Open in new window / Try shogun cloud
--- Log opened Sun Mar 09 00:00:30 2014
-!- sonne|osx [] has quit [Quit: sonne|osx]00:26
-!- HeikoS [] has joined #shogun00:37
-!- mode/#shogun [+o HeikoS] by ChanServ00:37
-!- funcode [~funcode@] has quit [Quit: Leaving]00:54
-!- gif [~gif@] has joined #shogun01:19
gifsonney2k: what version did you refer to? the compiler? the mac os?01:21
-!- HeikoS [] has quit [Ping timeout: 240 seconds]01:33
-!- sunil1337 [0e8bd057@gateway/web/freenode/ip.] has joined #shogun02:09
-!- sunil1337 [0e8bd057@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]02:22
-!- kislay [~kislay@] has joined #shogun03:38
shogun-buildbot_build #737 of nightly_default is complete: Failure [failed doc]  Build details are at
-!- KJainGSoC [783bc6ca@gateway/web/freenode/ip.] has joined #shogun03:50
-!- newlido [~quassel@] has joined #shogun03:55
-!- zxtx [] has quit [Ping timeout: 244 seconds]04:18
-!- zxtx [] has joined #shogun04:33
-!- newlido [~quassel@] has quit [Ping timeout: 240 seconds]04:39
-!- newlido [~quassel@] has joined #shogun04:41
-!- newlido [~quassel@] has quit [Remote host closed the connection]04:54
-!- newlido [~quassel@] has joined #shogun04:55
-!- kislay [~kislay@] has quit [Ping timeout: 265 seconds]05:08
-!- kislay [~kislay@] has joined #shogun05:31
-!- naywhayare [] has quit [Write error: Broken pipe]05:34
-!- naywhaya1e [] has joined #shogun05:35
-!- naywhaya1e is now known as naywhayare05:36
-!- abinash_panda [0e8b2997@gateway/web/freenode/ip.] has joined #shogun06:03
-!- newlido [~quassel@] has quit [Remote host closed the connection]06:22
-!- newlido [~quassel@] has joined #shogun06:22
-!- kislay [~kislay@] has quit [Read error: Connection reset by peer]06:22
-!- dhruv13_ [~dhruv13@] has joined #shogun06:23
-!- dhruv13_ [~dhruv13@] has quit [Client Quit]06:23
-!- dhruv13_ [~dhruv13@] has joined #shogun06:24
-!- dhruv13 [~dhruv13@] has quit [Excess Flood]06:25
-!- gif [~gif@] has quit [Remote host closed the connection]06:26
-!- kislay_ [~kislay@] has joined #shogun06:26
-!- sonne|osx [] has joined #shogun06:55
-!- newlido_ [~quassel@] has joined #shogun07:10
-!- newlido [~quassel@] has quit [Ping timeout: 244 seconds]07:11
-!- cameo54321 [2729cfbf@gateway/web/freenode/ip.] has joined #shogun07:20
shogun-buildbot_build #71 of osx2 - python is complete: Failure [failed test python modular]  Build details are at  blamelist: Parijat Mazumdar <>, Orion Poplawski <>, Soeren Sonnenburg <>, iglesias <>, Bj?rn Esser08:01
shogun-buildbot_<>, Thoralf Klein <>, lambday <>, Viktor Gal <>, Heiko Strathmann <>, tklein23 <>, Viktor Gal <>, Thoralf Klein <>, Sergey Lisitsyn <>, Fernando Iglesias08:01
shogun-buildbot_<>, Roman Votyakov <>08:01
-!- dhruv13_ [~dhruv13@] has quit [Ping timeout: 265 seconds]08:04
-!- cameo54321 [2729cfbf@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]08:06
-!- kislay_ [~kislay@] has quit [Ping timeout: 240 seconds]08:25
shogun-buildbot_build #72 of osx2 - python is complete: Failure [failed test python modular]  Build details are at  blamelist: Soeren Sonnenburg <>, Thoralf Klein <>08:30
-!- kislay [~kislay@] has joined #shogun08:40
shogun-buildbot_build #405 of osx1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: lambday <>, Soeren Sonnenburg <>, Viktor Gal <>, pl8787 <>, Viktor Gal <>, Heiko Strathmann08:48
-!- abinash_panda [0e8b2997@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:13
-!- akpraharaj [67157d4c@gateway/web/freenode/ip.] has joined #shogun09:22
-!- kislay [~kislay@] has quit [Read error: Connection reset by peer]09:22
-!- akpraharaj [67157d4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:28
-!- newlido [~quassel@] has joined #shogun09:30
-!- newlido_ [~quassel@] has quit [Ping timeout: 240 seconds]09:30
-!- kislay [~kislay@] has joined #shogun09:37
-!- kislay [~kislay@] has quit [Ping timeout: 264 seconds]10:14
-!- newlido [~quassel@] has quit [Remote host closed the connection]11:02
-!- kislay [~kislay@] has joined #shogun11:04
-!- kislay [~kislay@] has quit [Read error: Connection reset by peer]11:37
-!- PirosB3 [~pirosb3@] has joined #shogun11:48
-!- dhruv13 [~dhruv13@] has joined #shogun11:56
PirosB3hi all12:00
PirosB3anyone on?12:00
dhruv13PirosB3: hello!12:02
PirosB3hey dhruv13 hows it going?12:02
dhruv13PirosB3: not bad!12:02
PirosB3coin anything interesting?12:02
dhruv13PirosB3: kind of...12:03
dhruv13PirosB3: I'm new to a lot of stuff, but figuring it out seems fun!12:03
dhruv13PirosB3: (now)12:03
PirosB3what are you looking at?12:03
dhruv13PirosB3: K nearest neighbours implementation through kd trees12:04
PirosB3did you have expirence with ML before?12:04
dhruv13PirosB3: looking at a library which can be possibly bundled along with shogun, called nanoflann...12:04
dhruv13PirosB3: not much with implementing ML algorithms, though i have used them for various programming contests12:05
dhruv13PirosB3: like kaggle12:05
PirosB3oh cool!12:05
dhruv13PirosB3: :)12:05
dhruv13PirosB3: how about you?12:06
PirosB3looking up stuff for my proposal12:06
PirosB3have you done that yet?12:06
dhruv13PirosB3: not really... I'mstill getting a hang of how shogun works...12:06
PirosB3yeah me too12:07
dhruv13PirosB3: maybe once i'm done with this task, i'll hunt around for stuff ;-)12:07
PirosB3ahahah :)12:08
PirosB3anyways thx for your help last time12:08
dhruv13PirosB3: anytime!!12:08
PirosB3wiking: are you on by any chance?12:13
-!- iglesiasg [] has joined #shogun12:23
-!- mode/#shogun [+o iglesiasg] by ChanServ12:23
-!- khalednasr [2945ccf3@gateway/web/freenode/ip.] has joined #shogun12:36
-!- khalednasr [2945ccf3@gateway/web/freenode/ip.] has quit [Client Quit]12:37
-!- shogun-notifier- [] has joined #shogun12:44
shogun-notifier-shogun: Parijat Mazumdar :develop * 8d4eeef / / (4 files):
shogun-notifier-shogun: added tree structures for id312:44
shogun-notifier-shogun: Parijat Mazumdar :develop * eb4cb6e / / (14 files):
shogun-notifier-shogun: BinaryTreeMachineNode created12:44
shogun-notifier-shogun: Parijat Mazumdar :develop * b1d9a3d / / (7 files):
shogun-notifier-shogun: unit-tests added for valgrind check12:44
shogun-notifier-shogun: Parijat Mazumdar :develop * 56a6ff3 / tests/unit/multiclass/tree/ (3 files):
shogun-notifier-shogun: unittest names changed12:44
shogun-notifier-shogun: Parijat Mazumdar :develop * fba3a47 / src/shogun/multiclass/tree/ConditionalProbabilityTree.cpp:
shogun-notifier-shogun: REQUIRE condition removed to counter travis failure12:44
shogun-notifier-shogun: Fernando Iglesias :develop * 02a1773 / / (17 files):
shogun-notifier-shogun: Merge pull request #1950 from mazumdarparijat/pca12:44
shogun-notifier-shogun: added data structures for decision tree implementation12:44
PirosB3hey dhruv13 do you know anything about multinomial naive bayes?12:45
@wikingPirosB3: there's a lot of ML tutorials out there on the internet:
PirosB3lol, thx12:55
PirosB3wiking: do you have a sec? I wanted to ask you some questions regarding the SaaS. I have a better vision of it now and of it's resources. But I still have some doubts12:56
@wikingPirosB3: email plz12:57
@wikingpreferably mailing list12:57
shogun-buildbot_build #2754 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Parijat Mazumdar <>13:01
shogun-buildbot_build #2755 of deb1 - libshogun is complete: Success [build successful]  Build details are at
@iglesiasgall right, false alarm the fail in deb113:10
shogun-buildbot_build #406 of osx1 - libshogun is complete: Success [build successful]  Build details are at
-!- kislay [~kislay@] has joined #shogun13:20
dhruv13iglesiasg: hey! can i bother you with something?13:27
@iglesiasgdhruv13, tell me13:28
dhruv13iglesiasg: well, is there a significant difference between approximate knn and what needs to be implemented for the kd tree issue?13:28
dhruv13iglesiasg: nanoflann implements ann13:29
@iglesiasgdhruv13, so when I wrote the issue I was not thinking about ann actually13:29
dhruv13iglesiasg: please check
dhruv13iglesiasg:  section 1.613:29
@iglesiasgdhruv13, just a kd tree to do conventional exact nn search13:29
dhruv13iglesiasg: "what nanoflann can't do"13:30
@iglesiasgdhruv13, I was thinking about going on to ann once this was done13:30
@iglesiasgdhruv13, just because I think it is a more natural progression for the person working on that13:30
dhruv13iglesiasg: okay, then i think i will have to go with my own implementation of a kdtree13:30
@iglesiasglearn kd-tree, use it, see limitations on real data and the need of ann13:30
dhruv13iglesiasg: okay!13:30
@iglesiasgdhruv13, having said that13:30
@iglesiasgdhruv13, that was just my idea, it is not something set on stone13:30
-!- akp_ [67157e4c@gateway/web/freenode/ip.] has joined #shogun13:31
@iglesiasgdhruv13, ann is very useful. So the nanoflann integration could be very nice as well13:31
@iglesiasgdhruv13, I don't see anything about exact/approximate nn search in section 1.6 btw13:32
dhruv13iglesiasg: oh sorry, that part is in about13:33
dhruv13iglesiasg: "Queries for neighbors around any arbitrary location in space can then be solved quickly and efficiently using Approximate Nearest Neighbor (ANN) algorithms."13:33
@iglesiasgdhruv13, all right!13:34
dhruv13iglesiasg: the 1.6, point 3 confused me, so i referred to the paper13:34
dhruv13iglesiasg: the part about automatic configuration13:34
dhruv13iglesiasg: that seems to be in flann, but not nanoflann13:35
@iglesiasgdhruv13, all right13:35
dhruv13iglesiasg: so, for now, i think i will work on nanoflann integration13:36
@iglesiasgdhruv13, it sounds great!13:36
dhruv13iglesiasg: and then try out my own kdtree implementation13:36
dhruv13iglesiasg: it will take me a while, since i'm still working around the shogun feature class....13:37
@iglesiasgdhruv13, sure, no problem at all13:37
dhruv13iglesiasg: thanks!13:37
-!- PirosB3 [~pirosb3@] has quit [Quit: PirosB3]13:38
akp_are we looking for the basic knn algorithm using flann?13:38
shogun-notifier-shogun: Viktor Gal :develop * 5675b9d / / (22 files):
shogun-notifier-shogun: CMake-ize documentum generation13:39
shogun-notifier-shogun: Doxygen documentation was waaay out of date and it was still using the13:39
shogun-notifier-shogun: old make system. updated:13:39
shogun-notifier-shogun:  * documentation is generated via cmake i.e. `make doc`13:39
shogun-notifier-shogun:  * outdated doxygen descriptions are removed13:39
-!- travis-ci [] has joined #shogun13:39
travis-ci[travis-ci] it's Fernando Iglesias's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []13:39
-!- shogun-buildbot_ [] has quit [Quit: buildmaster reconfigured: bot disconnecting]13:41
-!- shogun-buildbot [] has joined #shogun13:41
@wikingshogun-buildbot: force build --branch=develop 'nightly_default'13:41
shogun-buildbotThe build has been queued, I'll give a shout when it starts13:42
akp_I wanted to discuss about the OpenCV interface to shogun toolkit.13:45
akp_Do we need a complete implementation here or simply training?>13:46
@wikingakp_: depends what u wanna do...13:47
akp_I am currently developing an interface for the dlib machine learning library.13:47
@wikingi guess you are then on the wrong channel mate13:48
@wikingthis is the shogun machine learning library channel13:48
akp_No. Sorry to mention the name of the library. But I think I can contribute to this library too.13:49
@wikingno worries13:49
@wikingthat's great13:49
akp_Before that13:49
-!- akp__ [67157f54@gateway/web/freenode/ip.] has joined #shogun13:51
akp__I just wanted toknow if shogun provides a basic interface for implementing SVMs13:52
-!- akp___ [67157e4c@gateway/web/freenode/ip.] has joined #shogun13:53
-!- akp_ [67157e4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]13:54
akp___I wanted to know if shogun provides interface for nonlinear svms13:54
akp___can you shed some light on that?13:55
@wikingit has everything there what u need to know about the interfaces13:55
-!- akp__ [67157f54@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]13:56
@wikinglisitsyn: btw ^13:56
akp___ignore my ignorance. What I think is a simple training interface using these non-linear svms13:56
@wikingakp___: as said... the documentation is quite extensive + examples13:57
@wikingi suggest check those13:57
akp___Well I think you misunderstood what I am saying. Apart from using the current features of shogun, I think we can integrate parts of OpenCV including feature extractio etc into shogun13:58
@wikingakp___: we do not really want to integrate parts of openc into shogun13:59
@wikingas shogun is a machine learning library13:59
@wikingwhilst opencv is rather an image processing library13:59
@wikingbut we'd like to have some example applications of opencv and shogun14:00
-!- cameo54321 [2729b2d3@gateway/web/freenode/ip.] has joined #shogun14:00
akp___yeah. That is what I am trying to say.14:00
@wikingsorry i did not get that...14:01
akp___ok. So apart from training, which can be used in examples such as ocr and image classification,14:02
akp___Many other examples can be demonstrated.14:03
@wikingyes indeed14:03
akp___I am working in the computer vision for past 6 months.14:04
akp___As i have said earlier, my notable implementations include a multi-object classifier and a text segmentation & recognition tool.14:05
akp___Having said that, can you say what specific applications of computer vision using your library are you looking for? Just some examples, or a specific example, like say a biometric application using14:07
akp___opencv and shogun14:07
@wikingakp___: well it's really up to you14:08
@wikingwe have not fixed any specifics ... some people already did eigenfaces examples14:08
@wikingbut it can be anything you like14:08
@wikingor you are interested in14:08
akp___ok, great!14:09
@wikingbut since there are like 2-3 students who started working on eigenfaces so that should be excluded from the possible options14:09
@wikingapart from that14:09
@wikingit's really up to u14:09
@wikingwe don't have limitations... but please make sure that if u start working on it that we expect you to create an ipython notebook out of that application14:10
akp___yeah. sure.14:11
akp___THe ipython should contain the appropriate prototype code at the time of proposal?14:12
akp___And apart from that,'14:13
akp___do we need to submit a demo application using OpenCV and Shogun at the time of proposal?14:14
akp___Can I discuss my idea with you in private?14:16
akp___chat I mean14:16
-!- akp_ [67157f4c@gateway/web/freenode/ip.] has joined #shogun14:18
@wikingwell for proposal you can use the mailing list14:18
@wikingand discuss there the idea14:18
@wikingwe don't expect you to create already a notebook for the idea... but you should start working on entrance issues of shogun to get yourself familiarised with shogun14:19
shogun-buildbotbuild #738 forced14:20
shogun-buildbotI'll give a shout when the build finishes14:20
-!- akp___ [67157e4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]14:20
akp_let me wrap up my idea. I'll get back to you after a while. Any problem?14:21
-!- cameo54321 [2729b2d3@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]14:21
@wikingakp_: use mailing list14:22
@wikingthat's read by everybody from the project so you'll get more feedback14:23
-!- akp__ [67157d4c@gateway/web/freenode/ip.] has joined #shogun14:23
-!- akp__ [67157d4c@gateway/web/freenode/ip.] has quit [Client Quit]14:24
-!- akp_ [67157f4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]14:25
-!- PirosB3 [~pirosb3@] has joined #shogun14:26
shogun-buildbotHey! build nightly_default #738 is complete: Exception [exception upload]14:28
shogun-buildbotBuild details are at
-!- shogun-buildbot [] has quit [Quit: buildmaster reconfigured: bot disconnecting]14:29
-!- shogun-buildbot [] has joined #shogun14:30
@wikingshogun-buildbot: force build --branch=develop 'nightly_default'14:30
shogun-buildbotThe build has been queued, I'll give a shout when it starts14:30
shogun-buildbotbuild #739 forced14:33
shogun-buildbotI'll give a shout when the build finishes14:33
shogun-buildbotbuild #1865 of cyg1 - libshogun is complete: Failure [failed compile test]  Build details are at  blamelist: Viktor Gal <>14:36
-!- dhruv13 [~dhruv13@] has quit [Ping timeout: 264 seconds]14:42
-!- travis-ci [] has joined #shogun15:06
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []15:06
shogun-buildbotHey! build nightly_default #739 is complete: Success [build successful]15:12
shogun-buildbotBuild details are at
-!- shogun-buildbot [] has quit [Quit: buildmaster reconfigured: bot disconnecting]15:19
-!- shogun-buildbot [] has joined #shogun15:19
shogun-notifier-shogun: lambday :develop * b6a9867 / doc/ipython-notebooks/logdet/logdet.ipynb:
shogun-notifier-shogun: fix failing input error in logdet notebook15:27
shogun-notifier-shogun: Viktor Gal :develop * 26aa304 / doc/ipython-notebooks/logdet/logdet.ipynb:
shogun-notifier-shogun: Merge pull request #1968 from lambday/develop15:27
shogun-notifier-shogun: fix failing input error in logdet notebook15:27
-!- iglesiasg [] has quit [Quit: Leaving]15:30
-!- PirosB3 [~pirosb3@] has quit [Quit: PirosB3]15:36
-!- HeikoS [] has joined #shogun15:37
-!- mode/#shogun [+o HeikoS] by ChanServ15:37
@wikingHeikoS: i'm just fixing documentation generation atm and i've just remembered last year you and i talking at TU-berlin to have ipython notebooks... it's fucking crazy how much of those we already have now15:41
@wikingas well as how crazy outdated the doxygen was...15:41
@HeikoSwiking: hey!15:42
@HeikoSwiking: yeah, the last year brought great improvements15:43
@HeikoSalso unit tests ...15:43
@HeikoSwiking: we should keep on pushing things :)15:43
@wikingHeikoS: tell me about lars15:44
@wikingHeikoS: what do u want there to be eigenized15:44
@HeikoSwiking: mainly I want to get rid of the un-readable lapack calls15:45
@HeikoSfor multiplications15:45
@HeikoSlike matrix-vector15:45
@HeikoSand matrix matrix15:45
@wikingHeikoS: so the agenda is that we get rid of lapack?15:45
@HeikoSwiking: I agree, we should do the factorisations via our own interface to exchange backends15:46
@HeikoSwiking: but for these lower level things, I think maybe lets just use eigen15:46
@HeikoSwiking: there have been many man ybugs caused by the confusing lapack api.15:46
@HeikoSwiking: with eigen, code is much more clear and therefore easier to verify15:46
@HeikoSwiking: lapack has all these parameters in the matrix-vector methods, people confuse them sometimes15:47
@HeikoSwiking: so its mainly readability15:47
@HeikoSwiking: also lapack is a pain to link against, eigen is much easier15:47
@wikingstill... there's magma which is exactly like lapack api ;)15:47
@HeikoSwiking: and I dont know whether it makes sense to overload the SGMatrix multiplication api15:47
@wikingso what r u going to do then?15:47
@HeikoSwiking: thats all hidden in the backend15:47
@wikingwell that's why i would do the same15:47
@HeikoSwiking: I want to have the algorithm code clean and easy15:48
@wikinglars should be engine independent15:48
@HeikoSwiking: do you also want that for low-level calls? such as matrix-.vector product15:48
@wikingif u have lapack u have lapack based lars15:48
@wikingif u have eigen u have that15:48
@wikingHeikoS: well on the long run...15:48
@HeikoSwiking: its very hard to have a unified interface for that15:48
@wikingHeikoS: think about doint dot product of a large matrix *vector in gpu15:49
@HeikoSwiking: makes things inefficient sometimes15:49
@HeikoSthink of transposes etc15:49
@HeikoSwiking: matrix-vector doesnt make sense on gpu15:49
@HeikoSwiking: maybe matrix matrix does15:49
@HeikoSwiking: but matrix matrix is alsready openmp'ed in eigen15:49
@HeikoSwiking: so I would go for that15:49
@wikingHeikoS: well say matrix is n x n15:49
@wikingand n is 1000000015:49
@HeikoSwiking: you cannot store that anyways15:49
@wikingHeikoS: well maybe shogun cannot15:50
@HeikoSwiking: the thing about the eigen3-ization also is that people wirte tests and review the code15:50
@HeikoSwiking: thats a good side effect15:50
@wikingbut come one a normal tesla can store amazing amount of data15:50
@HeikoSwiking: so I would vote for high level calls via our own backend15:50
@HeikoSwiking: and multiplications in eigen15:50
@HeikoSwiking: and maybe change that at some point once we have soe experience how the factorizations work15:50
@HeikoSwiking: moving from lapack to eigen already has many advantages (intro tasks, tests, bugfixes, much cleaner code)15:51
@wikingmmm well i just think that throwing out lapack is irresponsible15:51
-!- cameo54321 [2729ef55@gateway/web/freenode/ip.] has joined #shogun15:51
@wikingHeikoS: btw stan... dude that's really a mess15:52
@wikingHeikoS: nobody would like to have dependency on that ... not even u15:53
@wikingHeikoS: it has no good build system and everything is bundled15:53
-!- kprah [67157d4c@gateway/web/freenode/ip.] has joined #shogun15:54
@HeikoSwiking: wiking ok what about having some easy api for matrix-matrix and matrix-vector and do everything against that?15:54
@wikingHeikoS: SGMatrix has that15:54
@HeikoSwiking: Should not be part of SGMatrix but seperate SGMatrix is mainly a memory wrapper15:55
@wikingif not then it's a shame and should be implemented :D15:55
@HeikoSwiking: should be in linalg or so15:55
@wikingso like a static function15:55
@HeikoSwiking: although overloaded operators such as in eigen are much nicer15:55
@HeikoSwiking: what about transposes and stuff15:55
@wikingwhy not SGMatrix::operator*(sgmatrix)15:55
@HeikoSwiking: eigen does all that in place15:55
@wikingshogun-buildbot: force build --branch=develop 'nightly_default'15:56
@HeikoSwiking: mmh, I mean it would be *great* not to write against libs but against a general iunterface15:56
shogun-buildbotThe build has been queued, I'll give a shout when it starts15:56
@wikingHeikoS: i get that part15:56
@wikingand i'm up for that15:56
@HeikoSwiking: but I am also fine with only having high-level calls generalised15:57
@wiking(see the eigensolver story)15:57
@HeikoSwiking: since eigen is so good for low-level things15:57
-!- kprah_ [67157e4c@gateway/web/freenode/ip.] has joined #shogun15:57
@HeikoSwiking: and its alspo open-mp'ed for matrix-matrix15:57
@HeikoSwiking: we just have to activate that15:57
@HeikoSwiking: so maybe lets do that for now? Could do another project at some point doing the multiplications15:57
@wikingHeikoS: well if openmp is available then it should be already active.. or?15:57
@HeikoSwiking: one has to set number of threads in eigen /openmp or something15:58
@HeikoSthey have a howto page15:58
kprah_Hello guys. Anyone free?15:58
@HeikoSkprah, whats up?15:58
@HeikoSwiking: eigen3 also has views and all that, which is really helpful15:58
@HeikoSwiking: and lots of in-place magic15:58
@HeikoSthats very hard to put under a unified interface15:58
@HeikoSwiking: and most gain really is in high level calls15:58
kprah_The idea "LP/QP optimization framework caught my eye and I wanted to talk about that.15:59
@wikingkprah_: shoot15:59
@HeikoSwiking: vector products are so easy, that its not really a big deal with having different backends, and eigen3 works on parallel stuff15:59
kprah_I am a chemical engineering sophomore in my last semester.,15:59
@HeikoSkprah, for that you should try to catch one of the mentors, try writing to the mailing list and cc them15:59
@wikingHeikoS: well that's the thing that it does but not always the best solution...16:00
@wikingHeikoS: /me is the mentor ;)16:00
kprah_Previous semester in my winter internship, my work invloved16:00
@HeikoSwiking: haha ok then you deal with that :)16:00
cameo54321wiking: HeikoS thanks for bringing up the topic,16:00
@HeikoSwiking: I am always for step-wise improvements, so lets to the factorizations, that would already be amazing16:00
-!- kprah [67157d4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:00
@wikingHeikoS: we'll see how goes with the eigensolver unification16:01
kprah_usage of various solvers like bonmin and cplex to generate optimal prodUCTS FROM the given set of reactants.16:01
@HeikoSwiking: but I think those tasks to eigen3-ize things are good since they produce tests and get people involved :) ... We should however work against the new stuff for factorazations, yeah16:01
@HeikoScameo54321: what?16:01
@wikingkprah_: that sounds great16:01
kprah_I ended up using CPLEX and bonmin.16:02
kprah_My problem there was essentially a mixed integer quadratic programming problem.16:02
@wikingkprah_: in that task you would need to work a lot designing an interface that can be used regardless of the available solver backend16:02
@wikingso we could concentrate of course on a given problem as well16:03
@wikingbut the main idea is to do some unification of dealing with solvers16:03
@wikingas there's like kazzilion of them out there16:03
cameo54321I talked with wiking yesterday on bundling the stan issue.. Today I shifted more to exploring shogun's code, GP ipython notebook by roman and your kameleon mcmc implementation16:03
kprah_That depends a lot on the given constraints i.e. which solver you are using.16:03
kprah_One optimization solver that was one of the most principal one is OPTI toolbox16:04
kprah_it is open source and you can see it here. :
cameo54321@Heiko what if i start writing an ipython notebook for the kameleon mcmc by you?16:05
kprah_Basically it is a matlab toolbox. But it is very intelligent in itself and chooses the most appropriate solver depending on the condition that you are in16:06
kprah_I would like to port it to shogun , from matlab to c++.16:06
lisitsyndear HeikoS I would like to take an opportunity and participate in the project mentored by you, could you kindly tell me how to proceed next?16:07
-!- HeikoS [] has quit [Read error: Operation timed out]16:07
-!- kprah [67157d4c@gateway/web/freenode/ip.] has joined #shogun16:07
lisitsynkaboom I killed him!16:07
kprahwiking: what do you say?16:07
kprahAlso do we need mixed integer optimization here>16:08
@wikingkprah: sounds good but porting from matlab to c++ is always a bad idea16:09
-!- cameo54321 [2729ef55@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:09
kprahNo actually the code that the author wrote is actually just a wrapper.16:09
@wikingkprah: wrapper of?16:10
kpraha matlab wrapper for accessing all those solvers.16:10
-!- kprah_ [67157e4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:10
@wikingkprah: yeah that's what i'm saying as well16:11
@wikingit's a good start to have a look at it16:11
@wikingbut to use that in any way as a reference is a bad idea16:11
@wikingbecause you'll limit yourself and not use all the good stuff that c++ offers16:11
@wikinganyhow of course it's good to have an idea from it16:11
kprahWell actually there is a paper for it by the same author.16:12
-!- kprah_ [67157e4c@gateway/web/freenode/ip.] has joined #shogun16:13
-!- cameo54321 [2729ef55@gateway/web/freenode/ip.] has joined #shogun16:13
kprah_Well we can construct a similar wrapper here.16:13
kprah_Do you want to see the references?16:13
-!- cameo54321 [2729ef55@gateway/web/freenode/ip.] has quit [Client Quit]16:14
@wikingwell that's what i'm saying apart from having similar functionality it shouldn't be similar ;)16:14
@wikingkprah: since actually what we would like to achive is not only wrapper for all solvers16:14
-!- cameo54321 [2729ef55@gateway/web/freenode/ip.] has joined #shogun16:14
@wikingbut as well do create building blocks for LP/QP solvers16:14
@wikinglike KKT solving etc16:14
kprah_Yeah. It is there I think. We can construct it.16:15
@wikingsince there are several functionalities in a qp solver16:15
@wikingthat is the same16:15
@wikingregardless of the given QP solver...16:15
kprah_So what do you suggest finally then?16:16
@wikingand the idea was that one would create a reference cone LP/QP solver based on this framework..16:16
@wikingkprah: well in order to be able to do discuss something16:16
-!- kprah [67157d4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:16
@wikingplans should be made so we can iterate them16:16
@wikingbest of course to do these type of things on the mailing list16:16
@wikingsince then a lot of people can participate16:17
@wikingbut in the meanwhile you should as well get familiarised with shogun's code16:17
@wikingand start working on entrance tasks16:17
@wikingso that you get a grip of how development goes in shogun16:17
-!- kprah [67157d4c@gateway/web/freenode/ip.] has joined #shogun16:18
kprahI'll put up this on the mailing list then.16:19
kprahBTW you said cone programming16:19
-!- kprah_ [67157e4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:20
kpraha cone program is nothing a linear program with non linear constraints right?16:20
-!- kislay [~kislay@] has quit [Read error: Connection reset by peer]16:20
@wikingkprah: i mean one good example for this would to see how one could unify (reduce code redundancy) among libbmrm.cpp libp3bm.cpp libppbm.cpp in shogun16:20
cameo54321lisitsyn: can you please guide me on what happened recently?16:22
lisitsyncameo54321: ??16:22
cameo54321I mistakenly wrote @ with Heiko's name16:23
lisitsynah okay16:23
kprahSo you need a general conic solver?16:23
cameo54321at the same time a disconnection happened at my end16:23
cameo54321IRC logs say you killed somebody, was it me :-)16:25
kprahwiking : with any constraints and objectives?16:25
kprahwiking : I mean with unknown nature of constraints as well as the objective function?16:26
lisitsyncameo54321: no, that was heiko16:26
-!- kprah_ [67157d4c@gateway/web/freenode/ip.] has joined #shogun16:27
cameo54321lisitsyn Oh16:27
lisitsyncameo54321: just kidding nevermind ;) HeikoS will get back I guess16:27
shogun-buildbotbuild forced [ETA 38m51s]16:27
shogun-buildbotI'll give a shout when the build finishes16:27
lisitsynas he is sometimes having some bad connection16:28
-!- HeikoS [] has joined #shogun16:29
-!- mode/#shogun [+o HeikoS] by ChanServ16:29
shogun-buildbotbuild #1866 of cyg1 - libshogun is complete: Success [build successful]  Build details are at
kprah_wiking : so basically a cone program is a quadratic optimization problem with any given linear or non linear constraint, right?16:30
@wikingkprah_: are we talking about coneqp or conelp?16:30
cameo54321HeikoS Sorry I mistakenly wrote @ with your name16:31
-!- kprah [67157d4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:31
kprah_No both.16:31
kprah_It includes both16:31
kprah_a coneqp degenerates into a conelp if you set H = 016:32
kprah_that is a null matrix.16:32
kprah_It will accordingly change the algorithm used.16:33
-!- kprah [67157d4c@gateway/web/freenode/ip.] has joined #shogun16:33
@wikingHeikoS: u realise that eigenizing everything will be funny to use the shogun wrappers16:34
@wikingHeikoS: because then u will have to map back the matricies16:34
@HeikoSwiking: what do yo u mean?16:34
@wikingas obviously we dont want to support ::compute(Eigen::Matrix16:35
-!- kprah__ [67157f4c@gateway/web/freenode/ip.] has joined #shogun16:36
kprah__wiking : sorry, my connection with irc is a little unstable.16:36
@wikingkprah__: maybe mail?16:36
kprah__Sorry. Did you say anything before this?16:36
kprah__Yeah. We can try mail.16:37
-!- kprah_ [67157d4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:37
-!- kprah [67157d4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:37
@wikingHeikoS: so say u eigenize LARS. which means you map every matrix to an eigen::matrix irhgt?16:38
@HeikoSwiking: hey16:38
@HeikoSwiking: yes16:38
@wikingHeikoS: and then u say wanna call a QR factorization16:38
@HeikoSwiking: no QR should be done against our new interface16:38
@wikingyeah yeah16:38
@HeikoSjust for the algebra16:38
@wikingthat's what i'm talking about16:38
@wikingso when u did all the algebra16:38
@HeikoSwiking: eigen3 maps are O(1) to create16:39
@wikingand then u want to do QR16:39
@wikingthen u will have to map back16:39
@HeikoSand they directy modify the SGMatrix data16:39
@wikingok i'm just saying that it's a bit funky...16:39
kprah__wiking : so do you have anything to suggest? I am introducing myself on the mailing list.16:39
@HeikoSwiking: nono16:40
-!- travis-ci [] has joined #shogun16:40
travis-ci[travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun:
-!- travis-ci [] has left #shogun []16:40
@wikingkprah__: mmm just start the discussion about your idea16:40
@HeikoSwiking:  you dont map back16:40
kprah__ok. Thanks.16:41
@HeikoSwiking: you have an SGMatrix which you use to allocate memory, then you pass this to eigen which flls it16:41
@HeikoSwiking: or you have a matrix which you give to eigen and it modifies it16:41
@HeikoSwiking: so all good, thats why I like the lib so much16:41
shogun-notifier-shogun: = :develop * d6df009 / src/shogun/mathematics/linalg/ratapprox/logdet/ (2 files):
shogun-notifier-shogun: added constructors for LogDetEstimator16:43
shogun-notifier-shogun: Sunil Mahendrakar :develop * af28014 / / (3 files):
shogun-notifier-shogun: added constructors LogDetEstimator,unit tests for those and added get methods for private variables of LogDetEstimator16:43
shogun-notifier-shogun: Sunil Mahendrakar :develop * a4db061 / / (3 files):
shogun-notifier-shogun: fixed typos in #194016:43
shogun-notifier-shogun: Sunil Mahendrakar :develop * eb4ac93 / / (3 files):
shogun-notifier-shogun: added docs to #1940, unneccesary data from unit tests16:43
shogun-notifier-shogun: Sunil Mahendrakar :develop * 0ca5322 / / (3 files):
shogun-notifier-shogun: removed constructor for dense matrix16:43
shogun-notifier-shogun: Heiko Strathmann :develop * ce8102e / / (3 files):
shogun-notifier-shogun: Merge pull request #1940 from sunil1337/develop16:43
shogun-notifier-shogun: added constructors for LogDetEstimator16:43
cameo54321HeikoS would you like to discuss the kameleon mcmc code?16:44
-!- kprah__ [67157f4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]16:45
cameo54321I was thinking if it will be a good idea to write an ipython notebook for it/16:45
@HeikoScameo54321: I have to go now, but happy to discuss soon. Maybe already send some questions for the mailinglist?16:45
@HeikoScameo54321: yeah maybe, but keep in mind its not based on Shogun, so that it a point for after the MCMC framework is there16:46
@HeikoScameo54321: maybe think how the class structure can be translated to shogun16:46
cameo54321Ok Sure..16:46
cameo54321HeikoS: Last point, which class structure?16:48
@HeikoScameo54321: of the mcmc framework in kameleon mcmc and shogun-mcmc-prototype16:48
@HeikoSsee my repository16:48
cameo54321Ah ok.16:48
@HeikoScameo54321: running off, see you later16:48
cameo54321See you.. Thanks :-)16:49
-!- HeikoS [] has quit [Read error: Operation timed out]16:50
cameo54321lisitsyn: I still cannot believe it. The @ killed Heiko?16:51
cameo54321llistitsyn: It seems I'll have to learn how to use IRC..16:52
cameo54321Ok now I wrote the name wrong..16:53
-!- PirosB3 [~pirosb3@] has joined #shogun17:00
lisitsyncameo54321: no17:01
shogun-buildbotHey! build nightly_default #740 is complete: Success [build successful]17:04
shogun-buildbotBuild details are at
PirosB3hey guys, does anyone have a good example of init_shogun and exit_shogun written in python?17:08
@wikinglisitsyn: oh nice have u seen this:
lisitsynwiking: cool17:09
-!- kislay [~kislay@] has joined #shogun17:09
@wikinglisitsyn: got that tweet ? :D17:23
@wikinglisitsyn: seems somebody hacked his twitter acc17:24
lisitsynwiking: hmm yeah I guess this twitter is pwned17:24
-!- thoralf [] has joined #shogun17:25
thoralfHey :)17:25
-!- Jey [~chatzilla@2601:4:700:363:c507:abec:6e4:77d8] has joined #shogun17:26
-!- Jey [~chatzilla@2601:4:700:363:c507:abec:6e4:77d8] has left #shogun []17:27
-!- HeikoS [] has joined #shogun17:48
-!- mode/#shogun [+o HeikoS] by ChanServ17:48
thoralfHey Heiko17:49
@HeikoSthoralf: hey!17:49
thoralfHeikoS: Do you have a few minutes for discussing the Multilabel stuff?17:50
@HeikoSthoralf: yeah, good point17:50
@HeikoSjust going through the 50 emails today ;)17:50
@HeikoSbut almost done by noe17:51
@HeikoSthoralf: ok 2 mins, I might have to run off at some point17:51
@HeikoSthoralf: so about those arrays with dynamic length17:51
@HeikoSthoralf: this is a problem we often encountered, so it would be good to solve it somehow17:51
@HeikoSthoralf: because then things are serialisable etc17:52
@HeikoSthoralf: I think in fact there might already be a solution17:52
@HeikoSthoralf: this one also doesnt work for the modular interfaces17:52
thoralfOkay, first point: this array of vectors.  Can you give me a pointer?17:53
thoralfSecond point: About the modular interfaces, are you sure it doesn't work?17:53
thoralfOr what would prevent it from working in modular interfaces?17:54
@HeikoSthoralf: there is no datatype for those arrays17:59
@HeikoSthoralf: we can only map data that is possible to represent as shogun types18:00
@HeikoSthoralf: which means everything that you can register using SG_ADD18:00
@wikingHeikoS: "The '''compute(SGMatrix)''' method should just be a convenience wrapper for linear operators" why?18:01
@wikingHeikoS: since actually internally it's the other way round18:01
@wikingHeikoS: at least in case of DirectEigenSolver...18:02
@HeikoSwiking: ah for direct solvers it can be like that18:02
@HeikoSwiking: but then the direct solver has another interface than the sparse one18:02
@HeikoSwiking: mmh18:03
@HeikoSmaybe that should be like that even18:03
@HeikoSdense and sparse are two quite different cases18:03
@HeikoSwiking: for direct solvers, we assume that we can store the matrix18:03
@HeikoSwiking: but still, a linear operator is a more general object than a matrix18:03
@wikingyeah i got that18:03
@HeikoSwiking: and even direct eigensolvers can be applied to operators which are not matrices18:03
@HeikoSmatrices <= linear operators18:03
@wikingi mean i know that in case of an iterative solver it'll be better to have a linop18:03
@HeikoSwiking: but thats all not really needed18:04
@wikingHeikoS: yeah i've finished basic cs course back in the days of my bsc ;)18:04
@HeikoSwiking: haha ;)18:04
@HeikoSwiking: but maybe we should stick to conventions there18:04
@wikingHeikoS: well if we stick to linops18:04
@HeikoSwiking: which is: direct solvers take matrices, and sparse ones take sparse matrices or operators18:04
@HeikoSwiking: thats how matlab and numpy do it18:05
@wikingthen only the users will have to do18:05
@wikingwhich is of course a bit of a shit18:05
@wikingthat's why i've added those18:05
@wikingHeikoS: i mean what i'm trying to say is that obviously if the user is an idiot18:05
@wikingthat we cannot prevent :)18:05
@HeikoSwiking: yeah, its annoying if one actually *wants* to use a matrix, there should be methods for that, that hide all the linear operator crap18:06
@HeikoShaha ;)18:06
@wikingHeikoS: eyeah so i would go with18:06
@wikingEigenSolver has all sorts of pure virtual functions18:06
@wikingand then some of them are implemented and some not18:06
@wikingin the given implementation18:06
@wikingbut of course we can follow more strict rules18:07
@wikingof c++18:07
@HeikoSwiking: yeah18:07
@wikingand then only introduce compute(linop) in EigenSolver18:07
@HeikoSwiking: I mean we can have both right?18:07
@wikingand only in classs where it actually makes sense to do compute(SGMatrix) we introduce that18:07
@HeikoSthe direct eigensolver just can have another method18:07
@wikingHeikoS: yea hyeah sure18:08
@wikingHeikoS: just that then sometimes you'll have do do dynamic_cast18:08
@HeikoSwiking: yeah exactly, so if you know you are using a direct solver, you can use the compute (SGMatrix), but for things as lambday 's framework, we dont actually know, and need a common interface18:08
@wikingwhen u actually want to call that method that is only supported in a given child of EigenSolver18:08
@HeikoSwiking: yeah, but then you need to be sure what it is, which is fine18:08
-!- PirosB3 [~pirosb3@] has quit [Quit: PirosB3]18:08
@wikingi feel the former solution is more 'standard'18:09
@HeikoSwiking: prefer that a lot to having all methods in the base class and then get errors at runtime18:09
@wikingdo you?18:09
@wikingi'm just trying to say that i dont :D18:09
@wikingit's kind of awkward when there's an API but then you get a runtime error :S18:09
@wikingat least for me18:10
@wikinglisitsyn: ^18:10
@wikingbtw i think i cleared up some doxygen shit from the mid 2000s ;)18:10
@wikinglisitsyn: verdict?18:10
lisitsynwiking: I don't really get what's you are discussing18:11
@wikinglisitsyn: base abstract class has all the methods and then depending on the actual implementation some of those methods are not implemented vs abstract base class has only a few methods and then the inherited classes extends the api18:13
@HeikoSwiking: I think we should capture stuff at compile time18:13
@wikingin short18:13
lisitsynoh thanks18:13
lisitsynthat's really better ;)18:13
@HeikoSwiking: but its not the biggest deal on earth, more important to actually do this unification, however its done18:13
@HeikoSwiking: but run-time errors scare people, in contrast to compile errors which scare ourselves ;)18:13
lisitsynwhy do we need to have these non implemented methods?18:13
@wikingHeikoS: then the former one is the right way18:13
@wikingi mean latter18:14
@wikinglisitsyn: dont have to .. this is a design decision18:14
@wikinglisitsyn: you know one requires dynamic cast the other will just fucking fail runtime :)18:14
lisitsynwiking: what's the class we are talking about?18:15
@wikinglisitsyn: EigenSolver18:15
@wikinglisitsyn: MagicMachine2.018:15
@HeikoSwiking: I dont have hard feelings about either way, just a slight preference.18:16
@HeikoSwiking, lisitsyn how to deal with those datastructors of thoralf?18:16
@HeikoSarrays of different sized vectors18:16
@wikingHeikoS: std::vector? :)18:16
@HeikoSlisitsyn: you had a similar problem once right?18:16
@wikingor what?18:17
@HeikoSwiking: and interfaces?18:17
lisitsynHeikoS: yeah iirc18:17
@HeikoSwiking: how to pass this around?18:17
lisitsynHeikoS: actually fernando had it18:17
lisitsynand well I did some multilabel stuff as well18:17
@wikingHeikoS: if it's std::vector then it's easy T* &vector[0] :)18:17
@wikingHeikoS: gimmelinecode18:18
@wikingcan u tell me a file/line in the diff18:19
@HeikoSwiking: line 75 of MultilabelLabels.cpp18:20
@HeikoSwiking: yeah all that stuff18:20
@HeikoSwiking:  I dont like creating c -arrays . we moved to SG* for a reason18:20
@wikingoh yeah :D18:20
lisitsynargh we continue to do shoguncode18:21
@HeikoSlisitsyn, wiking it would be great if you joined the discussion in this PR18:21
@wikingHeikoS: DynArray?18:21
@HeikoSsince I dont really know how we should do all that stuff18:21
@HeikoSwiking: yeah that would be another solution18:21
lisitsynHeikoS: I am trying to follow this up18:21
@HeikoSand having different dimensions right?18:21
@wikingcan actually shogun/base/DynArray.h hold SGVector? :D18:22
lisitsyndynarray sgvector18:22
lisitsynsorry it is fuck o'clock18:23
@wikingnow i recall a moment18:23
@wikingwhen i had this shit as well :DDD18:23
@wikingbut cannot recall where was it actually18:23
@HeikoSwiking, lisitsyn gotta run off now, see you guys later18:25
-!- kprah [67157f4c@gateway/web/freenode/ip.] has joined #shogun18:27
-!- kprah_ [67157f4c@gateway/web/freenode/ip.] has joined #shogun18:30
-!- HeikoS [] has quit [Ping timeout: 264 seconds]18:30
kprah_wiking: where can I start?18:31
@wikingkprah_: i reckon u did not see my message there18:31
@wikingkprah_: r u familiar with bundle solvers?18:31
-!- kprah [67157f4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:31
-!- kprah__ [67157f4c@gateway/web/freenode/ip.] has joined #shogun18:32
@wikingkprah_: here are 3 different bundle solvers: libppbm, libbmrm and libp3bm they should share quite amount of code but they dont... so this could be a good start how to unify a bit the stuff there...18:33
-!- kprah___ [67157d4c@gateway/web/freenode/ip.] has joined #shogun18:33
kprah___i suppose yu are referring to solvers18:33
kprah___capable of solving multiple natured problems?18:34
-!- witness___ [uid10044@gateway/web/] has quit [Quit: Connection closed for inactivity]18:34
kprah___ignore the connection problems please.18:34
@wikingkprah_: i'm talking now about 3 given implementation in shogun18:34
@wikingthat should be worked on18:34
@wikingcheck the code of those solvers18:35
@wikingand see how you could unify them reducing code redundancy18:35
-!- kprah____ [67157f4c@gateway/web/freenode/ip.] has joined #shogun18:35
-!- kprah_ [67157f4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:35
kprah____wiking: am i right?18:35
@wikingkprah____: have u seen my messages?18:36
@wikingif not then i suggest i'll write an issue and u check it out there18:36
kprah____yeah, sure. But where am I supposed to find these issues?18:36
-!- kprah__ [67157f4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:37
@wikingu suppose to solve this issue :)18:37
-!- kprah [67157f4c@gateway/web/freenode/ip.] has joined #shogun18:37
-!- kprah___ [67157d4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:38
kprahwiking: sorry, but you didn't reply.18:39
@wikingkprah: i think i did18:39
@wikingmaybe during your bouncing18:40
@wikingyou've missed some of the stuff i've written18:40
@wikinghere's an entrance task18:40
@wikingwhy the fuck the analyzers are not being fired...:S18:40
-!- kprah____ [67157f4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:41
-!- shogun-buildbot [] has quit [Quit: buildmaster reconfigured: bot disconnecting]18:43
-!- shogun-buildbot_ [] has joined #shogun18:43
-!- kprah_ [67157e4c@gateway/web/freenode/ip.] has joined #shogun18:43
@wikingshogun-buildbot_: force build --branch=develop 'debian wheezy - memcheck'18:43
shogun-buildbot_build #212 forced18:43
shogun-buildbot_I'll give a shout when the build finishes18:43
-!- kprah [67157f4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:44
-!- kprah_ [67157e4c@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]18:47
thoralfwiking, lisitsyn: I missed how we can fix the issue Heiko had.18:53
thoralflisitsyn: Regarding the formatting issues: I used "bcpp" for this - If I fix this by-hand, then it will reverted the next time I call bcpp.18:54
lisitsynthoralf: no idea :D18:55
-!- KJainGSoC [783bc6ca@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:09
-!- thoralf [] has quit [Quit: Konversation terminated!]19:19
-!- shogun-notifier- [] has quit [Quit: transmission timeout]19:43
-!- iglesiasg [] has joined #shogun19:44
-!- mode/#shogun [+o iglesiasg] by ChanServ19:44
-!- newcastle [~newcastle@] has joined #shogun19:46
-!- cameo54321 [2729ef55@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:54
lisitsynbesser82: going to install f20 in a few minutes ;)19:56
lisitsynyou better hope I like it or I sue you!! :D19:56
-!- shogun-notifier- [] has joined #shogun19:58
shogun-notifier-shogun: Soeren Sonnenburg :develop * c6eb04c / doc/md/
shogun-notifier-shogun: remove german umlaut19:58
@sonney2klisitsyn, at least you no longer run around with your AK47 - besser you are pretty lucky ;)19:58
-!- newcastle [~newcastle@] has quit [K-Lined]19:58
lisitsynsonney2k: ohh welcome back :D19:59
lisitsynyou are apparently pretty rare here19:59
@sonney2klisitsyn, you know I am just a senior manager now20:06
lisitsynsonney2k: no, I don't know :)20:06
@sonney2kso I don't need to do any footwork ;)20:07
lisitsynsonney2k: yeah HeikoS and iglesiasg doing a lot of communication now20:07
@sonney2klisitsyn, I cannot say how glad I am that they / you all do this ;)20:09
-!- lisitsyn [] has quit [Quit: Leaving.]20:27
-!- lisitsyn [] has joined #shogun20:41
lisitsynbesser82: can't control display brightness! :D20:41
shogun-buildbot_build #212 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at
-!- sonne|osx [] has quit [Quit: sonne|osx]20:48
@iglesiasglisitsyn, apart from brightness, how does it feel?20:51
lisitsyniglesiasg: immediately rebooted back to ubuntu :D20:52
@wiking20:40 < lisitsyn> besser82: can't control display brightness! :D20:59
@wiking20:41 < lisitsyn> unacceptablee!!!!!!20:59
lisitsynwiking: as you like to say it is 201420:59
@wikingwell fuck it is21:00
lisitsynsatellites are evolving around the earth21:00
lisitsynand I still can't control brightness :D21:01
@wikinglisitsyn: brightness is overrated :D21:14
-!- tnosovich [~tnosovich@] has joined #shogun21:37
-!- tomy18 [~alex@] has joined #shogun21:40
-!- tomy18 [~alex@] has quit [Quit: Konversation terminated!]21:49
-!- thoralf [~thoralf@] has joined #shogun22:14
thoralfHey shogis.22:15
thoralf <-- can someone have a look on the PR of Piros?  He gets double-frees -- My suspect is using "exit_shogun()", since we don't use it in any of the python modular scripts...22:26
thoralfAny ideas?22:26
@iglesiasgI am off for today thoralf, sorry22:31
@iglesiasgtoo tired already22:31
thoralfNo problem.22:31
@iglesiasgI will read it tomorrow if it is still unsolved22:31
thoralfJust think if exit_shogun() is neccessary in python modular. ;)22:32
thoralfBecause I don't know.  The rest is up to Piros.22:32
@iglesiasgI have never seen/used it in a Shogun program from a modular interface22:32
@iglesiasgso I don't think it is necessary at all22:32
@iglesiasggood night!22:33
-!- iglesiasg [] has quit [Quit: Leaving]22:33
-!- shogun-notifier- [] has quit [Quit: transmission timeout]22:57
-!- kislay [~kislay@] has quit [Ping timeout: 240 seconds]23:07
-!- thoralf [~thoralf@] has quit [Ping timeout: 265 seconds]23:15
shogun-buildbot_build #213 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at  blamelist: Soeren Sonnenburg <>23:20
-!- HeikoS [] has joined #shogun23:37
-!- mode/#shogun [+o HeikoS] by ChanServ23:37
-!- HeikoS [] has quit [Client Quit]23:38
-!- HeikoS [] has joined #shogun23:39
-!- mode/#shogun [+o HeikoS] by ChanServ23:40
-!- lambday [67157f54@gateway/web/freenode/ip.] has joined #shogun23:44
lambdayHeikoS: hi!23:47
@HeikoSlambday: heya23:48
lambdayHeikoS: ummm I have a doubt23:48
lambdayHeikoS: in the B-test paper,23:48
@HeikoSlambday: yeah shoot23:49
lambdaythey show that the mmd estimates converges to Gaussian  - for both under alternative hypothesis (eq. 7, similar to the one for linear time mmd) and null hypothesis (eq. 10, similar to the one when we compute centered kernel matrices eigenvalues)23:50
lambdayHeikoS: which one should we have?23:50
@HeikoSlet me check23:50
@HeikoSlambday: null distribution23:55
@HeikoSthats the one we want23:55
@HeikoSso 1023:55
@HeikoSthe one with zero mean23:55
lambdayHeikoS: umm... but then we have to compute the whole kernel matrix, no?23:55
lambdayI mean, for streaming data23:56
lambdayit would be problematic23:56
@HeikoSshould not23:56
@HeikoSwhy do you think that?23:56
@HeikoSjust checking what C is23:56
lambdayjust above eq 1023:57
@HeikoSlambday: ah I see. so wait23:57
@HeikoSthese are population expressions23:57
@HeikoSbut we can estimate them in linear time23:57
@HeikoSthat what we have done in the kernel selection paper23:57
@HeikoSthere is a proof that if one constructs a test with a linear time estimate for the variance, this is correct23:57
@HeikoSyou dont need to udnerstand the proof (in fact, I dont really ;)23:57
@HeikoSlambday: but the estimation is very similar23:58
@HeikoSlambday: just an online variance estimate of the terms that are averaged over23:58
lambdayHeikoS: similar to the way linear mmd does it? estimate variance23:58
@HeikoSlambday: so the code should be almost the same as for the linear time mmd23:58
lambdayso that fits eq 7, doens't it23:58
lambdayoh no23:58
lambdaythe mean is different23:58
lambdaythere we have 0 mean23:58
@HeikoSlambday: how do they do it in their codes? constructing the threshold23:59
@HeikoSlambday: wait ill check23:59
lambdayHeikoS: var_est = (1 / m2) * cov(hh')23:59
--- Log closed Mon Mar 10 00:00:05 2014