Open in new window / Try shogun cloud
--- Log opened Tue Jun 19 00:00:41 2012
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has joined #shogun00:08
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has quit [Changing host]00:08
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun00:08
shogun-buildbot_build #612 of java_modular is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/java_modular/builds/612  blamelist: blackburn91@gmail.com00:11
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]00:36
-!- blackburn [~blackburn@62.106.111.221] has quit [Quit: Leaving.]01:31
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]01:55
-!- emrecelikten [~emre@176.40.254.205] has quit [Read error: Connection reset by peer]02:55
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun03:21
-!- cheng [80fa500f@gateway/web/freenode/ip.128.250.80.15] has joined #shogun03:54
-!- cheng [80fa500f@gateway/web/freenode/ip.128.250.80.15] has quit [Ping timeout: 245 seconds]05:21
-!- shogun-buildbot_ [~shogun-bu@7nn.de] has quit [Ping timeout: 245 seconds]05:28
-!- uricamic [~uricamic@2001:718:2:1634:397b:935c:85c:2381] has joined #shogun08:48
gsomixoptics!10:08
-!- gsomix [~gsomix@37.61.181.80] has quit [Quit: Ex-Chat]10:08
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun10:42
-!- emrecelikten [~emre@176.40.254.205] has joined #shogun12:48
-!- emrecelikten [~emre@176.40.254.205] has quit [Quit: Leaving.]13:59
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]14:06
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has joined #shogun14:56
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has quit [Changing host]14:56
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun14:56
-!- blackburn [d55783a2@gateway/web/freenode/ip.213.87.131.162] has joined #shogun15:11
blackburnwiking: you promised to send report ;)15:14
wikingsended15:14
wiking:)15:14
wikingshould b there in any seconds15:15
blackburnwhy didn't you do that yesterday? :)15:16
wikinghacking too late15:16
wikingi've just woken up man15:16
wiking :)15:16
blackburnit seems you like to get us worried :D15:16
wikingnono i like to work on this project15:17
wikingthat's why the all-nighters15:18
wikingcurrently trying to work with libqp somehow :S15:18
blackburnok15:18
wikingbut that's still shit :(15:18
wikingn4nd0: have u started on that15:18
wikingor u are still working with mosek?15:19
blackburnwhat are you doing this time?15:19
blackburnto get you in shape and make some progress we really would like to merge anything15:19
n4nd0wiking: libqp doesn't solve the problem I need t osolve15:19
wikingblackburn: well i'm trying already the SO implementation + m3e porting (http://jmlr.csail.mit.edu/proceedings/papers/v22/miller12/miller12.pdf)15:19
wikingn4nd0: how so?15:19
blackburnwiking: IIRC you had some simple latent svm before?15:20
wikingyeye15:20
wikingi should pull it though15:20
n4nd0wiking: we need the box constraints case with inequalities15:20
wikingso you can start marging true15:20
wikingn4nd0: s.t. \sum x_i <= b_k ?15:21
n4nd0wiking: check in your mail the pdf that nico sent15:22
wikingn4nd0: ok15:22
n4nd0wiking: or even better15:22
n4nd0http://iglesiashogun.wordpress.com/15:22
n4nd0second weekly report15:22
wikinghttp://s0.wp.com/latex.php?latex=s.t.~%5Cforall+i~%5Cforall+y+%5Cin+%5Cmathcal%7BY%7D+%5Csetminus+y_i+%3A+%5Cvec%7Bw%7D+%5Ccdot%5Cdelta%5CPsi_i%28y%29+%5Cge+%5CDelta%28y_i%2C+y%29+-+%5Cxi_i&bg=eeeae8&fg=4a4a49&s=0 ?15:24
wiking:D15:24
wikingmmm so does this mean that you'll actually stick with mosek for good?15:24
n4nd0it looks like there is no open source available for that problem15:28
n4nd0the plan is to use mosek now and later, if there's time, do a simple solver for it integrated into shogun15:29
wikingmmm15:29
blackburnn4nd0: you may try nlopt now btw15:30
blackburnno idea if it is any reasonable but at least open source15:30
wikingmmm shit then actually15:30
wikingbecause then my stuff with SO will have some problems :(((15:30
wikingwell anyhow it's not that major atm15:31
n4nd0blackburn: I will see if it can be used for this15:32
n4nd0blackburn: thank you :)15:32
blackburnn4nd0: it can be used I believe but performance would be bad probably15:32
blackburncause it is quite general15:32
wikingn4nd0: any plan for a pull request?15:33
n4nd0https://github.com/shogun-toolbox/shogun/pull/58015:34
wikingmmm15:34
wikingso it's pending15:34
wikingok15:34
wiking because now i'm cherry picking your so branch15:34
n4nd0yeah15:34
n4nd0does it give trouble for you that's not yet merged?15:35
wikingwell currently it's ok15:35
wikingjust stupid in a way15:35
n4nd0?15:35
wikingthat i'm maintaining 2 dev branches15:35
wiking1 for general latent stuff15:35
wikingand 1 for SO latent stuff based on your code15:36
n4nd0I see15:36
wikingsince the current SO in the main branch of shogun15:36
wikingis 'deprecated' ;)15:36
wikinganyhow as soon as it's in i could merge it afaik15:36
wiking*afais15:36
n4nd0it will be good if you start to use it15:37
n4nd0I want to see it in action15:37
n4nd0debug more15:37
n4nd0etc15:37
wikingwell i'm already having some fun with it15:38
wiking:)))15:38
wikingbut was hoping15:38
wikingyou'll be able to use libqp :s15:38
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun15:43
wikingn4nd0: u have some time now?15:45
n4nd0wiking: yes15:46
n4nd0tell me15:46
wikingcan u check this15:46
wikinghttp://jmlr.csail.mit.edu/proceedings/papers/v22/miller12/miller12.pdf15:46
wikingalgo 215:46
wiking(page 5)15:46
wikingso i need a solver for that :)15:47
n4nd0haha ok15:48
n4nd0I am looking to equation 1315:48
n4nd0definetely, it has some similarities with the primal sosvm15:48
wikingyep15:48
CIA-18shogun: Christian Widmer master * r59068d6 / (4 files in 3 dirs): extended SGSparseVector to make items accessible with operator[] - http://git.io/0tROCg15:49
CIA-18shogun: Christian Widmer master * rb5c22d6 / src/interfaces/octave_modular/swig_typemaps.i : updated type map for octave to use new index map - http://git.io/3K94aw15:49
CIA-18shogun: Christian Widmer master * rb89a58b / (2 files in 2 dirs): some clean-up - http://git.io/0WwzIg15:49
CIA-18shogun: Sergey Lisitsyn master * r43ea7ab / (5 files in 4 dirs): Merge pull request #588 from cwidmer/master - http://git.io/060Wsw15:49
n4nd0but it is not the same problem15:49
n4nd0I am not handling contrainst of type ||w - w_t ||^2 <= \mu15:50
n4nd0but I believe it should be possible to do15:50
n4nd0the norm is the common Euclidean one right?15:51
wikingyeps15:51
n4nd0||w - w_t||^2 = < w-w_t, w-w_t >15:51
n4nd0?15:51
n4nd0the dot product simply15:52
wikingsince it's a l2-norm15:53
n4nd0aham15:54
n4nd0so it's not the same that I solve :(15:54
n4nd0I am sorry15:54
n4nd0you can solve that with mosek though15:55
wikingheheh15:55
n4nd0http://docs.mosek.com/6.0/capi/node007.html#32974770415:55
wikingi know that :)15:55
n4nd0look there15:55
n4nd0yeah, wait15:55
n4nd0for my current problem15:55
n4nd0Q^k = 015:55
n4nd0but in your problem is not like that15:55
n4nd0do you see my point?15:56
n4nd0you have quadratic constraints I think15:56
n4nd0while in the primal so-svm those don't appear15:56
wikingthis is already some extra funky stuff the stuff i'll need with your solver is actually the algo 1 on page 4. afaik that's a simple structural svm (equation 9)15:57
n4nd0yes15:57
wikingi mean this other one i've mentioned before i'll need as well, but i thought i could go around your SO so that's why i've started that15:58
n4nd0that's my problem15:58
wikingok15:58
wikingactually15:58
n4nd0the second one is a bit different15:58
wikingit'll be great now15:58
wikingi thought it'll be a problem that the latent SO and the basic latent stuff15:58
wikingwill have very different implementation15:58
wikingbut now since you are depending on mosek15:59
wikingi think it's even good15:59
wikingthat they are so 'separate'15:59
n4nd0I don't understand what you mean15:59
n4nd0what do you mean with separate?15:59
wikingwell16:00
wikingthe basic latent svm implementation has nothing to do with your SO16:00
wikingobviously16:00
wikingbut now when i've started to use your SO framework16:00
wikingthe whole latent SO svm is a complete rewrite of the basic latent implementation16:01
wikingthere's almost nothing they share in the implementation part16:01
n4nd0aham ok16:01
wikingso for me it wasn't 'elegant' enough since essentially they are similar16:01
wikingso i wanted to have it implemented as one16:02
wikingbut now with the whole mosek story16:02
wikingi even actually like the idea that it's like this16:02
n4nd0but is it just mosek what makes them so 'separate'?16:02
wikingsince if one does not have mosek, he can still have basic latent svm16:02
wikingno of course not16:02
wikingit's the whole SO16:02
n4nd0ok16:03
-!- blackburn [d55783a2@gateway/web/freenode/ip.213.87.131.162] has quit [Ping timeout: 245 seconds]16:04
-!- blackburn [d55783a2@gateway/web/freenode/ip.213.87.131.162] has joined #shogun16:05
n4nd0wiking: I am going to be away some minutes now16:05
wikingthat's ok16:05
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun16:10
shogun-buildbotbuild #946 of r_static is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/946  blamelist: cwidmer@tuebingen.mpg.de16:15
shogun-buildbotbuild #857 of octave_static is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/857  blamelist: cwidmer@tuebingen.mpg.de16:21
wikingn4nd0: ping me when u r back16:21
wikingplz16:21
n4nd0wiking: back16:21
wikingn4nd0: do u have an example implementation of so?16:22
wikingso_multiclass.cpp?16:22
n4nd0yes16:23
wikingok so i could use that actually16:23
n4nd0what do you want to do?16:23
wikingwell to have a working example for so latent svm :)16:24
n4nd0ok16:24
wikingmmm16:24
n4nd0tell me16:24
wikingdata :D16:25
wikingi'd like to have a meaningful input for the example16:25
shogun-buildbotbuild #933 of python_static is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_static/builds/933  blamelist: cwidmer@tuebingen.mpg.de16:27
n4nd0what latent datasets are around there?16:27
-!- alexlovesdata [82955843@gateway/web/freenode/ip.130.149.88.67] has joined #shogun16:27
wikingnone :D16:28
n4nd0haha there must be something c'mon16:28
n4nd0btw, it looks like the mailing list has a looong delay or something :P16:28
blackburnit must be russian post16:29
wikingn4nd0: well there's one i've mentioned already16:30
wikinglast week16:30
wikingbut that's a simple latent svm example16:30
n4nd0aham ok16:31
wikingthere's supposedly 2 so svm16:31
n4nd0but it could be good to start with something simple and see if the results are ok16:31
wiking(to which the author of the paper refers to)16:31
wikingbut that's not available16:31
n4nd0remember that my implementation has not been thoroughly tested yet16:31
wikinganywhere... and there's still no answer from the author about the dataset16:32
wikingyeah i know that's why i actually want to use it on a dataset so that i could actually compare16:32
wikinghaaah!16:35
wikingbounced!16:35
alexlovesdatawe can create a latent structured problem easily16:40
alexlovesdatausing a small part of existing datasets16:41
alexlovesdatasmall enough by trainnig data size so that a generic solver would do it for testing code sanity and interfaces16:41
wiking2012-06-19 15:30:23 1SgzVe-0002PU-KF <= wiking@maeth.com H=wsip-184-185-108-33.ri.ri.cox.net [184.185.108.33] P=esmtpsa X=TLSv1:AES128-SHA:128 A=cram:wiking@info2k1.hu S=2102 id=B340BD45-8B1D-43AE-91ED-8E7F60B7C22F@maeth.com16:41
wiking2012-06-19 16:34:14 1SgzVe-0002PU-KF => shogun-list@shogun-toolbox.org R=dnslookup T=remote_smtp H=mail.nn7.de [85.214.94.156] X=TLSv1:DHE-RSA-AES256-SHA:25616:41
wiking2012-06-19 16:34:14 1SgzVe-0002PU-KF Completed16:41
wikingthis is what happened16:41
wiking:>16:41
wikingyes16:41
wikingalexlovesdata: how?16:42
alexlovesdatacan you explain pls what hapneed in detail ;)16:42
alexlovesdatahappened16:42
wikingthis is my exim's log16:43
wikingalexlovesdata: how would u create the dataset?16:43
alexlovesdatawhat is a exim log?16:43
wikingalexlovesdata: exim = my mailserver16:43
alexlovesdataahh, ok that has to do with the email delays ...  I was not interested in that, that is not so important for me16:44
n4nd0haha good answer16:44
alexlovesdatano going for code is more important for me16:44
wikingalexlovesdata: so my problem is that i want to be able to compare our results16:44
n4nd0alexlovesdata: have you some experience trying out SO for multiclass classification?16:45
alexlovesdatano, mostly I work on multilabel problems16:45
wikingalexlovesdata: so that's why i wanted to get the at least on of those latent so datasets that chun nam is referring to16:45
alexlovesdatathere is a paper about SO for multilabel problems by Christoph Lampert NIPS201216:45
n4nd0multilabel is not multiclass right?16:46
alexlovesdataright: multi-class = mutually exclusive16:46
alexlovesdatamulti-label: one image can have many labels at the same time16:46
alexlovesdataI have used the code of joachims for multi-class SO16:47
n4nd0ok16:47
n4nd0struct svm?16:47
alexlovesdatabut it was (as expected) not better than one-vs-all16:47
n4nd0svm struct probably...16:47
alexlovesdatayes16:47
n4nd0it is because my multiclass so kind of sucks a bit ...16:47
alexlovesdatawhat means sucks?16:47
n4nd0the results look bad to me16:47
n4nd0I don't know if it is normal or my implementation is messed up16:48
n4nd0or both ...16:48
alexlovesdatabad compared to what?16:48
alexlovesdatafor multi-class SO you have to use a higher regconst compared to one vs all16:48
alexlovesdatabecause you \|w\16:48
n4nd0one-vs-rest multiclass built with liblinear binary classifiers16:49
n4nd0wait a moment there16:49
alexlovesdata\|w\|^2 is a sum of sub-ws compared to one vs all16:49
n4nd0higher regconst?16:49
alexlovesdatayes, the C16:49
n4nd0the C in C*sum_i slack_i?16:49
alexlovesdatabecause \|w_{mc}\|^2 = \sum_c \|w_c\|^216:50
n4nd0mmm I don't think I understand what you mean, sorry :S16:50
alexlovesdataso as a quick fix: take a good C from 1vsall and multiply it with number of classes16:50
alexlovesdatayes the C in the slack sum16:51
wikingn4nd0: hahah i have to stop you there :)16:52
wikingalexlovesdata: so how to create the data set :)16:52
wikingn4nd0: i've taken over the priority for my mentor :> ask the rest later :P16:52
alexlovesdatamy experience, nando is: struct mc is in the same order as 1 vs all if you set the regcnost well16:52
alexlovesdataok wiking I listen16:52
wikingalexlovesdata: dataset for latent so svm16:52
n4nd0wiking: all yours ;)16:53
alexlovesdatamulti-class object detection or multi-label object detection16:53
wikingthere's the two that chun nam is talking about... but no access to them...16:53
alexlovesdatalatent variable is position16:53
wikingstupid question: how's that structural?16:53
alexlovesdataor: take images of object classes with background (pascal VOC) and rotate them ... latent variable would be rotation16:54
alexlovesdataor maybe: mosaics of images with structure16:54
wikingok the last one i get  :)16:54
alexlovesdataneed to demosaic the image (that would be a fun paper for a conference!)16:54
wikingbut like mult-class object detection16:54
wikinghow's that structural16:54
alexlovesdatait is structural in the same way as struct multi-class is structural16:55
wikingok so your loss function ...16:55
wikingok16:55
alexlovesdataor multi-task object detection ...16:55
alexlovesdatawe can check: http://pub.ist.ac.at/~chl/papers/lampert-nips2011.pdf16:56
wikingn4nd0: actually how's your perf for SO multi class vs svm one-vs-rest ?16:56
alexlovesdatawhether you can do an object detection version of it16:56
wikingn4nd0: i should read the log :)16:56
wikingok multi-label could be fun actually16:57
wikingalexlovesdata: but then again then we could actually stick to the dataset i've already have16:57
wikingas there i have 6 mammals16:57
wikingi.e. multi-class problem16:57
n4nd0naywhayare: perf?16:57
wikingn4nd0: performance16:57
alexlovesdataI have no problem with your mammals16:57
n4nd0wiking: I don't like at all16:57
wikingalexlovesdata: ok then i guess i need a loss function for that16:58
alexlovesdataif they have varying positions in the images16:58
wikingyep16:58
n4nd0SO-SVM: 38.10%16:58
n4nd0MC:     57.70%16:58
wikingn4nd0:  :(16:58
n4nd0that is using the same data both for training and testing16:58
n4nd010 Gaussians with 100 samples each16:58
alexlovesdatafor multi-label detection: you can use pascal voc datasets (a small part of it, say 5 object classes where one can expect correlations)16:58
wikingn4nd0: hahah i'll try my stuff then16:58
wikingn4nd0: meaning my mammals :D16:58
alexlovesdatan4ndo: which Cs did you use???16:58
wikingalexlovesdata: people + car?16:59
alexlovesdatan4ndo: you can try joachims svm multiclass with same parameters to countercheck your results16:59
n4nd0alexlovesdata: I am using C = number of classes in so16:59
n4nd0alexlovesdata: C = 1 in the other16:59
alexlovesdatafor seeing whether the problem sucks or you have a bug in your code16:59
alexlovesdataI guess in SO-SVM one class is dominating the predictions17:00
alexlovesdatan4ndo: did you check duality gap of your solutions???17:00
alexlovesdataa bad duality gap may hint at too early termination17:01
n4nd0alexlovesdata: I don't know what is the duality gap :S17:01
alexlovesdatathis might be important to check17:01
alexlovesdataduality gap is (primal objective - dual objective)/dual objective17:01
alexlovesdatafor a convex problem close to the optimum this should be very small17:02
alexlovesdataif it is 0.1 or larger your problem is not close to convergence17:02
n4nd0ok17:02
wikingalexlovesdata: so actually then one more question: how would a SOLabel look like17:02
n4nd0alexlovesdata: how can I compute the primal and udal objectives?17:03
alexlovesdataSOLabel for multi-class would be a vector17:03
alexlovesdatafor multi-label same a vector17:03
wikingalexlovesdata: yeah but how would u construct that vector17:03
alexlovesdatawell i-th entry = i-th class17:03
n4nd0wiking: that is already done I think17:03
n4nd0wiking: check MulticlassSOLabels17:04
n4nd0under structure17:04
wikingn4nd0: mmm17:04
wikinglesse17:04
n4nd0each label is a real number17:04
wikingn4nd0: i was checking your example17:04
alexlovesdataN4ndo: what is the precise mathemtical formulation of your SO-SVM for multi-class?17:04
alexlovesdatacan you give a pdf with its formulation?17:05
alexlovesdatafor a general treatment of duality gap17:05
alexlovesdataBoyd Vandenberghe (the book about convex optimization) is a good starting point17:05
-!- blackburn [d55783a2@gateway/web/freenode/ip.213.87.131.162] has quit [Quit: Page closed]17:05
n4nd0alexlovesdata: http://iglesiashogun.wordpress.com/2012/06/04/second-weekly-report-gsoc-2012/17:06
n4nd0I have there a formulation of the optimziation problem17:06
wikingn4nd0: so basically you create a SGVector and for i-tth example you set the i-th value of the vector to the given real number (that corresponds to the label)?17:06
n4nd0in any case I screwed before17:06
n4nd0because the C I was changing is the C in min w'*C*w17:06
n4nd0not in C*sum_i slack_i17:06
n4nd0wiking: it is not a SGVector17:07
wikingn4nd0: i know, but the way to create it...17:07
alexlovesdataoops! your C!17:07
wikingn4nd0: i was just checking the constructor17:07
n4nd0wiking: ok17:07
alexlovesdatathen you need: C_struct=1/numclasses in that case17:07
n4nd0wiking: I mean if you tell me you create an SGVector for the labels, I have to tell you that I don't, because it is not done ;)17:07
n4nd0alexlovesdata: are both Cs equivalent in that sense?17:08
alexlovesdatais C a scalar or a matrix???17:08
n4nd0one the inverse of the other?17:08
n4nd0matrix17:08
wikingn4nd0: no i was just asking how that SGVector in the ctor is interpreted17:08
alexlovesdatain one vs all you have w^2 + C/n \sum_i slack_i17:09
alexlovesdatawhy is your C a matrix???17:09
alexlovesdataa diagonal one?17:09
n4nd0alexlovesdata: that matrix is something that Nico wrote in a paper he did for the problem formulation, I have just seen it there17:10
n4nd0right now for me it is just the identity matrix17:10
alexlovesdatait cannot be a general symmetric matrix!!! then your problem gets degenerated IF C has negative eiogenvalues17:10
n4nd0so like if it was not there17:10
alexlovesdatacan you give me  link to rthe formula of nico? (Alex sighs)17:11
alexlovesdatanando: someting is strange about your formulation17:12
alexlovesdataif my C is a scalar --- how do you initialize the C matrix in yuor approach - I think this is the mistake17:12
-!- blackburn [d5578951@gateway/web/freenode/ip.213.87.137.81] has joined #shogun17:12
n4nd0https://dl.dropbox.com/u/11020840/shogun/ideas.pdf17:12
alexlovesdatathx nando17:12
wikingn4nd0: ok i think i've got it: so in the vector for each sample you set the each class' membership value? so if i have 3 classes and 10 samples then my input vector is 30, right?17:12
n4nd0is input vector = weight vector17:13
n4nd0=17:13
n4nd0?17:13
wikingn4nd0: yes for the CMulticlassSOLabels17:13
alexlovesdatan4ndo: pls try C= diag(1)* number of samples/number of classes17:13
alexlovesdatadiag(1) = identity matrix17:13
wikingn4nd0: figured this from the example it's in the repo17:14
n4nd0wiking: 3 classes with feature dim = 10, gives a weight vector of 3017:14
n4nd0not with 10 samples17:14
n4nd0alexlovesdata: ok, thank you, let me try17:14
wikingn4nd0: so_multiclass.cpp line 39->5417:15
wikingespecially labs vector17:15
wiking :)17:15
wikingSGVector< float64_t > labs(NUM_CLASSES*NUM_SAMPLES);17:15
alexlovesdatan4ndo: how did you set the C matrix before???17:15
wikingn4nd0: and u use this labs vector to construct CMulticlassSOLabels17:16
wikingthat's why i said that this vector will contain the membership value for each class for each sample17:16
n4nd0alexlovesdata: see init_opt, at the end, https://github.com/iglesias/shogun/blob/so/src/shogun/structure/MulticlassModel.cpp17:17
wikingn4nd0: or I'm flying here with the pink elephants here in my head17:18
n4nd0wiking: ???17:19
wikingn4nd0: http://www.pascalkolkman.com/delirium/images/image1.jpg17:19
wikingn4nd0: i'm trying to figure out what's the sgvector content for CMulticlassSOLabels17:20
n4nd0wiking: the labels17:20
wikingyes17:20
n4nd0wiking: the class labels17:20
n4nd0alexlovesdata: wow the results have improved!!17:21
n4nd0I never got a 40% haha17:21
wikingn4nd0: so for example if i 1 one sample and 2 labels then it would look like: 1,017:21
wikingor 1, 0.517:21
wikingi.e. the real values are there something like a membership value for a given class17:21
wikingor?17:22
n4nd0alexlovesdata: I think something is still wrong though17:22
alexlovesdatawith which C?17:22
alexlovesdatayou can compare against svm multiclass of joachims using an equivalent C17:22
n4nd0alexlovesdata: but joachims' is in the dual17:22
alexlovesdata(joachims has the C on the slacks)17:22
alexlovesdatano, he has a primal solver, too, I am sure17:22
alexlovesdatathis is a parameter in svm multiclass17:23
n4nd0alexlovesdata: I tried to use the primal but I didn't manage17:23
n4nd0wiking: I don't really understand what you mean man17:23
n4nd0wiking: we have labeled data17:23
wikingn4nd0: :DDDDd17:23
n4nd0wiking: the class labels are 0, 1, 2, 3 .....17:23
wikingthis cannot be that difficult as it is now17:23
wikingyes17:23
n4nd0wiking: from 0 to num_classes-117:23
n4nd0wiking: that's all17:24
wikingso i have 2 labels and 1 sample17:24
n4nd0but that is not possible17:24
wikingwhy?17:24
n4nd0if you have two samples17:24
n4nd0you need two labels17:24
wiking???17:24
wikingi'm sorry but what? :)17:24
n4nd0your training data would be then {(x1, y1), (x2, y2)}17:24
wikingi have one fucking example17:24
n4nd0I think we need to make a dictionary17:24
-!- uricamic [~uricamic@2001:718:2:1634:397b:935c:85c:2381] has quit [Quit: Leaving.]17:25
wiking{(x1, y1}}17:25
wikingthat's it17:25
wikingbut my |Y| = 217:25
wikingok?17:25
n4nd0ok17:25
blackburnone training example for two classes??17:25
wikinghow do you construct that sgvector: 1.0, 017:25
n4nd0you cannot17:25
wikingblackburn: doesn'tmatter dummy example17:25
n4nd0because there should be at least one training example for every class17:26
n4nd0call me crazy if you like ...17:26
wikingok17:27
wikingi think i've got it17:27
wikingbut i don't understand it17:27
wiking:D17:27
n4nd0....17:27
wikingso your label vector would actually be something like 0,0,1,117:27
wikingfor 2 labels 2 samples17:28
wikingor?17:28
n4nd0I think that what you are saying doesn't make much sense17:28
wikinglol :D17:28
n4nd0because you want to train a machine to recognize examples17:28
n4nd0for a class for which you are not giving any sample17:28
wiking{(x1,y1), (x1,y2)}17:28
wikingi don't want to do this at all17:28
wikingi just want to understand your input vector17:28
n4nd0that is what you were saying before17:28
wikingi don't want to create this17:29
n4nd0I have already told you what the input vector is17:29
n4nd0it's exactly the same as the one for labels/MulticlassLabels17:29
wikingbut i wanted to narrow down the misunderstanding factor to the most stupidest example17:29
wikingok17:29
wikingand then i've asked back17:29
wikingabout those lines17:29
wikingin the xample17:29
wikinglabs[c*NUM_SAMPLES+i] = c;17:30
n4nd0ok17:30
wikingwhether that is actually a membership value (that's why it's real variable) or why?17:30
n4nd0what's the problem there?17:30
wikingbut no17:30
n4nd0membership value?17:30
wikingapparently it's not a membership value of a sample for a class17:30
n4nd0what is taht?17:30
n4nd0the label?17:30
wikingn4nd0: http://en.wikipedia.org/wiki/Membership_function_(mathematics)17:30
wikingfuzzy?17:31
wikingyou have a membership value for a given class?17:31
n4nd0just tell me what you mean with it17:31
wikingi cannot it seems17:31
wikingbecause if you scroll back this conversation17:31
blackburnwhy labs is a matrix?17:31
wikingnothing makes sense17:31
alexlovesdatasorry I am with a physical student for 30 minutes17:31
blackburnwiking: c is the class number17:32
wikingblackburn: i've got that17:32
wikingso that it's not membership17:32
n4nd0labs[c*NUM_SAMPLES+i] = c17:32
wikingbut this way17:32
n4nd0NUM_SAMPLES = number of training examples of each class17:32
blackburnone thing I do not understand too is why labs is a matrix?17:32
n4nd0blackburn: it is not a matrix17:33
wikingn4nd0: it is17:33
blackburnah long vector17:33
n4nd0blackburn: I might be indexing it so it looks like that, but it is not17:33
n4nd0wiking: void gen_rand_data(SGVector< float64_t > labs, SGMatrix< float64_t > feats)17:33
wikingn4nd0: i'm reading that yes17:33
n4nd0(SGVector< float64_t > labs17:33
wikingyes17:33
wikingstill17:33
n4nd0labs[c*NUM_SAMPLES+i]17:33
wikingOMFG17:33
n4nd0is just a way to index17:33
wikingn4nd0: NUM_SAMPLES = number of training examples of each class17:33
blackburnso it has NUM_SAMPLES*NUM_CLASSES length?17:33
n4nd0it doesn't make it a matrix17:34
n4nd0blackburn: yes17:34
wikingn4nd0: ok now i've got it :D17:34
wikingi thought you have NUM_SAMPLES17:34
wikingin sum17:34
n4nd0wiking: dude .... you should read the full example before17:34
n4nd0:)17:34
wikingn4nd0: i've read it17:34
wikingn4nd0: but it's impossible that it wasn't understandable what i was asking17:34
blackburnok I think I got it too17:34
blackburnso you use it to simplify that setting17:34
blackburnto determine which examples are of some class17:35
blackburnright?17:35
n4nd0ok so I want to generate distinct data for each class17:35
blackburnI didn't read anything from the example but it may be confusing a little17:35
n4nd0I loop for each of the classes17:36
n4nd0generate a mean and a variance that I will use for that class (data distributed as Gaussian)17:36
n4nd0and later I loop again17:36
wikingbrb17:36
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]17:36
n4nd0generating the feature vectors for that class17:36
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 260 seconds]17:55
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has quit [Ping timeout: 245 seconds]17:57
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun18:01
wikingdoh n4nd0 is off18:01
wikingalexlovesdata: here?18:02
wikingalexlovesdata: so anyhow, about the multi-class data set i have: as said i have that mammals data set with 6 different mammals, so now that i *think* i understand how the CMulticlassSOLabels is constructed i'll try to use that dataset with latent so svm for testing. although i'll try it first with simple CPrimalMosekSOSVM to see the the accuracy w/o the latent var18:08
-!- gsomix [~gsomix@188.168.2.34] has joined #shogun18:18
gsomixhey18:18
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]18:31
-!- gsomix [~gsomix@188.168.2.34] has quit [Remote host closed the connection]18:49
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun18:58
wikingalexlovesdata: ping?19:06
@sonney2kblackburn, hmmhh why did you merge chris' pull request?19:21
@sonney2kit will explode sparsevector19:21
@sonney2kand uses std crap19:21
blackburnsonney2k: chris asked me to merge that today - I'll change it later19:22
blackburnhe needed that for some project19:22
blackburnI do not like it too but he asked and I'll change so no worries19:22
blackburnok?19:23
@sonney2kblackburn, I would say we introduce some big overhead sparse vector that can have chris' stuff19:27
@sonney2kas we discussed19:27
@sonney2kbut this should really not be in shogun until done properly19:27
@sonney2kblackburn, any news from n4nd0 ?19:28
@sonney2kcan I merge his SO stuff?19:28
blackburnsonney2k: I promise I'll remove that just after I get BSc;)19:28
blackburnsonney2k: he said results improved19:28
blackburntoday he played with C and it became better19:29
@sonney2kblackburn, I don't care about correctness (for now) just the framework19:29
@sonney2kothers need it so it is better to have it in main19:29
blackburnjust merge it then19:29
@sonney2kk19:29
wikingsonney2k: n4nd0 said that it's ok and it works/compiles and the example is running as well19:30
wiking(for me)19:30
CIA-18shogun: Soeren Sonnenburg master * r24aad4d / (38 files in 10 dirs): Merge pull request #580 from iglesias/so (+29 more commits...) - http://git.io/DO21UA19:31
@sonney2kwiking, alright - it is in then19:31
wikingok great19:31
wikingthnx a lot!19:32
@sonney2kwiking, so now I want to see your PR!19:32
wikingyep19:32
wikinglet's see if it merges... pulling19:32
wikingDUUUDE!!!19:34
wikingThe data for NPcoref can be found at:19:34
wiking\o/19:34
wikingfinally!19:34
* wiking has latent so dataset!19:34
wikingalexlovesdata: ping ping19:35
wikingalexlovesdata: chun nam just dropped me an email now finally!19:35
CIA-18shogun: Michal Uricar master * rd6c2696 / (6 files): DualLibQPBMSOSVM interface for libbmrm - http://git.io/J57Bbw19:36
CIA-18shogun: Michal Uricar master * rc504412 / (4 files): Function get_dim removed from RiskFunction (will be available in StructuredModel) - http://git.io/PL1h5A19:36
CIA-18shogun: Michal Uricar master * r4c424e4 / (4 files): Indentation fixed + getter/setter for bmrm parameters in CDualLibQPBMSOSVM - http://git.io/a9u69Q19:36
CIA-18shogun: Soeren Sonnenburg master * rc16fde6 / (6 files): Merge pull request #582 from uricamic/BM_SOL - http://git.io/3A8T9A19:36
shogun-buildbotbuild #1073 of libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/1073  blamelist: fernando.iglesiasg@gmail.com19:36
blackburnit is the merge day19:36
wikingdoh what is this build error?!19:36
@sonney2kwiking, you never know - maybe fernando forgot to use git commit --amend ...19:38
wikingmmm19:38
@sonney2kblackburn, do you know what the state of puffin's PR is?19:38
@sonney2kexcept for the live user in a git commit :D19:38
wikingshogun/kernel/ScatterKernelNormalizer.h19:38
wikingso it seems it's rather the kernel normalizer related stuff...19:39
@sonney2kn4nd0 certainly did not touch those19:40
@sonney2kblackburn, are the kernel normalizers now in kernel/normalizer ?19:40
wikingkmcquisten/master19:40
blackburnnot I didn't that yet19:40
wikingsonney2k: aa9ad6a30c9e4cb08aec27bdf00c39e4332834be19:40
wikingblackburn: it's your merge ;P19:40
blackburnsonney2k: we shall wait for puffin444 cause he was stucked with porting nlopt19:40
shogun-buildbotbuild #1074 of libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/107419:41
blackburnI have no access to git now so please fix ;)19:41
wikingbrb19:42
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]19:42
@sonney2kblackburn, uricamic's stuff is still in so folder19:45
shogun-buildbotbuild #1075 of libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/1075  blamelist: sonne@debian.org, uricar.michal@gmail.com19:45
@sonney2knot structure19:45
@sonney2kso will break anyways19:45
@sonney2kTest19:50
@sonney2kVery good works19:51
@sonney2kFinally online via mobile phone19:51
CIA-18shogun: Soeren Sonnenburg master * rc000355 / (12 files in 2 dirs): fix compile error due to so -> structure rename - http://git.io/a8cq_Q20:02
shogun-buildbotbuild #949 of r_static is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/94920:05
shogun-buildbotbuild #1076 of libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/107620:20
shogun-buildbotbuild #950 of r_static is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/950  blamelist: sonne@debian.org20:30
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has joined #shogun20:41
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has quit [Changing host]20:41
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun20:41
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]20:50
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has joined #shogun20:51
-!- wiking [~wiking@wsip-184-185-108-33.ri.ri.cox.net] has quit [Changing host]20:51
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun20:51
-!- emrecelikten [~emre@176.40.254.205] has joined #shogun21:10
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]21:29
-!- alexlovesdata [82955843@gateway/web/freenode/ip.130.149.88.67] has quit [Quit: Page closed]21:43
-!- blackburn [d5578951@gateway/web/freenode/ip.213.87.137.81] has quit [Ping timeout: 245 seconds]22:14
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has joined #shogun23:34
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun23:40
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]23:53
--- Log closed Wed Jun 20 00:00:41 2012