Open in new window / Try shogun cloud
--- Log opened Fri May 18 00:00:40 2012
-!- n4nd0 [] has joined #shogun00:08
n4nd0sonney2k, blackburn : I have seen in the logs you have been discussing about the labels hierarchy00:29
n4nd0so what is the conclusion?00:29
n4nd0will we use a base class for labels?00:29
blackburnheh no idea lets see what is he will come with00:30
n4nd0all right00:30
n4nd0I'll keep an eye on github00:30
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]00:48
-!- n4nd0 [] has quit [Quit: leaving]01:31
-!- av3ngr [av3ngr@nat/redhat/x-scprkyesygrtlzzt] has joined #shogun01:32
-!- av3ngr [av3ngr@nat/redhat/x-scprkyesygrtlzzt] has quit [Read error: Connection reset by peer]01:46
-!- av3ngr [av3ngr@nat/redhat/x-gckdfbduswynkgua] has joined #shogun01:46
-!- av3ngr [av3ngr@nat/redhat/x-gckdfbduswynkgua] has quit [Read error: Connection reset by peer]01:50
-!- abn_ [av3ngr@nat/redhat/x-siwxyjkjituyottb] has joined #shogun01:51
-!- av3ngr [av3ngr@nat/redhat/x-wgnhnksekfhutuhg] has joined #shogun02:16
-!- abn_ [av3ngr@nat/redhat/x-siwxyjkjituyottb] has quit [Ping timeout: 256 seconds]02:20
-!- abn_ [av3ngr@nat/redhat/x-edpkkfhhuuvrgzaj] has joined #shogun02:22
-!- av3ngr [av3ngr@nat/redhat/x-wgnhnksekfhutuhg] has quit [Ping timeout: 244 seconds]02:25
-!- Francis_Chan [~Adium@] has joined #shogun02:40
-!- Francis_Chan [~Adium@] has quit [Ping timeout: 256 seconds]03:31
-!- emrecelikten [~Anubis@] has quit [Quit: Leaving.]03:50
-!- Francis_Chan [~Adium@] has joined #shogun03:56
-!- abn__ [av3ngr@nat/redhat/x-ghucyoueatuuaupd] has joined #shogun04:05
-!- abn_ [av3ngr@nat/redhat/x-edpkkfhhuuvrgzaj] has quit [Ping timeout: 244 seconds]04:08
-!- abn__ [av3ngr@nat/redhat/x-ghucyoueatuuaupd] has quit [Ping timeout: 245 seconds]04:59
-!- wiking_ [] has joined #shogun05:34
-!- wiking_ [] has quit [Changing host]05:34
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun05:34
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 272 seconds]05:36
-!- wiking_ is now known as wiking05:36
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 245 seconds]05:46
-!- emrecelikten [~emrecelik@] has joined #shogun07:04
-!- emrecelikten [~emrecelik@] has quit [Client Quit]07:05
-!- vikram360 [~vikram360@] has quit [Ping timeout: 272 seconds]07:57
-!- vikram360 [~vikram360@] has joined #shogun08:21
-!- uricamic [~uricamic@2001:718:2:1634:1dd1:edc6:9e87:3324] has joined #shogun09:07
-!- blackburn [~blackburn@] has joined #shogun09:21
CIA-113shogun: iglesias master * r2e96c0f / src/shogun/lib/SGMatrix.h : * minor indent fix in SGMatrix -
-!- n4nd0 [] has joined #shogun09:31
n4nd0hey blackburn09:36
n4nd0good morning09:36
blackburnhow is it going?09:37
n4nd0fine, it's starting to be nice weather regularly around here :)09:37
n4nd0that makes me happy09:37
n4nd0what about you?09:37
blackburndid you receive welcome package yet?09:37
blackburnheh I have been happy with weather since late april I think09:38
blackburnthere was virtually no spring at all09:38
n4nd0wow, you're luckier in Russia in that sense that in Sweden09:38
blackburnjust baam and summer09:38
n4nd0I didn't receive anything yet09:38
blackburnno, but some guys did already according to gsoc ml09:39
n4nd0neither at home nor on the mail, I guess the ACM comes by mail09:39
n4nd0yes, it looks like that09:39
n4nd0they said we shouldn't worry until May 25th right?09:39
n4nd0can you believe that I have not gotten the grades for one of my exams yet?09:39
n4nd0I did it on April man, 28th I think09:40
shogun-buildbotbuild #871 of cmdline_static is complete: Failure [failed test_1]  Build details are at  blamelist: fernando.iglesiasg@gmail.com09:40
blackburnhahh that takes a while to rate your work damn09:40
n4nd0haha no way09:40
blackburnwhy do they do it so sloow?09:40
n4nd0this teacher is not very good09:40
n4nd0she just appeared the 50% of the lectures09:41
n4nd0and half an hour late, on average, the times she did it09:41
n4nd0anyhow, I think tomorrow it's the last day to publish so they should be ready by then09:41
blackburnhah I had had one phd student who 'teached' us calculus09:42
blackburnjust like your teacher - always late09:42
n4nd0too bad09:43
blackburnwe were so angry with him09:43
blackburnso we played kind of tennis09:43
blackburnright on lesson09:44
n4nd0haha, in the class?09:44
blackburnyes with books and paper09:44
shogun-buildbotbuild #850 of r_static is complete: Failure [failed test_1]  Build details are at  blamelist: fernando.iglesiasg@gmail.com09:44
blackburnand only after two minutes of play he told us (me and one other guy) to leave the room :D09:45
n4nd0shogun-buildbot: what did I do to you?09:45
n4nd0blackburn: do you where I can find an example of use of libqp in shogun?09:48
blackburnn4nd0: it is used in libocas09:49
blackburnonly I guess09:49
n4nd0blackburn: yeah, it seems that it's used there, thanks ;)09:50
-!- Francis_Chan [~Adium@] has left #shogun []10:12
-!- aphex [2e1fd566@gateway/web/freenode/ip.] has joined #shogun10:43
aphexhi all10:43
aphexI have a question about string kernel implemented in shogun10:44
aphexI would like to use CLocalAlignmentStringKernel on stringfeatures<char> with alphabet RAWBYTE. Is that compatible ,10:45
blackburnlet me check10:46
blackburnaphex: ok looks like it is suited for amino-acids so no idea whether it would work for rawbyte10:49
aphexThat's what I ve seen too.11:07
aphexDo you know what string kernel could be used for RAWBYTE alphabet ?11:07
blackburniirc WeightedDegreeStringKernel should work on rawbyte11:09
aphexblackburn: tx. Is there an example with this kernel ?11:11
blackburnfor which interface?11:11
aphexblackburn: note: I used string features of different sizes11:15
blackburnwell check libshogun examples11:16
blackburnall kernels are pretty similar11:16
aphexblackburn: ok thx11:17
-!- aphex [2e1fd566@gateway/web/freenode/ip.] has quit [Quit: Page closed]11:26
-!- n4nd0 [] has quit [Read error: Operation timed out]12:16
-!- pluskid [~pluskid@] has joined #shogun14:35
-!- eric_ [2e1fd566@gateway/web/freenode/ip.] has joined #shogun14:36
eric_hi again14:36
eric_blackburn: There is definitly somthg wrong when using Xvalidation with multiclass svm other than "CMulticlassLibLinear"14:37
eric_blackburn: CMulticlassLibSVM or CGMNPSVM with same linear kernel config give always XVal: [0.100000,0.100000] with alpha=0.050000, mean=0.10000014:38
-!- n4nd0 [] has joined #shogun14:53
n4nd0hey pluskid, how is it going?15:44
pluskidn4nd0: hi15:45
pluskidI was busy preparing docs for a VISA interview the last two days15:45
pluskidalmost done with ecoclib15:46
pluskidwill move to some papers Cheng mentioned to me15:46
pluskidhow's it going with SO?15:46
n4nd0I hope you visa interview goes good15:46
n4nd0it's going good, I find it fun15:46
pluskidthanks! it will be on May 2915:46
n4nd0you are going to US, right?15:46
pluskidn4nd0: cool15:47
pluskidyes, I'm going to the US15:47
pluskidin mid August15:47
n4nd0I would really love to go there to do my phd15:47
pluskidhave you decided where to get a PhD yet?15:48
n4nd0I expect to finish during June 2013, in about a year15:48
pluskidhmm, I see15:48
n4nd0when do you think I should start getting in touch with people there and prepare it seriously?15:49
pluskidso you will apply schools this Autumn15:49
pluskidn4nd0: I suggest you start to collect information about schools/profs. in the summer holiday15:49
pluskidthat would be a *huge* survey :D15:50
pluskidto get to know who and where you would be interested in15:50
n4nd0for the moment I know that I would like to go to a sunny state15:50
n4nd0I think I would really enjoy living in California15:50
pluskidAh, California is awesome, everybody wants to go there :D15:51
pluskidand there are many good universities15:51
pluskidwith strong machine learning groups15:51
pluskidn4nd0: what do you want to do for your PhD?15:52
pluskidI mean, what subarea of ml?15:52
n4nd0pluskid: I am interested in robotics, in particular in localization, planning, SLAM and the like15:54
pluskidhmm, interesting topic15:54
pluskidAndrew Ng has a project of a robot dog, which looks pretty cool15:55
pluskidvery robust on walking in various difficulty conditions15:55
n4nd0pluskid: btw, I was going to ask you about the labels hierarchy you suggested in a PR15:56
n4nd0did you get some feedback from sonney2k on that?15:56
-!- uricamic [~uricamic@2001:718:2:1634:1dd1:edc6:9e87:3324] has quit [Quit: Leaving.]16:02
pluskidin a PR?16:02
pluskidwhat is PR?16:03
pluskidanyway, the CLabels hierarchy I suggested was rejected by sonney2k16:03
n4nd0pull request16:03
n4nd0we'll have to wait and see how he wants to have it then :)16:04
pluskidcurrently MC works fine with the current labels16:04
pluskidso that hierarchy might makes thing complicated16:04
pluskidnot necessary16:04
pluskidso currently we will not refactor the labels system for MC16:04
pluskidbut I'm afraid something is needed for SO16:05
pluskidI think you know the best what is needed in SO16:05
pluskidmaybe we might consider refactoring after your SO labels are created16:05
n4nd0I have already created them16:07
pluskidoh, cool16:08
pluskidwhere are they?16:08
n4nd0but they are kind of out of hierarchy16:08
n4nd0take a look to my branch so in github16:08
n4nd0or in the pull request here,
pluskidwe might be able to incorporate them into the hierarchy16:08
n4nd0for the idea I have in mind what would be best is to have them under CLabels, something like CLabels <--- CSOLabels16:09
n4nd0but the current CLabels should change16:10
n4nd0since I am not interested to inherit in the SGVector< float64_t > for non-SO labels for example16:10
pluskidI didn't find it, what's the filename for your SOLabels?16:11
-!- wiking [] has joined #shogun16:14
-!- wiking [] has quit [Changing host]16:14
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun16:14
-!- emrecelikten [~emrecelik@] has joined #shogun16:16
n4nd0but there is almost nothing there yet, quite basic so far16:16
n4nd0I am waiting for the decision of the hierarchy16:16
pluskidyeah, it's difficulty to work with if CStructuredLabels is not inherited from CLabels16:18
pluskidbecause CSOMachine is interited from CMachine, right?16:18
pluskidso you are actually accepting CLabels for set_labels() method16:18
pluskidbut what you are expecting should be CStructuredLabels16:19
pluskidthe refactoring is non-trivial, since the implementations are different (SGVector vs. CDynamicObjectArray)16:19
pluskidI think you'd better discuss with sonney2k first16:20
n4nd0yesterday blackburn and sonney2k were discussing about it16:22
n4nd0if it made sense to have a base class CLabels16:22
pluskidso do they have any conclusion?16:22
pluskida base CLabels that is very empty?16:22
n4nd0that's is what I support16:23
n4nd0but there is no conclusion yet16:23
pluskidthe problem is this will need a lot of work :p16:25
pluskidto modify the existing code to perform explicit casting16:25
n4nd0yes, labels are used everywhere16:26
-!- eric_ [2e1fd566@gateway/web/freenode/ip.] has quit [Quit: Page closed]16:45
-!- pluskid [~pluskid@] has quit [Quit: Leaving]17:02
blackburnno not a lot of work17:03
blackburnpretty easy to do I think17:03
-!- emrecelikten [~emrecelik@] has quit [Quit: Leaving.]18:13
-!- n4nd0 [] has quit [Quit: leaving]18:17
-!- vikram360 [~vikram360@] has quit [Ping timeout: 256 seconds]18:23
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun18:28
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:45
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun19:17
puffin444Do you know when karlnapf is typically on irc? He wanted to chat with me about the Model Selection framework.19:20
blackburnpuffin444: he tends to appear rather randomly last days - he has exams - so no idea19:23
puffin444Oh okay.19:23
blackburnbut IIRC I have seen him there late (for me) evenings19:24
puffin444He had some ideas about extending the framework to suit hyperparameter learning.19:24
blackburnme? :D19:25
puffin444What timezone are you in?19:25
blackburnutc+4 I guess19:25
puffin444sry typo :)19:25
puffin444hmm.... so early afternoon for me.19:28
puffin444We have another GSoC IRC meeting the 23rd, yes?19:28
blackburnlet me check in logs19:30
blackburnpuffin444: 25th19:32
blackburn11 UTC19:32
blackburnpuffin444: did that presentation of you already take place?19:36
puffin444Oh not yet. Not until next week. I am still waiting to hear back for more information.19:38
blackburnI see19:38
puffin444I think I am going to go ahead and start making a presentation anyway because time is running out.19:38
CIA-113shogun: Soeren Sonnenburg master * r88f1152 / (3 files in 3 dirs): remove vec_index -
blackburnanother one bloody commit from soeren19:41
blackburnyeah he cuts off everything :D19:48
puffin444I see :)19:53
-!- vikram360 [~vikram360@] has joined #shogun20:03
-!- shogun-buildbot [] has quit [Quit: buildmaster reconfigured: bot disconnecting]20:19
-!- shogun-buildbot [] has joined #shogun20:19
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Quit: Page closed]20:25
@sonney2kblackburn, the solution we had yesterday doesn't work20:50
blackburnsonney2k: which one?20:50
@sonney2kblackburn, removal of CMachine apply()20:51
@sonney2kand have customized apply() stuff in sub-machines20:51
@sonney2kblackburn, because multiclass machine wants to call applyu20:51
@sonney2kfrom general machines...20:51
blackburnsonney2k: we can assume it is a binary machine20:52
@sonney2kbut machine has no apply -> dead20:52
@sonney2kwe don't have 'binary machines'20:52
blackburnwe should have binary machine20:52
@sonney2kwe discussed that yesterday kernelmachines20:52
@sonney2kare binary,regression, ...20:52
blackburnthen I see no way20:52
blackburnwell wait20:53
@sonney2kI think we should have apply() return RealLabels20:53
@sonney2kit won't work for SO though20:53
blackburnsonney2k: there is a way20:54
blackburnpretty easy20:54
blackburnsonney2k: for domain adaptation multiclass svm I extracted outputs of sub-machines to separate functions20:54
blackburnthey are virtual and could be overrided in linear and kernel multiclass machine20:54
blackburnwith proper casting20:55
blackburngeneral  multiclass is abstract at all so no problem there20:55
blackburnit could have no impl at all20:56
blackburnsonney2k: is there any other issue?20:56
@sonney2kblackburn, I have no idea what you are saying20:58
blackburnsonney2k: from the very beginning?20:58
blackburnsonney2k: all casts before apply in multiclass machine20:59
blackburncan go to linear and kernel multiclass machine20:59
blackburnwith casts to linear and kernel machines20:59
-!- n4nd0 [] has joined #shogun20:59
blackburnso if we add apply in both kernel and linear machine20:59
blackburnyour issue is solved20:59
blackburnsonney2k: got?20:59
@sonney2kI don't know what you are talking about21:01
@sonney2kthe only solution I see now is:21:01
@sonney2k- keep apply() in CMachine21:01
blackburnsonney2k: again..21:01
@sonney2k- check the label type of the returned label21:01
blackburncan we move apply to21:01
blackburnand linear21:01
@sonney2kand cast appropriately21:01
blackburnwhat is the issue you want to solve??21:02
@sonney2kbecause some machines work with any other machine that have an apply() method21:02
-!- vikram360 [~vikram360@] has quit [Ping timeout: 246 seconds]21:03
blackburnsonney2k: where you want to put apply?21:03
@sonney2kin CMachine if possible21:04
blackburnif not?21:04
@sonney2kin submachines21:04
blackburnwhich ones?21:04
@sonney2klinear, kernel, etc21:05
@sonney2kbut it is possible21:05
@sonney2konly problem is that interfaces that cannot cast just get some Label object21:05
blackburnwhy did you say we can't put apply to linearmachine?21:05
@sonney2kwhich sucks...21:05
@sonney2kblackburn, we have to have it in CMachine21:06
@sonney2kbecause e.g. multiclass machine calls apply() on general machines21:06
blackburnsonney2k: I said two times we can use apply on linear/kernel machines in multiclass machines21:07
blackburnsonney2k: put apply part to {linear,kernel}multiclassmachine21:07
@sonney2knot nice either21:08
blackburnsolves your issue21:08
@sonney2kbut you would need to determine the machine type in mc machine21:09
@sonney2kthen cast to linear/kernel machine and call apply there21:09
blackburnin linearmulticlassmachine all machines are linearmachines21:09
blackburnimplemented virtual method of {linear,kernel}multiclassmachine does the job of casting and applying21:10
blackburnnow we anyway cast it to CMachine21:10
blackburnanyway I need to made that apply part virtual21:11
blackburnbecause DA MC SVM needs to combine outputs of target and source domain submachines21:11
blackburnI will commit da mc svm soon21:12
@sonney2kbut I would try to get apply() to return just the baseclass21:12
blackburnbase class?21:12
blackburndidn't you want to remove base class there?21:12
@sonney2kwell I thought I had no other option21:12
blackburnis there a way now?21:13
@sonney2kbut I think we have an option21:13
@sonney2kyeah we always just return CLabel* on apply()21:13
blackburnis there any other issue except that multiclass?21:13
@sonney2kI mean if we construct a MulticlassLabel or so it will be cast to CLabel and can be cast back21:14
@sonney2kso no problem21:14
@sonney2kand it will work with StructuredOutputLabel too21:14
@sonney2kso CMachine has a nice interface21:14
blackburnyes I am ok with it too21:14
@sonney2kI think for swig one can hack sth up to return the real label type21:15
@sonney2kby using %ignore and %rename21:15
@sonney2kand potential helper functions21:15
@sonney2kthat do the casting to CRealLabel etc21:15
shogun-buildbotbuild #233 of nightly_none is complete: Success [build successful]  Build details are at
blackburn2140 warnings wat?!21:17
blackburndynamicarray warning for each type21:17
@sonney2kI think gsomix removed some warning ignore21:19
@sonney2kI think I will change everything back to CMachine -> CLabels* apply21:19
blackburnhah it seems you do some great changes in your .git21:20
n4nd0sonney2k, blackburn: hi guys21:20
blackburnn4nd0: hey21:20
n4nd0I don't think I really understand the plans you have for CLabels21:21
blackburndo you think you are the only one?21:21
n4nd0can you tell me about it more thoroughly? you know I am quite interested in it21:21
blackburnhmm I need some workforce21:22
shogun-buildbotbuild #241 of nightly_all is complete: Success [build successful]  Build details are at
CIA-113shogun: Sergey Lisitsyn master * r33fc1e4 / (10 files in 5 dirs): Added DomainAdaptationMulticlassLibLinear -
blackburnn4nd0: okay actually we need clabels or not21:56
blackburnI am not sure we really need it but it looks better with it21:56
shogun-buildbotbuild #153 of nightly_default is complete: Success [build successful]  Build details are at
-!- gsomix [~gsomix@] has joined #shogun22:31
gsomixhi all22:31
gsomixI'm alive.22:31
blackburnI sent you a cool pic22:32
gsomixI had a hard week. tomorrow I'm going with the news.22:32
gsomixblackburn, yeah22:32
blackburnwhere have you been?22:32
blackburnoptics trip? :D22:33
gsomixblackburn, in botan-land.22:33
gsomixsorry for that22:34
gsomixblackburn, about warnings. I'll remove it.22:34
shogun-buildbotbuild #873 of cmdline_static is complete: Success [build successful]  Build details are at
blackburnyes that could be cool to get rid of them22:35
gsomixaha, and about directors. I have read docs. tomorrow I'm going to experiment.22:36
gsomixgood night22:40
gsomixsonney2k, (^___^)/22:40
-!- gsomix [~gsomix@] has quit [Quit: Ex-Chat]22:43
shogun-buildbotbuild #852 of r_static is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #762 of octave_static is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #839 of python_static is complete: Success [build successful]  Build details are at
--- Log closed Sat May 19 00:00:40 2012