
-!- delcoder [~ishaan@101.63.49.239] has joined #shogun21:00
CIA-64shogun: Soeren Sonnenburg master * r013f6d2 / src/Makefile : fix make linecount - http://git.io/lNMR3Q21:17
CIA-64shogun: Soeren Sonnenburg master * radec872 / examples/undocumented/python_modular/graphical/interactive_clustering_demo.py : mark also cluster centers - http://git.io/9dUOdw21:17
CIA-64shogun: Soeren Sonnenburg master * re2a16a6 / src/shogun/clustering/KMeans.cpp : whitespace fix - http://git.io/J7hFgw21:17
shogun-buildbotbuild #509 of octave_static is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/509  blamelist: sonne@debian.org21:46
-!- blackburn [~qdrgsm@188.168.14.46] has joined #shogun21:59
-!- delcoder [~ishaan@101.63.49.239] has quit [Ping timeout: 244 seconds]22:02
shogun-buildbotbuild #510 of octave_static is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/51022:04
blackburnsonney2k: you found time to do interactive demos ;)22:05
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun22:25
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]22:43
-!- blackburn [~qdrgsm@188.168.14.46] has quit [Ping timeout: 252 seconds]22:43
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun22:53
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]23:36
--- Log closed Fri Mar 02 00:00:19 2012