Open in new window / Try shogun cloud
--- Log opened Mon Jun 18 00:00:41 2012
CIA-18shogun: Sergey Lisitsyn master * r058cc08 / src/shogun/lib/slep/slep_mt_lsr.cpp : Few fixes for SLEPs MT LS -
CIA-18shogun: Sergey Lisitsyn master * rd48a682 / examples/undocumented/python_modular/ : Added multitask group regression example -
CIA-18shogun: Sergey Lisitsyn master * rc3bfbbe / (2 files in 2 dirs): Merge branch 'slep' of git:// -
-!- Netsplit *.net <-> *.split quits: emrecelikten00:16
-!- Netsplit over, joins: emrecelikten00:24
blackburnsonney2k: ping00:38
shogun-buildbot_build #609 of lua_modular is complete: Success [build successful]  Build details are at
shogun-buildbot_build #634 of octave_modular is complete: Success [build successful]  Build details are at
CIA-18shogun: Sergey Lisitsyn master * r70d6259 / (32 files in 4 dirs): Moved normalizers to special directory in kernels -
shogun-buildbot_build #610 of lua_modular is complete: Failure [failed test_1]  Build details are at  blamelist: blackburn91@gmail.com02:12
blackburnwrong window hehe02:13
shogun-buildbot_build #635 of octave_modular is complete: Failure [failed compile]  Build details are at  blamelist: blackburn91@gmail.com02:18
CIA-18shogun: Sergey Lisitsyn master * r1b86df6 / (2 files in 2 dirs): Enforced parallel to use one thread in case of directors -
shogun-buildbot_build #610 of java_modular is complete: Failure [failed test_1]  Build details are at  blamelist: blackburn91@gmail.com02:26
CIA-18shogun: Sergey Lisitsyn master * r9a1b7ae / src/interfaces/octave_modular/swig_typemaps.i : Fixed octave compilation -
-!- av3ngr [av3ngr@nat/redhat/x-zivjfdcozmwbpgsi] has joined #shogun02:43
-!- av3ngr [av3ngr@nat/redhat/x-zivjfdcozmwbpgsi] has quit [Read error: Connection reset by peer]02:43
-!- blackburn [~blackburn@] has quit [Ping timeout: 244 seconds]02:52
shogun-buildbot_build #1065 of libshogun is complete: Failure [failed compile]  Build details are at  blamelist: blackburn91@gmail.com02:52
-!- emrecelikten [~emre@] has quit [Quit: Leaving.]06:35
-!- n4nd0 [] has joined #shogun07:04
-!- av3ngr [av3ngr@nat/redhat/x-zsvwajsquoqdnpzz] has joined #shogun07:55
-!- uricamic [~uricamic@2001:718:2:1634:7c47:92c4:30cb:d3e3] has joined #shogun08:49
-!- av3ngr [av3ngr@nat/redhat/x-zsvwajsquoqdnpzz] has quit [Quit: That's all folks!]10:00
-!- wiking [] has joined #shogun12:26
-!- wiking [] has quit [Changing host]12:26
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun12:26
-!- blackburn [~blackburn@] has joined #shogun12:45
-!- pluskid [~pluskid@] has joined #shogun12:48
blackburnwiking: pluskid: uricamic: n4nd0: one puzzle for you - will results change if we use fourier transformed features in SVM?12:56
pluskiddon't know12:57
uricamicwhat results? :)12:57
blackburnso the answer is no - parceval theorem12:58
blackburndot product in original space and dot product in fourier space are equal12:58
blackburnI've been so stupid I forgot about it and tried to classify in fourier domain12:58
blackburnIIRC some other integral tranforms meet parceval too13:02
n4nd0blackburn: I didn't know that it also preserves the result of the dot product, good to learn new stuff :)13:12
-!- n4nd0 [] has quit [Ping timeout: 276 seconds]13:55
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]14:00
-!- hoijui [] has joined #shogun14:32
-!- hoijui [] has quit [Client Quit]14:33
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]15:00
-!- blackburn [~blackburn@] has joined #shogun15:04
-!- nicococo [] has joined #shogun15:08
-!- wiking [] has joined #shogun15:44
-!- wiking [] has quit [Changing host]15:44
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun15:44
-!- n4nd0 [] has joined #shogun15:47
n4nd0nicococo: hey!15:48
n4nd0how is it going?15:48
nicococohi n4nd0.. its hot and i don't want to work anymore... :)15:49
nicococoi wrote you an email a couple of minutes ago..15:49
wikingit's weekly update day15:51
n4nd0nicococo: yeah, thanks, you made me woke up finally :)15:51
n4nd0the cell told me about your mail15:52
n4nd0so as I told you in the mail I did a couple of modifications and the results of multiclass classification improved a bit15:53
nicococodid you party all night?15:53
n4nd0they're *more or less* comparable to one-vs-rest15:53
n4nd0nicococo: not really, I just feel like taking a nap sometimes :P15:53
nicococo*more or less* seems reasonable enough to me15:54
n4nd0but I would like to do one more test about this to feel confident15:54
n4nd0since the results for this example change a lot with one-vs-rest using bias or not15:55
n4nd0I want to try training the so-svm with bias15:55
blackburnwiking: yes, is there anything to report? ;)15:55
nicococookay, there is enough time to do more tests.. especially if we want to test the coffin thingy with the mc-example15:55
n4nd0nicococo: so do you have to say something in special about doing it with bias?15:56
n4nd0nicococo: I thought of it and I think that it would be basically to add one more element to w15:56
n4nd0nicococo: and pad all the joint feature vectors psi with a 1 at the end15:57
nicococoyou add a one dimension for all phi_i(xi)15:57
n4nd0not psi?15:58
nicococoif you have 4 classes the joint feature vector is growing 4 dimensions, right?15:58
n4nd0the joint feature vector would be 4*dimension of each feature vector15:58
n4nd0ok, I think I see what you mean15:59
n4nd0I was thinking that it would be something that doesn't depend on the application at hand when it does depend15:59
n4nd0so the weight vector w is going to #classes more elements with a biased hyperplane16:00
nicococoso basically, there is a problem when using bias terms, if i understand correctly?16:02
n4nd0what problem?16:02
nicocococlassification performance problems?16:02
n4nd0without bias the results tend not to be very good when the number of examples and number of classes grow16:03
n4nd0but I think that's logical16:04
nicococothen lets try the same with coffin, okay..16:04
n4nd0about coffin, I read the paper again16:05
n4nd0but I have not thought how to introduce it to the framework16:05
n4nd0I have been busy with the debugging16:06
nicococowell, it is already implemented in shogun (afaik) and we just have to16:06
n4nd0I don't know if you have taken a look to what I changed in the code16:06
nicococo..replace  vector multiplications16:06
nicococowhich file?16:07
nicococoi think you just have to replace compute_loss_arg in order to use coffin16:09
n4nd0and probably in MulticlassModel, where the joint feature vector is computed16:10
n4nd0since there it is done explicitily16:10
nicococothe method compute_loss_arg should be moved into the model anyway..16:11
n4nd0I could do that16:12
n4nd0but the thing about coffin requires more work than that I think16:12
wikingblackburn: what happened with you :)16:12
n4nd0do you have clear what needs to be done?16:12
blackburnwiking: what do you mea?16:13
wikingblackburn: what happened with the soccer team of russia?!16:14
blackburnwiking: they are stupid bastards :D16:14
nicococon4nd0: probably.. there must be some setup/init routine but i don't have a clue but i have soerens telefon number and can make use of it ;)16:14
n4nd0nicococo: it could be cool if you can do that16:15
nicococoi try and will tell you tomorrow.16:16
blackburnwiking: our captain prefer selling Lay's but not to train I think :)16:17
-!- nicococo [] has left #shogun []16:17
wikingblackburn: arshavin? :)16:17
wikingwell i really thought they were good16:17
wikingbut seems they are still immature :(16:17
blackburnwell if we won we would have to beat germany16:18
blackburnno way16:18
wikingyeah i was already looking forward16:18
wikingfor seeing you and sonney2k here while russia and germany is playing soccer :)16:19
blackburnI think sonney2k doesn't care at all16:19
blackburnI wish portugal to win16:21
wikingtoday croatia vs spain :)16:21
wikingi wonder who'll win that one16:21
wikinggotta grab a coffee16:21
blackburnspain is pretty hmm strong16:22
wiking but i'm for croatia :)16:23
wikingcoffee time for real16:23
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]16:23
-!- wiking [] has joined #shogun16:45
-!- wiking [] has quit [Changing host]16:45
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun16:45
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit]16:48
-!- pluskid [~pluskid@] has quit [Quit: Leaving]17:00
-!- emrecelikten [~emre@] has joined #shogun17:01
-!- n4nd0 [] has quit [Quit: leaving]17:22
-!- uricamic [~uricamic@2001:718:2:1634:7c47:92c4:30cb:d3e3] has quit [Quit: Leaving.]18:05
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has joined #shogun18:34
blackburnpuffin444: hey18:38
blackburnare you using anything from NLOPT?18:38
puffin444blackburn, hey18:38
puffin444I hope to use the mma algorithm18:38
blackburnI think it is worth to add a nlopt wrapper here18:38
puffin444One problem is that a lot of the local gradient opt algorithms are fortran translated into c18:38
puffin444A nlopt wrapper?18:38
blackburnwhy do you care about its code?18:39
puffin444So we just interface with the code?18:39
puffin444Oh I was scared I would have to re-implement the whole algorithm and use nlopt as a benchmark for accuracy18:40
blackburnshould be better I believe?18:40
CIA-18shogun: Sergey Lisitsyn master * r4ce3514 / (2 files in 2 dirs): Added CSignal handling to MulticlassLibLinear -
blackburnI am currently adding it to configure so soon we will be able to use it inside18:40
puffin444I wasn't sure if actually just including nlopt with a wrapper was something the shogun community wanted18:41
blackburnhmm we should remove any reason to think so :)18:42
puffin444but just interfacing should make things vastly easier.18:42
blackburnI didn't even know there is a problem with that18:44
blackburnwe integrated some code like liblinear not because we want to have one million LoC18:45
blackburnbut because it wasn't applicable18:45
blackburnpuffin444: okay in your further code just guard your code with ifdef HAVE_NLOPT18:49
blackburnI will commit soon18:49
blackburnconfigure with nlopt detection18:49
puffin444blackburn, will do.18:49
blackburnpuffin444: here you go feel free to use it ;)18:52
CIA-18shogun: Sergey Lisitsyn master * rccfeb3a / src/configure : Added NLOPT detection -
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has joined #shogun19:30
KMcQuistenHey there folks19:30
KMcQuistenSo when the headers for the normalizers were moved, it broke some references in some SVMs and some kernels that had built-in normalizers.  I've branched off of the master and fixed them.  How should I go about putting in a request to have them pulled19:31
@sonney2kblackburn, can we have the normalizers in kernels/normalizers please?19:32
blackburnsonney2k: normalizers?19:32
blackburnwe have kernel not kernels19:32
blackburnso normalize19:32
blackburnnot normalizer19:32
@sonney2knormalizer then19:32
blackburnthat's the logic19:32
blackburnnormalizer is ok19:33
blackburnKMcQuisten: just push to your github and press pull request button19:33
KMcQuistenOk, the changes I ahve still have all the headers in the /normalize folder.  Do you want me to change it to /normalizer before I push?19:34
blackburnKMcQuisten: no let me do it after19:34
@sonney2kblackburn, btw good to see you being active again19:34
@sonney2kbtw what is nlopt?19:35
blackburnsonney2k: nonlinear optimization library19:35
blackburnit has funny method named COBYLA :D19:35
blackburnno idea whether your russian will remind you something19:36
blackburnif not - it stands for female horse19:36
blackburnsonney2k: both puffin444 and pluskid need it19:37
blackburnsonney2k: and unfortunately I'll become less active again for ~ two days19:37
blackburnor even dead as you said :)19:38
@sonney2kblackburn, no idea about the quality of the package ... no release for > 1 year19:38
blackburnsonney2k: oliver recommended it at some point if I understood that correctly19:39
blackburnthere are only a few gpl optimization libraries so not bad to have dependency on it19:40
blackburnhmm ceres!19:41
KMcQuistenblackburn:  I'm going to need a little guidance through this.19:42
blackburnKMcQuisten: sure19:42
KMcQuistenShould I merge my branch before I try to push?19:43
blackburnKMcQuisten: push to?19:44
KMcQuistenblackburn:  I'm not sure exactly.  I branched off of the master and made some fixes.  What's the process for getting those changes back to the repository?19:45
blackburnso you have the fork on github right?19:45
KMcQuistenI don't believe so, or at most I do not know.  I've kept everything local so far.19:45
blackburnaham so you did git clone19:46
blackburndid you commit your changes?19:47
KMcQuistento my local branch off master, yes19:47
blackburnso what you may do now is19:47
blackburnformat patch and send to me19:48
blackburnor fork, do these changes in your fork and do a pull request19:48
KMcQuistenOk, so i've made my fork on github.  Pull from there, then make the changes and push, then do a pull request?19:50
KMcQuistenSounds like a plan, thanks.19:52
blackburnlatter way is preferrable good you choose it :D19:52
@sonney2kblackburn, btw is puffin444's patch merged?19:54
@sonney2kblackburn, will n4nd0 be here this evening?19:54
blackburnsonney2k: no idea19:54
@sonney2kwish we could merge hsi19:54
@sonney2kblackburn, any news about gsomix? did the tornado do more harm there?19:55
blackburnsonney2k: no more harm, but he prepares for his exams I think19:56
@sonney2khmmhh then I hope he sends some status update19:57
blackburnsonney2k: yes I just reached him via jabber19:57
blackburnsonney2k: btw yesterday I empirically proved parceval theorem with director kernel19:59
blackburnhowever stupidity is not good anyway20:00
puffin444Are there tornadoes in Russia?20:00
blackburnsome kind of20:01
blackburnbut not like tornado rather hurricane I think20:01
@sonney2kblackburn, what's the difference?20:02
blackburnsonney2k: between?20:02
@sonney2kblackburn, tornado and hurricane20:03
blackburnno things like
blackburnbut very strong weather?20:03
@sonney2kblackburn, btw I saw that you changed sparse vectors / matri20:03
blackburnsonney2k: only matrix20:03
@sonney2kis it working now with references?20:03
puffin444looks like a tornado to me20:03
@sonney2kand what is a hurriance then20:04
puffin444A really mean tropical storm20:04
blackburnI wrote weather? :D20:04
blackburnI meant wind20:04
blackburnsonney2k: I'd say hurricane is strong rain and strong wind20:04
blackburnbut tornado is tornado and we have it very rarely20:05
blackburnin usa they emerge way more often20:05
@sonney2kso tornado is 'just' wind?20:05
blackburntornado is
puffin444A tornado is sort of like an intense air whirpool20:05
blackburnI've been waiting for this word20:06
puffin444They pop all the time in Illinois, but somehow I've never seen one in person.20:06
blackburnpuffin444: are you from IL originally?20:06
puffin444Yes on the very edge of the Chicago area20:07
@sonney2kI think I've seen some small ones in australia/namibia then20:07
@sonney2kwalked even through them20:07
blackburnsonney2k: have you been to namibia?20:07
puffin444I think those are dust devils, little mini tornados. A full tornado pulls chunks of houses into the sky sometimes.20:09
@sonney2kdust devils - yeah nice word20:09
@sonney2kblackburn, we need an application in shogun/applications called like that20:10
blackburndust devil?20:10
blackburnwhat should it do? break houses?20:10
KMcQuistenI live in Iowa, and we usually have a handfull of towns either damaged or destroeyed every spring and summer.20:10
@sonney2ksome windy thing20:10
KMcQuistenby tornadoes.  Dust devils are pretty harmless20:10
blackburneeeyeee lives in iowaaa20:11
KMcQuistenHad one go by me about 200m from my car once20:11
blackburnno 12 oz mouse fans here :(20:11
KMcQuistenHa!  Not so much.  I watch a bunch of other adult swim shows though20:13
KMcQuistenA few years back a town on a river got leveled by a tornado, and then the next week the river flooded and washed the debris away.  The whole town jsut vanished20:13
KMcQuistenOk, when I'm trying to push my changes to my fork, i'm getting a 403 error20:16
blackburnso you do 'git pull origin?'20:17
blackburnah I know20:19
blackburna min20:19
blackburngit remote update origin git://<YOUR_USER_NAME>/shogun20:20
blackburnsonney2k: actually I thought about normalizer name and normalize made more sense for me20:24
blackburnKMcQuisten: did that help?20:29
blackburnsonney2k: ping20:36
blackburnI want to check if PyObject is SGObject20:37
blackburncan it be done?20:37
@sonney2kblackburn, yes20:39
@sonney2kblackburn, from C++?20:39
@sonney2ksome instanceof thing20:39
blackburnsonney2k: I want to add [] -> CList typemap20:39
blackburndo you mind?20:40
@sonney2kblackburn, btw why do you use CList and not any of the DynamicObjectArrays etc20:40
blackburnI do not need indexing here :)20:40
blackburnand I do not like dynamic object arrays yes20:40
@sonney2kblackburn, hah I don't like lists :)20:41
blackburnsonney2k: do you remember how to get underlying instance of PyObject?20:42
blackburnI can't find the definition of PyObject :(20:42
KMcQuistenfatal: No such remote or remote group: git://
blackburnKMcQuisten: git remote set-url origin git://<YOUR_USER_NAME>/shogun20:44
blackburnand this way?20:44
KMcQuistenthat executes succesfully with no output20:45
blackburnand git push origin?20:45
KMcQuistenfatal: remote error:    You can't push to git://   Use
blackburnokay I think you know what to do now20:46
KMcQuistenPermission denied (publickey). fatal: The remote end hung up unexpectedly20:47
KMcQuisteni'm assuming there's some authorization setting I need to set?20:48
blackburnyes I think so20:48
blackburn this way20:49
blackburnKMcQuisten: did that help?21:13
KMcQuistenI'm in the process of finding out21:14
-!- puffin444 [62e3926e@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]21:16
KMcQuistenLooks like that worked.21:24
KMcQuistenblackburn:  everything went smooth as silk this time around.  Thanks for your help.21:43
blackburnKMcQuisten: so lets merge21:43
@sonney2kblackburn, are you working on getting things back to stable
KMcQuistenPutting in a pull request now21:43
blackburnsonney2k: sometimes21:43
KMcQuistenrequest sent21:45
blackburnsonney2k: I think KMcQuisten patch is all about that error21:45
CIA-18shogun: Sergey Lisitsyn master * raa9ad6a / (15 files in 3 dirs): Merge pull request #587 from kmcquisten/master -
CIA-18shogun: Kyle master * r0159ce6 / (15 files in 3 dirs): Fixed broken references to kernel normalizer headers, which have been moved -
KMcQuistenYes, I think my pull request fixed all that21:46
blackburnlets see21:47
shogun-buildbot_build #1069 of libshogun is complete: Success [build successful]  Build details are at
KMcQuistenGlad I could finally contribute directly21:58
blackburnyou are very welcome for further contributions21:59
blackburnKMcQuisten: it is actually very cool you are using shogun for some string stuff22:00
blackburnnobody from current dev team does any strings related thing..22:01
KMcQuistenYes indeed.  Now that kernels are working I can actually try a few ideas I've had bouncing around in my head for the last couple years22:01
KMcQuistenI see.  Well i'm glad that I can at least keep banging on that part of the code to see what might break :)22:02
blackburnKMcQuisten: did you manage to use some kind of external features btw?22:03
KMcQuistengsomix made the get_lhs and get_rhs returns polymorphic, so I don't have to bring in external features anymore.  Before I'd just been precomputing kernel matrices.22:04
blackburnI see22:04
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has quit [Quit: Page closed]22:12
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]22:24
-!- blackburn [~blackburn@] has joined #shogun22:28
-!- wiking [] has joined #shogun22:43
-!- wiking [] has quit [Changing host]22:43
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun22:43
@sonney2kwiking, please don't forget your weekly report22:53
wikingsonney2k: yep just writing it+sending22:53
@sonney2kwiking, ok22:53
@sonney2kwiking, btw - are you preparing any pull request?22:54
wikingi've seen your comment22:54
wikingthe only thing is (that you'll see now as well form the weekly report)22:54
wikingthat currently i'm running 2 dev branches22:54
@sonney2kwhy that?22:54
wikingsince i've started to try to merge SO22:55
wikingfrom fernando22:55
wikingbut that's still really early bird stage22:55
-!- gsomix [~gsomix@] has joined #shogun23:03
gsomixhi all23:03
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]23:13
shogun-buildbot_build #611 of java_modular is complete: Success [build successful]  Build details are at
shogun-buildbot_build #603 of ruby_modular is complete: Failure [failed test_1]  Build details are at  blamelist: kmcquisten@gmail.com23:39
-!- wiking [] has joined #shogun23:43
-!- wiking [] has quit [Changing host]23:43
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun23:43
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]23:50
-!- n4nd0 [] has joined #shogun23:53
--- Log closed Tue Jun 19 00:00:41 2012