Open in new window / Try shogun cloud
--- Log opened Sat Apr 07 00:00:15 2012
--- Day changed Sat Apr 07 2012
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]00:00
blackburnhey there00:02
blackburnanybody out there? :)00:02
-!- harshit_ [~harshit@] has joined #shogun00:03
-!- wiking [] has joined #shogun00:07
-!- wiking [] has quit [Changing host]00:07
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun00:07
-!- PhilTillet [] has quit [Remote host closed the connection]00:14
CIA-64shogun: iglesias master * rfbc2f8f / (src/shogun/classifier/KNN.cpp src/shogun/classifier/KNN.h): * convention and naming fixes -
CIA-64shogun: Soeren Sonnenburg master * r8d1fd5d / (4 files in 3 dirs):00:26
CIA-64shogun: Merge pull request #426 from iglesias/knn-covertree00:26
CIA-64shogun: CoverTree integration for KNN -
@sonney2kblackburn, look
@sonney2kheiko did it again :)00:27
blackburnkarlnapf: you rock00:27
@sonney2kkarlnapf, why does x-val get slower00:27
karlnapfsonney2k, blackburn, hi, thanks :)00:28
karlnapfsonney2k, because a subset is pushed in every fold, and then the active subset has to be recomputed00:28
karlnapfthat creates some overhead00:28
karlnapfesp for large number of folds, large number of repetitions compared to small amount of data (as in examples)00:29
blackburnwhat is overhead?00:29
@sonney2kkarlnapf, I would prefer 'add_subset' / 'remove_subset' just because we use the same terms with add_kernel / add_feature ...00:29
karlnapfbut should not be noticable in large scale real world examples00:29
karlnapfsonney2k, ok I can change this00:29
blackburnI thought this overhead is rather light00:29
@sonney2kkarlnapf, but isn't this a NOP when there is just 1 subset?00:29
@sonney2kI mean it doesn't need to recompute anything or?00:29
karlnapfoh yes, true00:30
karlnapfI should add that :)00:30
karlnapfcurrently it does (copy)00:30
karlnapfblackburn, I just noticed that for my examples it takes a bit longer00:31
@sonney2kkarlnapf, nice work!00:33
@sonney2kdo these 'fixes' and TODO's and merge at your own will :D00:33
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]00:33
karlnapfsonney2k, alright :)00:33
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun00:36
blackburnthat connection sucks00:36
blackburnhave I missed anything?00:36
blackburnI am curious whether commits can go to 2 github issues00:37
blackburnwith number # and pull request00:37
-!- PhilTillet [] has joined #shogun00:39
karlnapfblackburn, no i you did not miss anything :)00:39
karlnapfIll go to bed00:39
karlnapfblackburn, good night!00:39
karlnapfsonney2k, good night!00:39
blackburnkarlnapf: good night00:40
-!- karlnapf [] has left #shogun []00:43
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Quit: Page closed]00:54
-!- harshit_ [~harshit@] has quit [Remote host closed the connection]01:18
-!- pluskid [~chatzilla@] has joined #shogun02:23
shogun-buildbotbuild #452 of java_modular is complete: Failure [failed compile]  Build details are at  blamelist: fernando.iglesiasg@gmail.com02:30
shogun-buildbotbuild #448 of python_modular is complete: Failure [failed test_1]  Build details are at  blamelist: fernando.iglesiasg@gmail.com02:37
-!- PhilTillet [] has quit [Ping timeout: 252 seconds]02:50
-!- PhilTillet [] has joined #shogun02:52
-!- PhilTillet [] has quit [Ping timeout: 260 seconds]03:02
shogun-buildbotbuild #449 of python_modular is complete: Success [build successful]  Build details are at
-!- PhilTillet [] has joined #shogun03:36
-!- PhilTillet [] has quit [Ping timeout: 272 seconds]03:43
-!- pluskid [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120312181643]]03:51
-!- PhilTillet [~Philippe@] has joined #shogun03:55
-!- vikram360 [~vikram360@] has quit [Read error: Connection reset by peer]04:28
-!- PhilTillet [~Philippe@] has quit [Ping timeout: 272 seconds]04:44
-!- pluskid [~chatzilla@] has joined #shogun05:19
gsomixgood morning06:19
gsomixdamn, I was late for classes about complex analysis06:20
-!- n4nd0 [] has joined #shogun07:18
-!- iglesias [] has joined #shogun07:35
-!- n4nd0 [] has quit [Read error: Connection reset by peer]07:35
-!- n4nd0 [] has joined #shogun07:50
-!- iglesias [] has quit [Ping timeout: 260 seconds]07:53
-!- pythonroar [d2198538@gateway/web/freenode/ip.] has joined #shogun07:57
-!- n4nd0 [] has quit [Quit: leaving]07:57
-!- n4nd0 [] has joined #shogun07:57
-!- n4nd0 [] has quit [Client Quit]07:57
-!- n4nd0 [] has joined #shogun07:59
-!- pythonroar [d2198538@gateway/web/freenode/ip.] has quit [Quit: Page closed]08:20
n4nd0sonney2k: around?08:23
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]08:40
-!- n4nd0 [] has quit [Quit: leaving]09:23
-!- harshit_ [~harshit@] has joined #shogun09:39
-!- n4nd0 [] has joined #shogun10:30
-!- n4nd0 [] has quit [Client Quit]10:32
-!- gsomix [~gsomix@] has joined #shogun10:47
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]10:54
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]10:56
-!- gsomix [~gsomix@] has joined #shogun10:56
-!- wiking [] has joined #shogun11:00
-!- wiking [] has quit [Changing host]11:00
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun11:00
-!- blackburn [~qdrgsm@] has joined #shogun11:42
-!- harshit_ [~harshit@] has quit [Ping timeout: 240 seconds]11:45
-!- harshit_ [~harshit@] has joined #shogun11:57
wikingheheh it's kind of quite here since gsoc application ended :P12:17
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection]12:17
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun12:17
-!- n4nd0 [] has joined #shogun12:20
n4nd0sonney2k: hey! thank you for your answer, but I didn't get any qda file in data/testsuite/tests12:21
n4nd0maybe there is something wrong in my configuration12:21
wikingn4nd0: have you checked out the submodule?12:29
wikinggit submodule init && git submodule update12:29
n4nd0wiking: the only thing I have done related to the submodule is12:34
n4nd0git submodule update --init12:34
n4nd0I did that in my shogun local repository in order to set up the data directories12:34
wikingwell that should do it12:34
-!- PhilTillet [~Philippe@] has joined #shogun12:34
n4nd0oh wait12:35
n4nd0I have these files12:35
wikingand there's nothing in data/testsuite/tests ?12:35
wikingthere should be at least 276 files12:35
n4nd0where . is shogun/data/testsuite12:36
n4nd0I understood it wrong, I was looking into my shogun-data repository12:36
wikingif not more since i haven't updated my repo for a while...12:36
n4nd0so it looks like shogun/data is another repository right?12:37
n4nd0my git understands it like that at least12:37
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds]12:39
n4nd0how should I update this repo then?12:39
n4nd0I mean, upstream and origin can be used in the same way?12:39
n4nd0gtg, I will try to find more about this later, thank you!12:41
-!- n4nd0 [] has quit [Quit: leaving]12:41
@sonney2kuhh record number of students applied
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun12:51
PhilTilletThat's a lot12:53
blackburn>>> (48.0/6685)*110012:56
PhilTilletsonney2k, I forgot to mention in my pull request : ViennaCL only supports float and double right now (will probably evolve soon :p), that is why there is that viennacl_compatibility.h header :p12:58
blackburnsonney2k: first allocation should be 7-8 then :)12:59
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Ping timeout: 276 seconds]13:01
-!- harshit_ [~harshit@] has quit [Remote host closed the connection]13:08
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun13:13
-!- wiking_ [] has joined #shogun13:25
-!- wiking_ [] has quit [Changing host]13:25
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun13:25
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 248 seconds]13:28
-!- wiking_ is now known as wiking13:28
-!- gsomix [~gsomix@] has quit [Ping timeout: 246 seconds]13:42
-!- PhilTillet [~Philippe@] has quit [Ping timeout: 240 seconds]13:46
wikingmmm i really wonder if it's a good idea to use viennacl for opencl support :(13:54
-!- gsomix [~gsomix@] has joined #shogun13:55
-!- PhilTillet [] has joined #shogun13:58
-!- PhilTillet [] has quit [Ping timeout: 252 seconds]14:03
-!- muddo [~muddo@gateway/tor-sasl/muddo] has quit [Quit: Leaving]14:11
-!- muddo [~muddo@gateway/tor-sasl/muddo] has joined #shogun14:11
blackburnwiking: what are your thoughts on that?14:20
wikingi mean opencl it self has a standardized api14:20
wikingjust like opengl14:21
blackburnah yes14:21
wikingso currently i really don't see way is it necessary to add an extra abstraction layer to it14:21
wikingi mean yes14:22
wikingit for sure makes easier the implementation14:22
wikingbut then it's just another library dependency u will need if u want support for it14:22
blackburnyes I agree14:23
blackburnwiking: did you submit a proposal in time? ;)14:27
-!- pluskid [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120314124128]]14:57
-!- n4nd0 [] has joined #shogun15:04
n4nd0blackburn: how is it going?15:05
blackburnfine but getting busier15:05
n4nd0university, job?15:06
blackburnyeah I have to do some university stuff15:06
-!- PhilTillet [~android@] has joined #shogun15:06
n4nd0blackburn: I see, I was going to ask you about this data submodule stuff, but it can wait so no problem ;)15:06
blackburnn4nd0: no, ask :)15:06
n4nd0blackburn: so I've noticed that after doing this git submodule update --init15:07
n4nd0blackburn: I get like another repository in shogun/data15:07
n4nd0blackburn: I mean another repository because when I am in another directory of the project I see the branch where I'm (qda) and when I am under shogun/data I see another branch's name15:08
blackburnit is another repo15:08
n4nd0so how should I manage it?15:09
n4nd0for example this morning I did with tester.py15:09
n4nd0I generated a new example for qda15:09
blackburnprobably you'd need to fork it too15:09
blackburnsounds crazy I know15:09
n4nd0I see now a couple of new files classifier_qda_modular0.txt15:10
n4nd0and the same with modular115:10
n4nd0mmm but actually I forked shogun-data15:10
n4nd0I have it locally15:10
n4nd0but it is in another directory in my machine than the one I was telling you15:10
n4nd0in one placed I hace the forked I did of shogun-data15:11
n4nd0and inside shogun I have this shogun/data/ repository15:11
n4nd0I don't know if that really makes sense15:11
blackburnprobably you need to do PR with your shogun-data fork15:11
n4nd0but in my shogun-data fork it is not where I can see these new files+15:12
n4nd0this is getting weird :P15:12
blackburnjust put your files to your fork15:12
blackburnonce your PR is merged I'll update data hash in main shogun repo15:13
blackburnand it would be ok15:13
n4nd0I'll move the files and do the PR now, first I'll rebase and that stuff15:14
n4nd0btw I checked the contents of these files15:14
n4nd0really cryptic15:14
CIA-64shogun: Sergey Lisitsyn master * ra553686 / src/shogun/classifier/svm/NewtonSVM.cpp : Proper memory handling for #409 -
blackburnyeah but it would be worse to have data in the main repo15:15
wikingblackburn: heheh yes i did :)15:16
blackburnthat's nice then15:16
n4nd0blackburn: oh sorry, I think that was confusing ... I meant that the contents of these files (classifier_qda_modular[01].txt) I generated for the testsuite are really weird15:18
n4nd0blackburn: I cannot understand them at least15:18
n4nd0I suppose it has something to do with serialization15:18
blackburnn4nd0: is it similar to others ?15:18
n4nd0let me check15:19
n4nd0blackburn: yeah, it is as weird as classifier_knn_modular0.txt for example15:20
blackburnall ok then15:20
n4nd0blackburn: do you know sth about how the contents of these files work?15:20
n4nd0S?ren told me yesterday that it would be useful if I work with the testsuite, trying to fix the errors15:21
blackburnn4nd0: yes it is pickle15:21
n4nd0blackburn: python serialization right?15:21
n4nd0the thing is that with S?ren's suggestion, when some test is wrong, I execute the tester like this15:22
n4nd0./ -d wrong_test.py15:22
n4nd0that basically shows me the contents of those files15:22
n4nd0I don't know how one is supposed to find the errors from the contents there15:22
blackburnno way15:23
blackburndid he suggest to compare the outputs? he was pretty tired probably hah15:24
-!- PhilTillet [~android@] has quit [Read error: Connection reset by peer]15:24
blackburnn4nd0: ok if seriously you probably would need to read this stuff and compare parameters15:24
n4nd0blackburn: but I think that stuff is not readable at all :)15:25
n4nd0blackburn: check for example these new ones
blackburn-d should show some detailed sutff15:26
n4nd0that is what I get when I do -d15:26
n4nd0without -d is just ERROR15:27
blackburnsomething is wrong there then15:27
blackburncheck compare_dbg function15:27
blackburnI see15:28
blackburnline 57 is kinda stupid15:28
blackburnn4nd0: ok I think we need to change examples that return SGObjects15:29
blackburnthat makes no sense to me to compare that15:29
n4nd0blackburn: it makes no sense to me either15:30
n4nd0blackburn: also I think that something must be missing because S?ren told me sth like15:30
blackburnn4nd0: but numpy array compare is ok15:30
n4nd0with -d you can do debu15:30
n4nd0debug sorry15:30
n4nd0with print a you can print object a15:30
n4nd0with print b you can print object b15:30
n4nd0I have not entered in that kind of debug mode though :(15:31
blackburnyeah reading pickle is kind of mastery15:32
-!- PhilTillet [~android@] has joined #shogun15:37
-!- gsomix [~gsomix@] has quit [Ping timeout: 245 seconds]15:41
-!- gsomix [~gsomix@] has joined #shogun15:43
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]15:51
-!- n4nd0 [] has quit [Quit: leaving]15:56
-!- PhilTillet [~android@] has quit [Ping timeout: 252 seconds]15:57
-!- PhilTillet [] has joined #shogun16:19
PhilTilletblackburn, I think there is still a "float64_t" cast in the buggy pull request i sent a few weeks ago ( CMath::dot(float32_t*,float32_t*,int32_t))16:22
blackburnreally? why?16:23
PhilTilletI don't know, I rebased the upstream but there was still a float64_t cast16:23
PhilTilleteven though dsdot was back to sdot16:24
blackburndue to performance reasons we changed it back to sdot16:27
blackburnso what is the problem? :)16:27
PhilTilletint he for loop there is still the float64_t cast16:27
PhilTilletwhich halves performances :p16:27
blackburndo you think it should be removed?16:28
PhilTilletyes, I tested without it and peformances for float doubled :p16:29
PhilTilletI think multiplying a float with a double is a bad idea with SSE16:29
blackburnhowever on my machine blas is used probably16:30
PhilTillethmm, you probably have #define HAVE_LAPACK16:30
CIA-64shogun: Sergey Lisitsyn master * rca1f345 / src/shogun/mathematics/Math.cpp : Performance fix for single precision dot operation -
PhilTilletthat was a fail pull request of mine :D16:31
-!- gsomix [~gsomix@] has left #shogun ["????? ? ?? ??? (xchat 2.4.5 ??? ??????)"]16:46
-!- gsomix [~gsomix@] has joined #shogun16:49
-!- harshit_ [~harshit@] has joined #shogun16:50
-!- puneet [~puneet@] has joined #shogun16:53
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]17:14
-!- blackburn [~qdrgsm@] has joined #shogun17:18
-!- puneet [~puneet@] has quit [Ping timeout: 260 seconds]17:22
harshit_blackburn: hi !17:32
harshit_having some problems with CDotFeatures ..17:32
harshit_in implementing lbp17:32
-!- harshit_ [~harshit@] has quit [Quit: Leaving]17:34
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun18:02
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]18:10
-!- gsomix [~gsomix@] has joined #shogun18:10
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]18:21
-!- wiking [] has joined #shogun18:29
-!- wiking [] has quit [Changing host]18:29
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun18:29
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 252 seconds]18:42
-!- harshit_ [~harshit@] has joined #shogun18:59
harshit_sonney2k: around ?19:03
-!- blackburn [~qdrgsm@] has joined #shogun19:45
-!- n4nd0 [] has joined #shogun20:06
@sonney2kblackburn, I just learned that scikits-learn got 3 proposals20:06
n4nd0I am taking a look to the seg fault someone pointed out in the mailing list if you are not with it guys20:07
@sonney2kharshit_, now yes20:07
n4nd0sonney2k: they just got three proposals!?20:07
@sonney2kyes I guess these 3 are of high quality thouh!20:07
@sonney2kso everyone will be in20:07
n4nd0yeah sure ... but I don't know20:08
@sonney2kI would say we have 10-15 HQ students - I am still not through the whole list20:08
n4nd0what does HQ stand for? high quality maybe?20:08
CIA-64shogun: Soeren Sonnenburg master * r8b6902c / examples/undocumented/python_modular/ :20:09
CIA-64shogun: Merge pull request #429 from iglesias/qda20:09
CIA-64shogun: QDA for python_modular's testsuite -
@sonney2kn4nd0, did you understand a bit better how the testsuite is supposed to work?20:10
n4nd0sonney2k: well, just a little bit20:10
n4nd0sonney2k: I still don't really understand how one is supposed to fix the tests that are wrong :S20:11
harshit_sonney2k: Need to know how CDotFeatures operates internally ,20:11
harshit_As in I am implementing LBP features, so when user creates ClbpFeatures object by giving images and other argument, Then do I need to compute corresponding LBP feautes for those images there and then only ?20:11
harshit_or create LBP feature representation for one image when user request it .20:12
@sonney2kn4nd0, the way I do it is run -d bad_test.py20:13
n4nd0sonney2k: but the output there is quite unreadable20:13
@sonney2kthen when things are not the same I have these two lists a and b to investigate20:13
n4nd0sonney2k: do you work with that kind of format?20:13
@sonney2kwell I can see what is different20:13
@sonney2kif things are numpy arrays I can compute the deviation, e.g. numpy.max(numpy.abs(a[1]-b[1]))20:14
@sonney2kor so20:14
@sonney2kthe only thing where it is difficult is SGObjects20:14
n4nd0sonney2k: oh ok20:14
@sonney2kfor these I do pickle.dumps(a[1]) etc20:14
@sonney2kand then have a general llok20:14
@sonney2kfor SGObjects one can see though which arrays should be the same20:15
n4nd0sonney2k: let me show you just in case,
@sonney2kso for example if a[0] and b[0] are two objects derived from CSVM I can check if they are the same by computing the objective20:16
n4nd0sonney2k: that is the kind of output one should wait for?20:16
@sonney2kwell not the same actually - they might deviate slightly due to different numerical libs etc20:16
@sonney2kso I mark this as ok and run badscript.py20:16
n4nd0I see20:17
@sonney2kwait how do you get that output20:17
n4nd0./ -d classifier_larank_modular.py20:18
@sonney2k(but answer is yes if you do p a[0] and a[0] is an sgobject)20:18
n4nd0I choose that file because detects it as ERROR20:18
@sonney2kdammed some import pdb / pdb.set_trace() is missing20:19
n4nd0that is probably why I didn't see this interactive menu you told me about20:20
n4nd0to do print a, print b and so on20:20
@sonney2klooks like when I added these -m etc flags I did trash the pdb stuff20:20
harshit_sonney2k: In matlab example in libsvmocas for lbpfeatures, what is the format of image matrix, row major or column major20:21
@sonney2kharshit_, things are already halve in shogun btw LBPPyrDotFeatures.h20:22
n4nd0harshit_: AFAIK images coordinates are normally (0,0) for the top left corner and (M,N) for the bottom right20:23
@sonney2kharshit_, I am not sure - better ask vojtech about it (he is the author)20:23
@sonney2kn4nd0, yeah but you never know ...20:23
n4nd0harshit_: do you know if the images represent something you can see?20:24
harshit_sonney2k: okay i'll ask vojtech about format of it .. but can you tell me how exactly you want lbppyrDotFeatures need to be rebuilt20:25
n4nd0harshit_: if so, you could try to represent them with imagesc20:25
harshit_n4nd0: they are in a mat file20:25
n4nd0harshit_: that could give you a clue about the coordinates20:25
n4nd0harshit_: I meant if they are images of, idk, of horses for example20:25
harshit_500 is the no of features20:25
n4nd0harshit_: ok, so we are talking about feature vectors no raw images20:26
harshit_I dont know whether they are real images are not, have to ask vojtech20:26
CIA-64shogun: Soeren Sonnenburg master * r418a6b5 / testsuite/python_modular/ : enter debug mode for tester when -d is specified -
@sonney2kn4nd0,  now have a look again20:26
@sonney2kn4nd0, actually this particular svm you are testing is probabilistic...20:27
@sonney2kso it is no wonder this test fails...20:27
n4nd0sonney2k: aham! that would be make it more difficult to get the same output ...20:28
n4nd0yeah, now I get the pdb promt, cool20:28
@sonney2kn4nd0, so now you could do p a[0].get_labels()20:28
@sonney2kand see the values20:28
PhilTilletHello everybody :)20:29
n4nd0sonney2k: thank you .920:30
PhilTillet.9 is a pretty neat smiley :p20:30
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]20:38
-!- gsomix [~gsomix@] has joined #shogun20:38
blackburnsonney2k: yeah I have seen that in twitter of someone from scikits team20:39
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]20:41
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun20:44
-!- n4nd0 [] has quit [Read error: Connection reset by peer]20:47
-!- n4nd0 [] has joined #shogun20:48
@sonney2kif we get only 3 slots we are in serious trouble20:49
@sonney2kanyways we will know on april 1120:49
blackburnno way20:49
blackburnapril 11?20:49
blackburnsonney2k: why they would give us less than last year?20:50
@sonney2kapril 11 end of workday / pacific time20:50
blackburnwow they are pretty fast20:50
n4nd0sonney2k: will you let us know how many we get?20:50
blackburnhowever there is a waiting list20:51
@sonney2kit might be apr 12 though20:51
blackburnthat would become nervous if he didn't say that20:51
n4nd0I hope at least for 520:51
n4nd0you got 5 last year and everyone succeeded20:51
n4nd0I think it's fair20:51
@sonney2kI can actually enter how many amazing proposals we have (the pure minimum of #slots we want)20:51
@sonney2kand how many we would want if there were inf slots available20:52
blackburnsonney2k: enter to?20:52
@sonney2kproblem is that I am not through all the proposals yet so it is hard to say20:52
@sonney2kenter to melange20:52
blackburnare there special fields for that?20:52
gsomixsonney2k, and
blackburnsonney2k: and what do you think is a minimum?20:53
@sonney2kindependent of that each mentor can score students and there is some (hidden) discussion going on among mentors about students20:53
gsomixhi all20:53
@sonney2kblackburn, I don't know yet I would guess 7-920:54
blackburnsonney2k: heh and maximum is 9?20:55
@sonney2klets be realistic20:55
blackburnsonney2k: your opinion on that is changed ;)20:55
@sonney2kwe only have 10 mentors20:55
@sonney2kblackburn, what do you mean?20:55
blackburnsonney2k: yesterday you told me you think chances are we get 7-8 slots or so20:56
blackburnand now you think about 3 even :)20:56
@sonney2kno I don't think about 320:56
blackburnokay ;)20:57
@sonney2kI am just saying 3 would be catastrophic20:57
blackburn4,5 as well20:57
@sonney2kthen it is like rolling a dice to chose one of the very good students20:57
@sonney2kanyway since they will announce on apr 11 - I am under high pressure to actually get mentors to do sth over easter vacations20:58
@sonney2knot easy - I was hoping we have the tuesday too20:58
blackburndo you really need these rates? I mean you would have to choose ideas rather than students applying for the same idea20:59
n4nd0blackburn: btw did you see mailing list? pluskid suggested about using smart pointers as we talked about21:01
blackburnn4nd0: yes21:01
n4nd0sonney2k: what do you think about it?21:01
blackburnI have no opinion/expertise on that21:01
n4nd0blackburn: it's ok to be gay and use garbage collection :)21:02
blackburnn4nd0: too gay21:02
n4nd0blackburn: I'll never forget that man21:02
blackburnn4nd0: that video?21:02
blackburnhave you seen azis btw?21:02
n4nd0blackburn: both the video and your answer :D azis?21:02
blackburnoh you haven't seen that21:03
blackburnok a min21:03
blackburnthat's probably the most gay video in the universe21:03
gsomixazis *__*21:03
n4nd0talking about see, since we are in easter I think we all deserve a rest and watch star wars21:03
n4nd0telnet towel.blinkenlights.nl21:03
blackburnah yeah I remember that21:04
n4nd0blackburn: that video looks pretty hardcore, not a good moment to see it in the living room with other people .P21:04
blackburnn4nd0: hahahah lol21:04
n4nd0wtf is wrong with my keyboard today21:04
n4nd0that's better21:05
blackburnprobably not the best video to show gf's family21:05
blackburnoh lolz21:05
blackburnI imagine how could they react21:05
gsomixblackburn, have you seen the clip `Hop'? :321:06
blackburnis it as gay as azis is?21:08
blackburngsomix: no, I can't remember that21:08
blackburnI prefer hardcore gay video21:08
PhilTilletblackburn, like justin bieber?21:09
blackburnPhilTillet: exactly I am a belieber lol21:09
PhilTilletomg Azis.21:10
PhilTilletI did not know about him.21:10
blackburngsomix: while moving upstairs 15 minutes before I was thinking how to overload you a little more21:10
gsomixblackburn, and..?21:11
blackburngsomix: do you think it would be interesting for you to get shogun well-known in the .ru zone?21:12
blackburnPhilTillet: lucky21:12
PhilTilletwould be great if one day one gay commit had a sha1 starting with F4hQ4J4BFOM21:13
blackburnprobably it is the task for me but it would be useful for you21:13
gsomixblackburn, what do you mean? site? advertising?21:13
blackburngsomix: e.g. there is no russian page on Shogun in wiki21:14
blackburnhowever python3 stuff and other are in prior21:14
blackburngsomix: habrahabr would work as well :)21:15
gsomixblackburn, hmm. I can translate article from English.21:16
@sonney2kgoogle translate can too :D21:16
blackburnok let me describe my intention21:16
blackburnbut I am already used to some writing21:16
CIA-64shogun: Soeren Sonnenburg master * rc835442 / (2 files):21:17
CIA-64shogun: convert the float32 based dot function to return float32 - no need for21:17
CIA-64shogun: float64_t inflasion -
blackburnI can do that for sure21:17
blackburnbut you are not yet so it would be useful21:17
blackburnsonney2k: that function has been popular last days21:17
* PhilTillet feels embarassed21:18
blackburngsomix: so what do you think?21:18
gsomixsonney2k, Hulk sad. However, I translate better than I write. :)21:19
blackburngsomix: the more you do things the better you do it21:19
blackburnyear ago sonney2k usually had a headache while understanding my english21:20
gsomixblackburn, can we discuss details in private?21:21
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Quit: Page closed]21:21
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun21:22
blackburnbad connection21:22
blackburnsonney2k: I think we need to establish that # referring in github commits21:26
blackburnhowever risk is that we could forgot to refer issues too often21:27
blackburnone way to avoid this is to restrict commits w/o #21:28
blackburni.e. any commit should go to some issue21:28
blackburnn4nd0: what do you think?21:28
n4nd0blackburn: do you mean to use symbols for the commits?21:29
blackburnn4nd0: no, e.g. all doc things should go to #40821:30
blackburnonce you add #408 to the commit msg21:30
blackburn#418 sorry21:30
blackburnit goes to the issue21:30
n4nd0blackburn: ok, but later the commits can be added as usual, using PRs?21:31
blackburnn4nd0: could you please fix some doc and commit with #418?21:32
blackburnprobably it is only for me, Soeren and heiko commits21:32
n4nd0I should go something like21:32
blackburnI am curious whether it will appear in both PR and issue21:32
n4nd0fix doc #418 right?21:32
blackburnyeah something blablba #41821:33
n4nd0blackburn: done, in my branch qda21:36
n4nd0blackburn: yeah, it looks like it is included in Documentation improvements21:38
n4nd0I think that is cool21:38
blackburnthat's nice21:39
n4nd0but anyway21:39
n4nd0how does this work to merge later?21:39
blackburnit does nothing with merge/etc21:39
blackburnjust groups commits21:39
n4nd0jo the method is the same21:39
n4nd0do pull requests apart21:40
n4nd0I like it21:40
blackburnme too21:41
blackburnI will suggest it using the ML21:41
-!- PhilTillet [] has left #shogun ["Leaving"]21:42
-!- PhilTillet [] has joined #shogun21:42
shogun-buildbotbuild #446 of ruby_modular is complete: Failure [failed test_1]  Build details are at  blamelist: fernando.iglesiasg@gmail.com21:43
blackburngsomix: already? nice work21:44
blackburngsomix: is it a proof that it works?21:47
gsomixblackburn, yep.21:47
blackburnI believe it works ;)21:51
blackburngsomix: so you need next task?21:53
gsomixblackburn, yep.21:53
blackburngsomix: you prefer hardcore tasks, right? ;)21:56
gsomixblackburn, yep. for example, a researching of this:
blackburnI've never heard of your alco addiction ;)21:57
gsomixblackburn, hehe. it's a joke.21:58
blackburnmine addict is not ;(21:58
blackburngsomix: I know that custom kernel is wrong22:01
blackburnexact problem is in triangular from full22:02
blackburnor vice versa22:02
PhilTilletblackburn, I'm looking for some  tasks too. Would it be a good idea to implement more Kernels (like most of the dot kernels) in my opencl pull request?22:11
blackburnPhilTillet: I think not, do you mind doing something not opencl related?22:14
PhilTilletI don't mind22:14
PhilTilletI'd rather do something C++ related though22:14
blackburnyeah sure22:14
harshit_blackburn: Where is RealFeatures used in octave implemented in libshogun ?22:15
harshit_cant find any thing similar in libshogun22:15
blackburnharshit_: it is SimpleFeatures<float64_t>22:15
harshit_so if i have to refer SimpleFeatures<int32_t> then what do i need to use ?22:16
harshit_Is there any file where all these conversions are listed ?22:17
blackburnline 27622:17
harshit_okay thanks :)22:17
harshit_dam how can you remember line no too !!!22:17
harshit_you really are into shogun, must say :)22:18
blackburnI took a look22:18
blackburnno, I can't remember that stuff too :)22:18
harshit_but still you know alot about it :)22:19
blackburnI have to heh22:20
PhilTilletso, blackburn, any idea? :p22:21
blackburnPhilTillet: yeah thinking22:23
PhilTilleti couldn't find any size function for EFeatureType22:23
blackburnany preference?22:23
PhilTilletnot really22:23
PhilTilletyour question was kind of broad :p22:24
PhilTilletI can work on design issues, safety or some algorithm22:24
PhilTilletor implement garbage collection everywhere (who said troll? :p)22:25
blackburnknn regression?22:25
PhilTilletwhy not, i was more used to knn classification but it's ok I guess22:26
blackburnnearest centroid22:26
-!- n4nd0 [] has quit [Quit: leaving]22:26
PhilTilletboth seem fine22:27
shogun-buildbotbuild #202 of nightly_none is complete: Failure [failed compile]  Build details are at
harshit_blackburn: I can not use UIntFeatures, octave gives an error: `UIntFeatures' undefined near line 8 column 1322:30
harshit_but IntFeatures works fine22:30
PhilTilletblackburn, so let's say nearest centroid?22:32
blackburnPhilTillet: sure22:33
@sonney2kblackburn, I know you can trivially fix this issue
@sonney2kso please do :)22:33
blackburnsonney2k: git is currently unavailable for me22:33
PhilTilletguess I got to find inspiration on knn :)22:33
blackburnpublic access point and 22 is blocked22:33
@sonney2kblackburn, what happened?22:33
@sonney2kblackburn, then tell me what I have to do to fix this!22:34
blackburnjust add KNN_COVERTREE_POINT to ignores22:34
PhilTilletmetaprogramming errors22:35
PhilTilletinstantiated from ...22:35
-!- harshit_ [~harshit@] has quit [Quit: Leaving]22:36
@sonney2kblackburn, why does this error not occur w/ eg ISOMAP_COVERTREE_POINT22:37
blackburnsonney2k: isn't it ignored?22:37
blackburnsonney2k: the only difference is DOXYGEN_SHOULD_SKIP_THIS :D22:38
@sonney2kgsomix, why did you mark your patch work in progress?22:39
@sonney2kblackburn, that cannot be it22:39
gsomixsonney2k, a habit.22:40
blackburnsonney2k: how does swig select classes to include?22:40
@sonney2kblackburn, ahh I know - you define the class  in the .cpp file, n4nd0 in .h22:40
shogun-buildbotbuild #201 of nightly_all is complete: Failure [failed compile]  Build details are at
blackburnI just came over it as well22:41
@sonney2kgsomix, did you do a test? I mean set the data and get it and it matched?22:41
blackburnsonney2k: there are two ways usually - SERIALIZABLE_DUMMY and ignoe22:41
gsomixsonney2k, and
@sonney2kblackburn, so which one will work ?22:42
blackburnsonney2k: any but no need to have covertree point in the interface so just ignore22:42
@sonney2kgsomix, ok then great work - I don't understand why you mark this WIP then - it will not become any better than that.22:43
@sonney2kblackburn, can a covertree be serialized?22:44
blackburnsonney2k: currently no22:44
blackburnsonney2k: it is full of std stuff22:45
gsomixsonney2k, I removed the WIP status. :)22:45
@sonney2kargh then this won't work    SG_ADD((CSGObject**) &m_covertree, "m_covertree", "Member cover tree", MS_NOT_AVAILABLE);22:45
@sonney2ktoo bad22:46
blackburnsonney2k: mainly because covertree is not a descendant of CSGObject :D22:46
@sonney2knot only22:46
@sonney2kalright we have to bug n4ndo about this22:47
blackburnsonney2k: I am not really happy with that covertree22:47
@sonney2kwhat's up with JL's covertree?22:47
blackburnsonney2k: it is hard to integrate there22:47
blackburnboth I and gsomix tried22:47
blackburnand have died with a lot of leaks/bugs while integrating22:48
blackburnsonney2k: hard question22:49
gsomixnot hard. just somewhere I'm doing wrong. I need to be careful.22:49
@sonney2khmmhh - I expect JLs code to be fast at least22:49
blackburnyes it is faster22:49
blackburnso we would need to make another attempt22:49
@sonney2kso it would be really great to get it22:49
@sonney2khow long have you tried?22:49
blackburnsonney2k: yes I agree22:49
@sonney2kand serialization should be simpler too22:49
blackburnI spent a week or so on that22:49
@sonney2kbecause JL hates std:: crap22:49
blackburngsomix have spent a month or two22:50
blackburnbut not as active as me probably22:50
gsomixsonney2k, blackburn: I plan to return to work on covertree again.22:50
@sonney2kblackburn, gsomix has written the octave sparse matrix typemap - that would have taken you a year :D22:50
blackburnsonney2k: I do not really understand why we both failed22:51
blackburnsonney2k: I heard it was easy ;)22:51
@sonney2kactually :P22:51
@sonney2keverything is easy if you know how22:51
blackburnin the arpack.cpp22:52
blackburnI guess everybody is afraid to enter to that file22:52
blackburnsonney2k: I already did some sparse stuff you know22:52
blackburnsonney2k: I still do not understand why you hate std :)22:53
blackburnshogun/lib/ShogunException.cpp:29 is a funny line22:55
@sonney2kI saw this even!22:57
blackburnsonney2k: really? did you run shogun on some controller?22:57
@sonney2kno a 92G machine22:58
@sonney2kyou know big data ;D22:58
blackburnsonney2k: 92G and no place to allocate a string? lol22:59
CIA-64shogun: Soeren Sonnenburg master * r908c468 / (2 files in 2 dirs):23:00
CIA-64shogun: fix compilation of java_modular23:00
CIA-64shogun: - disable covertree serialization in KNN23:00
CIA-64shogun: - put KNN POINT type into modshogun ignores -
CIA-64shogun: Soeren Sonnenburg master * r46353c3 / (2 files in 2 dirs):23:00
CIA-64shogun: Merge pull request #430 from gsomix/sparse_octave23:00
CIA-64shogun: Sparse matrix typemap for octave modular. -
-!- n4nd0 [] has joined #shogun23:00
n4nd0sonney2k, blackburn I checked the logs23:06
n4nd0sonney2k: I had to define covertree_point in .h because of the covertree type member23:06
@sonney2kn4nd0, yeah I know23:07
@sonney2k(read your code)23:07
n4nd0ok :)23:07
n4nd0has that turned out to be an issue?23:07
n4nd0sorry for the non-required explanation :S23:07
@sonney2kwell it didn't compile and would crash when being serialized23:08
n4nd0not good23:10
@sonney2kn4nd0, yeah but not so easy to see23:10
@sonney2kn4nd0, btw how much faster is it on some realistic data set?23:10
shogun-buildbotbuild #447 of ruby_modular is complete: Success [build successful]  Build details are at
blackburnsonney2k: n4nd0: I'll let you know of speedup on GTSRB dataset (39029 vectors of 900d)23:11
n4nd0sonney2k: I proposed it yesterday to make a test on iris dataset to someone who was looking for a patch23:12
n4nd0blackburn: ok, great23:12
n4nd0yeah, why?23:12
blackburnI am afraid to be wrong but isn't it <1K of vectors?23:13
n4nd0150 instances is said on the webpage23:13
n4nd0mmm yeah, maybe that is too little23:14
blackburnyes it will be even slower may be23:14
n4nd0in octave example, super small, it was faster though23:14
blackburnslightly overheaded with tree :)23:14
blackburnok then23:14
n4nd0blackburn: do you have your own example to try it on GTSRB? somewhere in shogun?23:15
@sonney2kgsomix, do you want to continue to work on sth?23:15
blackburnn4nd0: no23:16
gsomixsonney2k, of course.23:17
blackburnn4nd0: I can share feature matrices and labels if you want23:22
n4nd0blackburn: yeah, sure23:23
n4nd0blackburn: how should we do it?23:23
blackburnn4nd0: just load it from numpy and use23:24
n4nd0blackburn: ok, thank you23:25
n4nd0wow, 174MB for one of them23:25
blackburnn4nd0: I don't remember dimensionality exactly but there are 39K of train vectors23:25
blackburnand 12K of test ones23:25
n4nd0blackburn: have you ever trained sth with this?23:26
blackburnn4nd0: yes multiclass svm23:26
n4nd0blackburn: how long did it take?23:27
blackburnkernel svm takes 3h or so23:27
blackburnGMNP essentially23:27
blackburnliblinear took a little23:27
blackburn5-10 minutes23:27
@sonney2kSo you could do either some  ASSERT()'s replacement with meaningful  SG_ERROR() messages or start some big task - SGVector23:28
blackburngsomix: it is for you23:28
@sonney2kfor SGVector you could either implement refcounting business and/or replace Array.h with SGVector23:29
@sonney2kthe latter would boil down to adding the functions from Array to SGVector and then replace23:30
@sonney2ksame fro Array2 -> SGMatrix, etc23:30
blackburnn4nd0: if you want to dig into details: features have been constructed using HOG w/o block normalization and prewitt gradient23:31
blackburnsome median filtering as well however it is really too detailed then23:32
n4nd0blackburn: are the same you use for the paper you showed me?23:34
blackburnn4nd0: bingo23:34
gsomixsonney2k, do I have the time to think? :)23:38
blackburngsomix: no you need to implement all the things *tonight*23:39
blackburnno kidding23:39
gsomixblackburn, ?????, ???. :]23:39
gsomixsorry for my Russian23:40
@sonney2kgsomix, well you can do other things to - just propose sth23:40
gsomixsonney2k, ok. now I'm working on custom_kernel.23:46
blackburnsonney2k: btw one more weka extension was published on mloss23:50
@sonney2kgsomix, what do you do there?23:51
blackburna kind of23:51
blackburnsonney2k: there is a bug still ;)23:51
@sonney2kblackburn, what are you talking about?23:52
blackburnsonney2k: some set_triangular_.... is wrong23:52
blackburngsomix should know details now23:52
blackburnsonney2k: old tester fails on custom kernel23:52
blackburnbecause of that bug23:53
--- Log closed Sun Apr 08 00:00:19 2012