Open in new window / Try shogun cloud
--- Log opened Tue Aug 14 00:00:14 2012
--- Day changed Tue Aug 14 2012
n4nd0the data one00:00
blackburnah00:00
blackburnwell nevermind I get it bakc00:00
n4nd0sorry about that :S00:01
CIA-39shogun: Sergey Lisitsyn master * r45a5ee4 / (40 files in 10 dirs): Merge pull request #684 from iglesias/so (+17 more commits...) - http://git.io/x0FI_A00:01
CIA-39shogun: Sergey Lisitsyn master * r122079a / data : Restored data - http://git.io/8Pl5mA00:02
n4nd0I think I do something wrong with the submodule00:02
n4nd0because my git status shows always00:02
n4nd0#modified:   data (untracked content)00:02
blackburnwhat is the output of git status in that dir?00:03
n4nd0http://pastebin.com/TQvfuw6600:04
n4nd0I don't want to put that file toy/hmsvm_data.mat in the main repo so I will delete it00:07
blackburnyeah00:07
blackburnthat's the reason00:07
n4nd0now I just get00:07
n4nd0# On branch master00:08
n4nd0# Your branch is ahead of 'origin/master' by 2 commits.00:08
n4nd0#00:08
n4nd0nothing to commit (working directory clean)00:08
n4nd0mmm but my last two commits are the same as the ones in shogun-data master00:08
shogun-buildbotbuild #345 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/345  blamelist: Viktor Gal <viktor.gal@maeth.com>00:09
blackburntry git reset --hard origin/master00:09
-!- zxtx [~zv@c-71-227-187-90.hsd1.wa.comcast.net] has quit [Ping timeout: 256 seconds]00:10
n4nd0blackburn: thanks, now everything is fine :)00:11
blackburnn4nd0: do you have a formulation of HM-SVM somewhere?00:13
blackburnprimal00:15
blackburnor is it only primal?00:15
n4nd0the optimization problem is the one used in generic SO00:16
blackburnloss?00:17
n4nd0the one in particular of the HM-SVM or the generic one?00:18
n4nd0for the HM-SVM we use Hamming loss00:18
n4nd0just count the number of elements that are different in two sequences of states00:18
n4nd0for the generic one we are using the Hinge loss00:19
blackburnn4nd0: so we need subgradient of hamming loss, right?00:19
n4nd0blackburn: do you mean for the risk function?00:19
blackburnexactly00:20
n4nd0I am not sure about that00:20
-!- gsomix [~gsomix@80.234.28.229] has quit [Ping timeout: 245 seconds]00:20
blackburnn4nd0: for a function h subgradient is a vector g00:21
blackburnsuch that for all w following holds true00:21
blackburnh(w') >= h(w) + g(w'-w)00:22
blackburnright?00:22
wikingblackburn: can u throw out the modular interface for LatentLabels ?00:23
wikingas the fix is not so much of a quickfix00:23
wikingand i gotta go to sleep now00:23
blackburnwhats the problem?00:24
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has quit [Ping timeout: 240 seconds]00:24
wikingsince i have to be awake in 3.5 hours00:24
blackburnI don't mind to wait for fix :)00:24
wikingah ok00:24
wikingthen i'll send it tomorrow around 12pm00:24
blackburnnp, ok00:24
wikingi gotta go now as i said00:24
wikinggnite00:24
blackburnnite00:25
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has joined #shogun00:25
blackburnn4nd0: what is the last message you received?00:25
n4nd0re00:25
blackburnhaha I totally missed that picture00:26
blackburnhttp://1.bp.blogspot.com/-4L8SUhLlmRk/T-DcRJSyvNI/AAAAAAAAAqk/Bg-RjlzijcM/s1600/image02.jpg00:26
blackburnthat's the very first puffin444 appearance00:27
blackburn:D00:27
shogun-buildbotbuild #346 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/346  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>00:27
n4nd0yeah00:27
n4nd0the last thing I got from was an exactly00:27
n4nd0I don't know if you got what I said after00:27
blackburnn4nd0: have you seen that pic?00:28
n4nd0yes00:28
blackburnah I see you +1 that00:28
blackburn:D00:28
n4nd0:)00:28
blackburnI wanted to see how puffin look like00:28
blackburnseems to be that his 5 kg notebook made him strong00:29
blackburnn4nd0: okay00:29
blackburnabout subgrad00:29
blackburndid you receive these messages?00:29
n4nd0which one?00:29
blackburn(02:21:40 AM) blackburn: n4nd0: for a function h subgradient is a vector g00:31
blackburn(02:21:52 AM) blackburn: such that for all w following holds true00:31
blackburn(02:22:08 AM) blackburn: h(w') >= h(w) + g(w'-w)00:31
blackburn(02:22:10 AM) blackburn: right?00:31
blackburnn4nd0: are you out again?00:32
blackburn:)00:32
n4nd0I am here00:34
blackburnn4nd0: so in HM-SVM primal is just00:35
blackburnl2 of w + hamming loss00:35
blackburnright?00:35
n4nd0the loss is not l2, it is the hinge loss00:36
n4nd0something similar to hinge loss to tell the truth, just max(0,z)00:36
n4nd0and it is not taken to w directly00:36
blackburnehmm I didn't say l2 loss00:36
n4nd0ok00:37
blackburnis regularizer ||w||^2_2?00:37
n4nd0what do you mean with l2 of w + hamming loss then?00:37
n4nd0where?00:37
n4nd0ok00:38
blackburnin your hm-svm :)00:38
blackburnhow do you define risk at all?00:38
n4nd0as they say in the paper :)00:39
blackburnyeah that's what I am begging for :D00:40
blackburnis that a paper? http://www.cs.brown.edu/~th/papers/AltTsoHof-ICML2003.pdf00:40
blackburnthe paper00:40
blackburnI mean00:40
blackburn:D00:40
n4nd0I understand it as a mean value of the loss function00:40
-!- zxtx [~zv@c-71-227-187-90.hsd1.wa.comcast.net] has joined #shogun00:40
n4nd0expected value transmits better the idea I think00:41
blackburnn4nd0: eq 18, right?00:41
n4nd0no00:41
blackburnwhich one?00:41
n4nd0that is the objective00:41
n4nd0in that paper they do not talk about risk functions00:42
n4nd0that definition pops up from a more general point of view, in generic SO00:42
blackburnobjective is the risk00:42
blackburnif we put s.t. part to the objective I mean00:42
n4nd0but those are the constraints, not the objective00:43
blackburnwe can put it into objective and get a risk function00:43
n4nd0http://jmlr.csail.mit.edu/papers/volume6/tsochantaridis05a/tsochantaridis05a.pdf00:43
n4nd0page 400:44
n4nd0that is the formal definition of the risk function00:44
n4nd0I don't really see the equivalence you say :S00:44
n4nd0putting the constraints in the objective to get the risk function00:45
blackburnrisk is regularizer+loss00:45
blackburnbinary svm can be seen as00:45
n4nd0aham maybe that is another definition then00:45
blackburn||w||^2+max(0,1-y*<w,x>)00:45
blackburnright?00:45
blackburnw/o s.t part at all00:45
n4nd0what I understood following this paper is that the risk is a measure of expected value of the loss00:46
n4nd0let's see what Michal has done00:46
blackburnn4nd0: no, it is an empirical risk00:47
blackburnwe optimize regularized risk00:47
shogun-buildbotbuild #347 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/347  blamelist: iglesias <fernando.iglesiasg@gmail.com>00:47
blackburnthat's objective w/o s.t. part00:47
n4nd0ok00:48
blackburnn4nd0: expected value is an actual risk00:48
blackburnn4nd0: one interesting point here00:49
blackburnn4nd0: do you know why one should introduce slack variable here?00:49
n4nd0to penalize how far we go from the region of correct classification00:50
blackburnn4nd0: I mean why not to make it use hinge loss as is?00:51
n4nd0what do you mean with hinge loss as is?00:51
blackburnmax(0,1-yz)00:51
n4nd0because in structured output, how do you define y?z?00:52
n4nd0y is not a number any more00:52
blackburnno, I don't mean that00:52
blackburnin typical svm00:52
n4nd0:)00:52
blackburnwhy do we introduce s.t. part00:52
blackburnbut not use max(0,1-yz) in objective?00:52
n4nd0you in typical svm you mean00:53
blackburnjust binary00:53
n4nd0min ||w||^2 +  \sum_i slack_i00:53
blackburnsvm without s.t. part00:54
blackburnis00:54
blackburnmin ||w||^2 + \sum_i max(0,1-y_i(<w,x_i>+b))00:54
blackburnagree?00:54
n4nd0agree00:54
blackburnno s.t. part at all00:54
blackburnright? :)00:55
n4nd0yes00:55
blackburnwhy not to solve that?00:55
n4nd0that is non linear00:55
blackburnyeah discontinuous I would say00:56
n4nd0because of the max00:56
n4nd0hehe ok00:56
blackburnthat's what I mean00:56
blackburnbut actually00:56
blackburnthat thing I wrote00:56
blackburnis the regularized risk of svm00:56
blackburnit can be solved via bmrm already00:57
n4nd0ok00:57
blackburnsee what I mean?00:58
n4nd0mmm not really00:58
blackburnso in your hm-svm thing00:58
blackburnyou have00:58
blackburnmin ||w||^2 + hamming(what?)00:59
n4nd0it is not directly like that00:59
blackburncan we come to that?00:59
n4nd0in generic SO we have01:01
n4nd0https://dl.dropbox.com/u/11020840/shogun/so.pdf01:01
n4nd0look at equation (1)01:01
n4nd0so-svm with margin rescaling01:01
n4nd0now01:02
n4nd0the l we are using is this modified hinge loss, i.e. max(0,z)01:02
blackburnno, for generic so we can't come up with subgradient at all I am afraid01:02
n4nd0yeah but substitute the missing pieces there for what I am telling you01:02
n4nd0and \Delta is the Hamming loss01:02
blackburnthat's clear01:03
blackburnwhat are s.t. statements?01:04
blackburncan we get rid of it?01:04
n4nd0just some very generic constraints01:04
n4nd0sonme of them are actually required01:04
n4nd0forget about those there in the paper01:05
n4nd0and the ones we really need are A?x <= b01:05
blackburnwhat is A and b?01:06
shogun-buildbotbuild #348 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/348  blamelist: iglesias <fernando.iglesiasg@gmail.com>, Sergey Lisitsyn <lisitsyn.s.o@gmail.com>01:06
n4nd0note that I use x and not w (x = [w auxiliary_variables slacks])01:06
n4nd0still here?01:06
blackburnhmm01:07
blackburnyes01:07
n4nd0they have several uses01:07
n4nd0one of them is particular to the cutting-plane algorithm I use01:08
n4nd0another one is for the hm-svm, to specify monotonicity in the emission functions learnt01:08
blackburnn4nd0: I got back to teo thesis01:09
n4nd0ok01:09
blackburnAppendix A.1 - isn't that what you need?01:09
n4nd0that is not telling me how to compute the risk in an hmm01:11
blackburnn4nd0: your loss is eq (A.2), subgradient is (A.8)01:11
blackburnrisk is loss value + regularizer value01:11
blackburnn4nd0: can you compute loss function value?01:12
n4nd0I think that the regularizer is not included in the risk01:13
blackburnyeah I tend to that too01:13
blackburnin that method it is I mean01:13
blackburngenerally it is included01:13
n4nd0at the beginning of the bmrm, in the definition they give, it is not included I think01:14
blackburngenerally I mean in ML at all :)01:14
blackburnI expect they regularize by L201:14
n4nd0ok :)01:14
n4nd0anyway they do not include it in Teo either01:15
n4nd0(A.2) for Teo is the loss (as you said) but it is also the risk01:15
blackburnrisk is sum of losses, yes01:15
n4nd0ok01:16
blackburnor average (depending on the implementation, I don't know exactly)01:16
blackburnbut it is not the issue right now01:16
n4nd0agree01:16
blackburnwith A.801:16
blackburnyou can compute subgradient01:16
n4nd0ok01:16
blackburnall you would need is to compute argmax in A.701:16
blackburnis it a hard task for hm-svm?01:16
n4nd0the subgradient of the loss01:16
n4nd0it is simply Viterbi01:17
n4nd0I already have that one there01:17
blackburnI am lame with markov models still but once you compute it you have everything you need01:17
n4nd0I think I start to see your point :)01:18
blackburnn4nd0: but you would need to run viterbi quite a few times, is it feasible?01:18
-!- zxtx [~zv@c-71-227-187-90.hsd1.wa.comcast.net] has quit [Ping timeout: 272 seconds]01:18
blackburn#of examples times to be exact01:18
n4nd0blackburn: it is the bottleneck in SO01:19
n4nd0the argmax01:19
blackburnn4nd0: btw delta is not included in subgradient here01:19
blackburnonly in argmax01:19
n4nd0yes01:20
n4nd0we could say too that it is included in the risk01:20
n4nd0not in the subgradient of rish though01:20
blackburnah right01:20
n4nd0look at this01:20
blackburnn4nd0: so do you get how to compute it now?01:20
n4nd0shogun/structure/MulticlassRiskFunction.h01:20
blackburnn4nd0: ha!01:21
n4nd0I think the doc Michal wrote there is precisely the important part we are discussing just now01:21
n4nd0and this deserves a huge LOL01:21
blackburnn4nd0: regularizer is included01:21
n4nd0mmm01:22
n4nd0blackburn: mmm where?01:22
blackburnwrong sorry01:22
blackburnn4nd0: everything is clear for me now01:23
blackburnomfg what does he do01:23
n4nd0there is something that doesn't fit I thin01:24
blackburnI have to change it now01:24
n4nd0think*01:24
blackburnwhat?01:24
blackburnhe compute feature vector and compute dot manually01:24
n4nd0\ell(y_i,y)01:24
n4nd0that should be the delta loss01:24
blackburnwhere?01:24
n4nd0doc in the header01:24
blackburnmulticlass risk/01:24
blackburn?01:24
n4nd0yeah01:25
blackburnso? it is implemented in line 9001:25
n4nd0ok01:25
n4nd0it was just a misleading in the notation01:26
n4nd0I am not saying is wrong01:26
n4nd0what do you want to put in there, DotFeatures?01:26
blackburnsure01:26
blackburnit is totally wrong to compute vectors here01:26
blackburnimagine sparse features with 1%  nnz01:27
blackburn:D01:27
blackburn100x speedup from nowhere01:27
n4nd0yes01:27
n4nd0maybe we should tell Michal instead01:27
n4nd0anyway, another thing01:27
n4nd0I don't know why the hell do we need a risk function for multiclass, another for hm-svm, another for ....01:28
blackburnhe merged his stuff already I think so no problem01:28
blackburnn4nd0: ehmm because it is simple for multiclass01:28
n4nd0but it is the same for all01:28
blackburnloop on line 82 is your argmax01:28
n4nd0just the argmax and the psi changes01:28
blackburnyes01:29
n4nd0I think this risk function should go in StructuredModel01:29
n4nd0and call the argmax and psi01:29
n4nd0do you see my point?01:30
n4nd0I mean it is like A.2 and A.801:30
blackburnyes that's misleading thing with our czech friend01:30
blackburnfriends :)01:30
blackburnthey tend to develop their own solutions01:30
n4nd0hehe01:31
n4nd0you have always to stop generalizing at some point...01:31
blackburnyes sure01:31
blackburnwait01:31
blackburnso you have everything you need already?01:31
blackburnboth psi and argmax?01:31
n4nd0what?01:32
blackburndo you have psi and argmax for hm-svm?01:32
n4nd0sure01:34
n4nd0you always need those01:34
blackburnso you can easily add risk function for that?01:34
n4nd0I think so01:34
n4nd0is what I told you01:34
n4nd0I would just add a RiskFunction in CStructuredModel01:34
blackburnbut what was the problem?01:34
n4nd0that I didn't know that the risk function was that...01:35
blackburnahh01:35
blackburnso it is all now easy, right?01:35
blackburnand we are happy? :)01:35
n4nd0very happy :)01:36
n4nd0I am actually suspicious01:36
blackburnwhy?01:36
n4nd0it seems far too easy right now01:36
blackburnjust try and compaer01:37
-!- n4nd0 [~nando@56.Red-81-33-212.dynamicIP.rima-tde.net] has quit [Quit: leaving]01:39
-!- zxtx [~zv@c-71-227-187-90.hsd1.wa.comcast.net] has joined #shogun02:37
CIA-39shogun: Sergey Lisitsyn master * rd16ae4b / (2 files): Made MulticlassRiskFunction use COFFIN operations and fixed reference count for data in dual lib qpbmsosvm - http://git.io/14c0SQ03:00
blackburnwiking: how can you write to twitter and sleep? :D03:01
shogun-buildbotbuild #54 of nightly_none is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/5403:02
-!- blackburn [~blackburn@109.226.80.43] has quit [Quit: Leaving.]03:22
shogun-buildbotbuild #62 of nightly_default is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/6203:24
shogun-buildbotbuild #50 of nightly_all is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/5003:45
-!- av3ngr [av3ngr@nat/redhat/x-ohnbmedkajymxfbl] has joined #shogun04:16
-!- av3ngr [av3ngr@nat/redhat/x-ohnbmedkajymxfbl] has quit [Client Quit]04:17
shogun-buildbotbuild #63 of nightly_default is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/6304:19
shogun-buildbotbuild #349 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/349  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>04:32
-!- uricamic [~uricamic@2001:718:2:1634:ad82:d30e:54f:2c11] has joined #shogun07:39
-!- zxtx [~zv@c-71-227-187-90.hsd1.wa.comcast.net] has quit [Ping timeout: 268 seconds]08:17
-!- gsomix [~gsomix@80.234.28.229] has joined #shogun08:22
-!- gsomix [~gsomix@80.234.28.229] has quit [Client Quit]08:23
uricamichi, does anyone know what shall I do, to let configuration script find the MOSEK?10:43
wikingyo11:11
wikingblackburn magic ;)11:11
wikingah u r not here :(11:11
wikingsonney2k: around?11:13
wikingsonney2k:  i need some advise11:13
wikinguricamic: here?11:16
uricamicwiking: yep11:16
wikinguricamic: n4nd0 had an idea yesterday to add the RiskFunction within the StructureModel class11:17
wikinguricamic: i.e. simply adding a new pure virtual function risk(...)11:17
uricamicwiking: yep, I think we had discussed that at the very beginning, so I am not against :)11:17
wikingi see11:17
wikingi'll check how that effects my code and i'll let you guys know whether i'm against it or supporting it :)11:17
uricamicthis divergence was done mainly, we had to submit some code and I was waiting for his part for too long, so I started to code it, to be able to continue, and we left this for later, which unfortunately never happened :)11:18
uricamicwiking: ok11:18
wikinghehe ok11:18
wikingi hope i can get back to this today11:18
uricamicI guess it is reasonable to have it together, because then user have to write just one derived class not many of them :)11:19
wikingindeed11:19
uricamicbut also it was better to have it separate because of simultaneous work11:19
uricamicbut I guess now we can merge it11:19
wikingmy problem is simply the mixture of latentmodel and structuremodel11:20
uricamicsince GSoC end is coming :)11:20
wikingthey share a lot of things together but not11:20
wikingso i'm still struggling with merging that too :))11:20
uricamic:)11:20
uricamicand maybe with merging also the support of user data for risk function will be simplified11:21
wikinguricamic: heheh that's for sure11:22
uricamici.e. we could than remove CRiskData class, and all user data needed could be defined in the StructuredModel11:22
wikinguricamic: i mean currently i hd the troulbe of having reference on some other functions11:22
uricamiceverything should be simplified then :D11:22
wikingsince i needed the reference on functions like argmax_{y} withing CRiskFunction11:22
wikingmmm ok letme try to sketch up that thing now11:23
wikingas otherwise i'm wating on blackburn to give me some advise11:23
uricamicok11:25
wikinguricamic: ok i think i'm done with it12:00
wiking:))12:00
wikingthe only thing is that i don't know what where to add the the MulticlassRiskFunction implementation12:02
wikinguricamic: i've sent you and n4nd0 an email with the patch and some comments....12:09
-!- zxtx [~zv@c-24-18-130-24.hsd1.wa.comcast.net] has joined #shogun12:32
uricamicwiking: ok, thanks, I will look on it12:58
uricamicwiking: yep, risk function is needed whenever you want to use BMRM solver13:01
-!- gsomix [~gsomix@80.234.28.229] has joined #shogun13:48
gsomixgood day13:48
wikinggsomix: where's blackburn? :)13:55
gsomixwiking, I dunno, lol. ;)13:55
-!- blackburn1 [~blackburn@109.226.80.43] has joined #shogun13:55
gsomixBLACKBURN!13:55
blackburn1hey there wiking uricamic13:57
uricamicblackburn1: hi13:58
wikingblackburn1: !!!14:00
blackburn1uricamic: we discussed things with n4nd0 last night14:00
uricamicyep, I have looked on the log :)14:01
blackburn1uricamic: I am curious why we do not include regularized?14:01
blackburn1zer*14:01
uricamicblackburn1: what do u mean?14:01
uricamicit is included in bmrm algorithm14:01
blackburn1uricamic: L2?14:01
uricamicyep14:01
blackburn1so L1 is not possible?14:01
uricamicI guess not in our implementation14:02
blackburn1I see14:02
blackburn1we need to merge things you all have done somehow :D14:02
uricamicthe problem with L1 is that it would not be that easily solvable14:02
blackburn1uricamic: are you ok with changes in multiclass risk?14:03
uricamicI think this was the general intention, we have discuss it with n4ndo when GSoC started, and then somehow forgotten about this idea to merge it together14:03
uricamicblackburn1: sure14:03
blackburn1that's really important to use dense_dot and add14:04
uricamicas I said, it was just fast not very clever implementation just to prove, that it works :)14:04
blackburn1I see14:04
uricamicand I thing the way how wiking wants to merge it together is nice14:05
uricamicand I guess also not very demanding or time consuming14:05
uricamicCRiskData will disappear, because it will not be needed anymore14:05
blackburn1right14:05
uricamicand in StructuredModel virtual function risk will appear14:05
wiking:)14:05
blackburn1there are way too many entities here :D14:05
uricamicyep, as I said, it was general intention to put it together eventually :)14:06
wikingyeah a patch has been proposed14:06
wikingnow i'm waiting n4nd0 and uricamic's comments14:06
uricamicbecause the risk function in fact uses argmax and loss function, which is already in StructuredModel14:06
uricamicwiking: I am ok with it like it is, just don't forget to add implementation of risk function in MulticlassModel14:07
wikingblackburn1: let me know when u have 5 mins14:07
wikinguricamic: okey that shouldn't be hard ;P14:07
uricamicnope, in fact it will be even easier now than it was :D14:08
wikinguricamic: i'll wait for n4nd0's feedback as well and then i'll fire a PR :P14:08
uricamicno need of using RiskData14:08
uricamicwiking: ok, looking forward :)14:08
blackburn1wiking: I will be back in 10 minutes14:08
blackburn1and then lets discuss14:08
wikingblackburn1: ok14:08
blackburn1wiking: re14:19
wikingblackburn1: rere14:20
wikingblackburn1: so it's regarding the latentlabels fix14:20
wikingso now i'm in a catch-22 state with the design14:20
blackburn1damn I should read that book, I stopped on page 10 or so :D14:21
wikingbasically what a latent machine has is LatentFeatures, LatentLabels and Labels14:21
wikingthe thing is that currently LatentLabels is derived from binarylabels thus it's 2 in one14:21
wikingthis is good but till the moment when u want to have structural labels14:21
wikingor any other types of labels, i.e. y_i14:22
wikingi have two choices a) i separate h_i (latentlabels) and y_i14:22
wikingor b) i add a CLabels object within LatentLabels to contain the y_i14:23
wikingsolution a) is cleaner i think but there comes the problem of standard API of CMachine14:23
blackburn1no, I like b) more if I understand it correctly14:23
wikingas it has CLabels* apply_<somekind> (CFeatures*)14:23
wikingwith that API i cannot return h_i and y_i14:24
wikingso i have to break the API of machine14:24
blackburn1I think latent labels should contain both h_i and y_i14:24
wikingif i go with solution b) then yes14:24
wikingthat way i won't break any API14:25
wikingbut somehow the API won't be so clear for the latentmachine14:25
wikingas the y_i will be 'hidden'14:25
blackburn1I think that's ok14:25
wikingi.e. it wont ever appear as a function argument for example when u create a latent machine14:25
blackburn1just add a class to contain h_i and y_i14:25
wikingblackburn1: ok then i'll just change the inheritance of CLatentLabels to CLabels14:26
blackburn1yeah14:26
wikingand basically add a CLabels within CLatentLabels to store the y_i14:26
blackburn1any separating would be worse here14:26
blackburn1hmm that makes sense too14:26
wikingok14:27
wikingthen i'll just stash now the structuredModel +risk refactoring until n4nd0 gives his comments...14:27
wikingor maybe i just send in a PR about that14:27
wikingand n4nd0 can comment there14:27
wikingi have a feeling that'll be faster  ;P just as in case of unit testing :D14:28
blackburn1I thought of latent as structured model14:29
blackburn1can't you come this way?14:29
wikingwell14:29
wikingthat's the next step somehow14:29
blackburn1I see14:29
wikingbut still cannot see a nice way to do it14:29
wikingyet14:29
wikingok i'm away for 30mins to do so errands14:32
wikingblackburn1: thnx for the input14:32
blackburn1np14:33
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun14:51
yoohi all14:51
blackburn1hey15:02
-!- blackburn1 [~blackburn@109.226.80.43] has quit [Quit: Leaving.]15:09
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Quit: Page closed]15:21
-!- CIA-39 [~CIA@204.152.223.100] has quit [Ping timeout: 265 seconds]15:51
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun16:10
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Client Quit]16:10
uricamicwiking: I almost forgot one thing16:16
uricamicin the patch, u can also change risk function from void to return float64_t, i.e. it will return value of the risk16:17
uricamicblackburn suggested this, and it is reasonable16:18
-!- CIA-21 [~CIA@204.152.223.100] has joined #shogun16:25
-!- heiko [~heiko@host86-176-126-57.range86-176.btcentralplus.com] has joined #shogun16:49
heikoblackburn16:49
heikooh not here16:49
CIA-21shogun: Heiko Strathmann master * r854ec32 / (2 files): minor updates - http://git.io/GtSCTg17:59
CIA-21shogun: Heiko Strathmann master * rf4c19cb / src/shogun/labels/BinaryLabels.cpp : bugfix - http://git.io/UK0PQQ17:59
CIA-21shogun: Heiko Strathmann master * rd75793a / (2 files): added example/test for fitting sigmoid on binary labels - http://git.io/CqWVLA17:59
CIA-21shogun: Heiko Strathmann master * r419f224 / (4 files in 2 dirs): Merge pull request #714 from karlnapf/master - http://git.io/WWKpjA17:59
CIA-21shogun: Heiko Strathmann master * r47bbe6a / (2 files): added john platt's method for computung calibrated posterior outputs for - http://git.io/l15Iiw17:59
-!- heiko [~heiko@host86-176-126-57.range86-176.btcentralplus.com] has quit [Quit: Leaving.]18:01
shogun-buildbotbuild #350 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/350  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>18:21
-!- blackburn [~blackburn@109.226.80.43] has joined #shogun18:35
-!- uricamic [~uricamic@2001:718:2:1634:ad82:d30e:54f:2c11] has quit [Quit: Leaving.]18:36
shogun-buildbotbuild #351 of deb3 - modular_interfaces is complete: Failure [failed compile java_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/351  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>18:46
@sonney2kwiking, could you please fix the compile error?19:27
@sonney2khey gsomix - how is 0-copy stuff proceeding?19:28
@sonney2kwiking, why do your latentlabels derive from denselabels?19:28
@sonney2kand not just labels?19:28
@sonney2kat least this is the reason for the compile error19:28
blackburnsonney2k: yeah he knows :)19:30
blackburnsonney2k: our SO guys need to make final effort to merge risk and models19:31
@sonney2kwiking, I am renaming your get_labels to get_latent_labels19:31
@sonney2kthen this is resolved19:32
@sonney2kwiking - hope that is what you had in mind19:32
gsomixsonney2k, good19:34
@sonney2kgsomix, so that means PR any minute?19:36
gsomixsonney2k, nope19:39
@sonney2kgsomix, what does it mean then?19:42
gsomixthat I figured out how to do it rightly. and now doing19:43
@sonney2kok19:44
CIA-21shogun: Soeren Sonnenburg master * r973c3d7 / (2 files): rename conflicting get_labels function to get_latent_labels in CLatentLabels - http://git.io/rFhFmw19:44
-!- gsomix_ [~gsomix@109.169.234.221] has joined #shogun19:56
-!- gsomix [~gsomix@80.234.28.229] has quit [Read error: Connection reset by peer]19:57
shogun-buildbotbuild #352 of deb3 - modular_interfaces is complete: Failure [failed compile csharp_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/352  blamelist: Soeren Sonnenburg <sonne@debian.org>20:07
blackburnhah20:22
blackburnthe two sgvectors problem20:23
blackburnmy favourite one20:23
CIA-21shogun: Sergey Lisitsyn master * rbbcbb01 / src/interfaces/modular/modshogun_ignores.i : Updated ignore statement to fix java building - http://git.io/QsvlLA21:07
-!- gsomix_ [~gsomix@109.169.234.221] has quit [Ping timeout: 244 seconds]21:25
shogun-buildbotbuild #353 of deb3 - modular_interfaces is complete: Failure [failed compile csharp_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/353  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>21:27
-!- gsomix_ [~gsomix@88.200.214.210] has joined #shogun21:38
CIA-21shogun: Sergey Lisitsyn master * rc6a588e / (4 files in 3 dirs): Attempt to fix csharp compilation - http://git.io/AI-xFg22:03
shogun-buildbotbuild #354 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/354  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>22:47
blackburnsonney2k: around?22:53
-!- n4nd0 [~nando@111.red-80-28-182.adsl.static.ccgg.telefonica.net] has joined #shogun23:01
n4nd0wiking: hey, I am checking your mail about the Risk Function right now (I've been out for the day)23:02
-!- n4nd0 [~nando@111.red-80-28-182.adsl.static.ccgg.telefonica.net] has quit [Ping timeout: 252 seconds]23:29
-!- gsomix_ [~gsomix@88.200.214.210] has quit [Ping timeout: 256 seconds]23:33
-!- n4nd0 [~nando@98.Red-2-137-39.dynamicIP.rima-tde.net] has joined #shogun23:37
CIA-21shogun: Sergey Lisitsyn master * r97768b9 / (11 files in 6 dirs): Fixed some regression static tester issues - http://git.io/Y8IVPA23:42
-!- n4nd0 [~nando@98.Red-2-137-39.dynamicIP.rima-tde.net] has quit [Ping timeout: 248 seconds]23:46
-!- n4nd0 [~nando@98.Red-2-137-39.dynamicIP.rima-tde.net] has joined #shogun23:47
-!- n4nd0 [~nando@98.Red-2-137-39.dynamicIP.rima-tde.net] has quit [Ping timeout: 240 seconds]23:51
-!- n4nd0 [~nando@98.Red-2-137-39.dynamicIP.rima-tde.net] has joined #shogun23:52
-!- n4nd0 [~nando@98.Red-2-137-39.dynamicIP.rima-tde.net] has quit [Client Quit]23:55
--- Log closed Wed Aug 15 00:00:17 2012