Open in new window / Try shogun cloud
--- Log opened Thu Jun 14 00:00:41 2012
-!- heiko [~heiko@host86-183-72-166.range86-183.btcentralplus.com] has joined #shogun00:08
-!- heiko [~heiko@host86-183-72-166.range86-183.btcentralplus.com] has quit [Ping timeout: 265 seconds]00:23
-!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: Connection timed out]00:49
-!- emrecelikten [~Anubis@176.41.25.77] has joined #shogun01:05
-!- emrecelikten [~Anubis@176.41.25.77] has quit [Quit: Leaving.]01:34
-!- emrecelikten [~emrecelik@176.41.25.77] has joined #shogun01:35
-!- gsomix [~gsomix@188.168.5.148] has quit [Ping timeout: 240 seconds]03:07
-!- gsomix [~gsomix@188.168.5.148] has joined #shogun03:09
-!- emrecelikten1 [~emrecelik@176.40.254.205] has joined #shogun03:12
-!- emrecelikten [~emrecelik@176.41.25.77] has quit [Ping timeout: 245 seconds]03:13
-!- emrecelikten1 is now known as emrecelikten03:20
-!- emrecelikten [~emrecelik@176.40.254.205] has quit [Quit: Leaving.]03:32
-!- gsomix [~gsomix@188.168.5.148] has quit [Ping timeout: 244 seconds]03:34
-!- gsomix [~gsomix@188.168.5.148] has joined #shogun03:34
-!- vikram360 [~vikram360@117.192.163.180] has quit [Ping timeout: 248 seconds]04:00
-!- vikram360 [~vikram360@117.192.176.63] has joined #shogun04:34
-!- vikram360 [~vikram360@117.192.176.63] has quit [Ping timeout: 248 seconds]04:43
-!- vikram360 [~vikram360@117.192.175.104] has joined #shogun04:44
-!- wiking [~wiking@ip191.67-202-72.static.steadfastdns.net] has joined #shogun04:47
-!- wiking [~wiking@ip191.67-202-72.static.steadfastdns.net] has quit [Changing host]04:47
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun04:47
-!- vikram360 [~vikram360@117.192.175.104] has quit [Ping timeout: 244 seconds]04:48
-!- vikram360 [~vikram360@117.192.175.104] has joined #shogun04:49
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 265 seconds]05:23
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun05:24
-!- vikram360 [~vikram360@117.192.175.104] has quit [Read error: Connection reset by peer]05:31
-!- vikram360 [~vikram360@117.192.175.104] has joined #shogun05:46
-!- gsomix [~gsomix@188.168.5.148] has quit [Ping timeout: 246 seconds]08:21
-!- gsomix [~gsomix@188.168.5.148] has joined #shogun08:31
-!- uricamic [~uricamic@2001:718:2:1634:5c39:53a4:ae34:8c40] has joined #shogun08:51
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun08:55
-!- gsomix [~gsomix@188.168.5.148] has quit [Ping timeout: 265 seconds]09:03
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]09:06
-!- heiko1 [~heiko@host86-181-80-133.range86-181.btcentralplus.com] has joined #shogun10:44
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun12:27
-!- vikram360 [~vikram360@117.192.175.104] has quit [Ping timeout: 240 seconds]12:28
n4nd0uricamic: hi! how is it going?12:29
uricamichi n4nd012:29
uricamicI have noticed that you have done some changes in the so framework12:29
n4nd0yes, I had a discussion with Nico, he didn't support the design separated into classes different classes for the argmax, the loss, etc12:30
n4nd0uricamic: does the change affects you a lot?12:31
uricamicok, in that case I think the class for the Risk function is not necessary and it would be better to provide the risk function in the same way as you did it with argmax, etc.12:31
n4nd0yes12:33
-!- vikram360 [~vikram360@117.192.175.104] has joined #shogun12:34
n4nd0but at first sight it shouldn't be problematic right?12:34
n4nd0I mean if you have some implementation already done in a risk function class12:35
n4nd0what we should do is12:35
n4nd01) add a method for the risk function in CStructuredModel12:35
n4nd02) the computation of the risk you may have done is moved to a class that inherits from CStructuredModel12:36
n4nd0do you agree?12:36
uricamicyes, I agree, the computation of the risk will be provided only in some examples, for example you are now working on the multiclass classifier, I can easily extend this example with the possibility to use dual BM solver instead of the primal one12:37
uricamicso, I can write the implementation of the risk function for the multiclass classifier12:38
uricamicnow I just need some of your pieces of so framework to be pushed into shogun, then I will be able to create some examples, and so on12:39
n4nd0ok12:40
n4nd0aobut the multiclass classification example12:40
n4nd0I would like to ask you about it12:41
n4nd0since my results are pretty bad right now12:41
n4nd0i.e. there must be something wrong :D12:41
uricamicok12:41
uricamicI will look for my implementation of risk function etc. in matlab12:41
n4nd0so I wonder if the election of the matrix C can be the problem12:41
n4nd0is this important?12:42
n4nd0right now I am just using the identity matrix12:42
uricamicC is the regularizer?12:42
n4nd0yes12:42
n4nd0it is the matrix in the quadratic term of the optimization problem12:43
uricamicwell, I was using just a single regularization constant12:43
n4nd0mmm12:43
uricamicI think for the BMRM it is possible to use such matrix also, but then one needs to compute the inverse of this matrix12:44
n4nd0I am just thinking how I should input it to mosek then12:44
n4nd0the problem in mosek is formulated as12:44
n4nd00.5?x'Q^0?x12:44
n4nd0Q^0 is all zeros except the left-top most part12:45
n4nd0that is the regularizer C12:45
n4nd0if I want to input a constant, then I guess I have to give a matrix of all terms equal?12:46
n4nd0or a diagonal matrix?12:46
n4nd0do you see what I mean?12:46
uricamicyep, but I need to see the definition of the problem in mosek12:47
n4nd0ok, I am sending it to you12:47
n4nd0http://docs.mosek.com/6.0/capi/node007.html#32974770412:48
uricamicI think it should be diagonal12:49
n4nd0ok12:49
n4nd0so probably to use the identity matrix is not the problem12:49
uricamicit shouldn't be, but it is important to choose some reasonable value of the regularization constant12:50
n4nd0how does one choose a reasonable value?12:50
uricamice.g. split training data to 3 parts, 1 part will be training on which you will try several values of this constant, like 1e-3, 1e-2, ... the next part will be validation, on which you will find out which regularization constant performs best on your data and the last one is for testing purposes, where you will use the classifier trained with the "optimal " regularizer12:51
n4nd0ok, like using cross validation then12:53
uricamicyep, exactly12:54
uricamicso for you it is sufficient now just to have 2 parts - one for training and the other one for validation of the regularization constant12:54
n4nd0ok12:55
uricamicyou will train the classifier with all regularization constants from you set and then according to the value of the objective function computed on the validation data you will select the regularizer which minimizes this12:55
n4nd0I will read again the code first though, just in case I can find any bug12:55
uricamicok, sure12:56
n4nd0I think that the results are pretty bad, I thought they should look better even if I am not performing cross validation12:56
uricamicI will have to go away for approx. 1 hour12:56
uricamiceverything depends on the value of the regularization constant12:57
uricamicif it is e.g. too high, then the results can be really bad12:57
n4nd0aham12:57
n4nd0I will try with other values then as well12:58
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 265 seconds]13:16
-!- vikram360 [~vikram360@117.192.175.104] has quit [Ping timeout: 244 seconds]13:30
-!- heiko1 [~heiko@host86-181-80-133.range86-181.btcentralplus.com] has quit [Ping timeout: 265 seconds]13:34
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun13:37
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has joined #shogun13:41
-!- cronor [~cronor@fb.ml.tu-berlin.de] has joined #shogun13:42
cronorHow can I enable debug output in Python? I get: terminate called after throwing an instance of 'shogun::ShogunException'13:45
n4nd0cronor: hi13:50
n4nd0cronor: you should run ./configure --enable-debug and re-compile shogun13:50
-!- blackburn [d5578d4d@gateway/web/freenode/ip.213.87.141.77] has joined #shogun13:52
nicococon4nd0: hi there13:53
n4nd0nicococo: hey13:53
nicococoso, the example is working, you wrote?13:54
n4nd0nicococo: but the results are not good13:55
blackburnhey I am a bit lost last days13:55
blackburn:D13:55
blackburnn4nd0: does everything go ok? :)13:55
n4nd0I am going to try with different values for the regularization C13:55
nicococon4nd0: thats a good idea.. they should have comparable results13:55
n4nd0blackburn: so so, I keep on working :)13:55
n4nd0nicococo: as I told you, I am just using C = I, is that a good election?13:56
n4nd0blackburn: done with your thesis and exams?13:56
blackburnn4nd0: just had hmm lets say pre-defence talk :)13:56
n4nd0blackburn: all right13:57
blackburnyesterday passed my exam about everything13:57
n4nd0blackburn: I bet you'll defend it good13:57
n4nd0nice, congrats13:57
nicococoyes, thats fine.. if you want to change the regularization just multiply the constant with the identity matrix R = reg*I13:57
blackburnthanks13:57
-!- vikram360 [~vikram360@117.192.175.104] has joined #shogun14:00
n4nd0nicococo: any good intuition for a value for reg?14:00
nicococo1 ;)14:01
n4nd0well14:02
n4nd0that is to use the identity matrix right?14:02
nicococotake the same as for the other method14:02
n4nd0that is what I was doing and getting bad results14:02
n4nd0the other method?14:02
nicococoone-vs-rest svm14:02
nicococois the formulation of the 1-vs-rest svm with or without biad term?14:03
nicococoops.. bias term14:03
blackburngmnp should support bias14:04
blackburnnot regularized - true bias I mean14:04
n4nd0with bias14:04
n4nd0blackburn: I am using liblinear14:04
nicococothat would be a reason.. we are currently not using a bias term, right?14:04
blackburnn4nd0: multiclassliblinear?14:04
n4nd0blackburn: no, one vs rest strategy built with liblinear14:05
blackburnahh14:05
blackburnok14:05
n4nd0blackburn: but those results are ok, perfect14:05
n4nd0it is multiclass so what fails14:05
n4nd0so my bad14:05
blackburnI understand14:05
blackburnI just wanted to say composition of liblinears won't exactly coincide14:05
n4nd0blackburn: ok, thank you :)14:07
n4nd0I don't think that is exactly the problem now though14:07
blackburnyeah got it14:07
n4nd0blackburn: why are the two different Cs in LibLinear btw?14:07
blackburnn4nd0: pos / neg14:07
n4nd0aham, all right14:07
n4nd0nicococo: it is set to 1 as well14:08
blackburnallows you to balance a little14:08
n4nd0nicococo: so there must be some bugs in the code definetely14:08
blackburne.g. if you have 10 pos and 10000000 negs14:08
blackburn:)14:08
n4nd0blackburn: yeah, I guess it can be good for applications with skewed training data14:08
n4nd0blackburn: exactly14:08
n4nd0nicococo: so it seems that the bias was playing an important role14:11
n4nd0nicococo: now the results of one-vs-rest are bad too14:12
n4nd0but still, they are very different14:12
n4nd0nicococo: https://dl.dropbox.com/u/11020840/results14:13
n4nd0if you want to see the results ^14:13
nicococoi would suggest to use c=2 classes .. the code looks okay to me14:13
n4nd0the first column is so-multiclass and the second one-vs-rest14:14
-!- pluskid [~pluskid@202.130.113.138] has joined #shogun14:14
n4nd0and at the beginning of each line is a number for the index of the training example that has no use :D14:15
nicococoseems that there are only two classes assigned (9 and 4)14:15
n4nd0nicococo: oh wait14:15
n4nd0nicococo: it looks that they are not using the same labeling or something, what I find pretty weird14:16
n4nd0but with an example for 2 classes14:16
n4nd0thre results are entirely swapped14:16
nicococoswapped means that sosvm delivers 1-vs-r-svm result *(-1) ??14:17
n4nd0yeah14:17
n4nd0nicococo: https://dl.dropbox.com/u/11020840/results14:18
nicococodid you assign -1 and +1 classes the right way for 1-vs-rest (seems that you swapped these classes)14:18
n4nd0nicococo: I just used some code in shogun for one-vs-rest, let me check14:19
n4nd0nicococo: eeeh in the last file I sent everything looks all right14:21
n4nd0ok14:23
n4nd0I have executed several times and it looks good14:23
n4nd0for two classes14:23
nicococook.14:23
nicococodo both algorithm train on the very same (same same) data (not just same distribution)14:24
n4nd0I am using the same same data14:24
nicococoand you test on the train set14:25
n4nd0nicococo: do you want to use the come for the example?14:25
n4nd0yes, I am testing in the training test14:25
nicococothen you also have corresponding slack variables.. right14:25
n4nd0corresponding?14:25
nicococofor each training example exist one slack variable14:26
n4nd0in sosvm we're doing it like that yes14:26
n4nd0I am not sure about one-vs-rest14:26
n4nd0I am using the primal for it14:26
nicococonot important.. but you can plot::    id: true_class pred_class slack for each example14:27
n4nd0the slack in sosvm?14:27
nicococoyes14:27
n4nd0what do you want me to print associated to each class exactly?14:29
n4nd0the variable introduced in the optimization vector for each training example?14:29
nicococothe outcome of the optimization is a vector x containing the w_i and slack_n  n=1..#train_exmaple14:30
n4nd0yes14:32
nicococohence, for each example you have the true class, the predicted class and the slack-value.. if you print this information instead of just id: pred_class we probably see more whats going on14:33
nicococowhen 90% of the example have a slack >0 we can assume that the regularization-value is too restrictive14:34
n4nd0aham, good idea14:35
n4nd0let me do that then14:35
n4nd0nicococo: we should use the slacks of the last time the optimization problem is solved?14:48
n4nd0with use I mean to print them on screen14:48
nicococoyes14:48
n4nd0nicococo: but one thing14:51
n4nd0so we are doing14:51
n4nd0min w'*C*w + sum_i slack_i14:51
n4nd0right?14:51
nicococojep14:52
nicococo0.5w'Cw ...14:52
n4nd0yeah14:52
n4nd0ok sorry, I thought one thing wrong :D14:53
n4nd0anyway14:53
n4nd0right now I get that all the slacks are equal to zero14:53
n4nd0I am going to check more carefully14:53
nicococoif they are zero then there shouldn't be any wrong label assignment14:53
n4nd0true14:54
cronorWhy is there no module Labels? i.e. I have to from shogun.Features import RegressionLabels, why not from shogun.Labels import RegressionLabels?15:07
n4nd0cronor: they are in features15:10
n4nd0cronor: the labels were placed inside this directory15:10
n4nd0cronor: the creation of the labels directory is somewhat new and regarding the modules they remain in the same place so far15:10
cronorah ok15:11
n4nd0nicococo: so the slacks were not exactly but most of them are very small15:11
n4nd0about e-1115:11
nicococoand the predicted labels == training labels?15:12
cronorn4nd0: I recompiled shogun with --enable-debug. How can I enable debug output now? I don't' see any debug messages so far15:15
-!- emrecelikten [~emre@176.40.254.205] has joined #shogun15:15
n4nd0nicococo: for the two class problem yes, https://dl.dropbox.com/u/11020840/results15:16
n4nd0nicococo: the order there is true label, predicted label, slack15:16
n4nd0cronor: I think you have to set the debug level somewhere too15:17
n4nd0blackburn: do you know how it works exactly?15:17
nicococowell, seems to work pretty good.. now the multiclass example15:17
n4nd0blackburn: cronor is trying to see the debug messages from python15:17
cronorI saw .io.set_loglevel(..) but what do i fill in?15:18
n4nd0nicococo: for three clases it looks pretty bad :( https://dl.dropbox.com/u/11020840/results15:18
n4nd0cronor: I am not totally sure because I have never used this functionality15:19
n4nd0cronor: but, set_loglevel accepts an EMessageType15:19
cronorfrom shogun.Regression import MSG_DEBUG; ...io.set_loglevel(MSG_DEBUG) could be possible15:20
n4nd0cronor: http://www.shogun-toolbox.org/doc/en/current/namespaceshogun.html#ad2ca0a3c84404274f8bfaa618869fdf915:20
n4nd0these ^ are the possible values of EMessageType15:20
cronori'll ty15:20
cronortry15:20
n4nd0but better check it in the code since the doc may not be updated15:20
nicococon4nd0: back in 1h..15:21
n4nd0nicococo: any clue about this?15:21
blackburn+uh15:22
blackburnn4nd0: okay shogun.Anything is just for legacy15:24
blackburnit is setup somewhere in python modular iirc15:24
blackburnset up*15:24
n4nd0yeah15:26
n4nd0the doubt was more related about the debug messages15:26
n4nd0but I think cronor figured it out already15:27
blackburnah sorry15:27
cronoryes i have debug output now15:27
blackburnI was asleep actually so just woke up with that msg sound :D15:27
n4nd0haha15:27
n4nd0I am sorry I woke you man15:27
blackburnno problem I shouldn't be asleep I think :)15:28
pluskidhi boys15:32
n4nd0hey15:33
pluskidis there any lib in shogun if I want to use gradient descent to solve an optimization problem?15:34
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun15:34
n4nd0I don't know about that :S15:35
pluskidblackburn: any suggestion?15:36
blackburnpluskid: in shogun?15:37
pluskidblackburn: yes, one of the algorithm needs a gradient descent in a step15:37
pluskidactually, it is using the "Nesterov's accelerated gradient method", but I guess ordinary gradient descent might also work15:38
blackburnhuh no I don't think we have something here15:41
blackburnI port some code from SLEP that is based on Nesterov's method but I don't think it is helpful..15:41
wikingasdf!@$%15:49
cronorI get a shogun Exception when I use MKLRegression.apply(). Training works fine. I can't find the error. Can someone check the debug output: http://pastebin.com/5Wp6emf0 ?15:58
pluskidso do you guys know there are some 3rd party libraries usable for my case?15:59
pluskidn4nd0: I heard you are using an optimization library16:00
pluskidwhat's it for?16:00
pluskidis it specifically for convex optimization? or more general?16:00
blackburnpluskid: what is type of problem you solve?16:00
n4nd0pluskid: I am using mosek to solve a QP16:01
pluskidit IS convex I guess16:01
n4nd0pluskid: I am not sure if they have methods for non-convex problems16:02
pluskidhmm, seems to be convex but not QP16:03
n4nd0pluskid: what?16:03
pluskidmy problem16:04
n4nd0your problem?16:04
n4nd0ok16:04
pluskidfrom ShareBoost16:04
n4nd0but mosek can be used for more apart from QP16:04
n4nd0how does the problem look like?16:04
pluskida little bit complicated, log \sum exp (blah blah)16:05
pluskidmaybe I should look at mosek16:05
nicococopluskid: log-sum-exp is a convex funktion.16:07
pluskidnicococo: ah, ok16:07
pluskidand thanks!16:07
pluskidthen I guess mosek can handle this if it handles general convex problem16:08
nicococoi think mosek won't help but it is a smooth function therefore quasi-newton methods16:08
nicococowould fit nice16:08
pluskidnicococo: should I implement quasi-newton myself or use some library?16:08
blackburngenerally I'd like to avoid mosek here if you can implement some (even a little worse) method by yourself16:09
nicococouse some library, i think there are open source implementations16:09
pluskidOK, I see16:10
pluskidI'll google for that16:10
pluskidor do you already know some famous or standard open source lib for this?16:10
nicococommhh.. sorry, i just use code snippets flying around my harddrive..16:12
pluskidhmm, thanks anyway16:12
nicococobut lookout for l-bfgs.. i guess they are implemented more or less the same (have a look at marc schmidts homepage .. http://www.di.ens.fr/~mschmidt/)16:13
-!- blackburn [d5578d4d@gateway/web/freenode/ip.213.87.141.77] has quit [Quit: Page closed]16:13
pluskidok16:13
pluskidseems most of his codes are for matlab?16:15
nicococoahh.. yes.. sorry.. we like c++  , i sometimes forget this ;)16:16
pluskid:D16:16
-!- pluskid [~pluskid@202.130.113.138] has quit [Quit: Leaving]16:45
n4nd0nicococo: do you know if there is another implementation of the SO framework similar to  what we have done so far?16:45
n4nd0I think it could be useful to find the bugs16:45
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun16:46
eric_hi16:46
nicococon4nd0: the hmsvm toolbox (from mloss.org) is quite similar16:46
n4nd0nicococo: aham, I thought it was exclusively for hmsvm16:49
ckwidmerhi16:50
ckwidmeranyone here?16:50
n4nd0ckwidmer: yes16:50
ckwidmerwhat's the new speak for labels->get_labels() for the current design?16:50
ckwidmercplex is still broken16:51
ckwidmertrying to fix cplex svm16:51
ckwidmerI haven't followed too closely how the new label classes work16:51
n4nd0I am not sure what you mean with new speak :S16:51
n4nd0probably my bad16:51
ckwidmerhow to express it now16:52
nicococon4nd0: yes, but the solver is still the same.. otherwise i don't know any implementation of multi-class in so-manner16:52
ckwidmerwhat was it replaced by16:52
nicococohi chrissss.16:52
ckwidmerhi nicocococococo16:52
ckwidmercoco16:52
-!- heiko [~heiko@host86-181-80-133.range86-181.btcentralplus.com] has joined #shogun16:52
ckwidmern4nd0, any idea?16:53
n4nd0ckwidmer: so now there's no get_labels in the base Labels class16:53
n4nd0ckwidmer: but I guess you are probably using it for something that needs DenseLabels?16:53
n4nd0then get_labels should work just fine16:54
ckwidmerah16:54
ckwidmerso, all the other SVMs use DenseLabels?16:54
n4nd0so maybe you just need to do an explicit casting16:54
n4nd0we have StructuredLabels, BinaryLabels and MulticlassLabels16:54
n4nd0and it looks like this16:55
n4nd0Labels <--- StructuredLabels16:55
n4nd0Labels <--- DenseLabels16:55
n4nd0DenseLabels <--- BinaryLabels16:55
n4nd0DenseLabels <--- MulticlassLabels16:55
ckwidmerand are binary labels actually binary16:55
ckwidmeras in 0 and 116:55
ckwidmeror int16:55
n4nd0-1 and 116:55
ckwidmeror can I get floats from them16:55
ckwidmerok16:55
n4nd0internally are stored with float64_t16:55
ckwidmerso for predictions16:56
ckwidmerwhich output scores16:56
ckwidmerI use BinaryLabels?16:56
n4nd0maybe RegressionLabels?16:56
n4nd0I forgot to tell you about those :O16:56
ckwidmerokok16:56
n4nd0DenseLabel <--- RegressionLabels16:56
ckwidmerso even svm *classifiers* output regression labels then?16:57
-!- uricamic [~uricamic@2001:718:2:1634:5c39:53a4:ae34:8c40] has quit [Quit: Leaving.]16:57
n4nd0it depends I think16:57
n4nd0in the problem you want to solve16:57
n4nd0I would say that most of them return BinaryLabels16:58
ckwidmerbut what if you don't want the sign(output), but rather output?16:58
n4nd0then I think you have to look to something different, not to the labels16:59
n4nd0let me check if I see it16:59
n4nd0mmm maybe you can get the weight vector after training and just perform the product?17:00
n4nd0I wonder if that can turn out to be too tedious17:01
ckwidmerah, it's ok for now17:01
ckwidmerdo you know what happened to SGMatrix?17:01
n4nd0what do you mean?17:01
ckwidmerseems there were some changes17:01
n4nd0I know that a bunch of methods from CMath where moved to SGMatrix and SGVector17:01
ckwidmere.g. H.destroy_matrix(); seems to be gone17:02
-!- vikram360 [~vikram360@117.192.175.104] has quit [Ping timeout: 244 seconds]17:02
n4nd0aham17:02
n4nd0I think they are not required any more17:02
n4nd0if you just create your matrix using true in the constructor17:02
ckwidmergreat :)17:03
n4nd0what's done by default17:03
n4nd0then it will be "automagically" dereferenced17:03
-!- vikram360 [~vikram360@117.192.168.146] has joined #shogun17:03
n4nd0if you want to it similar to how it used to be, then you should use free_data17:04
n4nd0I think this change comes from the introduction of SGReferencedData, a base class for SGVector and SGMatrix17:04
n4nd0nicococo: I am just thinking how we can debug the method effectively17:05
nicococon4nd0: so the two-class problem works (which is a great indicator) lets try the multi-class again.. check, the slacks and select the regularization .. i have to go now, but tomorrow we can discuss the results. if it is still not convincing we have to dive deeper17:09
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun17:11
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has left #shogun []17:12
ckwidmern4nd0, I fixed a few problems with cplex dependent classes and made a pull request17:15
ckwidmerthere are still problems left in SubGradientLPM, which need to be fixed (adjusting to the new label design)17:16
ckwidmercould you please take a look?17:16
-!- heiko [~heiko@host86-181-80-133.range86-181.btcentralplus.com] has left #shogun []17:33
n4nd0ckwidmer: sorry I am half-absent17:39
n4nd0ckwidmer: to the pull request or to the other problems to be fixed?17:39
ckwidmerideally both :)17:40
-!- eric_ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Quit: Page closed]17:40
n4nd0I have not power to merge pull requests so let's leave that to blackburn or sonney2k17:42
n4nd0I am not proffesional :P17:42
n4nd0and what are the problems with SubGradienLPM? at least it should compile, right?17:42
ckwidmerdoesn17:47
ckwidmert17:47
ckwidmercompile unfortunately17:47
ckwidmerthere are problems with get_labels all over the place17:47
ckwidmernot a biggie17:47
ckwidmerbut needs to be fixed17:47
ckwidmeris sonney2k around today?17:47
n4nd0I think he is a bit off these last days17:53
ckwidmerok17:55
n4nd0ckwidmer: is it only in SubGradientLPM where the compilation with cplex enabled fails or also in other parts?17:57
ckwidmeras far as I can tell, it's the only problem left17:57
ckwidmerI fixed two other problems17:57
ckwidmerin CPLEXSVM17:58
ckwidmerand LPM17:58
n4nd0I am saying it because it would be nice to have all these fixes together in a pull request17:58
n4nd0and in any case, how do you manage to continue with the work if the compilation still fails?17:59
ckwidmerwell18:09
ckwidmerdisable cplex18:09
ckwidmer;)18:09
puffin444Hey is it better to submit a pull request and ask about small changes to the submitted code, or would it be better to ask them beforehand?18:18
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has quit [Ping timeout: 245 seconds]18:19
-!- vikram360 [~vikram360@117.192.168.146] has quit [Ping timeout: 240 seconds]18:25
n4nd0puffin444: when they are small details I normally send it and just comment about them18:32
n4nd0but maybe that is not the right way :P18:32
puffin444okay18:32
puffin444I think it would help if people had the code to look at.18:33
puffin444Which would be available in a pull request.18:33
n4nd0sure18:34
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 244 seconds]18:38
CIA-18shogun: Soeren Sonnenburg master * r5477eb7 / (3 files in 2 dirs): Merge pull request #584 from cwidmer/master - http://git.io/oMXCUA18:45
CIA-18shogun: Christian Widmer master * r33a2582 / (3 files in 2 dirs): fixed a few problems with the new label architecture in CPLEX dependent classes - http://git.io/hHQc2Q18:45
@sonney2kpuffin444, I don't understand the question18:46
puffin444sonney2k, I have a chunk of code ready to be pulled into the main branch. I have some questions about some details, and I wonder whether I should ask them in the pull request or before submitting the request.18:47
@sonney2kpuffin444, as you wish18:48
@sonney2kdoesn't make a difference18:48
puffin444ok.18:48
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Quit: Page closed]19:16
shogun-buildbot_build #932 of r_static is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/932  blamelist: sonne@debian.org19:28
shogun-buildbot_build #843 of octave_static is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/843  blamelist: sonne@debian.org19:32
-!- gsomix [~gsomix@109.169.156.50] has joined #shogun19:33
gsomixhi all19:33
@sonney2khi gsomix19:33
gsomixphilosophy - passed19:33
gsomixtime to work19:33
-!- vikram360 [~vikram360@117.192.171.39] has joined #shogun19:35
-!- vikram360 [~vikram360@117.192.171.39] has quit [Ping timeout: 244 seconds]19:46
shogun-buildbot_build #626 of octave_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/62619:47
-!- cronor [~cronor@fb.ml.tu-berlin.de] has quit [Ping timeout: 244 seconds]19:52
shogun-buildbot_build #613 of csharp_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/61319:55
shogun-buildbot_build #601 of lua_modular is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/lua_modular/builds/601  blamelist: cwidmer@tuebingen.mpg.de20:02
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun20:37
@sonney2kckwidmer, does it compile w/ cplex now?20:47
ckwidmersonney2k, there are still a few problems with CLabels, I asked n4nd0 to take care of them20:48
ckwidmersonney2k, have you recently disabled cplex?20:48
ckwidmerthere were quite a few problems with it20:48
ckwidmersince changes were made to labels20:49
@sonney2kckwidmer, disabled? no.20:52
@sonney2kno one has cplex so we don't know if it works20:52
ckwidmerok, so you don't have it either20:53
ckwidmerthat's where i was going20:53
ckwidmerI guess one of the students should have it enabled20:53
ckwidmerto catch all the regressions in cplex dependent classes20:53
ckwidmerit's apparently free for academic use20:54
ckwidmerI wasn't using it myself, but marius wanted to use LPBoost20:54
@sonney2kckwidmer, anyway we didn't change anything that would make it hard to fix20:55
@sonney2konly Labels got split up20:55
ckwidmertrue20:55
@sonney2kand functions from CMath moved to reasonable places20:55
ckwidmerbut there is nothing wrong with checking if it still compiles20:55
ckwidmerit's certainly not hard to fix20:55
ckwidmerbut was broken20:55
@sonney2kckwidmer, sure bot none of us has cplex so we cannot20:58
ckwidmerI understand, but it might be worthwhile to have one of the students install it (it's free)20:58
@sonney2kckwidmer, they are all busy...21:05
ckwidmerwell, I guess then cplex is broken21:06
@sonney2kckwidmer, I wouldn't even mind removing that crap21:06
ckwidmerme neighter21:06
ckwidmerneither21:06
ckwidmerI guess LPBoost could be solved using GLPK anyway21:08
ckwidmerright?21:08
@sonney2kjust slow :D21:08
ckwidmerheh, yeah maybe21:08
@sonney2kLPBoost == LPM == SVM with L1 norm regularizer21:08
@sonney2kckwidmer, well it uses hotstarts...21:09
ckwidmerah21:09
ckwidmerand GLPK doesn't support that?21:09
@sonney2kexaclty21:09
ckwidmerI mean, I don't really use any of that21:09
ckwidmerI know Marius needs LPBoost21:09
@sonney2kI think liblinear has that too21:09
ckwidmerabout the other Cplex dependent stuff, I don't care at all21:10
@sonney2knot so sure21:10
ckwidmerI just think that if you choose to keep it in there21:10
ckwidmerit should at least compile21:10
@sonney2kwell if someone complains we can fix the errors one by one21:10
@sonney2kotherwise there is nothing we can do21:10
ckwidmerhmmm21:10
ckwidmerdon't you have a build server?21:11
@sonney2kyes21:11
@sonney2kbut it will only ever run oss21:11
ckwidmerso, why not install cplex there21:11
ckwidmerbut it would help to catch your errors21:11
ckwidmerif no one else has it installed21:12
ckwidmerin the end, it's up to you21:12
ckwidmerI think it's bad to deliver software that is sure to break if a particular tool is installed21:12
ckwidmerno if cplex is not maintained, at least disable it by default21:13
@sonney2kwell then fix it if you care so much :)21:14
@sonney2kit is just a matter of resources...21:14
ckwidmerI don't really care that much :P21:14
ckwidmeryeah, ok21:14
ckwidmerthen it should be disabled21:14
ckwidmeror checked automatically21:14
ckwidmeron a different note21:14
ckwidmeris there a sparse matrix class in shogun?21:14
@sonney2kyes21:15
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has quit [Ping timeout: 245 seconds]21:23
-!- gsomix [~gsomix@109.169.156.50] has quit [Ping timeout: 245 seconds]22:36
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun22:44
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has left #shogun []22:49
-!- ckwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun22:54
-!- blackburn [5bde8018@gateway/web/freenode/ip.91.222.128.24] has joined #shogun23:08
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun23:16
n4nd0sonney2k: hi there23:20
-!- cronor [~cronor@e178178115.adsl.alicedsl.de] has joined #shogun23:34
--- Log closed Fri Jun 15 00:00:41 2012