Open in new window / Try shogun cloud
--- Log opened Sun Apr 28 00:00:29 2013
NPHey!00:01
-!- NP [~NP@61.12.19.99] has left #shogun []00:03
-!- mikhailBelous [~towelenee@109.205.249.161] has quit []00:07
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds]00:36
-!- HeikoS [~heiko@90.219.232.242] has quit [Quit: Leaving.]01:20
-!- dvalcarce [5b757556@gateway/web/freenode/ip.91.117.117.86] has quit [Quit: Page closed]01:21
shogun-buildbotbuild #373 of nightly_default is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/37304:07
-!- kaushalb09 [~kaushalb0@cpe-74-77-131-130.buffalo.res.rr.com] has joined #shogun04:32
kaushalb09hi,my name is kaushal.I have a doubt regarding issue # 86304:36
kaushalb09Add sparse inverse covariance graphical example04:36
kaushalb09can i generate my own data or can i use the toy data provided04:37
kaushalb09hi , i am a GSoC applicant. is anyone there to help05:32
-!- kaushalb09 [~kaushalb0@cpe-74-77-131-130.buffalo.res.rr.com] has quit [Ping timeout: 246 seconds]07:01
-!- kaushalb09 [~kaushalb0@cpe-74-77-131-130.buffalo.res.rr.com] has joined #shogun07:15
-!- sids_aquarius [~sidi_@li400-124.members.linode.com] has joined #shogun08:37
-!- gsomix [~Miranda@83.149.21.112] has joined #shogun08:58
@wikingsonney2k: try to turn on python_modular interface on fbsd bot... although at the current state of shogun it will fail for sure :(09:42
-!- gsomix [~Miranda@83.149.21.112] has quit [Ping timeout: 240 seconds]09:45
-!- kaushalb09 [~kaushalb0@cpe-74-77-131-130.buffalo.res.rr.com] has quit [Ping timeout: 272 seconds]09:52
-!- deerishi [73f88294@gateway/web/freenode/ip.115.248.130.148] has joined #shogun11:03
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun11:08
-!- vivekn [318aa50a@gateway/web/freenode/ip.49.138.165.10] has joined #shogun11:16
-!- sijin [~smuxi@144.214.222.109] has joined #shogun11:26
vgorbatilisitsyn: hi, there?11:48
@lisitsynvgorbati: yes11:48
vgorbatilisitsyn: about t-SNE interface task, do I remember the steps correctly: 1. Add include in modular/Converter_includes.i . 2. Add rename in modular/Converter.i. 3. Add some graphical example in examples/undocumented/python_modular/graphical?11:50
@lisitsynvgorbati: better not graphical but simple in examples/undocumented/python_modular first11:51
-!- gsomix [~Miranda@83.149.21.112] has joined #shogun11:52
viveknHi, I am interested in implementing the Hashing Trick for GSoC, as I have used it for reducing memory usage with n-grams in one of my previous projects on sentiment classification. What are the specific features the shogun project is expecting for this?11:53
vgorbatilisitsyn: ok. Something like converter_factoranalysis_modular.py (with FactorAnalysis replaced by t-SNE class) should work, shouldn't it?11:54
@lisitsynvgorbati: exactly like that11:54
viveknHow can I start contributing and are there any entrance tasks pertaining to this idea?11:57
@lisitsynvivekn: no there is no particular task for this idea so feel free to contribute with anything you like11:59
deerishi@lisitsyn :i am having some problem with the integration test12:02
deerishihttps://github.com/deerishi/KernelPCA-test-check/commit/743d3d9bc373aa36b0ea4bb7065ed36ba8706a9312:03
deerishicould you check it?12:03
@lisitsyndeerishi: let me check12:03
@lisitsyndeerishi: looks like exactly like heiko said12:04
@lisitsynthe name of the function should be the same with the filename12:04
deerishiwhich function are we talking about?12:04
deerishidef generator?12:05
deerishior def get_fname?12:05
@lisitsyndeerishi: the function in the preprocessor_kpca.py12:05
-!- vivekn [318aa50a@gateway/web/freenode/ip.49.138.165.10] has quit [Quit: Page closed]12:07
deerishii renames the function in preprocessor_kpca as the same, but it still gives the same error12:10
deerishi@lisitsyn any comments?12:11
@lisitsyndeerishi: I haven't seen the file12:12
deerishii had sent a pull request https://github.com/shogun-toolbox/shogun/pull/101412:14
@lisitsyndeerishi: the function and the file have different names here12:16
deerishithis is the old pr.i tried renaming the function in the file preprocessor_kpca just now12:17
@lisitsynthen it should work please carefully check they are the same12:18
deerishione more thing,should i add the name preprocessor_kpca.py to the backlist of the file generator.py12:20
deerishi?12:20
@lisitsyndeerishi: only if it fails :)12:22
deerishii added that,which removed the error,but the 2 text files being created are broken12:25
-!- deerishi_ [73f88294@gateway/web/freenode/ip.115.248.130.148] has joined #shogun12:28
-!- deerishi [73f88294@gateway/web/freenode/ip.115.248.130.148] has quit [Ping timeout: 245 seconds]12:32
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds]12:38
-!- iglesiasg [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun12:39
-!- mode/#shogun [+o iglesiasg] by ChanServ12:40
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun13:00
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds]13:14
deerishi_@lisitsyn : i have tried renaming the the function as the filename,but it still is giving an error.only if i write the functioname in the backlist,i do not get any error.but then the text files that are created are broken13:15
deerishi_what could be the problem?13:18
deerishi_https://github.com/shogun-toolbox/shogun/pull/102413:19
-!- gsomix [~Miranda@83.149.21.112] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org]13:38
deerishi_?14:19
deerishi_?14:19
@iglesiasgdeerishi_: what is the error you get?14:33
deerishi_@iglesiasg : https://github.com/deerishi/KernelPCA-test-check/commit/743d3d9bc373aa36b0ea4bb7065ed36ba8706a9314:34
@iglesiasgdeerishi_: have you tried tracking the line where this fails?14:37
deerishi_yes,14:37
deerishi_in the generator.py14:37
@iglesiasgHeiko said that ir worked in his machine after renaming thoug14:37
@iglesiasgthough*14:37
deerishi_i tried that too.14:37
deerishi_i submited a pr after changing the function name14:38
@sonney2kshogun-buildbot, force build 'cyg1 - libshogun'15:16
shogun-buildbotbuild #713 forced15:16
shogun-buildbotI'll give a shout when the build finishes15:16
-!- gsomix [~Miranda@37.61.181.11] has joined #shogun15:22
gsomixhi15:22
-!- sids_aquarius [~sidi_@li400-124.members.linode.com] has quit [Remote host closed the connection]15:27
shogun-buildbotbuild #713 of cyg1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/71315:27
-!- alibezz [~alibezz@177.182.164.8] has joined #shogun15:28
alibezzhey yall15:29
@iglesiasghi alibezz15:29
alibezzI'm interested in working as an intern at GSoC 201315:29
alibezzjust sent my introduction to the mail list yesterday (Aline Bessa)15:29
@iglesiasgalibezz: yeah, I read your mails :) welcome around here15:29
alibezziglesiasg: thanks!15:30
-!- sids_aquarius [~sidi_@14.139.122.114] has joined #shogun15:30
@iglesiasgalibezz: I am also applying for a GSoC project15:30
alibezziglesiasg: Is it necessary to solve a bug to be able to participate?15:30
@iglesiasgalibezz: yes. You should submit a patch; that can be a bug fix, a new feature,...15:31
alibezziglesiasg: ok, any ideas about how I can find a good yet simple one?15:31
@iglesiasgalibezz: there are some tasks in github issues15:31
@iglesiasgalibezz: see for the ones labelled as entrance15:32
alibezziglesiasg: thanks! i have a paper deadline for Thursday, but will do my best to accomplish that. =)15:32
@iglesiasgalibezz: cool, also you are free to submit patches after the application deadline15:32
@iglesiasgIMHO, the more contributions you have, the stronger applicant you are15:33
alibezziglesiasg: chill, thanks for the advice. but it is important to have something **before** the application deadline, right? it's a pity I just became aware of GSoC late this year.15:33
@iglesiasgalibezz: mmm I don't think it is specially important to have something **before**15:34
alibezziglesiasg: I'll try to figure it out. ;) thanks, iglesiasg!15:34
@iglesiasgit is important to have it before the mentors make their decision15:34
@iglesiasgalibezz: you are welcome :)15:35
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun15:40
alibezziglesiasg: I'm gonna work on this entrance here: https://github.com/shogun-toolbox/shogun/issues/948 :)16:10
alibezz...or this one https://github.com/shogun-toolbox/shogun/issues/87316:12
alibezzbut actually, if I could find an entrance related to the dimred project http://shogun-toolbox.org/page/Events/gsoc2013_ideas#dimred, that'd be better16:13
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds]16:21
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has joined #shogun16:36
-!- kaushalb09 [~kaushalb0@cpe-74-77-131-130.buffalo.res.rr.com] has joined #shogun16:52
kaushalb09Hi ,I am a GSoC applicant. I have few doubts regarding issue# 863. Is anyone there16:56
-!- gsomix [~Miranda@37.61.181.11] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org]16:58
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has quit [Ping timeout: 245 seconds]17:40
-!- gsomix [~Miranda@185.6.239.174] has joined #shogun17:47
gsomixsonney2k: ping17:47
gsomixthere is no "pong" for me...17:57
@iglesiasgalibezz: sure, just pick the one you like the most17:58
alibezziglesiasg: are you aware of any that is specifically related to dimensionality reduction?17:58
@iglesiasgalibezz: mmm not really17:58
alibezziglesiasg: k ;)17:59
@iglesiasgalibezz: I know some DR methods that we are missing and the matlab DR toolbox has17:59
@iglesiasgsuch as maximum variance unfolding IIRC17:59
alibezziglesiasg: maybe that could be a starting point17:59
alibezziglesiasg: like, integrating these.17:59
@iglesiasgalibezz: in my opinion it could be indeed17:59
@iglesiasgalibezz: lisitsyn (mentor of the project) may be around and give his opinion about this18:00
alibezziglesiasg: k, i'm currently finishing my formal application. once i'm done, can i ping you to discuss it in further detail?18:00
alibezziglesiasg: oh cool, i'll ping him18:00
alibezziglesiasg: which project are you applying to?18:01
alibezziglesiasg: just out of curiosity18:01
-!- deerishi_ [73f88294@gateway/web/freenode/ip.115.248.130.148] has quit [Quit: Page closed]18:03
-!- deerishi [73f88294@gateway/web/freenode/ip.115.248.130.148] has joined #shogun18:03
@iglesiasgalibezz: structured learning18:03
alibezziglesiasg: chill. were you at NIPS last year?18:04
@lisitsynre18:19
@lisitsynalibezz: answering your mail: the sooner the better I think18:21
alibezzlisitsyn: you mean the patch?18:22
@lisitsynyes18:22
@lisitsynwe will have a pressure to decide rather fast after the deadline18:22
-!- mikhailBelous [~towelenee@109.205.249.161] has joined #shogun18:22
@lisitsynalibezz: just fyi you currently have one competitor for the dimension reduction idea :)18:23
alibezzchill. iglesiasg suggested me to try integrating some DM matlab algorithms to the shogun framework. do you think this is simple enough giving the time constraints? I'd like to send something until Thursday or Friday. Thing is: I have a conference deadline as well, so my time will be rather small.18:23
mikhailBelousHello, I would like to know, do you need some students who intrested in machine learning and have a huge c++ optimization expirience. Is some mentor for this task now online?18:24
alibezzlisitsyn: I'm also interested in the mldata project, given that I have some background with accessibility18:24
@lisitsynmikhailBelous: huge is like?18:24
alibezzlisitsyn: but seriously, if you could suggest me a task that you think that can be accomplished until Friday?18:25
alibezzlisitsyn: no "if you" haha just "could..."18:25
@lisitsynalibezz: I'll come with something and let you know18:25
alibezzlisitsyn: thank you VERY much! i am really interested in this project!18:26
deerishi@lisitsyn :the integration test passed :)18:27
@lisitsyndeerishi: nice18:27
-!- gsomix [~Miranda@185.6.239.174] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org]18:28
@lisitsynmikhailBelous: gsoc is mainly about searching long-term contributors - that's why we are pushing students to contribute18:29
deerishiturns out there was a  broken file in the folder18:29
-!- sids_aquarius_ [~sidi_@li400-124.members.linode.com] has joined #shogun18:31
-!- sids_aquarius [~sidi_@14.139.122.114] has quit [Read error: Connection reset by peer]18:31
@lisitsynmikhailBelous: so if you have such optimization skills just optimize something in shogun ;)18:31
alibezzlisitsyn: is there anything in the Issues entry at github related to the DimRed project?18:35
alibezzlisitsyn: I was thinking of starting today ;)18:36
@lisitsynalibezz: I am thinking if some of them are still left18:36
alibezzlisitsyn: yay #figerscrossed18:36
@iglesiasgalibezz: nah, no NIPS nor any other conference/journal for me yet :(18:37
@iglesiasgalibezz: what about you?18:37
alibezzalibezz: I've been in NIPS once. but not because of a paper. haha I submitted a project to Women in Machine Learning and it got approved. It was co-located with NIPS so I went to both18:38
alibezziglesiasg: DUH i wrote alibezz instead of your name. haha18:38
@iglesiasgalibezz: nice!18:38
@iglesiasgyeah18:38
@iglesiasglisitsyn: you see? I am not the only one around here talking to himself now :D18:39
@lisitsyniglesiasg: haha18:39
@lisitsynnice18:39
alibezzlolz18:39
@lisitsynalibezz: okay what about sammon mapping?18:39
@lisitsynI can describe it quite detailed18:39
alibezzlisitsyn: woot go for it please ;)18:40
@lisitsynalibezz: matlab code is available18:40
alibezzlisitsyn: where's this issue? the sammon mapping one? on github?18:40
@lisitsynalibezz: writing it18:40
@iglesiasgalibezz: just out of curiosity, what's about the paper you are preparing for this upcoming conference?18:44
alibezziglesiasg: according to the field of behavioral economics, everytime one looks at a list of items and is supposed to choose one of them, it does not depend only on the individual properties of each one of the items, but also on the universe of presented alternatives. so i model this mutual influence items exert on each other as pairwise factors, and individual qualities as individual factors. one can see it as a markov network, and then perform MAP inferenc18:48
@iglesiasgalibezz: it sounds interesting18:49
alibezziglesiasg: thx!18:49
@iglesiasgalibezz: do you have to perform parameter learning for the MRF too or are they given by some expert knowledge?18:50
alibezziglesiasg: actually, all factors are either spit out from a individual-item recommender learner, such as SVD++, and all pairwise factors are computed in a rather naive way so far, via maximum likelihood. I didn't go too far with the MRF anyways because it was too costly. I just show some experiments that illustrate that and show how the bayesian approximation is way faster, and also rather good, in practice.18:52
-!- sids_aquarius_ is now known as sids_aquarius18:53
-!- gsomix [~Miranda@188.168.3.32] has joined #shogun18:53
@iglesiasgalibezz: I see18:53
@lisitsynalibezz: https://github.com/shogun-toolbox/shogun/issues/102518:56
alibezzlisitsyn: chill, thx. i'll check it out asap.18:56
@lisitsynoops forgot one step18:56
@iglesiasglisitsyn: oh that's a nice description18:57
@iglesiasgdirectly to the game :)18:57
@lisitsyniglesiasg: I start to understand nobody give a damn about my code :D18:58
@iglesiasglisitsyn: c'mon, I don't think so18:59
@lisitsynalibezz: I tried to make it detailed as you are out of time19:03
@lisitsynbut ask if something is not clear19:03
@lisitsynalibezz: iglesiasg should be able to answer some questions too I think19:03
@lisitsynlala you are out touch I am out of time lala19:03
@lisitsyniglesiasg: know that song?19:04
@lisitsynfrom the vice city :D19:04
alibezzlisitsyn: thank you so much for being so understanding! in South America, this time of the year is always full of things... we don't actually don't have summer vacation in school here at this point of time because... uh... it's not summer! haha19:04
@iglesiasglisitsyn:  mmm I don't think I have heard it19:04
alibezzlisitsyn: iglesiasg: you guys are fun hahaha chill19:04
@lisitsyniglesiasg: hall and oates19:04
alibezziglesiasg: lisitsyn: yuppppp gotta dance like crazy with these guys.19:05
@lisitsynhttp://www.youtube.com/watch?feature=player_detailpage&v=s_8KR-n2fBQ iglesiasg19:05
@iglesiasglisitsyn: dafuq19:05
@lisitsyniglesiasg: I actually do like it more than some of ehm modern music19:06
@iglesiasgheh19:07
alibezziglesiasg: lisitsyn: the description of the issue is mad detailed yall. thanks!!!!19:09
alibezziglesiasg: lisitsyn: if you guys like music, you might enjoy this one. just guessing... hahaha http://www.youtube.com/watch?feature=player_embedded&v=T5Cp55MvX5419:11
@lisitsynget lucky is nice yes19:11
alibezziglesiasg: lisitsyn: one has to try ;)19:11
alibezzlisitsyn: but this is a chill cover for rainy days or so. ha19:11
@lisitsynyeah listening19:11
@iglesiasglisitsyn: maybe you can ask alibezz about music in Portuguese if you are still into it :)19:12
@lisitsynhaha yes19:12
@lisitsynnot sure what should I ask though19:12
alibezzlisitsyn: hell yea. I'm a big music nerd actually. hahaha and i love the music of my country, especially non-cliche shit19:12
@iglesiasgalibezz: the song is nice idd19:12
@lisitsynalibezz: I am a big fan of bossa like antonio carlos jobim19:13
@lisitsynseu jorge, etc19:13
alibezzlisitsyn: hey do you know Astrud Gilberto?19:13
@lisitsynof course19:13
-!- sids_aquarius [~sidi_@li400-124.members.linode.com] has left #shogun ["Off to code | sleep"]19:14
@lisitsyncaetano veloso19:14
@lisitsynto name a few19:14
@lisitsyn:D19:14
@lisitsynalibezz: this thing rocks http://www.youtube.com/watch?feature=player_detailpage&v=Q7xZ3cBPxIE19:14
alibezztim maia? erasmo carlos?19:14
@lisitsynI have seen 'tim maia' name somewhere in brazil drum'n'bass collections19:15
@lisitsynno idea about erasmo carlos though19:15
alibezzgonna think of something good to show ya. bout to eat something now. it's lunch time here. lemme know where you guys are from!19:20
deerishi@lisitsyn : how do go about merging?now that the integration test has passed19:20
@lisitsyndeerishi: let Heiko merge it19:20
@lisitsynalibezz: russia19:20
@iglesiasgalibezz: Spain, hidden in Sweden19:21
-!- kaushalb09 [~kaushalb0@cpe-74-77-131-130.buffalo.res.rr.com] has quit [Ping timeout: 260 seconds]19:23
-!- chintak [73f88294@gateway/web/freenode/ip.115.248.130.148] has joined #shogun19:23
-!- chintak [73f88294@gateway/web/freenode/ip.115.248.130.148] has quit [Client Quit]19:23
* gsomix missed all fun. Again. :(19:26
-!- kaushalb09 [~kaushalb0@cpe-74-77-131-130.buffalo.res.rr.com] has joined #shogun19:36
-!- HeikoS [~heiko@90.219.232.242] has joined #shogun19:36
-!- mode/#shogun [+o HeikoS] by ChanServ19:36
@lisitsynHeikoS: hey could you please take a look at deerishi PR?19:41
@HeikoSlisitsyn: yes, hwy?19:41
@HeikoSwhy19:41
deerishii 'll will send the revised pr in a few minutes19:41
@HeikoSdeerishi: ah hi19:41
deerishiintegration test passes19:41
@HeikoSI just sent you an email19:41
@HeikoSdid you manage to create a new test file?19:41
alibezzlisitsyn: iglesiasg: Could you please update the sammon mapping issue assigning it to me?19:42
@lisitsynalibezz: just write you will take care19:42
alibezzok!19:42
@lisitsynI can't assign it to arbitrary person19:42
@HeikoSdeerishi: so if the generator worked, try the tester.py with the same argument, if this passes, commit the test data and the data version in the development repository19:42
@lisitsynonly shogun-toolbox members are available there unfortunately19:42
deerishi@HeikoS : yes, 2 text files were created as you said19:42
@HeikoSdeerishi: nice, so add them to the commit19:43
@HeikoSand also add "data" in the main repo19:43
alibezzlisitsyn: no worries19:43
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has joined #shogun19:49
mikhailBelousCan you suggest any library that should be optimized?19:49
@iglesiasgmikhailBelous: what do you mean?19:50
mikhailBelousI wanna to do some contribution in optimization field, can you suggested something19:51
@iglesiasgwithin Shogun, or are you talking in general?19:52
mikhailBelousShogun19:52
@iglesiasgah ok so math optimization I guess you are talking about19:52
mikhailBelousOr worktime optimization19:53
@iglesiasgmikhailBelous: I implemented a method that you can find in structure/PrimalMosekSOSVM19:53
deerishi@HeikoS : i am having some problem in pushing the repo19:53
@iglesiasgmikhailBelous: which needs Mosek19:53
@iglesiasgmikhailBelous: it would be nice if we could drop this requirement by adding a QP solver with box contraints19:54
@HeikoSlisitsyn:  how does it work if someone who is not core dev wants to push to data?19:54
deerishii could not clone the repository,so i downloaded the  as zip folder and installed shogun19:54
@iglesiasgmikhailBelous: I can let you know more about the QP it needs to solve if you are interested19:54
deerishinow how do  i push the entire thing into a particular branch of the repository i had forked19:54
mikhailBelousIt will be nice19:54
@iglesiasgmikhailBelous: I wrote a bit about it here http://iglesiashogun.wordpress.com/2012/06/04/second-weekly-report-gsoc-2012/19:55
@lisitsynHeikoS: PR19:57
@lisitsynas usual19:57
@lisitsyn:)19:57
@HeikoSlisitsyn: so push to the forked data repo?19:57
@HeikoSand then PR the data?19:57
@lisitsynyes19:57
@iglesiasgmikhailBelous: in that post there is some application dependent information on structured learning19:57
@lisitsynthe data PR should be merged first HeikoS19:57
@HeikoSwell thats easy, deerishi, see above19:57
@iglesiasgmikhailBelous: but the idea is basically to get a solver like http://docs.mosek.com/6.0/capi/node007.html#25033079219:57
@HeikoSlisitsyn: thanks for the tip!19:57
@HeikoSindeed thats a good idea .)19:57
@iglesiasgmikhailBelous: however I have just used that as a black box, I have no idea how complex can it be to implement that. It might even a better idea to find some open source code able to do that and integrate it19:58
@lisitsynmikhailBelous: you may just learn any linear machine and benchmark how it computes the output on the test data20:01
@lisitsynand make it faster if you have any ideas20:01
@lisitsynwithout the loss of flexibility and crazy refactoring20:01
mikhailBelousIt looks simplier than qp realization20:02
@iglesiasgmikhailBelous: It may be a good idea to start off with something simpler ;)20:03
@lisitsynI wouldn't expect it is feasible to come up with something much better than libqp in a reasonable time20:04
deerishi@HeikoS: will do that ,and let you know soon20:04
mikhailBelousBut I think that QP can be solved by simple gradient method. But I may be wrong20:04
@iglesiasglisitsyn: the problem with libqp and this problem are the constraints20:06
@iglesiasgfor box constraints, it only handles an equlity contrainst20:06
@lisitsynall these libqp, mosek etc have been under development for a long long time20:06
@iglesiasgwhile this problem needs inequalities20:06
@iglesiasgIRRC20:06
@lisitsynso no that's not simple at all20:07
@iglesiasgindeed20:07
@HeikoSdeerishi: nice, Ill have a look tomorrow though since got some friends now visiting, see you!20:09
mikhailBelousWe have convex function on compact space, why it would be solved by gradient method?20:09
-!- HeikoS [~heiko@90.219.232.242] has quit [Quit: Leaving.]20:09
mikhailBelous*would not20:09
@lisitsynmikhailBelous: they are all gradient based but they are not simple at all20:11
mikhailBelousOkay, let me suggest this solution.20:11
@lisitsynit matters a lot how general the problem you solve is20:12
@lisitsynlibqp is quite general, liblinear optimizer is less general20:13
mikhailBelousDo you or do you not have a solution for this problem(math not code)?20:13
@lisitsynmikhailBelous: which problem?20:14
mikhailBelousQuadratic optimization with inequalities boundaries20:15
@lisitsynin libqp this method is implemented if you are interested: http://www.recognition.mccme.ru/pub/papers/SVM/keerthi00convergence.pdf20:15
@lisitsynI am not sure what the method the mosek stuff implements20:18
@iglesiasgme neither actually20:19
@lisitsynit depends on the formulation20:19
@lisitsyniglesiasg: is the qp problem in SO the same as in basic svm?20:19
@iglesiasglisitsyn: no, I don't think so. Let me ensure it20:20
@iglesiasglisitsyn: this version achieved with the cutting plane problem it is I think20:21
@iglesiasglisitsyn: the problem with SO is that you have a pretty large number of constraints, using a cutting plane algorithm you select the active ones and then you need to solve a QP20:22
@iglesiasgthis QP is pretty much like Crammer-Singer with the margin scaled20:23
@lisitsynoh yes yes20:24
@lisitsynC-S is the thing that made SO20:24
@lisitsyn:)20:24
@iglesiasg:)20:25
@lisitsynmikhailBelous: what is the function which gradient you would use there btw?20:27
mikhailBelousTo target function20:32
@lisitsynbut this doesn't handle the constraints right?20:33
@lisitsynI mean we have to use the lagrange function there as we have constraints20:33
mikhailBelousWe go along gradient vector until we reach boundary20:33
mikhailBelousWhat is lagrange fucnction when we use inequalities?20:34
@lisitsynmikhailBelous: please check the paper I send to you - QP is usually formulated through its dual form20:37
-!- sumit [73f91219@gateway/web/freenode/ip.115.249.18.25] has quit [Ping timeout: 245 seconds]20:49
-!- vivekn [65d75db0@gateway/web/freenode/ip.101.215.93.176] has joined #shogun20:49
-!- mikhailBelous [~towelenee@109.205.249.161] has quit [Ping timeout: 260 seconds]20:52
-!- deerishi [73f88294@gateway/web/freenode/ip.115.248.130.148] has quit [Ping timeout: 245 seconds]20:55
alibezzlisitsyn: I just submitted my GSoC proposal. If you could take a look at it when you have time and tell me if it's ok, I'd be glad!!20:56
alibezzlisitsyn: http://www.google-melange.com/gsoc/proposal/review/google/gsoc2013/alibezz/120:56
-!- mikhailBelous [~towelenee@109.205.249.161] has joined #shogun20:57
@iglesiasgalibezz: I think that if you want to apply for two different ideas you need to submit two proposals20:57
@iglesiasgeven if it is for the same org20:57
@iglesiasgjust mentioning because I saw this Idea 1 and Idea 2, sorry if you already knew that :)20:58
alibezziglesiasg: I think I'll just apply for the DimRed project. I'm much more interested in this one!20:58
alibezziglesiasg: ok, I'll fix this later20:58
alibezz;)20:58
alibezzlater today.20:58
mikhailBelousPlease, give me an advise how to watch project in some using linux?20:58
@iglesiasgmikhailBelous: what do you mean?20:58
mikhailBelousI need IDE in order to read code, what do you use in linux?20:59
gsomixmikhailBelous: vim21:00
@iglesiasgmikhailBelous: ah ok. I am not an IDE guy, I use terminal and vim as editor21:00
@iglesiasgmikhailBelous: eclipse of code::blocks might be a choice21:01
mikhailBelousAnd how do you find function declaration and implementation?21:03
@iglesiasgmikhailBelous: I grep all the way :D21:03
@lisitsynmikhailBelous: ctags + vim supports that quite ok21:04
@lisitsyneclipse supports that natively so feel free to use it21:04
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun21:05
alibezziglesiasg: lisitsyn: fixed! this is the current version of my application http://www.google-melange.com/gsoc/proposal/review/google/gsoc2013/alibezz/1 if you guys have any suggestions, please let me know21:07
@lisitsynalibezz: bahh you need a new notebook ;)21:09
@iglesiasghehe21:10
alibezzlisitsyn: seriously? well, I can always ssh a better machine in my university21:10
@lisitsynalibezz: just kidding21:10
@lisitsynI mean it is pretty old :)21:10
alibezzlisitsyn: lolz YEAH IT IS hahaha my newer one broke.21:10
@iglesiasgtoo bad21:11
alibezzlisitsyn: and now i myself am super low on cash... hahaha21:11
alibezzlisitsyn: gonna buy a new one, maybe a mac-os, when I make some dough. ;)21:11
vgorbatilisitsyn: hi there. So I implemented all the steps, also run generator.py on the created examle (it succeeded). But, when I run tests, my new example gives an 'EXCEPTION No module named modshogun', what is kind a strange, regarding that all other tests work ok. Any suggestions what this can be related to?21:12
@lisitsynvgorbati: hey21:12
@lisitsynnot really21:12
gsomixalibezz: interesting proposal :)21:15
@lisitsynalibezz: vgorbati: fight! :D21:16
@lisitsynyou are the competitors21:16
@lisitsyn*mortal kombat theme*21:16
alibezzhahahahah well i might be losing already because I just arrived.21:16
@iglesiasgso epic one21:16
@lisitsynalibezz: you submitted the proposal first though21:17
alibezzwell, no I'm gonna work on my paper, and then on the issue! ;)21:18
alibezzsee yall later!21:18
@lisitsynsee you21:18
@iglesiasgbye bye21:18
-!- alibezz [~alibezz@177.182.164.8] has left #shogun []21:18
@lisitsyniglesiasg: vgorbati: a puzzle!21:19
@lisitsynhttps://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/lib/tapkee/traits/callbacks_traits.hpp21:19
@lisitsynwhat is the C++ idiom here in lines 20-31?21:19
-!- deerishi [73f88294@gateway/web/freenode/ip.115.248.130.148] has joined #shogun21:20
@iglesiasghaha no idea -- let me see if I come up with anything to say :D21:20
@iglesiasgthe dummy pattern21:20
@lisitsynhaha21:21
@lisitsynno21:21
@iglesiasglisitsyn: so what is it?21:26
@lisitsyniglesiasg: sfinae21:26
-!- deerishi [73f88294@gateway/web/freenode/ip.115.248.130.148] has quit [Quit: Page closed]21:40
vgorbatilisitsyn: my mind is wrecked:) The purpose of this code is to determine whether a type T has a 'dummy' inside it, isn't it?21:40
@lisitsynvgorbati: yes21:40
vgorbatilisitsyn: that is why I do not like C++)21:41
@lisitsynwhy?21:41
vgorbatilisitsyn: there are too much possibilities to shoot yourself in the foot:)21:41
@lisitsynit is important that it is compile-time21:41
@lisitsynTBH I am not in the trend of C++ haters :D21:43
* sonney2k qualifies to be a java hater21:43
@lisitsynhahah21:43
@lisitsynwell C++ sucks when you don't get it21:43
@lisitsynthat's for sure21:43
@lisitsynsucks much more than everything else probably21:44
gsomixsonney2k: hey!21:44
@iglesiasgwe are all matlab haters, aren't we? :D21:44
gsomixiglesiasg: nope21:44
gsomix:321:44
@iglesiasggsomix: but.. but, python!21:44
vgorbatilisitsyn: maybe it is like a very sophisticated tool, but personally I prefer understandable tools21:45
gsomixiglesiasg: there is not python in my lab.21:45
gsomix*no21:45
@lisitsynvgorbati: what is understandable for you?21:45
@iglesiasggsomix: wow so the pay for matlab licenses but don't download python? funny21:46
vgorbatilisitsyn: hard to explain:) python is understandable tool, pure C is understandable:)21:47
@lisitsyniglesiasg: pay? I don't think so :D21:47
vgorbatilisitsyn: the only thing that I miss in pure C is templates21:47
@iglesiasghehe21:47
@lisitsynvgorbati: what else is left? classes/21:48
gsomixiglesiasg: so, such is the objective reality21:48
gsomixlisitsyn: tsssss21:48
@sonney2kgsomix, hey21:48
gsomixsonney2k: yay! glad to see you21:48
gsomixsonney2k: have a minute?21:48
gsomixor second...21:49
vgorbatilisitsyn: personally I do not see a big difference between calling myVector.sort() and sort(myVector).21:49
@lisitsynvgorbati: for big projects it matters a lot21:49
gsomixlisitsyn: vgorbati: http://www.youtube.com/watch?v=I6knCdiUBz421:50
@sonney2kgsomix, ask21:50
vgorbatilisitsyn: maybe, maybe:) btw, have you seen the fake Stroustrup interview?:)21:50
gsomixsonney2k: https://dl.dropboxusercontent.com/u/19029407/ideas%20for%20gsoc%202013.txt21:50
@lisitsynvgorbati: which?21:50
vgorbatilisitsyn: http://www.snopes.com/computer/program/stroustrup.asp this one:)21:51
vgorbatilisitsyn: even though it is fake, it is kind a interesting:)_21:51
@sonney2kgsomix, all certainly valid problems21:51
@sonney2kgsomix, but might be too difficult21:51
@sonney2kgsomix, e.g. r-modular is not sth I think you can fix nor the swig -> multiple modules21:52
@lisitsynhah vgorbati I do not get the point of hating C++ a lot21:52
@lisitsynI always hear things like 'I do not like pointers'21:53
@lisitsynor 'it is very complex'21:53
@lisitsynbut nobody forces to use it, right? :D21:53
-!- mikhailBelous [~towelenee@109.205.249.161] has quit []21:54
-!- vivekn [65d75db0@gateway/web/freenode/ip.101.215.93.176] has quit [Ping timeout: 245 seconds]21:54
vgorbatilisitsyn: well, I do not actually hate it, I just think that some thing could be done better, that's all:) Like in python (and when you will need the speed - just implement this part in C and use it as a linked library)21:55
@lisitsynvgorbati: do you really have any improvement to C++?21:55
@lisitsynsome things do feel wrong (like syntax)21:56
vgorbatilisitsyn: you mean python vs C++?21:56
gsomixsonney2k: sad. are there another ideas for project?21:56
@lisitsynbut I do not have any suggestion at all21:56
vgorbatilisitsyn: actually yes, I guess I am mainly concerned about syntax - after python I always fill that I write too much code when in C++)21:58
@sonney2kvgorbati, I totally agree21:59
@sonney2kgsomix, well we need file i/o stuff21:59
@sonney2knot sure if this is of interest to you21:59
@lisitsynwell that's not the intrinsic property of C++21:59
@sonney2klisitsyn, c'mon it is21:59
@lisitsynsonney2k: example?22:00
@sonney2kpython is like 10 characters and c++ would be 1000 lines22:00
@lisitsynif you avoid libraries or what?22:00
gsomixsonney2k: http://shogun-toolbox.org/page/Events/gsoc2013_ideas#fastio this?22:00
@sonney2kgsomix, yes22:00
@sonney2klisitsyn, sure I am exaggerating but its like it22:01
@lisitsynsonney2k: usually people just do not exploit capabilities and write C+- instead of C++22:02
gsomixsonney2k: hm, nice project22:02
@lisitsynsonney2k: for example people keep writing inner loops to say compute vector norm22:05
@lisitsynit is not only eigen that provides such things but blitz++ and armadillo and etc22:05
vgorbatilisitsyn: I guess, I've just lost one point as a GSoC applicant :)22:07
@lisitsynsonney2k: furthermore you are assuming stl is wrong - but it does reduce the # of LoC22:07
@lisitsynvgorbati: no, why?22:08
vgorbatilisitsyn: because I do not like C++ as much as you do:) just kidding)22:08
@lisitsynvgorbati: am I stalin or what? :D22:09
@lisitsynvgorbati: I do like all the languages they are all cool22:09
gsomixsonney2k: I think it's interesting idea. should/can I add something in it from my list of ideas?22:11
@iglesiasgok, one more application to the batch :)22:11
vgorbatilisitsyn: let's stop the holywar on this point:)22:11
@lisitsynhaha22:11
@lisitsynsonney2k: we with iglesiasg do need a quick idea for some conference paper :D22:15
@iglesiasglisitsyn: coffin is structured svm is the way :)22:19
@lisitsynI do not get how it can help it though22:19
@lisitsyniglesiasg: http://cs409823.vk.me/v409823955/1836/D_Ps3YSxsak.jpg new photo from the city of accident black holes22:21
vgorbatilisitsyn: ok, it looks like I've managed to finish the task - the test passes. Is it ok if I submit a PR now?22:22
@lisitsynvgorbati: yes22:22
@lisitsynvgorbati: btw t-sne has local minima so I'd blacklist it for now22:23
@iglesiasglisitsyn: wtf22:23
vgorbatilisitsyn: but then we need a totally different test:)22:24
@iglesiasglisitsyn: you could help in lot of stuff for it. You understand coffin, you are a coding master, etc :)22:24
@lisitsynvgorbati: yes - all converters are blacklisted IIRC22:24
@lisitsynvgorbati: it is only PCA that produces the same result and even it can vary a little22:25
vgorbatilisitsyn: :D yes, I guess some tests are 'blacklisted' in interesting manner - they call 'converter.apply(features)' and then 'return features' - that's how the result will be similar every time:)22:27
vgorbatilisitsyn: but at list they are checking, whether the apply method does not call an exception22:27
vgorbatilisitsyn: is it ok if my test will be 'black-listed' in a similar manner?22:28
@lisitsynvgorbati: just blacklist it for real22:28
vgorbatilisitsyn: ok22:29
@lisitsynhowever22:29
@lisitsynno idea22:29
@lisitsynrather do a PR now22:29
@lisitsyn:)22:29
@lisitsynbut yes, probably blacklisting it from the very beginning is the way22:29
vgorbatilisitsyn: ok, but currently test works as I described - it return original features - that is how it passes22:29
@lisitsynvgorbati: no, it produces the same result then22:30
@lisitsynreturned features are serialized with generator.py22:30
@lisitsynthen de-serialized and checked22:30
vgorbatilisitsyn: yes, and the test passes (if only the apply method throws an exception)22:31
@lisitsynif it was producing the different result22:31
@lisitsynit would fail22:31
vgorbatilisitsyn: does not throw*22:31
vgorbatilisitsyn: the thing is, that calling 'converter.apply(features)' does not change the 'features' itself - it returns the embedding. So, 'return features' will just return original data every time - that's why the test won't fail22:33
vgorbatilisitsyn: anyway, I got it - I will add it to blacklist and send a PR22:34
@lisitsynahhh22:34
@lisitsyngot it22:34
@lisitsynyes you right22:34
@lisitsynvgorbati: it came from the old time when it did change the feautres22:35
vgorbatilisitsyn: so this tests should probably be refactored?22:35
@lisitsynwell they won't start working as tests anyway22:36
@lisitsynvgorbati: even for say LLE22:36
@lisitsynall the features in the embedded space22:37
@lisitsyncan be both multiplied by +1 or -122:37
@lisitsynand it will be correct22:37
@lisitsynI mean sign of the eigenvector doesn't matter22:37
vgorbatilisitsyn: I meant that they should be gradually refactored so that they actually do return an embedding instead of original data + that they will be tested 'the right way'22:38
@lisitsynyes but sounds like a long way22:39
@lisitsynand it actually could become a unit test instead of integration test :D22:39
vgorbatilisitsyn: So the integration test can only test for equality of results, right?22:39
@lisitsynno but it is supposed to do something like that22:40
vgorbatilisitsyn: ok, I got to go now - will submit a blacklisted test tomorrow22:40
@lisitsynalright22:40
vgorbatisee you22:41
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Quit: Page closed]22:41
@lisitsyn10 proposals so far22:41
@iglesiasggood number22:53
--- Log closed Mon Apr 29 00:00:29 2013