IRC logs of #shogun for Saturday, 2012-08-04

« Friday, 2012-08-03 Index Sunday, 2012-08-05 »
--- Log opened Sat Aug 04 00:00:17 2012
shogun-buildbot_build #253 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/253  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>00:22
CIA-18shogun: Sergey Lisitsyn master * r3d62578 / (4 files in 2 dirs): Revert "drop GMM - it abuses SGVector to store Gaussians" - https://github.com/shogun-toolbox/shogun/commit/3d625786d9fbcd8273b197a0e11744bd81da42db00:34
CIA-18shogun: Sergey Lisitsyn master * ra79170d / (4 files in 2 dirs): Get GMM back to work again - https://github.com/shogun-toolbox/shogun/commit/a79170d593d99745a6ce8cbcd5f88cdf064c83b300:34
shogun-buildbot_build #244 of deb1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/244  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>00:36
shogun-buildbot_build #245 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/24500:36
shogun-buildbot_build #254 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/254  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>01:10
shogun-buildbot_build #210 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/210  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>01:12
CIA-18shogun: Sergey Lisitsyn master * ra686d8c / (6 files in 4 dirs): Introduced Kernel Mean Matching - https://github.com/shogun-toolbox/shogun/commit/a686d8c75783f49c6ada03ea3204fd74a999dd7502:09
CIA-18shogun: Sergey Lisitsyn master * rebf738c / examples/undocumented/python_modular/statistics_kmm.py : Added KMM example - https://github.com/shogun-toolbox/shogun/commit/ebf738c3fca7d6524c1beecd9bac6048c139514a02:10
CIA-18shogun: Sergey Lisitsyn master * r9bf02e0 / src/shogun/regression/gp/InferenceMethod.cpp : Fixed inference method crasher - https://github.com/shogun-toolbox/shogun/commit/9bf02e0995f378926b1a9a5799f61aaf9faa082002:14
shogun-buildbot_build #255 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/255  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>02:33
shogun-buildbot_build #211 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/211  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>02:37
blackburn???02:38
-!- blackburn [~blackburn@109.226.80.43] has quit [Quit: Leaving.]02:49
shogun-buildbot_build #256 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/256  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>02:53
shogun-buildbot_build #212 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/212  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>02:55
shogun-buildbot_build #43 of nightly_none is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/4303:10
shogun-buildbot_build #49 of nightly_default is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/4903:33
shogun-buildbot_build #39 of nightly_all is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/3903:56
-!- something_someth [~tejaswidp@117.202.17.244] has joined #shogun07:03
-!- something_someth [~tejaswidp@117.202.17.244] has left #shogun ["Leaving"]07:04
-!- zxtx [~zv@75-59-237-124.lightspeed.sntcca.sbcglobal.net] has joined #shogun07:21
-!- zxtx [~zv@75-59-237-124.lightspeed.sntcca.sbcglobal.net] has quit [Ping timeout: 272 seconds]08:14
-!- pluskid [~pluskid@108.171.196.83] has joined #shogun08:36
pluskidoops! it seems Euclidean is typo-ed as Euclidian in shogun...10:31
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has joined #shogun12:46
n4nd0:-O I cannot believe my eyes12:47
n4nd0sonney2k is not online!12:47
-!- blackburn [~blackburn@109.226.80.43] has joined #shogun13:25
n4nd0blackburn: hey!13:35
blackburnn4nd0: hey13:35
n4nd0where is the captain?!13:35
blackburnhow was your flight?13:35
n4nd0blackburn: it was fine, no problem13:36
blackburnhmm no idea there was some net split I guess13:36
n4nd0the plne was half empty13:36
n4nd0I had a full row (three seats) just for me :)13:36
n4nd0you know, I had never really thought of it before13:36
n4nd0but Spain is warm .... pretty warm13:37
n4nd0:D13:37
blackburnheh13:38
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has quit [Quit: leaving]14:04
CIA-18shogun: Sergey Lisitsyn master * rb65dad3 / (2 files in 2 dirs): Removed size() of sgvector - https://github.com/shogun-toolbox/shogun/commit/b65dad36802264fc5cb63ded2436831f23178ebd14:05
blackburnpluskid: I can add primal and dual formulation of crammer-singer (I derived it once) to the multiclass part14:05
pluskidblackburn: cool!14:06
pluskidI'm about to asking you about that :D14:06
blackburnpluskid: do you think we should have detailed derivations of such stuff?14:07
pluskidso a netsplit cause half of the plane into another space? :D14:07
blackburnactually in papers and books they are usually do not derive it14:08
blackburnyes probably14:08
pluskidblackburn: I don't know the crammer-singer derivation, I think we'd better only add short derivations14:08
pluskidblackburn: btw, I saw the skeleton of the "essentials" part, you are not going to finish all those before you die, haha :D14:08
blackburnwhy? I am pretty interested in that stuff14:09
pluskidblackburn: I mean there are soooooo many contents14:09
blackburnwell 10-20 pages :D14:09
pluskidno, you should be able to write 600 pages for that part :D14:10
pluskidlike vapnik's book14:10
blackburnlets see :)14:10
shogun-buildbot_build #248 of deb1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/248  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>14:10
blackburnvapnik's EDBED is ~250 pages long IIRC so we should get into that :)14:11
CIA-18shogun: Sergey Lisitsyn master * rc07a8a1 / src/shogun/statistics/KernelMeanMatching.cpp : Fixed wrong size() call - https://github.com/shogun-toolbox/shogun/commit/c07a8a1f62cbd4b5897d8eafdf8b098ef06a59ce14:12
blackburnpluskid: so yuo are going to MIT?14:13
pluskidblackburn: right after GSoC14:13
blackburnpluskid: what are you going to research there?14:13
pluskidmachine learning :D14:13
pluskidspecific topic not decided yet14:14
blackburnI hope you will manage to use shogun and extend it through research14:14
pluskidyeah, hope so :)14:15
shogun-buildbot_build #249 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/24914:17
blackburnpluskid: I see you are interested in VC too :D14:17
pluskidblackburn: VC you mean VC-theory?14:19
blackburnyeah14:19
pluskidblackburn: yeah, I'm pretty interested in that :D14:19
blackburnpluskid: however google translates your blog pretty bad14:21
pluskidblackburn: haha, let me try it14:21
blackburnpluskid: well actually sometimes pretty well14:23
blackburnbut sometimes I find it hard to parse :)14:23
pluskidoops, beautiful margin-notes and equations disappeared :D14:24
pluskidone of my classmate actually use google translate literally for an assignment of paper translation (into chinese)14:24
pluskidand the teacher didn't notice it14:25
pluskid(though I guess the teacher never looked at the assignments)14:25
blackburnheh14:26
pluskidahh... I tried to read some fragments, and feel completely unreadable...14:26
pluskidbut google translate is the state of the art, right?14:27
blackburnno idea about that actually14:28
blackburnCIA should has something better to translate secret chinese docs :D14:29
blackburnargh have*14:29
pluskidno no14:29
pluskidthey must have human translators14:30
shogun-buildbot_build #257 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/257  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>14:38
shogun-buildbot_build #213 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/213  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>14:41
-!- blackburn [~blackburn@109.226.80.43] has quit [Quit: Leaving.]14:58
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has joined #shogun15:37
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has quit [Ping timeout: 272 seconds]16:38
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has joined #shogun16:40
-!- pluskid [~pluskid@108.171.196.83] has quit [Quit: Leaving]16:42
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has quit [Quit: leaving]18:15
-!- sonney2k [~shogun@7nn.de] has joined #shogun18:42
-!- mode/#shogun [+o sonney2k] by ChanServ18:42
-!- blackburn [~blackburn@109.226.80.43] has joined #shogun19:07
@sonney2kblackburn, why did you remove size() from sgvector?19:22
blackburnsonney2k: it conflicts with csharp stuff19:22
blackburnI'd prefer not to touch csharp :)19:23
blackburncould break things19:23
blackburnbut actually I'd like to have stl-like interface in vector19:23
blackburnsonney2k: btw gmm is back19:25
@sonney2kblackburn, ok so we can just put it back and rename it for c# then19:26
blackburnsonney2k: yeah but I don't want to touch csharp typemaps actually :)19:26
CIA-18shogun: Soeren Sonnenburg master * rd7a2b11 / src/Makefile.template : fix path to regression tests in makefile - https://github.com/shogun-toolbox/shogun/commit/d7a2b116bb3fefb75ccff12817919d49c15d0ecb19:27
CIA-18shogun: Soeren Sonnenburg master * r18b4d74 / src/shogun/base/Parameter.cpp : add '\n' at end of error msg - https://github.com/shogun-toolbox/shogun/commit/18b4d74636d1bb95dc3e5f5f898cced4ba44143519:27
@sonney2kblackburn, seen that - nice work!19:27
blackburnsonney2k: but with std::vector actually :D19:27
blackburninstead of sgvector19:27
blackburnhowever could be changed19:27
@sonney2kseen that too and I guess I will convert it to dynarray at some point :-]19:27
blackburnshould be easy actually19:28
blackburnif dynarray supports []19:28
@sonney2kit does19:28
@sonney2khowever std:: is a nogo certainly for header files - for the rest I guess I will have to live with whatever you do...19:29
@sonney2kblackburn, I hope the buildbot gets happy again...19:29
blackburnsonney2k: why not to put into headers?19:30
blackburnsonney2k: well for serialized fields yes for sure19:30
blackburnsonney2k: do you want to write a makefile for tutorial? ;)19:32
@sonney2kcompile time++19:32
blackburnahh19:32
@sonney2kblackburn, why me?19:32
blackburnsonney2k: I am lame with it still (crying)19:33
@sonney2kblackburn, I first have to take care of the PRs19:34
@sonney2kblackburn, is n4nd0's PR OK?19:34
@sonney2kblackburn, but we shouldn't merge anything before we get the buildbot back to green19:34
blackburnhmm let me check it again19:34
blackburnisn't it all green now?19:34
@sonney2knothing is19:35
blackburnsonney2k: no, n4nd0  should get USE_MOSEK guards back19:35
blackburnnot to merge it right now19:35
@sonney2kblackburn, err? I think he did?19:35
blackburnahh19:35
blackburnright19:35
blackburnsure19:35
blackburnI was confused with no commit after19:36
blackburnsonney2k: http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/213/steps/test%20cmdline_static/logs/stdio that fails for no real reason19:37
@sonney2kblackburn, well due to the test restructuring of wiking/heiko19:37
@sonney2kI tried to fix it just now19:37
blackburnI see19:38
@sonney2kblackburn, do you understand the MatrixFeatures n4nd0 introduced?19:38
@sonney2kI dont' understand the doc19:38
blackburnsonney2k: well yeah a little19:38
blackburnsonney2k: feature vector = matrix19:38
@sonney2kwhy such a weird name?19:42
blackburnsonney2k: because it contains a set of matrices19:42
@sonney2kI guess underneath it is treated as vector again19:42
@sonney2kblackburn, no I mean feature vector19:43
@sonney2kI think he should also rather use sth like NDArray to set the matrices19:43
@sonney2kno idea how to do it for matrices of different size though19:43
@sonney2k(and we have no typemaps to work with these features...)19:44
shogun-buildbot_build #258 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/258  blamelist: Soeren Sonnenburg <sonne@debian.org>19:46
blackburnyeah it is a problem19:46
blackburnehm19:46
blackburnwhy does it faill still?19:46
shogun-buildbot_build #214 of deb2 - static_interfaces is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/21419:49
@sonney2kblackburn, there is this one typemap I have no idea how to fix for c#19:51
@sonney2kblackburn, when you have some method that gets >1 SGVector or so as argument19:51
@sonney2kthen the build will fail for c#19:51
blackburnoooh19:51
blackburninteresting19:51
@sonney2kso the only fix I have is to %ignore that method for c#19:51
@sonney2kblackburn, I guess you didn't revert that typemap %ignore thing...19:52
@sonney2kit should have been in interfaces/modular/modshogun_ignores.i19:52
@sonney2kin the #ifdef SWIGCSHARP section19:52
blackburnokay I see19:53
blackburnwill add it then19:53
blackburnsonney2k: so I can get size back then19:54
@sonney2khehe19:54
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has joined #shogun19:55
n4nd0hi sonney2k19:55
n4nd0let me ask you, how do you prefer that I introduce the changes you have commented in the pull request?19:56
n4nd0new commit or ammending?19:56
n4nd0I remember once you said something about amend better19:56
@sonney2kblackburn, please also have a look at pluskid's PR again - he explained why he needs a kernel factory19:56
n4nd0however, that would require to force the push and I am not really confident whether it is good to do that19:56
@sonney2kblackburn, in principle he needs to *create* kernels of a certain type19:57
@sonney2kthough I dont' really understand why19:57
@sonney2kn4nd0, amend is better ... there are certainly enough commits already in this PR19:58
@sonney2kn4nd0, btw good that you are here19:58
@sonney2kn4nd0, I don't understand the matrix features doc19:58
blackburnthat kind of factory could be replaced by copy constructor I think19:58
blackburnor some clone()19:58
n4nd0sonney2k: probably the documentation about it is not very elaborated :S19:59
@sonney2kblackburn, I mean if pluskid needs to create a new kernel then it will have no arguments set!19:59
@sonney2kso it cannot give reasonable results !?19:59
@sonney2kI must be missing sth20:00
@sonney2kn4nd0, so please explain again :D20:00
n4nd0sonney2k: there is a type in the doc now that I read it again20:00
@sonney2ktypo?20:00
n4nd0sonney2k: it is caused by the first variable length matrix features I created, and later I went to MatrixFeatures20:00
n4nd0yeah, typo, sorry20:00
n4nd0sonney2k: however ammending you won't be able to note the new changes20:03
@sonney2kn4nd0, your PR is just too big20:04
shogun-buildbot_build #259 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/259  blamelist: Soeren Sonnenburg <sonne@debian.org>20:04
@sonney2kwhy didn't you just send a PR for matrixfeautres?20:04
n4nd0sorry about that :(20:04
@sonney2kand then others for isolated changes?20:05
n4nd0because I couldn't develop first the whole MatrixFeatures and later the HM-SVM, etc20:05
-!- zxtx [~zv@75-59-237-124.lightspeed.sntcca.sbcglobal.net] has joined #shogun20:07
CIA-18shogun: Sergey Lisitsyn master * r4efd4e9 / (2 files in 2 dirs): Restored size(), ignored cast operators and ignored KMM constructor in csharp - https://github.com/shogun-toolbox/shogun/commit/4efd4e966518e928daac51e48d215e95e24f3d1b20:18
-!- zxtx [~zv@75-59-237-124.lightspeed.sntcca.sbcglobal.net] has quit [Read error: Operation timed out]20:20
@sonney2kn4nd0, what is a twostate model?20:22
n4nd0sonney2k: I have introduced the changes you told me in github and modify the MatrixFeatures doc, please tell me if now is more comprehensible20:22
n4nd0sonney2k: are you familiarized with hmsvm toolbox?20:23
@sonney2kn4nd0, no - I know the theory / concepts though20:23
n4nd0sonney2k: ok, so the TwoState model is used for the state representation of the markov model20:25
n4nd0sonney2k: the TwoState model is the simplest example20:25
n4nd0sonney2k: it is needed because the labels do not represent directly the states in the model20:25
n4nd0though the labels are strictly related with the states20:26
shogun-buildbot_build #260 of deb3 - modular_interfaces is complete: Failure [failed compile python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/260  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>20:26
blackburn:(20:26
@sonney2kn4nd0, so a toy example right?20:28
CIA-18shogun: Sergey Lisitsyn master * rf513ae0 / src/interfaces/modular/modshogun_ignores.i : Fixed syntax error in includes - https://github.com/shogun-toolbox/shogun/commit/f513ae0cc314956dc39a5a7f8b66639f062e44aa20:28
n4nd0sonney2k: not really20:29
n4nd0it covers any HM-SVM application where the labels can take two different values20:30
n4nd0sonney2k: do you understand what I mean?20:33
@sonney2kyes20:35
@sonney2kn4nd0, so it could be used for sth like background vs. foreground in images etc right?20:35
@sonney2kimage segmentation I mean20:35
n4nd0sonney2k: yes, for example20:35
n4nd0yes20:36
@sonney2kn4nd0, btw everything currently only works with mosek right?20:36
n4nd0sonney2k: yes, but according to what wiking and uricamic told me yesterday20:36
n4nd0if I define a risk function for the HMSVM model20:37
blackburnexplain! how can I represent image as markov model?20:37
n4nd0I could use uricamic bundle methods, that is really cool20:37
n4nd0blackburn: it is not really an image what you can represent as a markov model20:37
n4nd0blackburn: because there is no time variance there20:37
n4nd0blackburn: but for example imagine a video sequence20:37
blackburnokay sequence, what's next?20:38
n4nd0blackburn: take a window of the video sequence20:38
n4nd0you could model that as an markov model20:38
blackburnehmmmmm20:38
@sonney2kn4nd0, that should really have a high priority then - otherwise the code you added cannot be used :/20:38
n4nd0say I want to know whether this region is in foreground or backbround20:38
blackburnwindow like some area of image right?20:38
n4nd0blackburn: sure20:39
blackburnso you just go through all image20:39
n4nd0maybe you could do that to20:39
n4nd0but what I was thinking is20:39
blackburnhmm I see20:39
n4nd0you fix your window20:39
n4nd0and run the video sequence20:39
blackburnbut actually I agree with soeren we should make it usable not only with mosek20:40
n4nd0for each frame (or image) you model if that window is in background or foreground20:40
n4nd0blackburn, sonney2k: I agree with you guys20:40
blackburnokay so you get something like20:40
blackburnfore - back - fore - back - back20:40
blackburnfor frames20:40
blackburnright?20:40
n4nd0yes20:40
n4nd0those would be the hidden states20:41
@sonney2kn4nd0, I still don't get the description20:41
n4nd0sonney2k: of MatrixFeatures?20:41
@sonney2kfor matrix feats20:41
@sonney2kyes20:41
@sonney2kn4nd0, can you just explain here please?20:41
n4nd0sonney2k: sure20:41
n4nd0so with dot features we have things like20:42
n4nd0x1 = [1, 0, -1, 4, 5]20:42
n4nd0x2 = [-4, 9, 7, 6, 5]20:42
n4nd0x3 = ...20:42
n4nd0each xi is a feature vector20:42
n4nd0sonney2k: ok?20:42
n4nd0the feature 1 of x1 takes the value 1, of x2 takes the value -420:43
CIA-18shogun: Soeren Sonnenburg master * r4ec77d8 / src/interfaces/modular/modshogun_ignores.i : fix %ignore - SGVector is of type index_t not float... - https://github.com/shogun-toolbox/shogun/commit/4ec77d80012857ffdeb8eae2c6314699f160a68320:44
@sonney2kn4nd0, yes20:44
@sonney2kblackburn, I hope this commit fixes the build finally...20:45
blackburnargh20:45
blackburndamn20:45
blackburnsure index t20:45
blackburnthnask20:45
n4nd0in matrix features each xi is a matrix20:46
n4nd0x1 = [ 1,  0,  4;20:46
n4nd0       2, -1,  3]20:46
n4nd0x2 = [ 4, -3,  2,  5;20:46
n4nd0      -6,  5,  0, -3]20:46
n4nd0again each xi is a feature vector20:46
n4nd0I don't really like to say feature vector because now they are represented as matrices20:46
n4nd0I like to use more example20:46
n4nd0anyway20:46
n4nd0now20:46
n4nd0the feature 1 of x1 is [1 0 4], and of x2 [4 -3 2 5]20:47
@sonney2kn4nd0, ok so you basically have matrices as examples right?20:48
@sonney2kdo they all have the same size?20:48
n4nd0NOTE: in MatrixFeatures each of the features are the columns, while in this toy example they are the rows20:48
@sonney2kI mean same nr of rows/cols?20:48
n4nd0sonney2k: now about the sizes20:48
n4nd0they must have the same number of features20:48
@sonney2kfeatures?20:48
n4nd0BUT the features don't need to have the same length among examples20:48
n4nd0yes20:48
@sonney2k== rows?20:49
n4nd0in my example ^, the same number of rows20:49
n4nd0in MatrixFeatures, the same number of columns20:49
n4nd0but as you see above, the other dimension is not the same for all the examples20:50
n4nd0sonney2k: am I making it clear or not?20:50
@sonney2kn4nd0, why do they have to have the same number of columns?20:51
n4nd0there is another point of view that may make more sense20:51
n4nd0sonney2k: where? in the example above or in MatrixFeatures?20:51
@sonney2kit is not crystal clear yet20:51
@sonney2kin MatrixFeatures20:51
@sonney2kn4nd0, I first thought it is basiclly an array of matrices of same size20:51
n4nd0it is almost that20:52
shogun-buildbot_build #261 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/261  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>20:52
n4nd0it is an array of matrices with the same number of columns20:52
@sonney2know you say it is basically an array of matrices where the matrices have the same number of columns20:52
n4nd0yes, it is that20:52
@sonney2kthat is much much easier to understand20:52
@sonney2kbut still doesn't explain why this restriction is needed!20:52
@sonney2kand what one can do with this20:52
n4nd0the restriction is needed because you need to have the same number of features for all the examples20:53
n4nd0imagine dotfeatures where x1 has 4 elements and x2 three20:53
@sonney2kthe term features creates a parse error for me here20:53
n4nd0yeah20:53
n4nd0features != feature vector20:54
n4nd0so what I said20:55
n4nd0in DotFeatures20:55
n4nd0all the examples have the same number of features20:55
n4nd0(otherwise we are talking about StringFeatures)20:55
n4nd0in my MatrixFeatures this is still the same, same number of features20:55
n4nd0therefore, the restriction of the same number of columns arises20:56
n4nd02) about what can one do with this: train HM-SVMs :)20:56
n4nd0sonney2k: closer to crystal clear?20:57
@sonney2kn4nd0, closer but what do the different nr of rows stand for?21:00
@sonney2kn4nd0, I am currently thinking if it would not be better to have a general matrix features type with that restriction removed (or at least optional)21:00
n4nd0sonney2k: sure it could be21:01
n4nd0I don't think that actually I coded this restriction in the MatrixFeatures21:01
n4nd0although I use it in the HM-SVM21:02
n4nd0because there the restriction exists21:02
n4nd0but I am sure about this21:03
n4nd0*not* sure21:05
n4nd0sonney2k: so, are we ok now or should I explain something else?21:09
@sonney2kn4nd0, I understand that now - though the doc is not too clear21:12
@sonney2kbut I would really like to have these matrix features to be more general21:12
@sonney2kotherwise we need to have one more class21:12
n4nd0sonney2k: yes, you are right21:12
n4nd0sonney2k: I am opening an issue in github for me, ok?21:13
@sonney2kone thing that is not so easy to understand to me21:13
n4nd0tell me21:13
@sonney2kmomplz21:13
@sonney2kn4nd0, ok back21:21
@sonney2kn4nd0, how do we feed data into the matrixfeatures?21:21
@sonney2kI mean I could imagine we use SGNDArray to put in matrices of same size21:22
-!- zxtx [~zv@75-59-237-124.lightspeed.sntcca.sbcglobal.net] has joined #shogun21:22
n4nd0sonney2k: one way you can currently do is feeding matrix by matrix21:22
@sonney2kn4nd0, I guess you might not see the problem but it is related to how all the interfaces currently work - we have typemaps aka conversion routines that convert e.g. a numpy array into shogun's dense features21:23
n4nd0sonney2k: if you don't want to restrict that the matrices are of the same size, then I think that an SGNDArray is not appropriate21:23
@sonney2kn4nd0, so I guess one way to do it would be to put in a list of numpy matrices21:23
n4nd0yes21:23
@sonney2kbut we could as well have a new SG* data type for that then21:24
@sonney2ke.g. for strings we have SGStringList21:24
@sonney2kso we could have SGMatrixList21:24
n4nd0yeah, I think that makes sense21:25
shogun-buildbot_build #262 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/262  blamelist: Soeren Sonnenburg <sonne@debian.org>21:25
n4nd0sonney2k: do you think it would be better having an SGMatrixList <--- SGReferencedData rather than the SGMatrix*?21:25
-!- zxtx [~zv@75-59-237-124.lightspeed.sntcca.sbcglobal.net] has quit [Ping timeout: 244 seconds]21:29
@sonney2kn4nd0, yes21:29
@sonney2kn4nd0, we do this already for SGSparseMatrix21:31
@sonney2k...and will do it for stringfeautres too so this would be the consistent way21:31
@sonney2kblackburn, ^ seen the buildbot is still unhappy21:32
@sonney2kblackburn, new error though21:33
blackburnsonney2k: I do not understand it21:33
blackburntests related again as I can see21:33
@sonney2kblackburn, yeah and wiking hides somewhere and doesn't want to fix the bugs he introduced :D21:33
@sonney2kn4nd0, can you do these changes (introducing SGMatrixList?)21:35
n4nd0sonney2k: is "in the long run" a valid answer? :-O21:36
n4nd0I'd prefer to get the plifs, coffin, try uricamic's bundle methods first21:36
@sonney2kn4nd0, c'mon this will take you 4 hours max21:38
@sonney2kgtg now sry21:38
-!- n4nd0 [~nando@114.Red-2-137-6.dynamicIP.rima-tde.net] has quit [Quit: leaving]21:49
-!- zxtx [~zv@75-59-237-124.lightspeed.sntcca.sbcglobal.net] has joined #shogun22:36
--- Log closed Sun Aug 05 00:00:17 2012
« Friday, 2012-08-03 Index Sunday, 2012-08-05 »

Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!