Open in new window / Try shogun cloud
--- Log opened Wed Aug 29 00:00:17 2012
-!- n4nd0 [] has quit [Quit: leaving]00:14
-!- sr___ [u5548@gateway/web/] has joined #shogun01:35
shogun-buildbotbuild #80 of nightly_default is complete: Failure [failed test]  Build details are at
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has joined #shogun08:43
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has quit [Client Quit]08:47
-!- n4nd0 [] has joined #shogun08:48
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun09:39
-!- n4nd0 [] has quit [Quit: leaving]09:44
-!- av3ngr [] has joined #shogun10:07
-!- av3ngr [] has left #shogun []10:07
-!- zxtx [] has quit [Ping timeout: 246 seconds]11:52
-!- zxtx [] has joined #shogun12:06
-!- fjig [] has joined #shogun14:06
-!- fjig is now known as n4nd014:06
blackburnn4nd0: new nickname? :)14:07
n4nd0blackburn: connected from university :)14:09
n4nd0irssi took my username in this network14:09
n4nd0blackburn: when are we planning to release btw?14:11
n4nd0do you think everything is prepared?14:11
blackburnn4nd0: heiko made me happy about serialization14:12
blackburnit is broken :D14:12
n4nd0fuck... really?14:12
n4nd0everything related to serialization?14:12
n4nd0it sounds like a big issue14:12
blackburnI am unsure what exactly is broken14:12
blackburnworst thing heiko can't take care of that14:13
n4nd0mmm I think I could try to look at it tonight14:15
n4nd0is there any test where it breaks? to have sth to start with14:15
blackburnn4nd0: I asked heiko - lets wait for answer14:22
blackburnn4nd0: we have quite a few serialization examples in examples/libshogun14:34
blackburnyou could try to track errors with valgrind14:34
blackburnI am going to get home soon so I will try to track down errors tonight too14:48
n4nd0I will see you there then14:58
n4nd0I am going to class now14:59
-!- n4nd0 [] has quit [Quit: leaving]14:59
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]15:08
CIA-52shogun: Soeren Sonnenburg master * r5398c6e / (3 files in 3 dirs): fix up clustering evaluation example -
-!- n4nd0 [] has joined #shogun17:33
-!- gsomix [~gsomix@] has joined #shogun17:33
gsomixgood evening17:33
-!- gsomix [~gsomix@] has quit [Remote host closed the connection]17:44
@sonney2kshogun-buildbot, force build nightly_default17:45
shogun-buildbotbuild forced [ETA 37m15s]17:45
shogun-buildbotI'll give a shout when the build finishes17:45
n4nd0hi sonney2k17:55
n4nd0did you read about this issue we were talking about yesterday related to using MKL with pnorm larger than one?17:55
-!- gsomix [~gsomix@] has joined #shogun18:13
shogun-buildbotbuild #81 of nightly_default is complete: Success [build successful]  Build details are at
-!- audy [~audy@unaffiliated/audy] has quit [Quit: ZNC -]18:58
-!- audy [] has joined #shogun19:00
-!- n4nd0 [] has quit [Quit: leaving]19:08
-!- blackburn [~blackburn@] has joined #shogun19:13
@sonney2kblackburn, hey... what is wrong with the string kernels? why are they suddently marked unstable?19:14
blackburnsonney2k: because they were and are failing tests19:14
@sonney2kblackburn, the old regression tests? or which?19:14
blackburnyes, old ones19:15
@sonney2kdid they fail back then too?19:15
@sonney2k(like when we released 1.0 last year?)19:15
blackburnfor 1.0 and 1.1 yes19:15
@sonney2kI can tell that noone can report anything19:16
@sonney2kthese are too involved19:16
blackburnsonney2k: I am going through serialization stuff now - not really understand anything19:16
blackburnsonney2k: yes and because of it I think they are unstable19:17
@sonney2kblackburn, whats wrong with serializaton?19:17
blackburnsonney2k: heiko says it is broken somehow19:17
@sonney2kin which way?19:17
@sonney2kwhat is not working?19:17
blackburnI want to understand that too19:17
blackburnwell it is leaking19:17
@sonney2kblackburn, do the 'new' regression tests work?19:17
@sonney2k'just' memory leaks?19:17
blackburnsonney2k: yes seems to be that only leaks19:18
blackburnsonney2k: unit tests are working yes19:18
@sonney2kahh ok then I know19:18
blackburnsonney2k: is there a code where sgvector is being created?19:18
@sonney2kit is because of SGVector and friends I think19:18
blackburnon deserialization?19:18
@sonney2kthe reference count needs to be saved too IMHO19:18
blackburnyes it should19:19
blackburnI am confused with a lot of code there19:19
@sonney2kblackburn, any other issues that need attention?19:20
blackburnno, I don't think so19:20
blackburnsonney2k: do you have time to check serialization?19:21
@sonney2kblackburn, btw I learned that our 'regression' tests are called 'integration' tests19:21
@sonney2kso I would rather rename them now19:21
@sonney2kblackburn, no19:21
@sonney2kthat is too involved19:22
@sonney2kI am afraid only heiko can fix things19:22
blackburnheiko is out now19:22
@sonney2kblackburn, when is his deadline for handing in the masters thesis?19:22
blackburnhe is busy till 8th19:23
@sonney2k+ 2 days to recover19:23
@sonney2kblackburn, I realize that we have quite some crappy examples under python_modular19:23
@sonney2knot really functions19:23
@sonney2kso not useful for integration tests19:24
@sonney2ksome even pull stuff of the web19:24
blackburnthat's all pretty bad19:24
@sonney2kalso some have documentation on top19:24
@sonney2kinstead of descriptions/ folder19:24
blackburnsonney2k: are you back to your job already too?19:25
blackburnwhen will you?19:25
@sonney2kbut next week19:25
blackburnI see19:25
@sonney2kwill mean less time (can that be?)19:25
@sonney2ktoo bad I got sick now that I in principle would have had time19:25
@sonney2kat least I can type again ;)19:26
blackburnI don't know whether we could be ready19:26
@sonney2kno idea either19:28
@sonney2kif not then not19:28
blackburnsonney2k: I like rc concept so may be we could release 2.0rc19:29
blackburnit is actually seems to be a good idea for me19:30
CIA-52shogun: Soeren Sonnenburg master * r550727e / (264 files in 30 dirs): rename regression test directory into integration tests to avoid -
blackburnbecause no way to release  2.0 without bugs we would have to fix in next 2-3 weeks19:30
@sonney2kbut why release RC then if we know the bugs :D19:31
@sonney2kotherwise I am fine with rc's19:31
blackburnsonney2k: no we don't know all bugs19:31
blackburnI expect more anyway19:31
blackburnsonney2k: uhhhhh ref count currently is pain in the ass19:33
blackburnmore or less I realized how that works19:33
shogun-buildbotbuild #381 of deb2 - static_interfaces is complete: Failure [failed test cmdline_static]  Build details are at  blamelist: Soeren Sonnenburg <>19:38
blackburnsonney2k: I don't know how to handle that at all19:40
blackburnsgvector could be referenced from basically every other object19:41
CIA-52shogun: Soeren Sonnenburg master * rfeafbf7 / src/Makefile.template : - adjust regression -> integration in makefile -
@sonney2kblackburn, yes19:43
blackburnthat means we can't determine actual refcount19:44
blackburnactually after deserialization it seems to be 119:45
blackburnfor every field19:45
blackburnI do not have any scenario with refcount=2 initially19:45
@sonney2kblackburn, the only thing I can think of is that one has two RealFeatures objects sharing the same SGMatrix or so19:46
@sonney2k( or CLabels)19:46
@sonney2kthen current serialization would save things twice19:46
blackburnyeah that's one19:46
blackburnafter deserialization it will become correct again19:47
@sonney2kblackburn, I think the proper fix would be to use gsomix' CMap class to store all the things that are to be mapped first19:47
@sonney2kthen later save them19:47
@sonney2kthis way it can be deduplicated19:47
@sonney2kand refcounts can be computed19:47
blackburnno I don't want to implement that now :D19:48
blackburnlets assume it is 119:48
blackburnafter deserialization19:48
@sonney2kI didnt' say that this should be done *now*19:48
@sonney2kbut I think that this is the only clean way19:48
@sonney2kotherwise we save things twice or even more often19:49
@sonney2kand of course behavior changes since objects that were just pointers to the same memory location19:49
@sonney2kare now clones19:49
blackburnsonney2k: okay I will add reset_refcount method and call it after deserialization19:51
blackburnthat should fix the process a little19:51
@sonney2kblackburn, why reset?19:51
@sonney2k1 is OK19:51
blackburnsonney2k: reset = set to 119:51
@sonney2kblackburn, yeah but it is one no?19:51
blackburnwhy? where is that set?19:52
@sonney2kwhat is it?19:52
blackburnokay you are creating machine19:52
blackburnplay around with w19:52
blackburnw refcounter is 919:52
blackburnthen you load it19:52
blackburnthen no need to touch it??19:53
blackburnokay I lost the idea what I need to fix19:53
@sonney2kyeah I also don't know19:54
@sonney2kit should not create leaks19:54
@sonney2kor heiko means some other leaks19:54
@sonney2kmaybe *when* (de)serializing?19:55
@sonney2kI mean in that code?19:55
blackburnsonney2k: if you run any base or serialization or parameter libshogun example it leaks19:56
blackburnsonney2k: okay so what about 2.0rc on 1st of September and 2.0 in the end of September?20:01
shogun-buildbotbuild #483 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Soeren Sonnenburg <>20:04
blackburnwiking: please fix latent svm unit test20:05
shogun-buildbotbuild #382 of deb2 - static_interfaces is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #484 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at
wikingblackburn: mmmm20:52
@sonney2kwiking, what are the required packages to run the unit tests and how?21:05
@sonney2kblackburn, do you also get an error in regression_gaussian_process_modular.py21:05
wikingsonney2k: ubuntu?21:05
wikingsonney2k: you need google-mock21:06
wikingand need to install as there's no package for it under ubuntu (just -dev package)21:06
@sonney2kwiking, both debs?21:07
wikingsonney2k: haven't checked21:07
wikingmaybe debian has it21:07
@sonney2klibgtest-dev ?21:07
wikingsonney2k: have u managed to get the static-build work?21:07
wikingsonney2k: u would need libgtest021:07
wikingso debian has it21:08
@sonney2k(wheezy has it)21:08
@sonney2kwiking, now what?21:08
wikingok after this21:08
wikingrun ./configure21:08
wikingand hopefully it'll detect the mocking framework for you21:09
wikingcan u paste me here the configure.log part?21:09
@sonney2kI installed google-mock21:09
@sonney2k/usr/bin/ld: cannot find -lgmock21:09
@sonney2k/usr/bin/ld: cannot find -lgtest21:09
wikingyeah google-mock will install libgtest-dev21:09
wikingbut not libgtest021:10
wikingcan u send me the compiling code?21:10
wikingwhat are the actual flags21:10
@sonney2kwiking, I have /usr/lib/
@sonney2kbut no libgmock or so21:11
wikingdpkg -L google-mock21:11
wikingdoes not install libgmock ?121:11
wikingis this deb wheezy?21:12
@sonney2kdpkg -L google-mock | grep .so21:12
blackburnsonney2k: I'll check but it was failing before, yes21:12
wikingwhat does it have then?21:12
@sonney2kgmock_gen and a buch of headers21:12
wikingdoes it have maybe then libgmock ?21:13
@sonney2kThe Google C++ Mocking Framework uses conditional compilation for some21:13
@sonney2kthings.  Because of the C++ "One Definition Rule", gmock must be21:13
@sonney2kcompiled with exactly the same flags as your C++ code under test.21:13
@sonney2kBecause this is hard to manage, upstream no longer recommends using21:13
@sonney2kprecompiled libraries.21:13
@sonney2kthe new recommendation is:21:14
@sonney2kg++ -I/usr/src/gmock -c /usr/src/gmock/src/gmock-all.cc21:14
@sonney2kg++ -I/usr/src/gtest -c /usr/src/gtest/src/gtest-all.cc21:14
@sonney2kar -rv libgmock.a gmock-all.o gtest-all.o21:14
@sonney2kto build libgmock statically!21:14
@sonney2keveryone has to do it before running tests21:14
@sonney2kwiking, same with libgtest btw!21:15
wikingyeah i gathered now21:15
wikingthat this is why there's no libgtest021:15
@sonney2kactually squeeze still had libgtest21:16
@sonney2kwheezy does not21:16
blackburnsonney2k: yeah failing still21:30
blackburnsonney2k: we are waiting for jacob's fix21:35
blackburnokay leaving now21:35
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]21:36
naywhayarehello there, can someone tell me if I am performing this task in the most computationally efficient way? (task to follow)22:06
naywhayareI have one dataset (no test/train split) and wish the find the K nearest neighbors (index of neighbors and distances) of each point in the dataset22:07
naywhayarethe only way I can seem to find to do this is to use the KNN classifier, which is more than I want to do; I just want the neighbors, no classification22:07
naywhayareis there a way to do this?  currently I am using the command-line interface but it seems to me like that has less functionality than some other interfaces to the library22:08
naywhayarethanks in advance for any help.22:09
gsomixsonney2k, hey. :)22:21
@sonney2knaywhayare, I guess the most effictive would be to create the kd-tree like k-nn does yourself but that would mean you have to get your hands dirty too22:26
@sonney2kand write C++ cod22:26
@sonney2kgsomix, hey...22:26
@sonney2kwiking, do you intend to fix the gtest/gmock stuff?22:27
gsomixsonney2k, how can I add something in python_modular/tools?22:28
@sonney2kjust add it there...22:34
@sonney2kbut wnat do you need?22:34
naywhayaresonney2k: I don't mind getting my hands dirty; if there's not a nice way to do it through one of the interfaces I'll just write up a little C++ utility22:34
naywhayarethanks for th einfo22:34
naywhayare*the info22:34
@sonney2knaywhayare, we actually don't mind you do it and share the code later on via a PR such that it can be used from some other interface22:35
naywhayareokay, I'll share the code but it'll be brutally simple and probably not useful to anybody but me22:36
gsomixsonney2k, add in repo.22:36
gsomixhm... I want -> tools/multiclass_shared.py22:36
@sonney2kgsomix, git mv  tools/multiclass_shared.py22:40
gsomixsonney2k, ok22:40
* sonney2k Zzzzz22:40
gsomixsonney2k, nite22:41
naywhayaream I correct in understanding that the CKNN class either uses cover trees or does the naive O(n^2) computation?  I don't see kd-trees being created there22:53
naywhayareforgive me if I've missed something simple as I am unfamiliar with shogun internals22:53
--- Log closed Thu Aug 30 00:00:17 2012