Open in new window / Try shogun cloud
--- Log opened Wed Aug 22 00:00:05 2012
--- Day changed Wed Aug 22 2012
n4nd0(I don't know how to write most today...)00:00
blackburnit appear in risk00:00
n4nd0well ok ... it looks like we are not going to agree today00:00
n4nd0my guess is that you need to explicitily look for what is the value of that max_y00:01
blackburnrisk uses violating object00:01
n4nd0and not say that it is going to be given by argmax_y00:01
blackburndecision function does not00:01
blackburnin case of correct delta it should comply with max of the second term00:02
n4nd0I don't think so00:02
n4nd0I mean, I do not agree00:02
blackburnwhat can I say it works for HMM :)00:02
n4nd0a good model will give you for a training pair (xi,yi)00:03
blackburnwill commit things pretty soon00:03
n4nd0an argmax_y for xi that is close to yi00:03
n4nd0then the delta will be close to zero00:03
blackburnwhy we want to make delta zero00:03
blackburnif we maximize it?00:03
n4nd0blackburn: commit so I can take it a look00:03
blackburnwill do once I check examples00:04
yooorolf this yml format makes me crayz .. "The node does not represent a user object"00:04
blackburnI hope I will manage to wake up tomorrow :D00:05
blackburntoday I was late for an hour00:05
yoooyour commits today make me fell that one hour late didnt prevent you working on shogun during the day ^^00:07
blackburnI am trying to find a balance between not being fired and shoguning00:08
blackburnn4nd0: how many iterations mosek does on that hm svm example?00:09
n4nd0blackburn: I am not sure right now, I think between 6-10, let me check00:10
blackburnbmrm does 1400:10
blackburntakes 0.63s here00:10
n4nd0it takes quite longer here00:10
n4nd09 iterations00:12
CIA-52shogun: Sergey Lisitsyn master * r45840d7 / (12 files in 5 dirs): Refactored SO machine to not use its own features reference, fixed generic risk and bmrm result type -
n4nd026 seconds00:12
blackburnhere you go00:12
n4nd0wow that's a large lambda :)00:12
n4nd0how did you guess that value?00:12
CIA-52shogun: Sergey Lisitsyn master * rf253152 / (2 files in 2 dirs): Improved multiclass tree guided logistic regression -
blackburnno idea just entered big value00:13
blackburnit is not the best value btw00:14
blackburnobjective tends to be pretty big this way00:14
blackburnokay 5 millions is too big00:14
blackburn500 thousands tends to be overregularize too00:15
blackburnhmm 50000 works just like 500000:15
blackburn0 is da best00:15
blackburnwill not converge00:15
blackburnand furthermore will break you computer00:16
blackburnI changed one already00:16
blackburnn4nd0: with 0.01 accuracy is 0.998900:16
n4nd0blackburn: nice00:16
blackburnwith 1e-5 it is 0.999100:16
blackburncan it be 100%?00:17
blackburnlets try 1e-3700:17
blackburnoh no00:17
blackburn0.9998 and crazy numbers00:18
blackburnthat is primal objective LOL00:18
n4nd0that looks pretty weird00:18
n4nd0maybe my accuracy thing is doing something wrong00:18
blackburnthat is ok for 1e-37 lambda00:19
blackburnn4nd0: w contain e+41 elements this way00:20
blackburnn4nd0: okay let you compare results now :)00:20
n4nd0n4nd0: e+41 elements?00:20
n4nd0talking to myself again...00:20
blackburnn4nd0: not number, but order00:20
n4nd0does the objective in bmrm minimize too?00:21
blackburnn4nd0: do you understand what they sing about00:21
blackburnn4nd0: hmmmm really.. I don't know00:21
blackburnlooks like it maximiaze00:21
n4nd0blackburn: can you explain again why it is the argmax? :P00:23
n4nd0I just realized it is not feasible to check all the combinations y \in Y in HM-SVM00:23
n4nd0it just make sense to check all in the multiclass example00:23
blackburnn4nd0: why loss is argmax?00:24
blackburnbecause we find a most violating label for the vector00:24
n4nd0why max_y [ loss(yi,y) + < w, Psi(xi,y)-Psi(xi,yi) > ]00:24
n4nd0why the maximum is found for the argmax00:24
n4nd0why the maximum of the equation I have just written00:25
n4nd0is achieved with the argmax_y function00:25
n4nd0but the definition of the argmax is not to find the most violated label00:25
n4nd0it is to maximize <w, Psi(x,y)>00:25
blackburnmax_y [ loss(yi,y) + < w, Psi(xi,y)-Psi(xi,yi) > ] = max_y [ loss(yi,y) + < w, Psi(xi,y)> ] - <w, Psi(xi,yi) >00:27
n4nd0now why00:28
n4nd0argmax_y [ loss(yi,y) + < w, Psi(xi,y) > ] = argmax_y < w, Psi(xi,y) > ??00:29
blackburnyes it is00:29
blackburnfor proper loss it is for sure00:29
blackburnthey are both losses00:30
blackburnfirst is 'label' loss00:30
blackburnsecond is feature space 'loss'00:30
n4nd0that's not really a reason...00:30
blackburnlet me get a tea00:30
n4nd0we can talk about it tomorrow otherwise00:31
n4nd0maybe I am just stupid right now00:31
n4nd0and so stubborn as to see it00:31
n4nd0but this idea seems so simple00:31
n4nd0argmax_y should give a y close to yi for the input xi00:32
blackburnI am kind of practical guy so I prefer to check if it works first :D00:32
n4nd0and a y close to yi implies Delta(yi,y) close to zero00:32
n4nd0blackburn: I cannot compile here00:33
n4nd0line 3000:35
n4nd0that constructor is not in00:35
blackburnuh ok00:35
n4nd0did you forget to add any of the files or?00:36
blackburnI removed features - did you forgot? :)00:36
n4nd0yeah I know you did that :)00:36
n4nd0but since the build doesn't compile00:37
-!- naywhaya1e is now known as naywhayare00:37
blackburnI have to do a blind fix00:38
shogun-buildbot_build #431 of deb3 - modular_interfaces is complete: Failure [failed compile csharp_modular]  Build details are at  blamelist: Sergey Lisitsyn <>00:38
n4nd0blackburn: why a blind fix?00:39
blackburnhave no mercy^W mosek00:39
n4nd0aaah ok00:39
n4nd0true fact00:39
blackburnwhicho ne?00:39
n4nd0I can apply that if you want00:39
n4nd0it looks like that was the only one00:40
blackburnokay let you check it00:40
blackburnn4nd0: please try now00:41
CIA-52shogun: Sergey Lisitsyn master * rff37c90 / (5 files in 2 dirs): Updated PrimalMosekSOSVM -
blackburnn4nd0: inequality just after A.8 in teo00:44
blackburnthis one should make you believe00:45
blackburnI didn't though about that00:45
blackburn1 + 2 is majorizing thing for 100:45
blackburnso it is a major of risk00:46
n4nd0but I don't understand why the loss is being majorized too!00:46
n4nd0I am becoming crazy00:46
blackburnthere is a description out there00:46
blackburnfirst of all dot product is positive okay? ;)00:47
n4nd0ok :)00:47
blackburnsecond ineq comes from max00:47
blackburnare you convinced now?00:49
n4nd0I am reading the explanation around here00:49
n4nd0Note that (A.2) majorizes ?(y, y * ), where y * := argmaxy w, ?(x, y ) [Tsochantaridis00:49
n4nd0et al., 200500:49
n4nd0haha, magic copying from PDFs...00:50
blackburnI have no idea if it has something formal00:51
blackburnI have a  gut feeling about that :D00:51
blackburnI know that would not work in science haha00:51
CIA-52shogun: Sergey Lisitsyn master * r416d360 / (2 files in 2 dirs): Doc fixes -
n4nd0blackburn: why not?00:52
n4nd0do you mean you?00:52
blackburnwhat do you mean now?00:52
n4nd0< blackburn> I know that would not work in science haha00:52
blackburnI mean no one would believe me if I said I have a gut feeling :D00:53
blackburnabout quantum mechanics or economics or anything00:53
blackburnokay no moar mosek00:53
blackburnI'm satisfied now00:54
shogun-buildbot_build #432 of deb3 - modular_interfaces is complete: Failure [failed compile csharp_modular]  Build details are at  blamelist: Sergey Lisitsyn <>01:00
n4nd0blackburn: I think the same still!01:01
n4nd0I believe strongly that another argmax is required01:02
blackburnit is infeasible01:02
n4nd0one that includes the term Delta(yi,y)01:02
n4nd0it depends on the application01:02
n4nd0for the HM-SVM it is probably feasible01:02
blackburninfeasible for HM, right?01:03
blackburnokay, infeasible for CRF?01:03
n4nd0in the same way that argmax_y < w, Psi(x,y) >01:03
n4nd0blackburn: my CRF knowledge is a bit limited for this01:03
n4nd0but my guess is that it is feasible for CRF too01:03
n4nd0since they have been used in real applications01:04
blackburnyou should taklk to some SO expert01:04
n4nd0blackburn: do you give me some confidence at least? :D01:04
blackburnI am pretty sure it is correct now01:04
n4nd0but you are just basing it in the result of the accuracy01:05
n4nd0take into account that the dataset is not difficult01:05
n4nd0the underlying distribution is not really difficult01:05
blackburnno, I am basing it on my understanding of all these things01:05
n4nd0and we are providing lot of training examples01:05
n4nd0but look to equation A.7 in Teo01:06
n4nd0a hint could be that they do not use the same to denot01:06
n4nd0and y_*01:07
n4nd0y_* = argmax_y < w, phi(x,y) >01:07
n4nd0and y_bar is the thing of A.701:07
blackburnoh I have to make SO locked training work01:07
blackburnit is not supported now01:08
n4nd0time to sleep now01:09
blackburngood idea01:10
n4nd0good night & good job :)01:10
blackburngood night01:10
-!- n4nd0 [] has quit [Quit: leaving]01:10
-!- zxtx [] has quit [Ping timeout: 252 seconds]01:16
-!- blackburn1 [~blackburn@] has joined #shogun01:20
-!- blackburn [~blackburn@] has quit [Read error: Operation timed out]01:21
-!- zxtx [] has joined #shogun01:27
CIA-52shogun: Sergey Lisitsyn master * rd6b3a7d / (9 files in 3 dirs): Refactored apply of latent machines -
-!- blackburn1 [~blackburn@] has quit [Quit: Leaving.]01:38
shogun-buildbot_build #433 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at
-!- yooo [575b08cb@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]02:41
shogun-buildbot_build #62 of nightly_none is complete: Failure [failed compile]  Build details are at
-!- av3ngr [av3ngr@nat/redhat/x-swaimdyxogxxwdlk] has joined #shogun05:57
CIA-52shogun: Sergey Lisitsyn master * rcf43532 / (2 files): Fixed compilation of data generator in case of no lapack available -
CIA-52shogun: Sergey Lisitsyn master * r40564ac / src/shogun/latent/LatentSOSVM.cpp : Added missed lambda usage in LatentSOSVM -
CIA-52shogun: Sergey Lisitsyn master * r157fba3 / src/interfaces/modular/modshogun_ignores.i : An attempt to ignore display methods of SG datatypes -
-!- uricamic [~uricamic@2001:718:2:1634:155e:7544:fba4:9878] has joined #shogun09:03
-!- n4nd0 [] has joined #shogun09:32
n4nd0hey uricamic09:44
n4nd0how are you doing?09:44
_____________uricamic: hey - you should convince n4nd0 today :D09:49
n4nd0_____________: uricamic and I will convince you :D09:50
n4nd0_____________: let me ask you09:53
n4nd0why do you do CDotFeatures* in 47 of PrimalMosekSOSVM.cpp09:53
_____________n4nd0: why not CFeatures?09:54
n4nd0_____________: yeah09:54
_____________because I didn't know what is restriction09:54
_____________please change if it is not correct09:54
n4nd0just wanted to ensure that there was no special reason09:54
n4nd0but remember this restriction was the one that started all the refactoring09:55
n4nd0because in BMRM it was CDotFeatures and we needed CFeatures there09:55
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has joined #shogun10:13
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has quit [Client Quit]10:13
CIA-52shogun: iglesias master * rb985ee1 / src/shogun/structure/PrimalMosekSOSVM.cpp : * fix compilation error with MOSEK support -
CIA-52shogun: Sergey Lisitsyn master * r8122bd8 / src/shogun/structure/PrimalMosekSOSVM.cpp : Merge pull request #751 from iglesias/master -
uricamichi, sorry, I have been away10:28
_____________uricamic: we were arguing HARD yesterday10:28
_____________uricamic: help us to resolve that :)10:28
_____________uricamic: okay - the problem is10:29
uricamicok :)10:29
_____________is argmax of <w, Phi(x_i,y)>10:29
_____________equal to argmax of10:29
_____________\Delta(y_i, y) + <w, Phi(x_i,y) - Phi(x_i,y_i)>10:29
_____________see what I mean?10:29
n4nd0uricamic: put some light into here please :D10:30
uricamicI am not sure what u mean, but wait a sec, I will figure it out :)10:30
_____________uricamic: I've added generic risk function10:30
_____________to structured model10:30
n4nd0uricamic: when one needs to compute the risk function10:30
_____________have you seen?10:30
_____________it works like10:30
uricamicnot yet, but I will check it now10:30
_____________for each feature vector: find argmax, update risk and update subgrad10:31
_____________but n4nd0 thinks it is wrong to find argmax without loss10:31
uricamicyep, it is wrong10:31
uricamicu need to add loss there, because u are trying to find the most violated constraint there10:32
uricamicand adapt to it10:32
uricamicit is similar to perceptron algorithm10:32
n4nd0and the argmax doesn't give you the most violated10:32
uricamicit can if u add loss term there10:33
uricamicto all examples10:33
_____________okay it is a luck it works now then :)10:33
uricamice.g. I have been using it in one of my applications of so-svm, I have designed argmax function in matlab with the possibility to pass also the losses10:33
uricamic_____________: and on which instance u have checked that?10:34
_____________HM-SVM and multiclass10:34
_____________n4nd0: is it feasible to implement argmax with loss?10:34
_____________for HM-SVM?10:34
uricamicit should be possible10:35
_____________n4nd0: can you fix my code then?10:36
_____________what is needed is10:36
n4nd0_____________: it should be done modifying the Viterbi I implemented there10:36
_____________to add argmax_loss or so10:36
n4nd0uricamic: I am trying out the DualLibQPBMSOSVM and it never ends :(10:36
n4nd0I am doing something wrong for sure10:37
_____________change risk to call that, remove loss addition in risk10:37
uricamicn4nd0: I haven't checked your code for HM-SVM yet, but wouldn't it be possible to just ass loss to the features in argmax?10:37
_____________and then implement argmax_loss for multiclass and HM10:37
uricamicn4nd0: which algorithm and what lambda have u tryied?10:37
n4nd0uricamic: I have not set any algorithm explicitely to tell the truth10:39
uricamic_____________: btw: I wanted to ask about the changes in my code for modular interfaces, because of java_modular failure10:39
uricamicso it should use BMRM10:39
n4nd0uricamic: I thought that would use BMRM by default10:39
_____________I fixed that10:39
uricamicthat one has proven convergency10:39
n4nd0uricamic: lamda equal to 0.0110:39
n4nd0it doesn't finish though10:39
uricamicI see, that could be the problem10:39
_____________uricamic: do I understand right C is 1/lambda?10:40
uricamicLike it goes through 1000 iterations and did not stop?10:40
_____________uricamic: ah one ask for your code too10:40
_____________could you please add !CSignal::cancel_computations() to while condition?10:40
n4nd0uricamic: 294 iterations and it just froze10:40
uricamicI guess not just 1/lambda, it depends on the formulation of the task with C, do you have it somewhere? I guess it could be n/lambda10:41
_____________to make it possible to Ctrl+C that10:41
uricamichmm, that is strange10:41
uricamicit might mean that the qp task is really hard to solve10:41
uricamicand it could take really a long wjile10:41
n4nd0I am using the so_multiclass example, in libshogun10:42
uricamicn4nd0: it does not converge for so_multiclass ?10:44
_____________it was converging on my machine ehm10:44
_____________very fast10:44
n4nd0I am doing something wrong for sure then10:45
n4nd0but what?10:45
n4nd0just find bundle to look for the parts that use bmrm10:46
uricamictry to increase lambda10:47
uricamica lot10:47
uricamicbecause, I probably haven't mentioned it anywhere yet10:47
uricamicbut because of P3BMRM10:47
uricamicwe are using unnormalized risk10:47
uricamici.e. without division by number of examples10:47
uricamicwhich changes the lambda a lot10:48
n4nd0to what number?10:48
uricamictry something really big for start and then u can try lower10:48
n4nd0trying with 100010:48
uricamiclike 1e410:48
uricamicyep, or that one10:48
uricamicthe problem is that without this unnormalized risk, P3BMRM would not be identical to BMRM10:49
n4nd0ok it finished now, thank you :)10:49
uricamicit would require division of number of cp_models used depending also on the number of examples in the cp_model10:49
uricamicn4nd0: you are welcome10:49
n4nd0I get this accuracy10:50
n4nd0SO-SVM: 82.80%10:50
n4nd0BMRM:   69.90%10:50
n4nd0is that normal?10:50
uricamicbut when using unnormalized risk, they are equal10:50
uricamiccould be, try lower lambda now :)10:50
n4nd0model selection for this would be nice :D10:51
uricamicI have finished benchamrk for MNIST data yesterday, but I will probably have to change my code a bit to be able to send it in PR10:52
uricamicI mean  I need to return history of Fp, Fd and wdist from the solver, but in current form it crashes java_modular10:53
uricamicso I guess, it will be sufficient to add getters for these to DualLibQPBMSOSVM class and everything should be fine then10:54
_____________uricamic: it works now10:56
_____________I believe it does10:57
uricamic_____________: how come?  I guess it is impossible :)10:57
_____________uricamic: you didn't declare constructor and it was not inherited from sgobject10:57
uricamicbecause now in modular there is no support to return bmrm_return_T10:57
_____________so I had to declare ctor/dtor and add dummy save/load_serializable10:57
_____________I restored it10:57
uricamicoh, I see10:58
_____________just check latest10:58
uricamicI haven't noticed it10:58
uricamicok, thanks, I will check it10:58
_____________it was crashing python graphical example of multiclass10:58
_____________but now it works without any glitches10:58
_____________uricamic: could you implement a multiclass agrmax with loss?10:59
uricamic_____________: I will look on it10:59
uricamicso the design will be to have overloaded argmax one without loss and the other with it?11:00
n4nd0I think that is the idea11:02
n4nd0I'd like to skype with Nico first to ensure it11:02
_____________I was pretty sure it would work without loss11:03
_____________but okay if it is not I don't mind11:03
uricamicwell, I am still wondering that it was working, but it could happen :)11:04
uricamicbut I am sure, that the loss has to be there before argmax11:05
_____________n4nd0 thought it is only because of simple cases11:05
_____________I could believe so11:06
uricamicyep, that is possible11:06
n4nd0I am preparing this case for multiclass classification to compare11:06
_____________ hahah11:07
wikingblackburn why u no break somebody else's code? :D11:07
_____________man it is easy to apply that patch11:08
n4nd0the lol was for the picture :D11:08
wikingwtf man11:08
wikingyou broke the example :)11:08
wikingi'll paste bt of gdb11:08
wiking[DEBUG] entering LatentSVM::apply(LatentFeatures at 0x10260bd90)11:09
wikinglibc++abi.dylib: terminate called throwing an exception11:09
wiking[ERROR] Unknown problem type11:09
wikingProgram received signal SIGABRT, Aborted.11:09
wiking0x00007fff8caff212 in __pthread_kill ()11:09
_____________oh okay11:09
_____________that can be let11:09
_____________me fix it11:09
_____________wiking: well you are not very active with that code so I didn't think I could make it hard to rebase :)11:10
_____________all changes in your code are just11:11
_____________apply -> apply_latent11:11
wikingit doesn't mean that i'm not preparing a patch11:11
wikingyeah i know11:11
n4nd0uricamic: I don't manage to make so_multiclass_BMRM work properly11:11
_____________that is the right way to do that so just rename it and you wouldn't really need to rebase11:11
wikingand as well the fuckup of yesterday's primalmosek :)11:11
n4nd0uricamic: I execute on command line ./so_multiclass_BMRM data.out 2 1000 500 1 0.01 BMRM11:11
uricamicn4nd0: what do u mean properly?11:11
n4nd0and it stays in iteration 0 forever11:11
wikingi mean the only thing is that please commit if it compiles11:12
wikingand not when it doesn't11:12
wikingand not having mosek is not an excuse when u edit that code ;)11:12
* wiking feeling bossy today :)11:12
CIA-52shogun: Sergey Lisitsyn master * r5347c30 / src/shogun/machine/Machine.cpp : Added missed latent problem handling in apply of machine -
n4nd0_____________: I have to support wiking there, I am sorry :S11:12
uricamicn4nd0: hmm,  I think I have changed that example to expect data in svmllight format11:12
uricamicbut let me check that11:12
n4nd0uricamic: aaahm I am just using the data from the example I run11:13
uricamicI am just compiling the latest code11:13
n4nd0uricamic: ok, let me know someting11:13
_____________well guys I have nothing to say if you are so worried with breaking your code :)11:13
n4nd0I guess that we all want to rush a bit the things in order to be ready for release11:14
wikingbtw what has happened with fixing up the static analyzer problems?11:14
_____________I have been waiting for generic risk function for a while and nobody did it11:15
uricamichmm, it seems that I have to modified that example, cos' it refused to compile now :D11:15
_____________furthermore we had wrong interface with redundant features11:15
wiking_____________: a PR + wait for comment would be nice... it really worked out for me till now quite well11:15
_____________so having no mosek is excuse, it is not project of mine but I did that11:15
wiking_____________: but you've broken the code, and you couldn't see the problem as that code never compiled on your machine11:16
n4nd0uricamic: oh yes, you have to change the constructor11:16
n4nd0uricamic: just leave out features11:16
n4nd0uricamic: and thank _____________  :P11:16
uricamicn4nd0: I see :D11:16
wikinguricamic: :DDD11:16
_____________wiking: I know what to do better11:17
wiking_____________: u mean u know better? :)11:17
_____________breaking mosek is not the issue when we are in rush to release sooner11:17
wikingbut then we have a broken release :)11:17
wikingespecially that mosek is not tested on any of the bots11:18
uricamicn4nd0:  ok it works for me now11:18
n4nd0_____________: we have to face that it will probably be difficult to have everything working with bundle methods for release11:18
wikingi.e. nobody else will pick up this error just who really wants it to work :)11:18
_____________n4nd0 will always test it because he works with it11:18
_____________n4nd0: I can't see any difficult there11:18
_____________you said HM argmax with loss is easy?11:18
wikingok i was mostly joking till now11:18
n4nd0I didn't say it was easy11:19
n4nd0I am just guessing11:19
_____________so you can't do that?11:19
wikingbut seriously is it really hard to wait 6 hours for comments on a PR _____________ ?11:19
wikingif yeas11:19
wikingthen mea culpa11:19
uricamicn4nd0: so the input data are expected in svmlight format11:19
n4nd0_____________: I didn't say I can't11:19
_____________I have no fucking idea when will you appear wiking11:19
wiking_____________: i dont have to appear to comment on a PR on githug11:20
_____________what to wait for?11:20
n4nd0_____________: I just say that maybe it cannot be done for today11:20
_____________I will do that then11:20
uricamici.e. on each line first number is the label and then non-zero elements of feature vectors in format idx:value11:20
wikingespecially if you add a @<github username> into the comment11:20
wikingas that person will get an email most probably11:20
wikingas that's the default setting of github...11:20
_____________wiking: what should I wait for?11:21
wiking_____________: comment on your intended PR11:21
wikingfrom the author of that actual code11:21
_____________which PR?11:21
wiking_____________: i'm talking about future11:21
wikingthat if u change code11:21
wikingwhich u really dont feel sure about11:21
wikingthen maybe go with a PR and wait 3-6 hours11:21
_____________in future I would never get you back11:21
_____________what is unsure in that PR?11:22
wikingand if there's no comment11:22
_____________it should be apply_latent but not apply11:22
_____________did you ever check how machine works?11:22
wikingit's not about that given commit11:22
wikingsame goes with primal mosek per se11:22
wikingand this is not a personal vendetta11:22
_____________did you check logs where I asked n4nd0 to check if it compiles?11:22
wikingso please do not start to patronize me with "did you ever check how machine works?"11:23
wikingi'm just suggesting11:23
wikingthat maybe it would be better...11:23
wikingbut if you know better11:23
wikingthen it's ok11:23
_____________should I spend time on installing mosek just to check if it compiles when I could ask n4nd0 and he don't mind to do that?11:23
wikingsorry mate to bringing up this idea11:23
_____________I do not understand what you do blame me about11:23
wikingi dont blame u11:24
wikingi'm just saying that it would be great that if you start changing code that you haven't authored and it's not really an obvious assertation11:24
wikingthen maybe it would be great that you somehow assure that it's ok11:24
_____________what is not obvious? removing features or renaming apply to apply_latent?11:25
wiking_____________: i've given there the example: not obvious = not an assertation11:25
_____________you claim I do that - where did I something that needs an author comment?11:25
wikingall these commits11:26
wikingthat actually broken something11:26
_____________which these?11:26
wikingsee the mentioned ones above11:26
wikingor now you really want to have sha-1 lines?11:26
wikingi mean seriously you are now going to take this as a personal matter11:26
_____________your example should work now it is just a missed line11:26
wikingbecause sorry mate i really like you so i rather just shut up11:27
_____________I do not understand why making it correct is so much an issue for you11:27
_____________yes because you are blaming me in things I didn't do11:27
_____________I am spending time on things I am unsure you will spend on11:28
_____________lately I have been fixing doc - wiking your code too - I don't know whether you would do that11:28
n4nd0this conversation makes me sad :(11:28
_____________as for your example I planned to run it later anywhay11:29
_____________I just wanted to make your projects more suitable before release that is coming already11:30
wikingok 2 more questions: a) have you commited code that broke examples and compilation of primalmosek? (and you have your answer on your line of 'blaming you') b) why do you feel attacked here, i think this really a work matter and nothing personal? but anyhow if you feel in any ways offended, then i'm honestly sorry and let's just forget this... it's really not worth it...11:30
_____________you are not in active development (at least it looks like that for me)11:30
_____________a) yes, and fixed that after n4nd0's report11:31
_____________b) because you are attacking me - pretty obvious11:31
_____________n4nd0: can you finally assure wiking I didn't break your code without your participation at all?11:32
wikingas said in b) and earlier it wasn't meant to be attacking at all just a suggestion what would be great...11:32
_____________or you think too I am the breaker that makes your projects worse?11:32
wikingyou completely take this the wrong way11:33
wikingnobody ever told that "I am the breaker"11:33
wikingand that you are making the project worse11:33
wikingyou are taking this out of proportion11:33
wikingi haven't told any of this11:33
n4nd0uricamic, _____________ : nice! I finally finished the test11:34
uricamicn4nd0: great :)11:34
wikingand wasn't even nearly implying it...11:34
n4nd0uricamic, _____________ : so I have compared the error when using the CMulticlassModel::risk and the error when using CStructuredModel::risk11:34
n4nd0the same data, same epsilon, same lambda, everything the same except from that function11:35
n4nd0when using CMulticlassModel::risk we get an error equal to 21.4%11:35
n4nd0and when using the other around 40%11:35
n4nd0_____________: so I finally conclude that bot argmax are not the same :)11:36
_____________n4nd0: I don't mind that but just do that correct11:36
uricamicn4nd0: yep, of course they can't be :)11:36
uricamicI am going for lunch now, will be back hopefully in less than 1 hour11:36
n4nd0uricamic: bye11:37
n4nd0_____________: c'mon man, don't say it so :(11:37
_____________how so?11:37
n4nd0never mind11:37
_____________I just mean if you have no time for that11:38
_____________just tell me11:38
_____________I will try to do by myself11:38
n4nd0look _____________, I have time for that11:38
n4nd0but time in my time frame11:38
n4nd0I may have done it by the end of this week, tested and everything11:38
n4nd0because first I want to consult with Nico11:39
n4nd0if you do not like that ..... just do it yourself11:39
_____________I want that to be in release I am sorry I am pushing you11:39
n4nd0_____________: I will do my best then11:39
_____________it seems to be the best new feature of shogun11:40
n4nd0_____________: then I will get more motivation :)11:40
_____________so I don't want to fuck up there11:40
n4nd0all right11:40
_____________so argmax is just viterbi in hm?11:41
_____________can't you just put losses in that loss matrix?11:41
_____________I am newbie in HM still but just ask11:41
n4nd0let me check one thing in the hmsvm toolbox11:42
_____________it is easy to implement for multiclass11:43
CIA-52shogun: Soeren Sonnenburg master * r6d43e98 / src/shogun/features/DataGenerator.cpp : include lib/config.h when checking for a DEFINE_TO_BE_SET -
_____________really easy11:43
n4nd0take a look to that code11:44
_____________sonney2k: it is already included in .h - why?11:44
n4nd0_____________: you can see there are two calls to best_path11:45
_____________hah so it uses shogun's viterbi?11:45
n4nd0in the past yes, not in that version11:45
_____________because our HMM is broken11:45
_____________and not because of me wiking11:45
n4nd0I think it was just to make it standalone...11:46
n4nd0at least they say so in the doc11:46
_____________n4nd0: 59-64 lines are the thing, right?11:47
n4nd0and that is actually what I do in my code11:47
n4nd0I think that my argmax is looking for the MMV11:47
_____________what is MMV?11:48
n4nd0it is always doing the pred_path_mmv and not the pred_path11:48
n4nd0maximal margin violater11:48
n4nd0it appears in that snippet, in a comment11:48
wiking_____________ as i said earlier i'm sorry that i've brought up this whole topic; it was completely unrespectful from me to say anything like that....11:48
_____________ah right sure11:48
_____________wiking: you are right in what you are saying11:49
_____________I didn't break someone code if it is not obvious fix11:49
_____________or doc fix11:49
_____________or I didn't talk to author11:49
_____________obvious fix is your case11:49
_____________in case of mosek svm I was in touch with n4nd0 - no idea why he refuses to say that :D11:50
n4nd0because I do not agree in the way you are handling this11:50
n4nd0but anyway ... let's move on, this is not so important11:51
_____________okay one more guy against me11:51
wikingas i said it was my fault so i hope we can get over this11:51
n4nd0_____________: you see, it is not against you or with you man11:51
_____________why didn't you say I should install mosek and do not touch it yesterday?11:51
n4nd0_____________: I don't know how you say that if I am not even taking part ....11:51
_____________you said okay I will compile'11:51
_____________didn't you say that?11:51
-!- emrecelikten [] has joined #shogun11:52
n4nd0all right guys11:52
wikinglet's just get over this and i'm sorry for bringing this up11:52
wikingit's not worth any of this11:52
n4nd0sonney2k: captain, bring some peace please :)11:53
_____________I give up guys11:53
wiking_____________ there's nothing to give up, keep on as is and it's all good11:53
_____________let you work on that separetely then if you are so hurted11:53
-!- emrecelikten [] has quit [Read error: Connection reset by peer]11:53
wiking_____________ nobody is hurt here....11:54
wikingexcept maybe you11:54
_____________with commits I have to do without much checks11:54
wikingbut that wasn't any of my intention11:54
wikingand since it escalated into this that you think i'm saying that you are makeing bad for the project11:54
wikingwhich is OBVIOUSLY not the case... it's the opposite11:55
wikingyou are pushing this project like nobody else here11:55
wikingwhich is awesome11:55
wikingso please just try to forget my comments and
_____________I partly do understand your unsatisfaction11:57
_____________but I do not understand what made you so worried about that11:57
wikingneither of that has happened in me :)11:57
wikingneither unsatisfaction nor worried11:57
wikingit was really just a comment11:57
wikingno more no less11:58
wikingand u do whatever you want with it11:58
wikingjust please dont take it as a personal vendetta against you and your personality11:58
_____________n4nd0: why didn't you say it is bad to do that blind fix?11:59
_____________and why do you feel it so bad if you could test that - as I asked you12:00
_____________I do not understand that12:00
n4nd0_____________: I think I told you I could fix the MOSEK thing to ensure that it will compile, but you insisted12:01
wiking*booom* and the last 20 minutes disappeared :DDD12:01
n4nd0I don't think it is bad to the blind fix either12:01
_____________insisted what?12:01
n4nd0bad sounds quite bad12:01
_____________I wanted you to test if it compiles after removing features12:01
wiking(do not let you speeak :DDD )12:02
_____________is that wrong for you?12:02
n4nd0of course not12:02
_____________what is?12:02
_____________I want to understand what is you feel wrong in handling12:03
_____________you said you disagree12:03
_____________with what?12:03
n4nd0how you have reacted12:04
n4nd0but it was nothing bad12:04
wikinghave i told that ever since a chick moved into my office my office became a geek magnet and all the guys are coming into my office chatting... :))12:05
wikingi'll be a true cockblocker :)12:05
wikingi still need to work on some lines :)12:06
n4nd0_____________: seriously, I don't think you did wrong doing any fix there12:06
_____________then I am completely lost what I am wrong with12:06
wiking_____________: nothing!12:07
wikingthat's the point :)12:07
wikingand i was just practicing my trolling skills :D12:07
_____________I understand you are comfortable with your own code and it is rather you should do changes12:07
_____________but it would take a while and I want to push latest must have features before release12:08
_____________and make it correct12:08
n4nd0_____________: agree12:08
wikingso is it over yet?12:08
n4nd0let's work together and have as many stable things as possible before release12:08
_____________I'll let you polish your code by yourselves then12:09
_____________wiking: so is your example worknig now?12:12
wiking_____________ yeah awesom-ooo-ly thnx!12:12
_____________wiking: why latentsosvm has training but has no apply?12:13
wiking_____________ because it's WIP12:13
wikingit'll be no WIP by this weekend12:13
_____________I see12:14
wikingwe have this 'little' ongoing problem with structmodel and latentmodel and merging them12:14
wikingas currently creating a working latentSOSVM12:14
wikingtakes too much user defined stuff12:14
wikingyou have to define at least 4 different classes12:14
wikingand i think we shoudl take it down to 312:15
_____________I am sorry I am a bit offensive but I still don't really get why did you both blame me12:15
wikingu r the MISUNDERSTANDER!12:16
wikingdo not be that one :>12:16
wikingcomeon man seriously12:16
wikingcan we get over it?12:16
_____________yes but I want you to understand I am not a psycho12:17
wikingheheh i know you are not12:18
wikingnever was the case12:18
wikingyou are just passionate about shogun12:18
wikingthat would make you a geek but i would really like to think that you are not that type of a geek who would come into my office to try to chat with this chick :)))12:19
_____________it is hard to talk about structured output machine with chick12:20
wikingi hope they won't find these chat logs12:20
wikingor maybe i hope they'll find it :DD12:20
_____________I am sure no chick would ever try to read IRC LOG of MACHINE LEARNING TOOLBOX, I am pretty sure12:21
* wiking works at a very unpleasant place so that's why his implied complaning :)12:21
_____________which office you are talking about? your university?12:21
wikingheheheh she is doing phd in maths...12:21
wikingi mean it's not soooo unprobable12:21
wikingyeah my uni :)12:22
wikingah yeah cool12:22
wikingi know how i'll cockblock12:22
wikingi'll use banners12:22
wikingi'll print some banners and just raise them sometimes when i get annoyed with the jibberish :)12:22
n4nd0_____________: I think that the argmax I implemented does: argmax_y [ Delta(yi,y) + <w, Psi(xi,y)> ] in training12:23
n4nd0_____________: and argmax_y [ <w, Psi(xi,y)> ] during prediction12:23
wikingn4nd0: it should do that yes12:23
wikingso if you have some good ideas for the banners let me know :)12:23
n4nd0wiking: but the definition of argmax is: y* = argmax_y <w, psi(xi,y)> or?12:24
wikingn4nd0: well that's the thing12:24
_____________n4nd0: can't we assume training is for +LOSS and not-training is w/o loss?12:24
n4nd0_____________: do you mean for every model, or just for HM-SVM?12:24
wikingn4nd0: it should be that in your case but i think you are trying to find the most violating ones12:24
_____________for all models12:24
n4nd0_____________: I am not sure about all models12:24
n4nd0_____________: let me check what I did in the multiclass one12:25
wikingnot-training w/o loss?12:25
wikingno comply in me please explain12:25
_____________wiking: there are two modes in argmax12:25
_____________first is training one12:25
n4nd0_____________: in the multiclass the loss is not taking into account12:25
_____________n4nd0: we should change that then12:25
wikingbut classification (not-training?) is just w*psi(x,y)12:26
_____________n4nd0: okay wait - so HM is correct because it was correct alreadY?12:26
_____________wiking: yes12:26
n4nd0_____________: why? I don't know which one is the correct :D12:26
n4nd0wiking: yes12:26
_____________n4nd0: you said it does loss+w max12:26
wikingso what else is not-training? :)))12:26
wikingwell then u don't need that12:26
n4nd0when you don't have the true label, you cannot get the loss12:26
wikingit's just w*psi()12:26
_____________no, it is argmax of w*psi12:27
wikingah ok12:27
wikingi get it12:27
_____________most probable label12:27
wikingor i would get it in latent case12:27
wikingbut in simple struct as well?12:27
n4nd0look guys at algorithm 112:27
n4nd0oh shit I am getting crazy confused12:28
_____________what to do with it?12:28
_____________n4nd0: if you are optimizing with loss term already12:28
_____________lets just remove it from generic risk12:28
_____________and put loss to multiclass argmax on training=true12:29
n4nd0I don't understand, sorry12:29
_____________n4nd0: you said you compute argmax with loss already, right?12:29
n4nd0in the HM-SVM12:30
n4nd0not in the multiclass12:30
_____________we can fix multiclass - it is not the issue12:30
_____________I mean currently it is being added to risk12:30
_____________in generic risk12:30
_____________ line 17012:31
_____________n4nd0: if we assume argmax includes loss on training we remove line 17012:31
_____________and add loss to multiclass12:31
_____________then everything becomes correct and we are all happy12:31
n4nd0the thing is that I am not sure whether the argmax should include the loss12:32
n4nd0even if we are in training12:32
_____________you was sure yesterday12:32
_____________and uricamic said it should12:32
_____________I was unsure it should12:32
wikingwell risk is being used in bmrm code12:33
_____________okay I am off for food12:33
wikingand there the primal objective in some way will include using argmax12:33
wikingalas the risk12:33
n4nd0_____________: yesterday I was sure that the argmax doesn't include the loss12:33
n4nd0_____________: that's why I said that it is not the same the argmax that was already implemented in the model than the one we need for risk12:34
wikingn4nd0 what i dont get is that how u want to have a generic risk, where what the bmrm solves is actually labmda*||w|| + RISK12:34
wikingso it pretty much depends on you what RISK -> p.o. is12:34
n4nd0but the risk can be written in terms of an argmax and the psi function12:35
wikingwell it depends what's RISK's form12:35
n4nd0so the generic risk calls these model dependent functions12:35
wikingyeah i get it12:35
wikingbut the risk can be pretty much anything12:35
n4nd0R({\bf w}) = \sum_{i=1}^{m} \max_{y \in \mathcal{Y}} \left[ \ell(y_i, y) + \langle {\bf w}, \Psi(x_i, y) - \Psi(x_i, y_i)  \rangle  \right]12:36
wikinguntil it's convex... ;)12:36
n4nd0I thought the risk is that, and only that12:36
wikingthis is from multiclass right/12:36
n4nd0(paste here for better reading)12:36
wikingi know this one12:37
n4nd0wiking: why just the multiclass one?12:37
wikingbut from where did you get this comment?12:37
wikingafaik it's a good risk function for multiclass12:37
n4nd0it is the same that appears in Teo12:38
wikingyeah i know :)12:38
n4nd0page 103, equation A.212:38
wikingand a lot of other places with SO12:38
n4nd0they don't talk there just about multiclass, right?12:38
wikingwait a sec12:38
wikingi'll just dig something out12:39
wikingok so here's another thing12:39
wikinghow fucking lame is this :)))12:39
wikingthey all want the pussy :)))12:39
wikingso yeah n4nd0 look12:40
wikingpage 3 eq(7)12:40
wikingfor me that is the RISK12:40
wikingwith the -C\sum_....12:40
wikingthe whole thing after 1/2||w||12:41
n4nd0yeah, that is definitely different12:41
n4nd0that term is new12:41
wikingand the Dual solver12:41
wikingbasically is intented to solve12:41
wikinganything that is convex and in labmda*||w||^@ + RISK12:42
wikingw/o constraints12:42
wikinguntil risk is convex12:42
wikingyou can have anything12:42
wikingas risk12:42
n4nd0I see12:42
wikingthis is my understanding and that's why i thought that StructuredModel::risk should not be implemented12:42
wikingn4nd0: but in multiclass case12:42
wikingyou are totally right though12:42
wikingyou can actually define the risk function with the help of psi and argmax12:43
wikingbut what if is your risk is something completely different but still convex12:43
wikingsee for example my case12:44
n4nd0but maybe this risk function we use in multiclass, and probably in HM-SVM too, may work as well12:44
wikingand yes in my case \hat{y} and \hat{h} should be defined/given by argmax12:44
wikingyeah it will work12:44
wikinguntil you want that risk function12:44
n4nd0then we just make it virtual12:45
wikingbut i think you should note it somewhere that this is just an example ;)12:45
n4nd0giving the possibility to be overridden12:45
wikingyeah as until now12:45
n4nd0wiking: sure12:45
n4nd0in any case12:45
n4nd0probably it is not that great to have this generic risk for efficiency reasons12:46
wikingbut i thought that it's a bit 'narrowing down' when you already implement a risk function in an abstract class whilst your risk function could be anything12:46
wikinganything = the PO is convex...12:46
n4nd0yes, I see your point and I agree with you12:46
n4nd0it is not conceptually right I think12:47
wikingthere could be a protected default risk function ?:)12:47
wikingin structuredmodel12:47
wikingand then anybody who inherits from this class12:47
wikingwill have already a function at hand12:47
wikingand he can just chuck that in into his risk function12:48
wiking(call it from it) if he really doesn't know what is he doing :)))12:48
n4nd0sounds like a good idea12:48
wikingor just wants to compare12:48
wikingbmrm vs mosek12:48
wikingbtw have any of you run that ?12:48
wikingwhat's the accuracy difference between mosek and bmrm12:49
wikingon one given data set12:49
n4nd0not using an important data set12:49
wikinghehe yeah12:49
n4nd0just randomly generated Gaussian12:49
wikingbut any dataset12:49
n4nd0for the multiclass example12:49
wikingand what are the numbers?12:49
n4nd0yeah, I have prepared that before12:49
wikinghow far r they from each other ?12:49
n4nd0running again right now ..12:50
n4nd0but I don't think these results are very trustworthy12:50
n4nd0I just plugged a lambda at random12:50
n4nd0SO-SVM: 92.60%12:50
n4nd0BMRM:   88.60%12:50
n4nd0MC:     40.00%12:50
wikingso bmrm is not so far at all12:51
wikingbtw: labmda = 1/c12:51
wikingwhat's your c in primal?12:51
wikingjust use lambda = 1/c12:51
n4nd0wiking: michal said something about that BUT also that he was not regularizing with the #examples12:51
n4nd0C = 1 in multiclass classification her12:52
wikingbut it's pretty cool though12:52
n4nd0for the PrimalMosekSOSVM12:52
wikingand for mosek?12:52
wikingah ok12:52
wikingthen use lambda =112:52
n4nd0I am interested in knowing how faster it is12:52
wikingn4nd0: CMath::time12:52
wikingbut still it's awesome that there's bmrm that can actually give similar values as mosek but it's fully opensource12:53
n4nd0it is pretty slow with lambda = 112:53
wikingok i'm off to pick up something to eat12:54
wikingbrb in 1012:55
n4nd0for a new run with other data and lambda = 1000 (#examples)12:55
n4nd0SO-SVM: 85.30%12:55
n4nd0BMRM:   79.20%12:55
n4nd0MC:     54.10%12:55
n4nd0>>>> PrimalMosekSOSVM trained in   22.071813:01
n4nd0>>>> BMRM trained in    8.632813:01
n4nd0SO-SVM: 85.10%13:01
n4nd0BMRM:   81.90%13:01
n4nd0MC:     46.30%13:01
uricamicback again, I had to read quite a big log :)13:07
uricamicso for the argmax question13:07
uricamicI said, that it is reasonable to have overloaded argmax function, which in training will get loss and therefore it will find the most violated constraint13:08
uricamicand the standard argmax function which should be used for prediction of with trained W vector13:09
_____________uricamic: should it argmax including loss?13:09
uricamiconly in training13:10
_____________n4nd0: what is MC?13:10
_____________uricamic: sure13:10
_____________uricamic: my suggestion is13:10
_____________there is a training flag in argmax13:10
uricamicbut then for predictions it shouldn't use it13:10
uricamic_____________: ok13:10
_____________in case of training it includes loss13:10
n4nd0_____________: linearMulticlassMachine with LibLinear13:10
_____________in other cases no13:10
n4nd0_____________: see libshogun/so_multiclass.cpp13:11
_____________uricamic: good for you?13:11
uricamic_____________: ok, that should be fine then13:11
_____________n4nd0: agree?13:11
n4nd0then I had not fucking idea what the argmax should do13:11
n4nd0I feel stupid13:11
_____________ehhm I do not understand13:11
_____________I was wrong not you13:11
uricamicbecause then it fits the design and it is easy to understand what is going on there13:11
n4nd0_____________: I thought the loss is not included in the argmax13:12
n4nd0so uricamic13:12
_____________what we were arguing about last night?13:12
n4nd0uricamic: can you please check MulticlassModel.cpp13:12
uricamicargmax should do exactly what the name says - find argument where the dot product is maximal13:12
_____________I was pretty sure my position is that argmax with loss is equal to argmax without13:12
_____________but it was wrong13:12
n4nd0_____________: yeah, but I am not talking about that13:12
n4nd0I thought that the argmax must not include the loss13:13
_____________to score?13:13
uricamicand in training it has to find this maximum enhanced with losses -> i.e. find the most violated example13:13
n4nd0I understand13:13
uricamicn4nd0: the version which now in shogun or somewhere else?13:13
CIA-52shogun: Sergey Lisitsyn master * r50ba31c / src/shogun/structure/StructuredModel.cpp : Removed redundant loss in generic risk -
n4nd0uricamic: let me paste it for you13:14
uricamicn4nd0: ok13:14
n4nd0that is used for training currently13:14
n4nd0and if I understand everything properly now13:14
n4nd0that is kind of wrong13:14
n4nd0since for training13:14
n4nd0the score should be13:15
_____________n4nd0: uricamic so should we add return_value->delta in risk?13:15
n4nd0uricamic: what is computed in that snippet in line 8 + the loss13:15
_____________n4nd0: yes it is wrong13:15
_____________it is not wrong for *applying*13:15
_____________but it is wrong for *training*13:15
uricamicn4nd0: I miss the loss there13:16
n4nd0uricamic: exactly13:16
_____________uricamic: n4nd0 only in case of training right?13:16
n4nd0_____________: I believe so13:16
_____________I can fix that13:17
uricamicit is ok for prediction but for training u have to add loss there to find the maximum score on features with loss13:17
_____________or anybody is willing to?13:17
n4nd0_____________: I can do it13:17
_____________okay please do then13:17
n4nd0uricamic: should we expect better classification results after this change?13:17
_____________n4nd0: I am unsure we need delta parameter in result type13:17
n4nd0_____________: that was designed by Nico13:18
_____________if it is being added in case of training to score13:18
_____________we won't use it anyway13:18
uricamicn4nd0: well I would expect that, because if you haven't counted with loss in training u actually trained different classifier13:18
n4nd0_____________: I am not sure if it is used somewhere right now13:18
n4nd0_____________: did you check?13:18
uricamic_____________: what do u mean by resturn_value->delta?13:19
_____________it is being set but it won't be used by risk13:19
_____________because score contains it13:19
_____________uricamic: 10913:19
_____________we don't need it, right?13:19
uricamicprobably not13:20
_____________because we don't need in when applying13:20
_____________and it is included when training13:20
_____________n4nd0: agree?13:20
n4nd0I wouldn't remove it right away13:21
n4nd0I guess Nico put it there for some reason13:21
n4nd0but no idea13:21
_____________okay no need to remove it13:21
_____________but lets remove it's setting in multiclass model13:21
_____________okay I'll wait for your code :)13:26
_____________n4nd0: we may do that with delta though13:29
_____________n4nd0: we then would need to add it in risk function but subtract from maximum score13:29
_____________see what I mean?13:29
wikingn4nd0: just a little bit of comment on CStructuredModel. dont u wanna pass const SGVector<float64_t>& w in argmax?13:29
wikingn4nd0: just to reserve some space on the heap13:30
wikingi mean stack13:30
wikingotherwise copy constructor is being called etc etc etc13:30
_____________please do that somebody13:30
_____________or I will13:30
wikingand actually13:31
wikingin risk function13:31
_____________what is in risk function?13:31
wikingdon't we want to use SGVector<float64_t> instead of the simple pointers?13:32
wikingfloat64_t risk(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info=0);13:32
wikingfloat64_t* -> SGVector13:32
_____________why to use it?13:32
_____________we could but any specific reason?13:32
wikingsince we have sgvector as a wrapper for simple type arrays13:33
wikingno other reason13:33
_____________wiking: to make it explicit I would keep it13:34
_____________it makes you think you are going to write to given memory13:34
_____________we need to add director model before release13:37
_____________I will do that13:37
_____________oops I am at job13:38
wiking_____________: working again?13:38
_____________yes starting from this week13:39
wikingsame place?13:39
_____________it is worth to mention that netcracker supports shogun implicitly :D13:39
wikingput it in the release notes :>13:40
_____________because I do shogun instead of my job13:40
wiking(got that part ;P)13:41
wikingn4nd0: wtf is with CLatentData CData and CStructedData?13:43
n4nd0wiking: you talkead with sonney2k about that right?13:45
n4nd0he didn't like CData I think13:45
n4nd0uricamic: does BMRM use argmax somewhere?13:45
_____________n4nd0: ?? risk13:46
uricamicBMRM uses risk function, and risk function should use argmax inside13:46
n4nd0call directly to argmax13:46
wikingn4nd0: i remember sonney2k telling us that he doesn't like the naming... that was as far as i understood :)13:46
wikingn4nd0: that's why i'm asking to what now?13:46
wiking_____________: dont u wanna fix this? :)13:47
n4nd0wiking: I don't know ... I am ok with anything regarding that to tell the truth :)13:47
wikingand i'm TOOTALy not being ironic here :)13:47
_____________wiking: fix what?13:47
wikingas much as it sounds :>13:47
wiking_____________: so that we have this shit going on13:47
wiking_____________: we have CLatentData (actually we had now it's CData) and CStructuredData13:47
wikingand basically they are serving the same exact purpose13:48
n4nd0uricamic, _____________: so ... the accuracy doesn't change at all in this multiclass example13:48
n4nd0_____________, uricamic: training with argmax using or not using loss13:48
uricamicyou mean with argmax containing losses in training?13:48
wiking_____________: so i thought it would be great to remove this redundancy and merge CLatentData and CStructuredData and just have CData13:48
wiking_____________: but sonney2k didn't like the name13:48
wiking_____________ so we need a new name and that's it basically ;)13:49
_____________n4nd0: it might be for multiclass (as far as I understand)13:49
_____________wiking: why latent is not structureD?13:49
wiking_____________ because it's not :)13:49
n4nd0uricamic: yes13:49
uricamicn4nd0:  but it could be true just for some particular example, definitely not generally13:49
n4nd0uricamic: ok13:49
wikingi mean they are semantically not the same13:49
_____________wiking: what is different in latent?13:49
n4nd0uricamic: I am using also loss equal to 1 or 0 so it might normal that it doesn't make a huge difference13:49
wiking_____________: that it's latent and not fixed contrary to structured13:49
wiking_____________: other than that it's the exact same implementation :)13:49
uricamicn4nd0: yep, that is another important fact13:50
wiking_____________: it's just semantics13:50
wiking_____________: but imho important one...13:50
uricamicwhen u use e.g. L2 loss the difference should be more obvious13:50
_____________wiking: okay lets get back to that later13:51
n4nd0uricamic: it converges faster including the loss though, for the PrimalMosekSOSVM case13:52
_____________n4nd0: what did you change now?13:53
uricamicn4nd0: you mean less number of iterations?13:53
n4nd0I am just comparing what happens when the argmax includes the loss or not13:53
n4nd0uricamic: I am just looking at the training time13:53
n4nd0uricamic: but I guess it is because the number of iterations is smaller13:54
uricamicn4nd0: I see, but it should be reasonable since without loss it could happen that sometimes not true most violated constraint is found and therefore it needs more iterations to converge13:54
n4nd0uricamic: yeah, could be13:55
n4nd0uricamic: you are an expert in this man!13:55
uricamicn4nd0: :D I wouldn't call myself one, but I have already written one paper about that13:56
uricamicactually 2 but unfortunately the second one was not accepted, since someone has already done something similar -> axed by reviewers immedeately13:56
n4nd0too bad13:57
uricamicit is life :)13:57
n4nd0uricamic: btw, I would really really prefer if the MulticlassModel::risk calls to argmax and delta loss13:57
n4nd0uricamic: what do you think?13:57
uricamicn4nd0: sure, it is no problem13:57
uricamicI will get to this hopefully soon13:57
n4nd0uricamic: do you want to change that?13:57
uricamicn4nd0: yep I can, but if u want to do that feel free :)13:58
uricamicsince I have to write TR for my work which could be then referenced by shogun tutorial13:58
n4nd0I think I am going to stop soon of shoguning today ...13:59
n4nd0I should have done something for the job this morning13:59
_____________wiking: yes that made all russian twitter day yesterdday14:00
_____________n4nd0: uricamic where did you stop?14:01
_____________is it ok to include loss in multiclass now?14:01
_____________can I commit that?14:01
uricamic_____________: you mean loss in multiclass argmax?14:01
uricamic_____________: then yep14:02
_____________if nobody of you are willing to do that I'll do14:02
_____________n4nd0: job?14:02
n4nd0_____________: yeah14:03
n4nd0my professor at KTH gave me some things to do to help out in the robotics course this year too :)14:04
_____________teaching is great14:04
n4nd0he didn't tell me if he needs me in the lab this year too, for the moment I am just setting up some material14:05
n4nd0so to sum up14:06
n4nd0now both Multiclass and HM-SVM should work good with BMRM methods14:06
n4nd0uricamic, _____________: agree?14:06
_____________n4nd0: well14:10
_____________if HM-SVM argmaxes with loss14:10
_____________does it?14:10
_____________okay nice14:11
_____________then yes, but multiclass needs to be changed14:12
_____________n4nd0: does produce 99% still with latest changes?14:12
n4nd0it is changed14:12
_____________changed where?14:12
_____________by you - you mean?14:13
_____________if so let me merge it ;)14:13
n4nd0PR going14:13
n4nd0python structure_hmsvm_bmrm.py14:15
n4nd0Accuracy = 0.998914:15
_____________I ask because I removed loss from risk14:15
_____________I assume it is correct and score includes it14:16
n4nd0let me think about it a moment14:16
n4nd0but I think one term needs to be substracted from the risk14:18
n4nd0psi_truth in particular14:18
_____________multiclass has that problem14:18
n4nd0well not just psi truth14:19
n4nd0but <w, psi(xi,yi)>14:19
_____________is it true for HMModel too?14:20
n4nd0I am not sure for that14:20
n4nd0for example14:20
n4nd0I don't see directly that the score in the HMMModel includes the loss14:20
n4nd0although it may be since a matrix used for Viterbi decoding includes the loss term...14:21
n4nd0I am not sure regarding this14:21
n4nd0I will ask Nico about it tomorrow14:21
n4nd0or maybe uricamic can tell us about it14:21
_____________n4nd0: why did you make first -INFTY?14:22
_____________this way you disallow it to select 0th class14:23
_____________if you do it this way then c=0 initially14:23
-!- av3ngr [av3ngr@nat/redhat/x-swaimdyxogxxwdlk] has quit [Quit: That's all folks!]14:24
n4nd0mmm but that is weird14:29
n4nd0the accuracy remained the same after that fix14:29
n4nd0yeah of course it'd remain the same if I don't compile ....14:30
wikingn4nd0: you talking to me? (c) ?14:31
uricamicn4nd0: sorry, I have been away for a while, what is the question?14:31
n4nd0looking better now14:31
n4nd0wiking: talking alone I think...14:31
wikingn4nd0: hehe it was a quote from the taxi driver14:32
_____________n4nd0: c=014:32
n4nd0uricamic: I need a rest now ...14:32
_____________please :)14:32
uricamicn4nd0: ok :)14:32
_____________n4nd0: where is your power?14:32
n4nd0_____________: I updated the PR14:32
wikingn4nd0: for your rest ;)14:32
n4nd0I lost my power14:33
CIA-52shogun: iglesias master * r1a87bf0 / (3 files): * fix MulticlassModel argmax, the loss must be included in training to find the MMV -
CIA-52shogun: iglesias master * r80f82b5 / (3 files): + extension in so_multiclass example and * fix in multiclass BMRM -
CIA-52shogun: iglesias master * rbbb3457 / src/shogun/structure/MulticlassModel.cpp : * wrong looop initialization -
CIA-52shogun: Sergey Lisitsyn master * r8cec5d0 / (6 files in 2 dirs): Merge pull request #752 from iglesias/master -
n4nd0I think i'll do the dishes during my rest instead14:33
shogun-buildbot_build #333 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: iglesias <>, Sergey Lisitsyn <>14:44
CIA-52shogun: Sergey Lisitsyn master * r2f999e4 / examples/undocumented/libshogun/so_multiclass_BMRM.cpp : Update examples/undocumented/libshogun/so_multiclass_BMRM.cpp -
wikingmay the ILSVRC people rot in hell14:48
wikingand capital14:48
wikingwtf :))))14:48
_____________oracle loves to capitalize words14:49
wikingand putting tars in tar is really a smart idea...14:49
_____________putting a tar archiver into a tar is way smarter14:49
wikingyeah that'dbe14:49
wikingbut still now i'm running low actually on space on the cluster :)14:50
wikingi mean not the big hdfs14:50
wikingbut the local little /home fs14:50
_____________how big it is untared?14:50
wikingwell i guess it's the same size14:50
_____________yeah makes sense14:50
wikingso i would need another +138 gigs14:50
wikingbut i think i've found a way around here :)14:50
wikingsince the / has 800 gigs free on one node14:51
wikingthat is part of the free space of hdfs14:51
wikingbuuut i could use that for temp time14:51
wikingthey put 1000 tars within that big tar14:53
wikingi think the 1000 tars is the different classes14:53
wikingthis is nice14:55
wikingactually they've provided a matlab file with all the dsift features within a class14:55
* wiking wonders if it's all the images in the class14:55
_____________dense sift? it must be way too big to be large scale14:55
wikingit is but still14:56
wikingit's there14:56
_____________however what is the other way?14:57
_____________or bow like yeah14:57
wikingmmm lol yeah14:57
wikingwe have already all the sifts if we want it :)14:57
_____________wiking: what is deadline for ilsvrc?14:58
wiking23rd of sept14:58
_____________I see14:58
-!- _____________ [5bdfb203@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]15:07
shogun-buildbot_build #351 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: iglesias <>, Sergey Lisitsyn <>15:20
shogun-buildbot_build #444 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: iglesias <>, Sergey Lisitsyn <>15:45
shogun-buildbot_build #334 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Sergey Lisitsyn <>15:50
-!- alexlovesdata [] has joined #shogun15:58
shogun-buildbot_build #445 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>16:10
-!- blackburn [~blackburn@] has joined #shogun16:12
shogun-buildbot_build #352 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>16:12
blackburntime to fix things16:13
uricamicblackburn: btw I have checked IRC log, BMRM cannot be used with lambda=016:16
blackburnuricamic: sure16:16
blackburnthat's clear16:16
uricamicand also, it should not be even close to 016:16
blackburnwell it converges with small lambda16:16
blackburnbut it makes no sense16:17
blackburnit is just overfitting16:17
uricamicblackburn: yep, I just wanted to point that out just for sure :)16:17
blackburnthe only reason I did that16:17
blackburnto try to fit model perfectly16:17
blackburnon training data16:17
blackburnwith no regularization at all16:17
uricamicwell, it can happen that for some instance and features the regularization term is not needed, by then it is not good idea to use BMRM16:17
uricamicblackburn: I see, but BMRM is not designed for such things, lambda has to be >016:18
blackburnuricamic: btw multiclass model is not typical in bmrm, right?16:18
blackburnI mean it is L2 of whole w16:18
uricamicI am not sure that I understand the question now16:19
blackburnahhh nevermind16:19
blackburnit is ||w||^2 so it is the same16:19
blackburnI meant that in crammer-singer or weston one16:20
blackburn\sum ||w|| is used16:20
uricamicyou mean the way how it is implemented in shogun now?16:20
blackburnbut it is the same in cases of ||w||^216:20
uricamicI see16:20
blackburnread ^ as \sum ||w_i||^216:20
blackburnuricamic: so are we ok now with multiclass bmrm?16:22
blackburnuricamic: btw what is the difference between all these solvers?16:22
uricamicI will check multiclass bmrm, soon, but I guess it should be ok now16:23
uricamicthe difference16:23
uricamicBMRM is the standard method as described by Teo et al.16:23
uricamiconly the inactive cutting plane (ICP) removal strategy is added there16:24
blackburnhow one should choose a solver?16:24
uricamicPPBMRM is BMRM enhanced by prox-term16:24
blackburnkind of initial guess?16:25
uricamicwhich is our method how to enforce possibility of some initial solution by restricting the L2 distance between consecutive W_t16:25
blackburnour - your and vojta?16:26
uricamicand the last one P3BMRM is the same  as PPBMRM plus it can use multiple cutting plane models, it is similar to Joachim's nslack16:26
uricamicwhen cp_models = 1 in P3BMRM then it is reduced to PPBMRM16:27
uricamicand when you set K = 0 PPBMRM should reduce back to BMRM (but I haven't checked this particular one yet)16:27
blackburnwe have to make your example support default values16:27
uricamicI mean in Shogun's implementation16:27
uricamicyou mean cpp example?16:28
uricamicok, I put it there just for my convenience to test all algorithms quickly16:28
blackburnI don't mind command line commands actually16:28
blackburnbut we need it able to run as is16:29
uricamiclike ./so_multiclass_BMRM and that's all?16:29
blackburnMakefile does that16:29
blackburnwhen testing16:29
uricamici.e. to use randomly generated data and default settings16:29
blackburnyeah that would work I think16:29
blackburncan you do that next time?16:30
blackburnthe earlier the better because currently test is broken:
uricamicI think I will do this either in the evening or tomorrow16:30
uricamicahh, ok16:30
uricamicso back to the solvers, if u are tuning the optimal lambda, it is nice to have some reasonable range of lambda's and in the cycle use PPBMRM or P3BMRM, since then it will use solution from the last lambda as initial one and converges much faster, especially for lower values of lambda16:33
uricamicto put it in numbers, on MNIST data (10 classes, 784 feature dimension), 60k training examples16:34
uricamicthe difference between BMRM and P3BMRM on lambda=116:34
blackburnkind of hot start?16:35
uricamicwas from 2.01 hours to 2.09 hours for P3BMRM16:35
uricamic0.209 for P3BMRM :D16:35
uricamicotherwise it would be so impressive :D16:35
uricamicoh, again, wouldn't :D16:36
blackburn2 hrs?16:36
uricamicin iterations it was from 5932 iterations to 40816:36
blackburnuricamic: 0.209h sounds like it is pretty fast, something near to liblinear16:37
uricamicand the other thing about P3BMRM is that it could be quite easily written to use more cores16:37
uricamicblackburn: I haven't tried liblinear yet :)16:38
uricamicbut this multicore thing has probably reason only when computation of the risk takes a long time16:39
blackburnI see16:39
uricamicso in multiclass it would probably not give any speedup at all16:39
blackburnthat's clear16:39
uricamicanother thing is that both PPBMRM and P3BMRM tends to converge to more precise solution than BMRM because of our alpha setting strategy and termination condition16:40
blackburnlast two weeks I've been trying to employ L1/L2 tree regularization16:41
blackburnno success :(16:41
-!- blackburn1 [~blackburn@] has joined #shogun16:45
-!- blackburn [~blackburn@] has quit [Ping timeout: 250 seconds]16:47
blackburn1at some point users must claim us as an OSS having best support ever16:50
blackburn1I will eat them if not16:55
-!- n4nd0 [] has quit [Quit: leaving]17:02
-!- blackburn [~blackburn@] has joined #shogun17:03
-!- blackburn1 [~blackburn@] has quit [Ping timeout: 244 seconds]17:04
-!- blackburn1 [~blackburn@] has joined #shogun17:18
CIA-52shogun: Sergey Lisitsyn master * r90d4424 / (9 files in 4 dirs): Refactored multiclass machines to support C parameter selection -
-!- blackburn [~blackburn@] has quit [Ping timeout: 252 seconds]17:21
-!- uricamic [~uricamic@2001:718:2:1634:155e:7544:fba4:9878] has quit [Quit: Leaving.]17:22
shogun-buildbot_build #335 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Sergey Lisitsyn <>17:27
wiking"otherwise my concern would be -- where is the unittest?   logic-wise17:41
wikingit should indeed accomplish the desired goal but imho test would help to17:41
wiking'guarantee' that ;)17:41
wikingpeople are damanind :)17:42
wikingdemanding :P17:42
wikingso yeah where's the unit test :>17:42
shogun-buildbot_build #446 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>17:55
shogun-buildbot_build #353 of deb2 - static_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>17:57
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has joined #shogun18:02
yoohi all18:02
-!- yoo [2eda6d52@gateway/web/freenode/ip.] has quit [Quit: Page closed]18:28
-!- emrecelikten [] has joined #shogun18:34
-!- Netsplit *.net <-> *.split quits: shogun-buildbot_, yoh, audy, emrecelikten, @sonney2k, naywhayare, sr___, wiking, zxtx, CIA-52, (+1 more, use /NETSPLIT to show all of them)19:38
-!- Netsplit over, joins: CIA-52, sonney2k, audy, shogun-buildbot_, sr___, wiking19:41
-!- naywhaya1e [] has joined #shogun19:41
-!- Netsplit over, joins: yoh, emrecelikten, blackburn1, zxtx19:41
-!- ServerMode/#shogun [+o sonney2k] by moorcock.freenode.net19:41
shogun-buildbot_build #336 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Sergey Lisitsyn <>19:43
-!- Netsplit *.net <-> *.split quits: yoh19:45
-!- Netsplit over, joins: yoh19:45
wikingblackburn1: we were in a split19:45
wikingso any ideas there?19:46
wikingah the func arg19:47
wikinghave you got my two example cases for diff dims?19:47
wiking and
wikingare two actual applications19:47
blackburn1this is *not* different dims case19:48
blackburn1just like in multiclass it is just sparse19:48
wikingblackburn1: noup
wikingcheck this one19:48
wikingscore += sm->w[cur_class*sparm->size_hog+i+1]*hog[i];19:48
wikingw = num_class*sparm->size_hog19:49
blackburn1it is just multiclass hog19:49
wikingi mean the dimension of w19:49
wikingotherwise the indexing would not work19:49
blackburn1dimension is W is always equal to dimension of Psi, but Psi can be sparse19:49
blackburn1that is the case19:49
blackburn1and multiclass is the case19:49
wikingu r right19:50
blackburn1dimension of Psi here is not equal to dimension of hog19:50
blackburn1it is dim of hog * n_classes19:50
blackburn1but everything out given class is 019:51
blackburn1so it is not computed at all19:51
wikingmy bad19:51
blackburn1actually if you have that code HOGSO Model is straightforward19:53
wikingmmm now i'm in trouble...19:55
wikingi need to calculate this for all examples: argmax_{(ybar,hbar)} [<w,psi(x,ybar,hbar)> + loss(y,ybar,hbar)].19:55
wikingand i want to keep the COFFIN fw19:56
blackburn1line 25 is a dense dot19:56
blackburn1wiking: it is argmax alreadyy19:58
wikingi mean now i'm writing a generic implementation i'm not talking about those pastebins19:58
wikingit's just that currnetly i have 1 method for psi19:58
wikingis that CDotFeatures* get_psi....()19:58
wikingi guess i could generate all the different CDotFeatures and generate a matrix of the w,psi(x,ybar,hbar)> + loss(y,ybar,hbar) values20:01
wikingand do there a max value search for each row/column (depending on who do i store [w,psi(x,ybar,hbar)> + loss(y,ybar,hbar)]) and then sum it up20:02
wikingit's just a lot of memory waste...20:02
wikingblackburn1: input?20:03
blackburn1yeah sounds inefficient20:04
wikingbut i dont have something like SGVector<float64_t> get_psi(x,y,h)20:04
wikingas sparse vector wouldn't work then20:05
blackburn1why don't you have it?20:05
wikingbecause i couldn't have sparse vector20:05
wikingthis is why i threw out the first place the get_psi(x,y,h) and have only CDotFeatures* get_psi_vectors()20:06
blackburn1I see20:11
shogun-buildbot_build #447 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>20:12
-!- n4nd0 [] has joined #shogun20:20
n4nd0wiking: I just read the the thread with the problem with risk function args :D20:22
-!- blackburn [~blackburn@] has joined #shogun20:42
-!- blackburn1 [~blackburn@] has quit [Ping timeout: 248 seconds]20:43
-!- sr___ [u5548@gateway/web/] has quit [Read error: Connection reset by peer]21:03
-!- sr___ [u5548@gateway/web/] has joined #shogun21:03
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]21:15
-!- naywhaya1e is now known as naywhayare21:38
-!- n4nd0 [] has quit [Read error: Connection reset by peer]21:56
-!- n4nd0 [] has joined #shogun21:58
n4nd0any clue why that crash is happening in Arthur's question?22:03
n4nd0mailing list22:03
wikingn4nd0: swig version could be a possible problem22:11
wikingi mean i have 2.0.822:11
wiking1.3.29 seems a bit old22:11
n4nd0wiking: that's true22:12
n4nd0wiking: should I answer him or do you want to do it?22:13
wikingmmm i think that should be tested though22:13
wikingand if that's the case then we should find the minimum swig version shogun requires to be able to compile the modular interfaces22:14
wikingthe problem is that i have 0 knowledge about swig22:15
wikingi think gsomix would be the most competent to say anything about this22:15
n4nd0wiking: I remember once my compilation failed and sonney2k told me to update swig22:16
n4nd0it worked after that22:16
n4nd0I will just tell him in case it can solve his problem22:16
@sonney2kn4nd0, wiking yes we need newer swig - but IIRC configure should fail if swig < 2.0.422:17
wikingyeah the problem is i think that he would need to manually install a new version of swig as he now uses the standard distrib supplied swig22:17
wikingooh yeah22:19
n4nd0I could do it via apt-get easily22:19
wiking_swig_version=`${SWIG} -version 2>/dev/null | grep Version | cut -f 3 -d ' '`22:19
wikingif assert_version swig $_swig $_swig_version 2.0.4; then22:19
wikingechores "$_swig_version"22:19
wikingn4nd0: it's RH :)22:19
n4nd0wiking: RH?22:19
wikingaccording to his email he uses redhat => no apt-get ;P22:20
wikinglol his configure log is22:21
wiking============ Checking for SWIG ============22:21
wikingResult is: 1.3.2922:21
wikingand the assert_version doesn't fail22:21
-!- blackburn [~blackburn@] has joined #shogun22:21
wikingso there's something wrong with that if line22:22
blackburndiscussing swig issue? I have to note he is using 1.1.022:23
blackburnI do not remember if it requires >2.0 but probably it does22:23
wikingblackburn: just joined 1 minute later22:23
wikingblackburn: yeah accoding to ./configure script it should actually fail because it's not 2.0.4 or later22:24
wikingbut for some reason that if branch keeps being true22:24
blackburnhe seems to be a hardcore guy22:24
blackburnusing RHEL22:24
blackburnn4nd0: did you graduate from kindness school? :)22:26
n4nd0blackburn: haha why so?22:26
n4nd0it is free to be kind, not expensive :D22:27
wikingif i set _swig_version by hand the configure script fails22:27
wikingi think he is using a rather old shogun version22:28
wikingwhere this check is not yet present in ./configure script22:28
wikingas Result is: gcc 4.1.2, ok22:28
wikingand afaik that should fail as well with < 4.322:28
wikingi wonder why they want to use shogun @ ernst & young ;P22:36
blackburnhey really22:37
blackburnwhat the heck22:37
wikingmaybe for tax auditing :DDD22:37
blackburnwhy anybody in ernst & young would need shogun22:37
wikingi guess some classification stuff on tax auditing data ;P22:37
wikingwould be great to hear more what's his idea for application22:38
blackburnhey we must get more E&Y users22:38
blackburnI bet they have money22:38
blackburnI don't mind to become rich :D22:38
wikingyeah you should do a shogun spin-off ;P22:39
wikingand make money of shogun support :)))))22:39
blackburnn4nd0: did you notice22:39
wikingi mean this guy should just fucking update his linux distrib22:39
blackburnhis name is sean22:39
blackburnbut email is arthur edge22:39
wikingah lol22:40
wikingmaybe he is disguising himself ;P22:40
n4nd0haha yeah I noticed22:40
wikinghow old is RHEL 5.8 actually?22:40
blackburnI like footer part22:40
n4nd0I surprised myself when I thought, I am going to change the signature just in case it is not the same as the mail22:40
blackburnwiking: 2010 probably22:42
n4nd0to check* not to change22:43
blackburnAny U.S. tax advice contained in the body of this e-mail was not intended or written to be used, and cannot be used, by the recipient for the purpose of avoiding penalties that may be imposed under the Internal Revenue Code or applicable state or local tax law provisions.22:45
blackburnwiking: really?22:46
blackburnlatest is 6.3 IIRC22:47
blackburnI had to work with 5.3 yesterday on dev server22:48
wikingand it's a 2012 february news22:48
wikingredhat ships with fucking gcc 4.1.x in 2012?22:48
blackburnit is ENTERPRISE redhat22:49
blackburnvery slow22:49
blackburnvery very slow22:49
wikingso it must be filled with acient stuff? :)22:49
blackburnvery very slow developing22:49
wikingi guess they are following: 'the older the better'22:49
blackburnthey would use win9522:49
blackburnif m$ did support it still22:49
wikingah lol22:49
wikingthere's gcc44 package22:49
wikingso he should be ok22:49
blackburnwiking: I know a company which forces my company to use ClearCase22:49
blackburnheard about that?22:49
wikinghahahah YES22:49
wikingthat's like fucking 200022:49
wikingyes unfortunately22:49
blackburnthat's like22:49
blackburnthat's like nothing on the earth22:49
blackburnit is pure love22:49
wikingyeah it's a pure beauty22:49
blackburnit is when you lose your mind of happiness22:50
blackburnbut serious, did they really think that dance is a good  way to develop22:50
blackburnyou have to create an integration stream to view sources22:50
blackburnthen you add activity there you specify what do you want22:50
wikingi dont want to hear about that shit!22:50
wikingi had physical pain22:50
blackburnand then you need to create a development stream22:50
wikingwhen i had to use it22:50
blackburnand join project with it22:50
blackburnthen checkout or hijack the file where you want to remove 2 lines22:50
blackburncommit (or how is it called out there)22:50
wikingi mean seriously why would u not use git+github in 2012?22:50
blackburnand write down 40 pages description of the commit22:50
blackburnwe are using svn22:50
wikingblackburn: and i hope u use git-svn ;)22:50
blackburnbut one project uses clearcase still22:50
wikingblackburn: just to make them cry :))))22:50
blackburnthat is restricted and I didn't say that but it is TELUS22:50
blackburnwiking: you are the only one but me developer of shogun22:50
blackburnwho have a sexual contact with CC22:51
blackburnI have still though22:51
wikingfucking clearcase22:51
wikingi know a guy22:51
wikinghe just had to start to use it22:51
blackburnin 201222:51
wikingyep yep22:51
blackburnsounds like torquemada is back22:51
wiking2 weeks ago or something22:51
blackburnand he is full of will to punish somebody22:51
wikinghe seemed desparate on facebook22:51
blackburnthat's quite normal to get a depression after using CC for a while22:52
blackburnI am pretty sure there was a case of commiting suicide after trying to get that thing to get your changes22:54
wikingi think a git-cc project would never be possible22:54
blackburnI am not using git-svn btw22:54
wikingi was proved WRONG22:54
blackburnI actually feel comfortable with svn as well22:55
wikingI wrote this purely for fun and to see if I could stop use Clearcase at work22:55
wikingonce and for all."22:55
blackburnfor the thing we develop it is ok22:55
wikingblackburn: yeah i just don't like that u actually cannot commit a code in svn w/o having internet connection22:55
wiking"Actually what I would love to see22:56
wikingmore is for Clearcase to die"22:56
blackburnwiking: that is ok when you have pretty reliable internet connection at your job22:56
wikinglol this guy did feel the pain22:56
blackburnI share his pain22:56
blackburnokay I am going to sleep now probably22:58
blackburnI am going to be at job 10 (11) o'clock :D22:59
blackburnor sth like 10*22:59
n4nd0blackburn: good night23:00
-!- blackburn1 [~blackburn@] has joined #shogun23:00
blackburn1oops my cable didn't like that23:00
blackburn1wiking: I think git-cc is too dangerous to even try23:00
blackburn1CC looks like a space ship command panel23:01
blackburn1who knows may be one button could destroy the whole world23:01
wikingyeah i think it opens a new dimension if u try it :)))23:01
blackburn1one may end with nuclear war or some chemical disaster23:01
wikingif saddam would have used gitcc23:02
n4nd0haha you guys scare me sometimes :D23:02
wikingmaybe they would have found some weapons of mass destructions :D23:02
blackburn1yeah CC with development stream23:02
blackburn1with all files hijacked23:02
blackburn1n4nd0: you should try CC23:03
blackburn1to achieve zen23:03
blackburn1in hofstadter book a theory that zen's stuff is just for destroy of all means23:03
-!- blackburn [~blackburn@] has quit [Ping timeout: 265 seconds]23:04
blackburn1CC is just the same, it destroys development as we know it23:04
n4nd0cc is a version control system right?23:04
wikingn4nd0: nonono23:04
wikingblackburn1: dont confuse him23:04
blackburn1spaceship missile control system23:04
n4nd0I was confused at the beginning with the command cc23:04
blackburn1with version control support23:04
wikingn4nd0: it's like a nuclear reactor man23:04
blackburn1have you seen star trek?23:05
blackburn1may be gayniggers from the outer space?23:06
blackburn1they had planet destruction stuff23:07
blackburn1it comes out of box with clear case23:07
n4nd0I don't know what the heck are you talking about :)23:07
wikingn4nd0: never ever touch clearcase man23:08
wikingnot even in your dreams23:08
n4nd0all right, I promise you that :)23:08
n4nd0I had heard about subversion and cvs but not about clearcase23:08
n4nd0when was that used?23:08
blackburn1n4nd0: in 1400-145023:09
blackburn1when the inquisition was like fashion23:09
wikingblackburn1: columbus tried to teach CC for the indians23:09
wikingand they told him: i rather fucking die23:09
wikingok i know i might be a bit offensive with this comment, wasn't inteded23:10
n4nd0then it was probably invented by Spanish people23:10
blackburn1n4nd0: it is not true jeanne d'arc was burnt23:10
n4nd0Spain was quite an important country back then23:10
blackburn1she had to commit to UNOPENABLE DEVELOPMENT STREAM23:10
n4nd0is that like a code in that system?23:11
wikingn4nd0: i think they thought that using CC would be the solution... that's why your ship sank against nelson admiral23:11
blackburn1yeah, they just forgot to complete VOBs on integration views23:12
blackburn1that usually leads to ship sink or sth like that23:12
blackburn1chernobyl is the similar case23:12
blackburn1somebody joined wrong project23:12
wikingblackburn1: i bet gorbachov thought that using CC would be the solution as well23:12
n4nd0I think we should move shogun to CC23:12
blackburn1haha lol23:12
n4nd0fuck off with git23:13
n4nd0CC sounds exciting23:13
blackburn1it is23:13
alexlovesdatabye guys23:13
-!- alexlovesdata [] has left #shogun []23:13
wikingCC is a masterpiece23:13
blackburn1already leaving?23:13
n4nd0we scared alex!23:13
wikingi must check for CC!!!23:13
blackburn1okay now finally leaving you too23:14
n4nd0CC vs git23:14
blackburn1have a nice night I will push you tomorrow to develop more code with clearcase23:14
blackburn1lock before edit is a pros23:15
blackburn1is in pros*23:15
blackburn1that's really nice23:15
blackburn1wiking: fits for your issue about breaking code :D23:15
n4nd0First, I would answer that comparing git to UCM is unfair to ClearCase, and not only from the point of view of performance!23:16
blackburn1nobody would break your code because you can LOCK IT23:16
blackburn1:D ohmfg23:16
blackburn1n4nd0: they all sound like dinosaurs23:16
blackburn1I've seen one BIG COMPANY WITH MANAGERS claimed that open source solutions sucks23:17
blackburn1in 201223:17
wikingblackburn1: yes23:17
wikingblackburn1: that's pretty much normal :D23:17
blackburn1okay okay23:17
n4nd0I found an answer you would like more in another forum23:17
blackburn1see you tomorrow23:17
n4nd0Now, there is some potential in ClearCase that might make it compare favourably to Git.23:17
wiking" If not, I'll probably need to have some internal discussion before proceeding.23:17
n4nd0But it is not commonly known or used, and would require some support at last from IBM.23:17
wikingthese guys are trying to use23:18
n4nd0I love the last part23:18
wikingshogun @ ey23:18
blackburn1n4nd0: yeah nuclear missile button is absent in git23:18
wikingblackburn1: :DDDDDDDDDDDDDDDDDDDDD23:19
blackburn1okay see you :)23:19
-!- blackburn1 [~blackburn@] has quit [Quit: Leaving.]23:19
wikinglet's support EY to use shoung23:20
wikingerm shogun ;P23:20
n4nd0wiking: do you know it gcc 4.1 will go well with shogun v.1.1.0?23:22
wikingi'm just checking23:23
wikingthat why actually we had that bug23:23
wikingi must dig out the stupid commit log23:23
wikingok yeah htis one23:27
wikingn4nd0: can u open this one?23:27
n4nd0it took a bit long though23:28
wikingyeah here too23:28
wikingmmm what was the code that actually caused this one23:28
wikingah git grep will tell me23:28
n4nd0oh didn't know there is a grep in git23:28
wikingit's fucking great23:29
n4nd0sound really useful yes23:29
n4nd0thank you!23:29
wikingmmm do we have tags?23:29
wikingah yeah cool we do23:29
n4nd0what are you executing exactly?23:30
wikingtrying to find out23:30
wikingwhether that bug actually present in 1.1.0 version23:30
n4nd0what command are you using? just to learn more about it23:30
wikingwell now i've done this23:30
wikinggit grep "\.~"23:31
wikingthat'll give me back where the destructor was called explicitly in the code23:31
wikingand now i wanna do the same23:31
wikingon the 1.1.0 tag23:31
wikingbut i cannot find tag for that23:31
n4nd0aham I see23:31
wikinggit tag23:31
n4nd0did you go back to that commit?23:31
wikingn4nd0: yeah tag is for that23:32
n4nd0what was the gcc bug exactly?23:32
wikingn4nd0: that you cannot call explicitly a destructor23:32
wikingon a templated class23:32
n4nd0all right23:32
wikingi.e. something.~SGVector<float64_t>()23:32
wikingwill not compile23:32
wikingthe problem is that there's no tag for 1.1.0 release23:33
n4nd0too bad23:33
wikingi'll check 1.0.023:33
n4nd0maybe with the release date we can guess the last commit done by that time?23:33
wikingok in 1.0.0 there was no such thing23:34
wikingmmm i cannot see it from the git log23:35
wikingwhich one supposed to be the 1.1.0 release23:35
wikinghow the fuck he downloaded 1.1.0 ?23:36
wikingah ok23:36
wikingi was checking the wrong thingy23:36
wikingi really dont feel for downloading the tar.gz and do a grep -r on it23:37
wikingi'll just tell him to try to compile it with 4.123:37
wikingand it might fail23:37
wikingif it does23:37
wikingthen it's because we execute that bug :D23:37
n4nd0wasn't the bug fixed in gcc 4?23:37
wikingn4nd0: after 4.323:38
n4nd0ahh al right23:38
n4nd0I thought it was after 3.x23:38
wiking4.3 and later23:38
wikingn4nd0: check the comments on the bug page23:38
wikingok this guy might be lucky23:39
wikingi've just checked it :D23:39
n4nd0let's see23:40
wikingsupport email sent23:42
n4nd0good job, nice mail!23:43
wikinghahah another one23:44
wikingit should fucking work23:44
wikingok shit i forgot to cc the mailinglist23:45
wikingwonder what acient osx is he using23:46
n4nd0time to sleep for me23:47
n4nd0good night!23:47
-!- n4nd0 [] has quit [Quit: leaving]23:47
--- Log closed Thu Aug 23 00:00:17 2012