Open in new window / Try shogun cloud
--- Log opened Wed Feb 13 00:00:45 2013
-!- SCV__ [~FSCV@] has quit [Quit: Leaving]01:05
-!- n4nd0 [] has quit [Quit: leaving]01:42
-!- sumit [75e210ac@gateway/web/freenode/ip.] has joined #shogun03:02
-!- sumit [75e210ac@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]03:11
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has joined #shogun04:29
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]05:21
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has joined #shogun05:38
-!- blackburn [~blackburn@] has joined #shogun06:38
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]06:49
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]07:09
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has joined #shogun08:11
-!- lambday [3ba2174d@gateway/web/freenode/ip.] has joined #shogun08:54
-!- sumit [ca4eaca2@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]09:13
-!- blackburn [] has joined #shogun09:43
-!- shogun-notifier- [] has joined #shogun09:43
shogun-notifier-shogun: iglesias :master * c4443b0 / src/shogun/converter/KernelLocallyLinearEmbedding.cpp:
shogun-notifier-shogun: small fix in KLLE to avoid valgrind errors09:43
shogun-notifier-shogun: iglesias :master * ded1104 / src/shogun/converter/DiffusionMaps.cpp:
shogun-notifier-shogun: + Gaussian kernel width initialization in DiffusionMaps09:43
shogun-notifier-shogun: iglesias :master * 3f87a2e / src/shogun/converter/NeighborhoodPreservingEmbedding.cpp:
shogun-notifier-shogun: * minor fix in NPE to avoid valgrind complains09:43
shogun-notifier-shogun: Sergey Lisitsyn :master * db0a595 / src/shogun/converter/ (3 files):
shogun-notifier-shogun: Merge pull request #862 from iglesias/tapkee09:43
shogun-notifier-shogun: [WIP] Small fixes in converters to avoid valgrind errors09:43
shogun-buildbot_build #805 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: iglesias <>09:55
-!- n4nd0 [] has joined #shogun09:56
blackburnn4nd0: thanks for issuing a PR09:56
blackburnis everything fixed now?09:56
n4nd0no, not everything yet09:56
n4nd0I am working in the LLTSA errors now09:56
n4nd0I tried with the usual path, setting m_nullspace_shift but it didn't solve it09:57
n4nd0I tried also a couple of things more but didn't succeed09:57
n4nd0I will look to it now :)09:57
blackburnif you have some time for that09:58
blackburnI'd appreciate09:58
n4nd0mmm SGVectorTest failed09:59
n4nd0blackburn: btw I was thinking yesterday10:02
n4nd0it could be good if we think of a small set of tasks10:02
n4nd0and write them down somewhere10:02
blackburnn4nd0: github issues10:02
n4nd0blackburn: yes, it is a good idea to write them there10:03
blackburnn4nd0: I added 'entrance' label10:04
blackburnif you have any ideas please put it10:04
n4nd0will do, thank you!10:04
n4nd0blackburn: why is it that lltsa needs the features to be set both in parameters.kernel and parameters.features?10:08
blackburnn4nd0: it uses both kernel and feature access callbacks10:10
blackburnn4nd0: so I asked for help in the ML10:21
n4nd0blackburn: yeah, just read it10:21
n4nd0that's nice, let's see if we get more suggestions10:21
blackburnn4nd0: 3 already! :)10:26
lambdayhello, this is soumyajit. I want to get involved with shogun.. I've already cloned the source code and built it.. (cmdline static and python interfaces).. But I am not sure how to get started with such a huge codebase.. could you please suggest me something?10:32
n4nd0lambday: you have come in a good moment :)10:33
n4nd0blackburn has just created an entrance section on github issues10:33
lambdayokay. i'm checking..10:33
n4nd0it was announced in the mailing list a few minutes ago too10:34
-!- proKrammer [73fe7104@gateway/web/freenode/ip.] has joined #shogun10:34
lambdayokay.. I'm not yet registered to the mailing list.. will do10:35
lambdayand also, is there any kernel related development task currently within the scope? I'm doing some reading on string kernels and I'm really interested to work with one10:40
n4nd0lambday: you could check if we already have implemented that kernel10:40
n4nd0if it is not, you are welcome to include it10:41
n4nd0lambday: see src/shogun/kernel/string10:41
lambdayn4nd0: thanks.. I'm checking10:42
n4nd0what kernel is it btw the one you are interested to work with?10:42
lambdayn4nd0: well, I was reading this paper "Fast Kernels for String and Tree matching" by Vishwanathan et al. as a part of my coursework10:45
blackburnI'd recommend Soeren's thesis
n4nd0blackburn: everything in there is probably already in shogun though10:50
n4nd0I am not sure, just guessing10:50
lambdayblackburn: are these the kernels that are already implemented in shogun?10:50
blackburnlambday: I think so10:51
lambdayoh I see the subsequence kernel10:51
lambdaybut it implemented the paper by lodhi..10:51
lambdayblackburn: well, can you suggest me something then, any development related ideas (better if kernel related).. I'll do some reading and will be in touch...10:55
lambdayand also, how should I go about understanding the architecture of shogun?10:56
blackburnlambday: we are now collecting ideas
blackburnthere should be some kernel related for sure - we just started10:56
n4nd0lambday: get to know the code superficially, browse different classes and so on10:57
lambdayblackburn: okay.. thanks :)10:57
blackburnit is a difficult question about architecture - I have nothing general about that10:57
lambdayn4nd0: alright.. I'll keep in touch.. meanwhile, I'll go through the thesis by Soeren11:00
n4nd0wiking: haha ok ok11:14
n4nd0I don't think that in this case to compile with -lpthread -pthread would make a mistake anyway11:15
n4nd0wiking: who can take a look to configure script?11:15
wikingn4nd0: i did now... it's a bit problematic11:15
wikingi mean the configure script for -pthread or -lpthread is quite easy11:16
wikingthe problem is more about that i actually do not understand11:16
wikingwhy -pthread is not working for unit test compilation11:16
n4nd0yeah, I reckon that feeling when looking at configure :)11:16
wikingbut works fine for libshogun11:16
sonne|workn4nd0: regarding this list - is LDA multiclass already?11:17
n4nd0sonne|work: LDA multiclass and change QDA to eigen were in my head11:17
n4nd0I am going to write them soon11:18
n4nd0in the entrance issues I mean11:18
sonne|workn4nd0: yes please thx11:18
wikingn4nd0: what is the exact compilation that does not work for you for unit test?11:18
n4nd0wiking: give me a couple of minutes to check11:18
wikingn4nd0: ok11:19
n4nd0sonne|work: did you talk with c-base people?11:19
n4nd0I don't exactly recall when was the exact11:19
n4nd0around 15 IIRC11:19
n4nd0exact date*11:19
blackburn14 I think11:19
blackburnlove meeting11:19
n4nd0haha yeah11:19
sonne|work14th but it looks like I don't have time that evening :/11:20
sonne|workand gunnar == /dev/null11:20
wikingah shit the linking line is actually not there :(11:23
wikingn4nd0: can u edit the Makefile under tests/unit11:24
wikingand there at line 2811:24
n4nd0wiking: sure11:25
wikingremove the @ in front of @$(LINK)11:25
n4nd0it seems that sonne|work has found an answer though hehe11:25
wikingn4nd0: so just remove that @11:25
wikingand do make again11:25
wikingand copy paste the linking line11:25
wikingthe last c++ line11:25
wikingsonne|work: afaik there's a better fix i just need to see the linking line11:26
wikingi think the order of -pthread matters11:26
proKrammerhi n4nd0 , I have spoken to you yesterday regarding the installation, i am done with it, but what about the GUI part?11:26
sonne|workwiking: true11:26
wikingn4nd0: u have it?11:26
n4nd0c++ -o shogun-unit-test ./classifier/svm/ ./multiclass/ ./base/ ./latent/ ./lib/ -Xlinker --no-undefined -L../../src/shogun -lshogun libgmock.a11:27
n4nd0/usr/bin/ld: libgmock.a(gtest-all.o): undefined reference to symbol 'pthread_key_delete@@GLIBC_2.2.5'11:27
n4nd0/usr/bin/ld: note: 'pthread_key_delete@@GLIBC_2.2.5' is defined in DSO /usr/lib/ so try adding it to the linker command line11:27
n4nd0/usr/lib/ could not read symbols: Invalid operation11:27
n4nd0collect2: error: ld returned 1 exit status11:27
n4nd0make: *** [shogun-unit-test] Error11:27
wikingi wonder why there's no pthread there11:27
wikingok just a sec11:27
wikingsonne|work: shouldn't $(LINKFLAGS) contain -pthread?11:28
wikingsonne|work: POSTLINKFLAGS11:28
wikingthat should be it?!11:28
wikingn4nd0: can u add to that line 28 this variable11:28
wiking$(LINK) -o shogun-unit-test $(TESTOBJFILES) $(LINKFLAGS) -L$(SHOGUNSRCTOP)/shogun -lshogun libgmock.a $(LINKFLAGS_GMOCK) $(POSTLINKFLAGS)11:29
wikingso it should look like something like this ^11:29
wikingand run make again11:29
wikingand paste here the c++ line11:29
n4nd0wiking: that made it11:30
wikingok how does the line look like?11:30
n4nd0c++ -o shogun-unit-test ./classifier/svm/ ./multiclass/ ./base/ ./latent/ ./lib/ -Xlinker --no-undefined -L../../src/shogun -lshogun libgmock.a  -lm -pthread -ljson  -lxml2  -llapack -lcblas -larpack -lz -lbz2 -llzma11:30
n4nd0-pthread is in da house now11:31
wikingand it did not complain?11:31
wikingran fine?11:31
n4nd0all tests passed11:31
wikingcommit and PR it11:31
n4nd0just a few warnings11:31
wikingi'll merge it11:31
wikingthat's ok that's because of MOCK11:31
n4nd0should I put back the @ in front of @(LINK)?11:31
n4nd0compiling again with @ to ensure11:32
wikingyeah that just silences the compile line11:32
wikingso that it doesn't get printed on stdout11:32
wikingbut should work fine still11:32
n4nd0yeah it did11:32
wikingok commit + pr :>11:32
n4nd0wiking: the pull request is a bit ugly because of the previous commit and pull request I had issued :S11:39
n4nd0wiking: take a look11:39
n4nd0three commits in there11:40
wikingaah yeah ok11:40
wikingi got it why :)11:40
n4nd0was there any way to fix that11:40
wikingcan u do --amend?11:40
n4nd0I should always work from a branch11:40
wikinggit commit --amend .11:40
n4nd0it was already pushed in my repo, since I did the PR11:41
wikingah yeah11:41
n4nd0git commit --amend would make a problem?11:41
wikingi think then it would break your branch11:41
wikingbut that's ok we can merge it like this no sonne|work ?11:41
wikingi mean yeah 3 commits for 1 fix but still it's workng :>11:41
shogun-notifier-shogun: iglesias :master * 021237f / tests/unit/Makefile:
shogun-notifier-shogun: add -lpthread flag to build tests correctly11:42
shogun-notifier-shogun: iglesias :master * a954c4f / tests/unit/Makefile:
shogun-notifier-shogun: minor fix: add POSTLINKFLAGS in linking gtest11:42
shogun-notifier-shogun: iglesias :master * c553667 / :
shogun-notifier-shogun: Merge branch 'master' of
shogun-notifier-shogun: Conflicts:11:42
shogun-notifier-shogun: tests/unit/Makefile11:42
shogun-notifier-shogun: Viktor Gal :master * da53770 / tests/unit/Makefile:
shogun-notifier-shogun: Merge pull request #867 from iglesias/master11:42
shogun-notifier-shogun: minor fix: add POSTLINKFLAGS in linking gtest11:42
sonne|workwhy 3 commits not one?11:42
n4nd0wiking: thank you for finding out the error :)11:42
wikingsonne|work: see ^11:43
n4nd0sonne|work: becuase of the previous PR with one commit11:43
n4nd0I did that from master :S11:43
n4nd0I shouldn't have done it, my bad11:43
sonne|workn4nd0, wiking now please get shogun in release shape and we are good11:43
n4nd0sonne|work: I am currently taking a look to converter errors detected by valgrind11:44
blackburnfixing my mistakes hah11:47
shogun-buildbot_build #807 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: iglesias <>11:48
n4nd0well you are actually telling me where the errors are most of the time :D11:48
wikingwhy did u faaaail11:48
wikingah ok11:49
n4nd0wiking: I think the test was already broken before11:49
wikingSGVectorTest.misc failed11:49
wikingok yeah we need to use EXPECT_DOUBLE_EQ :>11:50
blackburnwiking: if you want to get more tests for shogun and have ideas - please add them to 'entrance' issues at github11:50
wikingblackburn: oh good idea!11:51
blackburnbut it would need to describe what exactly to test11:54
shogun-buildbot_build #808 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Viktor Gal <>, iglesias <>11:54
blackburnour candidates could get stucked with concepts that are well-known for us11:55
blackburnthat's probably normal but we should keep that in mind :)11:56
wikingblackburn: yeah i'll try to make a module list and put them up one by one11:56
n4nd0blackburn: hey12:43
n4nd0maybe this makes sense to you12:43
n4nd0in lltsa to set parameters.eigenshift equal to m_nullspace_shift (default -1e-9)12:44
n4nd0gives more valgrind errors thatn12:44
n4nd0to set it to m_reconstruction_shift (default 1e-3)12:44
-!- n4nd0 [] has quit [Ping timeout: 272 seconds]13:08
-!- n4nd0 [] has joined #shogun13:19
blackburnn4nd0: that's out of sense for me13:24
n4nd0too bad13:26
n4nd0I will look at it again later today or tom13:28
blackburnI hope I can take a look tonight13:30
sonne|workwiking: urgs -have you seen the ocas test failing?13:51
wikingsonne|work: noup13:54
wikingsonne|work: where?13:54
sonne|workwiking: don't you get the buildbot emails ?13:55
-!- proKrammer [73fe7104@gateway/web/freenode/ip.] has quit [Quit: Page closed]13:55
wikingi do but not always i check them ;P13:55
wikingsonne|work: i reckon there's some problem there with an SGVector function13:56
wikingsonne|work: since we had the exact same errors when the ::max was bogus13:56
sonne|workwiking: no idea but please check. if we don't address the issues tests are useless...13:58
wikingsonne|work: working on it13:59
blackburn7 entrance tasks so far!14:02
wikingsonne|work: btw what is the proper method of right away applying patches of mine into shogun? or do i should go still with the commit local repot -> pr -> merge ?14:02
n4nd0blackburn: we need now 7 students! hehe14:03
wikingblackburn: ^14:03
blackburnwiking: my position is that for small changes/bugfixes it is ok to commit directly14:03
wikingblackburn: how?14:03
blackburnwiking: hmm you can't?14:04
wikingblackburn: no i dont know how :)14:04
blackburnwiking: I think you need to change remote14:04
sonne|workwiking: yeah that is best, but merge yourself14:05
blackburnyou probably have git://
sonne|workwiking: I will switch to doing that too14:05
blackburnas a remote14:05
sonne|workdon't directly work on shogun master14:05
blackburnI do not think we should do PRs for every change we make14:06
sonne|workblackburn: and I do :)14:06
wikingmm this is interesting14:06
wikingsonne|work: any ideas ^14:06
sonne|workbut we had that discussion already.14:06
wikingwe use EXPECT_DOUBLE_EQ in both cases14:06
wikingbut still it complains14:06
blackburnsonne|work: you know once I get POWER and AUTHORITY..14:07
sonne|workwiking: then maybe we need a slightly higher epsilon14:07
sonne|workfloat stuff is certainly scary14:08
sonne|workblackburn: you will I am sure but it is better that everyone's damage to the code might be looked at14:09
blackburnsonne|work: well we actually should not work in master branch14:10
blackburnI do not feel the difference between one commit + PR and one commit14:10
blackburnboth can be looked at14:10
blackburnand the damage is done14:10
sonne|workblackburn: yeah dev branches might be a good idea14:11
sonne|workonce we have a release we should attempt to switch14:11
wikingfucking google test blog answers the same shit14:19
n4nd0wiking: link?14:19
wikingsonne|work: can u enable unit testing on freebsd bot plz?14:21
wikingsonne|work: ah ok it is14:21
sonne|workwiking: so why not use say 1e-12 as precision?14:24
wikingsonne|work: no i was wrong14:26
wikingwhat is interesting though14:26
wikingthat the same bot14:26
wikinghas some weird things14:26
wikingso there the SGVector.misc fails14:27
wikingbut then in the next build14:27
wikingit is all good14:27
wikingbut the multiocas fails14:27
wikingcan we force a new build on this bot14:27
wikingjust out of curiosity14:27
sonne|workwiking: sure just do it14:28
wikingsonne|work: what is the command? :D14:28
sonne|workwiking: I guess it is due to parallelization...14:28
wikingshogun-buildbot_: force build deb114:29
sonne|workshogun-buildbot_: force build "deb1 - libshogun"14:29
shogun-buildbot_no such builder 'deb1'14:29
shogun-buildbot_build forced [ETA 1m27s]14:29
shogun-buildbot_I'll give a shout when the build finishes14:29
wikingalmost :)14:29
shogun-buildbot_build #809 of deb1 - libshogun is complete: Success [build successful]  Build details are at
wikingthis is mega insteresting14:30
wikingas now it succeeded14:31
sonne|work(02:28:20 PM) sonne|work: wiking: I guess it is due to parallelization...14:32
wikingyeps seems so14:32
-!- shogun-notifier- [] has quit [Quit: transmission timeout]14:42
-!- FSCV [~FSCV@] has joined #shogun14:53
sonne|workwiking: so then just limit it to use a single core15:31
n4nd0sonne|work, wiking : why does parallelization make it fail?15:36
sonne|workn4nd0: not sure if it fails or just finds some other valid solution (given objective w/i epsilon)15:37
wikinglol now i have a failure with:15:44
wikingValue of: SGVector<float64_t>::arg_max_abs(a.vector, 1, a.vlen, __null) Actual: 015:44
wikingExpected: arg_max_abs15:44
wikingWhich is: 915:44
wikingsame with max_abs15:44
sonne|workwiking: valgrind....15:46
wikingok i see why15:46
wikingit's an actual bug indeed15:47
wikingbtw where's github bot?15:50
-!- shogun-notifier- [] has joined #shogun15:51
shogun-notifier-shogun: Viktor Gal :master * 59e15f5 / tests/unit/ (2 files):
shogun-notifier-shogun: Fix unit tests for SGVector::max_abs and MOCASSVM15:51
shogun-notifier-shogun: there was a bug when calculating max_abs in an SGVector15:51
shogun-notifier-shogun: and MulticlassOCAS has been set to use only 1 thread15:51
shogun-notifier-shogun: Viktor Gal :master * 18ea338 / tests/unit/ (2 files):
shogun-notifier-shogun: Merge pull request #872 from vigsterkr/master15:51
shogun-notifier-shogun: Fix unit tests for SGVector::max_abs and MOCASSVM15:51
-!- blackburn [] has quit [Quit: Leaving.]15:51
wikinginteresting that until now that max_abs error has not surfaced... there's no fool proof unit-testing as this one proves it well ;)15:54
sonne|workwiking: what is the github bot?15:54
wikingsonne|work: we used to have github hook for commits and pull requests15:55
wikingthat was set up with a hook15:55
shogun-buildbot_build #811 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Viktor Gal <>16:00
wikingseems 1 thread did not help16:00
sonne|workwiking: no longer working with github :/16:01
wikingreally? :) i guess then it's a bug of github that we should report16:03
-!- tom__ [2eda6d52@gateway/web/freenode/ip.] has joined #shogun16:04
sonne|workwiking: or they shoudl just install irker on their side...16:04
sonne|workwiking: about the build fail      - could you please print the objective value in the test?16:05
tom__Quesion about saving/loading mklmachine : I use 'CSerializableAsciiFile' but when loaded from file, mkl->getsubkernelweights() return 0. Is there code sample for that ?16:06
sonne|worktom__: please give us a minimal example to reproduce16:07
sonne|workthen we can create a test out of this and fix it!16:07
tom__sonne|work: ok16:08
tom__sonne|work: here it is
tom__my first question is: does the mkl saved in file contain all kernels, machine params and weights ?16:16
-!- n4nd0 [] has quit [Quit: leaving]16:33
tom__sonne|work: I just re-use the mkl example from shogun examples and add save and load function. Ths for help16:52
sonne|worktom__: yes it should16:53
sonne|worktom__: it should be self containing16:53
tom__then no need to save kernels and weights separatly ?16:53
sonne|workjust save the core mklmachine16:53
sonne|workthen all kernels/labels/features/parameters *should* be saved too16:54
sonne|workif not -> bug16:54
tom__that's what I am doing in the example I gave you. it is c++, maybe I misunderstood smthg16:54
tom__how to access to combined kernel and weights from mkl pointer ?16:55
sonne|workfor example the multiclass classifier for the hand written digit recognition is stored exactly like that (the one from here )16:55
sonne|worktom__: gtg now might have a look later16:55
tom__ok thx anyway i wil lokk into samples16:56
-!- tom__ [2eda6d52@gateway/web/freenode/ip.] has quit [Quit: Page closed]16:56
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has joined #shogun17:35
KMcQuistenHi folks17:36
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:41
-!- ipondering [] has joined #shogun18:07
-!- ipondering [] has left #shogun []18:08
-!- shogun-notifier- [] has quit [Quit: transmission timeout]18:51
-!- blackburn [~blackburn@] has joined #shogun18:51
blackburnsonney2k: around?19:26
-!- heiko1 [] has joined #shogun20:02
-!- n4nd0 [] has joined #shogun20:16
blackburnheiko1: really?20:19
blackburnNP-hard graph colouring?20:19
heiko1blackburn, greedy version allowed :)20:21
heiko1blackburn, updated20:22
blackburnheiko1: what do you need it for btw?20:22
heiko1just need *a* graph colouring20:22
heiko1for estimating the trace of a matrix20:22
heiko1 tr A = E_s [s^T A s] where s is a random variable with zero mean and identity covariance20:23
heiko1the distribution of s determines the variance of this estimator20:23
heiko1and there is a clever way to choose it20:24
heiko1based on a graph colouring of the matrix20:24
heiko1blackburn, its a bit weird, but gives really cool results20:24
heiko1Google Hutchison estimator for the basic ideas20:24
heiko1(without graph colouring)20:24
blackburnokay anyway I don't mind :)20:24
heiko1hey blackburn a question20:25
heiko1c++ /swig /python20:25
heiko1I have a class which inherits from a templated class20:25
heiko1so class B: A<float64_t>20:26
heiko1fixing the template arugment20:26
heiko1I can call any methods from B<T> in C++20:26
heiko1but not from python modular20:26
heiko1says they dont exist20:26
heiko1blackburn, any idea how to solve that?20:27
heiko1I could make B also a template and then pass the parameter, create a bunch of names for pyhton for all different types20:27
heiko1this works20:27
heiko1but I dont want to do all the template hell since not needed anyway20:27
blackburnwait wait let me understand20:27
blackburnnicht schiessen20:28
heiko1I will commit this then you can try it out20:28
blackburnthat sounds strange for me actually20:28
blackburndidn't know about that20:29
heiko1maybe I did a mistake20:29
blackburnheiko1: what if A->B->C20:30
blackburnand B is templated20:30
blackburnwhile A and C not20:30
blackburndoes C have methods of A?20:30
heiko1dont know20:31
heiko1but thats not my problem :D20:31
blackburnwait wait20:31
blackburnwe have such example20:31
blackburnheiko1: CPNorm20:31
blackburnheiko1: I think it works nice - nothing lost20:32
-!- shogun-notifier- [] has joined #shogun20:33
shogun-notifier-shogun: Heiko Strathmann :master * 7e6bc9c / src/shogun/features/streaming/generators/ (2 files):
shogun-notifier-shogun: fixed typo20:33
shogun-notifier-shogun: Heiko Strathmann :master * 822d40a / src/shogun/features/streaming/generators/ (2 files):
shogun-notifier-shogun: added new streaming data generator class for creating Gaussian blobs as in NIPS paper for optimal kernel choice for linear time MMD20:33
shogun-notifier-shogun: Heiko Strathmann :master * 773664d / examples/undocumented/libshogun/streaming_generators.cpp:
shogun-notifier-shogun: added example for new gaussian blobs data generator20:33
shogun-notifier-shogun: Heiko Strathmann :master * d9bb5b8 / / (7 files):
shogun-notifier-shogun: made CMeanShiftDataGenerator not be a template anymore since useless and annoying20:33
shogun-notifier-shogun: Heiko Strathmann :master * 0a21ab6 / / (9 files):
shogun-notifier-shogun: Merge pull request #874 from karlnapf/master20:33
shogun-notifier-shogun: New class for data generation20:33
heiko1blackburn, try in python to call the method get_streamed_features of CMeanShiftDataGenerator20:34
heiko1probably just me being stupid :)20:34
heiko1ah just read a hint in compile warnings20:35
blackburnheiko1: what is it?20:35
heiko1../../shogun/features/streaming/generators/MeanShiftDataGenerator.h:28: Warning 401: Nothing known about base class 'CStreamingDenseFeatures< float64_t >'. Ignored.20:36
heiko1../../shogun/features/streaming/generators/MeanShiftDataGenerator.h:28: Warning 401: Maybe you forgot to instantiate 'CStreamingDenseFeatures< float64_t >' using %template.20:36
blackburnso here you go then :)20:36
heiko1this was not in the interface20:36
heiko1it IS in there20:37
heiko1#ifdef USE_FLOAT6420:37
heiko1    %template(StreamingRealFeatures) CStreamingDenseFeatures<float64_t>;20:37
heiko1I think the problem is that the wrong class name is used20:37
heiko1when compiling the interface20:37
heiko1any ideas how to fix that?20:38
shogun-buildbot_build #812 of deb1 - libshogun is complete: Success [build successful]  Build details are at
heiko1this is the class20:38
heiko1class CMeanShiftDataGenerator: public CStreamingDenseFeatures<float64_t>20:38
heiko1I think fixing the base class here is the problem20:38
heiko1python thinks that CStreamingDenseFeatures<float64_t> should be "real"densefeaturesblabla20:38
shogun-buildbot_build #814 of deb1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>20:38
blackburntwo ideas20:39
blackburnidea one - forward declaration of that class in meanshift...h20:39
blackburnlike template <class T> CStreamingDenseFeatures<T>20:39
blackburntemplate <class T> CStreamingDenseFeatures<T>;20:39
blackburnjust declaration20:40
@sonney2kblackburn, yes %template is missing20:40
heiko1in fact, it all compiled on my machine20:40
blackburnsonney2k: where?20:40
blackburnsonney2k: who could lead vowpal wabbit?20:41
blackburnI put it to myself for onw20:41
blackburnit is of great priority IMO20:42
shogun-notifier-shogun: Heiko Strathmann :master * e15edc3 / src/shogun/features/streaming/generators/MeanShiftDataGenerator.cpp:
shogun-notifier-shogun: added forgotten template arugment in base class20:42
shogun-notifier-shogun: Heiko Strathmann :master * 8d72b29 / src/shogun/features/streaming/generators/MeanShiftDataGenerator.cpp:
shogun-notifier-shogun: Merge pull request #875 from karlnapf/master20:42
shogun-notifier-shogun: quickfix20:42
blackburnsonney2k: it is something like true large scale20:43
blackburnheiko1: whoa!20:43
blackburnhow did that compile :D20:43
heiko1blackburn ?20:43
heiko1no idea :)20:43
blackburnheiko1: wonderful world of C++20:43
heiko1sonney2k, blackburn, how to fix this template issue?20:44
heiko1blackburn, yes :)20:44
shogun-buildbot_build #813 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>20:44
heiko1hehe , great comic blackburn :)20:45
heiko1not responsible for the above fail20:45
-!- n4nd0 [] has quit [Ping timeout: 245 seconds]20:46
@sonney2kblackburn, I don't know - it is overhyped20:46
@sonney2kit is just sgd20:46
blackburnsonney2k: really? I don't know for sure20:46
@sonney2kbut yes it can run on kiloclusters20:46
@sonney2kthat is the cool part about it20:47
@sonney2kIMHO any truely largescale stuff is just (distributed) sgd20:47
blackburnsonney2k: I added ideas page - feel free to update if you have some horsepower20:48
@sonney2kno more horses left :(20:48
blackburnI think I do not so will continue tomorrow20:49
@sonney2kheiko1, what did you do? you derived from a templated class without specifying the type?20:49
@sonney2kyou crazy?20:50
heiko1sonney2k, class CMeanShiftDataGenerator: public CStreamingDenseFeatures<float64_t>20:50
heiko1works fine in C++20:50
heiko1I think its an issue of renaming CStreamingDenseFeatures<float64_t> to StreamingRealFeatures20:51
heiko1when processing Features.i20:51
heiko1it does work when I do20:52
heiko1template <class T> class CMeanShiftDataGenerator: public CStreamingDenseFeatures<T>20:52
heiko1but then I have to create all these names for the different types20:53
heiko1which I would like to avoid20:53
shogun-buildbot_build #815 of deb1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Heiko Strathmann <>20:57
@sonney2kheiko1, I have to see the error20:57
blackburnsee you tomorrow20:58
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]20:58
-!- lambday [3ba2174d@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]20:58
heiko1sonney2k, ok will show soon, first have to fix this thing here, which is strange since compiles on my machine20:59
heiko1gotta go now, will look after the build in a few hours when home21:02
-!- FSCV_ [~FSCV@] has joined #shogun21:04
-!- FSCV [~FSCV@] has quit [Ping timeout: 245 seconds]21:06
-!- syst3mw0rm [] has quit [Ping timeout: 255 seconds]21:06
-!- syst3mw0rm [] has joined #shogun21:09
shogun-notifier-shogun: Soeren Sonnenburg :master * b80a07c / src/ (3 files):
shogun-notifier-shogun: fix %template warning and potential compile error21:12
@sonney2kwiking, so how do I write a test21:13
@sonney2kwiking, I want to write one for MKL21:13
@sonney2kerr CombinedKernel21:13
shogun-buildbot_build #816 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Soeren Sonnenburg <>21:24
wikingsonney2k: well follow my lead :)21:25
-!- heiko1 [] has left #shogun []21:25
wikingsee latentsvm for example21:25
-!- n4nd0 [] has joined #shogun21:27
shogun-notifier-shogun: Soeren Sonnenburg :master * f509af5 / tests/unit/classifier/svm/
shogun-notifier-shogun: add objective test for ocas21:30
@sonney2kwiking, done :)21:30
shogun-notifier-shogun: Soeren Sonnenburg :master * 9580b2d / tests/unit/kernel/
shogun-notifier-shogun: add test for combined kernel weights getter/setter21:30
shogun-notifier-shogun: Soeren Sonnenburg :master * bba7f87 / tests/unit/multiclass/
shogun-notifier-shogun: set #threads to one *before* training and epsilon to sth very low21:34
shogun-buildbot_build #817 of deb1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Soeren Sonnenburg <>21:39
n4nd0sonney2k: hey, what do you think about the project idea "Further design and implementation of an API for structured output learning"?21:40
shogun-notifier-shogun: Soeren Sonnenburg :master * d41b313 / tests/unit/kernel/
shogun-notifier-shogun: add serialization test for combined kernel21:45
@sonney2kwiking, gtest stuff is really a cool and convenient framework!!21:46
n4nd0sonney2k: do you know anyone who could mentor that?21:46
@sonney2kn4nd0, nah needs a totally different name21:46
@sonney2kn4nd0, did you ask georg?21:46
n4nd0sonney2k: aham, such as?21:46
n4nd0sonney2k: yep21:46
n4nd0he is going to contact us by this weekend with his ideas21:47
@sonney2kn4nd0, names that continue some work are not a good thing21:47
n4nd0aham I understand21:47
n4nd0and I think it makes sense21:47
@sonney2kso sth like implement CRFs, Markov Random Field, Conintuos variable SO would be good names :)21:48
n4nd0got it21:48
* sonney2k wonders what the issue was with _tom's problem - the tests work just fine...21:49
@sonney2kwho was it who said sparse features + serialization doesn't work?21:49
n4nd0not me I think :) AFAIK SGNDArray + serialization doesn't work, other than that everything works :)21:50
@sonney2kthis one21:58
@sonney2kI've just written a test21:58
@sonney2kand it indeed fails21:58
@sonney2kwiking, how can I run just a single test e.g. via valgrind?21:58
@sonney2kwith gdb?21:58
@sonney2kit is the inplace destructor22:05
-!- n4nd0 [] has quit [Ping timeout: 255 seconds]22:16
-!- n4nd0 [] has joined #shogun22:17
shogun-notifier-shogun: Soeren Sonnenburg :master * 8cad60c / tests/unit/features/
shogun-notifier-shogun: add sparse features serialization test22:18
shogun-notifier-shogun: Soeren Sonnenburg :master * 2f80763 / tests/unit/Makefile:
shogun-notifier-shogun: add valgrind target22:18
-!- cwidmer [8ca3fe9d@gateway/web/freenode/ip.] has joined #shogun22:21
cwidmerhi all22:21
cwidmerwhat, no blackburn?22:21
n4nd0cwidmer: he said good night some time ago hehe22:22
n4nd0I am surprised too! :)22:23
-!- zxtx_ [] has quit [Ping timeout: 244 seconds]22:23
cwidmern4nd0, wow, that seems unusual22:23
cwidmerjust sent you an email22:23
cwidmerlooking into SGMatrixList after all22:23
cwidmerI did some typemaps some time ago22:24
cwidmerthis one seems to be more involved, though22:24
n4nd0then you know more about typemaps than I do :)22:24
n4nd0but sonney2k seems very active this evening22:24
n4nd0maybe he can help you?22:24
n4nd0I am taking a look in any case22:25
cwidmersonney2k, around?22:25
cwidmertrying to add typemap for SGMatrixList22:25
n4nd0cwidmer: curiosity, what is it you didn't like of the SGNDArray interface?22:29
n4nd0cwidmer: stupid question, it hits me that there is matrix_list_from_numpy but no matrix_list_to_numpy22:31
n4nd0as for SGMatrix, SGVector, etc22:31
cwidmerthat's right, its one directional at the moment22:32
n4nd0not needed yet?22:32
cwidmeronly an IN typemap22:32
cwidmeras for SGNDArray, it doesn't let me access the matrices as nicely using operator()22:32
cwidmerI mean, SGMatrix now has operator()22:33
n4nd0SGNDArray is a T* actually, isn't it?22:33
cwidmerSGMatrixList seems to have operator[]22:33
n4nd0yeah, it does22:34
cwidmerso it's just a matter of nicely accessing it22:34
n4nd0I understand22:35
n4nd0cwidmer: did you see SGNDArray::get_matrix()?22:38
n4nd0I think it should be better done returning an SGMatrix instead of a T* but still22:38
n4nd0good night then guys22:50
-!- sumit [73f91219@gateway/web/freenode/ip.] has joined #shogun22:51
-!- n4nd0 [] has quit [Quit: leaving]22:51
--- Log closed Thu Feb 14 00:00:45 2013