Open in new window / Try shogun cloud
--- Log opened Fri Jun 08 00:00:41 2012
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 245 seconds]00:33
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has joined #shogun00:35
KMcQuistenHello all00:35
KMcQuistenI have some questions.  I've been trying to use the new DirectorKernels, but I'm having some trouble.00:35
KMcQuistenAnyone awake to give me a hand?00:35
@sonney2kKMcQuisten, well ask... if no one answers maybe tomorrow someone will00:36
KMcQuistenAllrighty00:36
KMcQuistenI noticed in the example provided, the kernel function is accessing a global variable in order to calculate the kernel function in terms of the indices00:37
KMcQuistenThis is really horrible coding style, and I was hoping there was a way to acess data for the kernel internally using the init() function and the acessors built in to the kernel object00:38
KMcQuistenBut when I use get_lhs() to get one of my features, it has no get_feature_vector method since it's returned only as a CFeatures, not the descended class that I instantiated.00:38
@sonney2ktrue00:39
KMcQuistenSo how can I access the data to overload kernel_function?00:39
@sonney2kthat's a limitation of going from type aware to type unaware languages00:39
@sonney2konly fix is to introduce (in C++) a function that dyn_casts into the right type00:40
@sonney2kthere really is no other way...00:41
n4nd0I remember we talked about these functions for doing casts00:41
KMcQuistenWhat are the chances of that getting added in?  It seems like without it the DircetorKernel framework isn't as useful as it could be00:41
n4nd0sonney2k: isn't what we did for the labels?00:41
@sonney2kn4nd0, yes00:41
@sonney2ksomeone just has to write those...00:42
@sonney2kwe do this in CBinaryLabels etc too00:42
@sonney2kanyway00:42
* sonney2k Zzzzz....00:42
KMcQuistenHa.  I know it's late over there00:42
n4nd0sonney2k: good night then00:42
n4nd0KMcQuisten: would you like to add this functionality?00:43
n4nd0there is something equivalent for labels already done00:43
KMcQuistenn4nd0, how much work would it take to include that sort of thing for feature accessors within kernels and for kernel accessors within SVMs?00:43
n4nd0the methods are called obtain_from_generic00:43
KMcQuistenI could give it a shot00:43
KMcQuistenif I had the example from the labels implementation i could probably figure it out00:44
n4nd0yeah, the methods are really short00:44
KMcQuisteni'm assuming it's in the definition for Lavbels?00:44
n4nd0do you have the code locally?00:44
KMcQuistenYes I do, but I don't have authorization to commit to the repository00:44
n4nd0most of us don't have it :)00:44
KMcQuistenI"ll see if I can implement it, then we can worry about that, though00:44
n4nd0we have a fork in github and do pull requests there00:45
KMcQuistenWell yeah, don't want all kinds of pushes walking all over each other00:45
n4nd0the method is implemented in the children of CLabels00:45
n4nd0C{Binary, Multiclass, Regression}Labels00:45
KMcQuistenSo i'd have to implement it in all the Cfeatures children i want to use00:45
KMcQuistenAnd similarly if I want to pull kernels from SVMs00:46
n4nd0I think so00:46
n4nd0I am not totally sure though, let me think about it a moment00:46
n4nd0for the info you have written I would say so00:47
n4nd0but I suggest that you first implement it for the class you were talking about and try it out00:47
KMcQuistenOh of course00:47
KMcQuistenNow, I have pulled down the code previously using git00:48
KMcQuistenbut I pulled down the master00:48
KMcQuistenis there another branch I should pull before I try this00:48
KMcQuisten?00:48
n4nd0no00:49
n4nd0just create a new branch and do your stuff there00:49
KMcQuistenSounds good00:49
KMcQuistenif/when I get it working I'll come back here and figure out how to proceed00:50
n4nd0I am checking a thing for you first00:50
KMcQuistenOk00:50
n4nd0since in labels we implemented this LabelsType class00:50
n4nd0that is precisely used in obtain_from_generic00:51
n4nd0there is no such a FeaturesType class though00:51
KMcQuistenSo it would have to be written as well00:51
n4nd0I said nothing00:51
KMcQuisten?00:52
n4nd0that class exists :)00:52
KMcQuistenAh!00:52
n4nd0shogun/features/FeatureTypes.h00:52
KMcQuistenAh, yes, with the Enums00:52
n4nd0I think that you may use here the EFeatureClass tyep00:52
n4nd0type*00:52
n4nd0that you can get it using the method get_feature_class() that should be defined for all the features hierarchy00:53
KMcQuistenYes, I do have access to that from cfeatures00:54
n4nd0nice00:54
n4nd0a btw, if it is not a huge trouble00:55
n4nd0mind to do commits that leave the code in compilable state00:55
n4nd0the buildbot compiles commit by commit so in that way the proccess goes smooth and faster00:56
KMcQuistenYes.  CI is a good thing00:56
n4nd0I am used to small commits and I broke it some times at the beginning :D00:57
n4nd0it is not a big deal in any case00:57
KMcQuistenI'll only commit to my local branch one I get the thing working00:57
n4nd0that's the spirit00:57
KMcQuistenMy advisor's a stickler for making sure those of us in the group use good version control habits00:58
n4nd0that's very very good00:59
n4nd0I still cannot believe that they didn't tell me about VC in any of my courses00:59
KMcQuistenSo i'll branch the code off of the master, make my edits, commit them only to my branch, then come back here and go over how to properly merge them in01:00
KMcQuistenSound like a plan?01:00
n4nd0yes01:01
n4nd0before you start get the final version of the code in the repo01:01
KMcQuistenYes, i'll pull the latest before I start01:02
n4nd0in master git pull --rebase upstream master or equivalent01:02
KMcQuistenThe whole line?  I'm not yet totally familiar with all of git's command line options01:03
KMcQuistenNM, I got it01:04
KMcQuisten:)01:04
n4nd0from master it would be enough just to do01:04
n4nd0git pull upstream master01:04
n4nd0since you don't normally use master as a branch for development01:04
n4nd0but I think it is a good practice to use the rebase so you keep it in your head for the moments when it really does matter01:05
KMcQuistenRight01:05
n4nd0e.g. when you have working in your branch for some days and want to sync with master before merging01:05
n4nd0then it is good that you rebase instead of merging with git pull01:05
KMcQuistenAh, yes01:05
n4nd0I am far from being a git expert so be critic towards my comments too ;)01:06
KMcQuistenNo worries.  I fully plan to RTFM01:06
KMcQuistenSo in the most recent build, Is the RealFeatures wrapper for python_modular named that way for backwards compatibility?  Is the code in the DenseFeatures.cpp file, since there is no RealFeatrues.cpp?01:24
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has quit [Quit: Page closed]01:31
n4nd0ups sorry I missed that01:37
n4nd0I guess that is RealFeatures because it has not been updated, a bug01:37
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]03:18
-!- romi_ [~mizobe@187.66.121.115] has quit [Remote host closed the connection]06:47
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun07:15
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host]07:15
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun07:15
-!- gsomix [~gsomix@109.169.255.59] has joined #shogun07:37
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun07:56
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]08:13
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun08:24
-!- gsomix_ [~gsomix@109.169.153.39] has joined #shogun08:26
-!- gsomix [~gsomix@109.169.255.59] has quit [Ping timeout: 252 seconds]08:29
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun08:53
-!- gsomix__ [~gsomix@109.169.230.79] has joined #shogun08:56
-!- gsomix_ [~gsomix@109.169.153.39] has quit [Ping timeout: 245 seconds]09:00
-!- uricamic [~uricamic@2001:718:2:1634:284e:b30f:f03f:5117] has joined #shogun09:00
-!- av3ngr [av3ngr@nat/redhat/x-oexcxshsepjrscvy] has joined #shogun09:05
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]09:32
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:39
-!- av3ngr [av3ngr@nat/redhat/x-oexcxshsepjrscvy] has quit [Quit: That's all folks!]09:45
-!- av3ngr [~av3ngr@60-241-222-244.static.tpgi.com.au] has joined #shogun10:10
-!- av3ngr [~av3ngr@60-241-222-244.static.tpgi.com.au] has quit [Remote host closed the connection]10:10
-!- blackburn [~blackburn@31.28.59.65] has quit [Ping timeout: 260 seconds]10:10
gsomix__sonney2k, hey10:10
-!- gsomix__ is now known as gsomix10:10
-!- foo__ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has joined #shogun10:21
foo__hi10:21
foo__I would like to know if you planed to add into multiclass evaluation, equal error rate for each class ? thx.10:22
-!- blackburn [6da9c935@gateway/web/freenode/ip.109.169.201.53] has joined #shogun10:40
-!- blackburn [6da9c935@gateway/web/freenode/ip.109.169.201.53] has quit [Ping timeout: 245 seconds]11:05
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun11:21
n4nd0gsomix: is it really that written http://9gag.com/gag/4410715 :D?11:33
gsomixn4nd0, hah, everything is possible in Russia11:36
n4nd0haha so really?11:37
gsomixIt's example of stupid transliteration.11:41
gsomix"Oh god NO! That should be ice cream in assortment."11:41
-!- ckwidmer [~chris@HSI-KBW-046-005-237-106.hsi8.kabel-badenwuerttemberg.de] has joined #shogun11:51
-!- emrecelikten [~emrecelik@176.40.238.218] has joined #shogun12:02
-!- hoijui [~hoijui@dslb-088-074-106-085.pools.arcor-ip.net] has joined #shogun12:03
-!- blackburn [6da9c935@gateway/web/freenode/ip.109.169.201.53] has joined #shogun12:08
-!- blackburn [6da9c935@gateway/web/freenode/ip.109.169.201.53] has quit [Quit: Page closed]12:27
n4nd0sonney2k: around?13:15
n4nd0does someone know how to add a module for modular interfaces?13:29
n4nd0I created a directory so with some classes inside13:29
n4nd0I have modified the files for swig insid interfaces/modular to include these files13:29
n4nd0but I still got ImportError: No module named SO in my python program13:29
ckwidmerhi n4nd013:31
n4nd0ckwidmer: hi13:31
ckwidmerhave you trying to replicate the steps needed for another module13:31
ckwidmer?13:31
ckwidmerusually what I do when e.g. adding a new class13:32
n4nd0yes, but it seems that there must be a step that I am missing13:32
ckwidmercheck where another particular class name occures in interface files etc13:32
ckwidmerhave you grepped through all the files?13:32
n4nd0yeah I am grepping all the way :)13:32
ckwidmerhmm13:33
n4nd0but I must be missing something13:33
ckwidmerthe other (rather obvious) thing that comes to my mind is make clean13:33
ckwidmerbefore compiling?13:33
n4nd0I thought that they key was to add my SO.i and SO_includes.i in interfaces/modular/modushogun.i but didn't work13:33
n4nd0I actually changed a make file too13:34
n4nd0so I think that it built everything again but I am going to try forcing the clean13:34
ckwidmeroh btw13:36
ckwidmerhave you talked to nico about the cplex error?13:36
n4nd0oh I remember he pasted for me on a mail about an error13:37
ckwidmeryes13:37
n4nd0but I didn't look into it, sorry13:37
ckwidmerthere is some problem with CLabels13:37
ckwidmerif cplex compile flag is set13:38
n4nd0ahm13:38
n4nd0can you paste the error for me again please?13:38
gsomixn4nd0, do you use python_modula?13:38
n4nd0in the mail it appears with quite a few of strange characters13:38
ckwidmerwell, you can reproduce it if you have cplex13:39
n4nd0gsomix: yeah, configure is run for python-modular13:39
ckwidmerotherwise it'll be hard to do13:39
n4nd0ckwidmer: ok, I'll do that then once I am done with my multiclass example for SO13:39
gsomixah, yes. n4nd0, maybe you should look at makefiles?13:39
ckwidmerso, back to your module problem, so you've included SO.i and SO_includes.i from modshogun.i?13:40
n4nd0ckwidmer: yes13:40
ckwidmerand it compiles just fine, but you don't see it from python?13:40
n4nd0gsomix: so in shogun/Makefile I added SO in a for loop inside install-python-modular13:40
n4nd0ckwidmer: make and make install work well13:41
n4nd0ckwidmer: but from shogun.SO import bla bla bla doesn't work13:41
n4nd0there must be somewhere where to define explicitily a module for modular interfaces13:42
n4nd0mmm I have just found something that could be the answer to my problem in shogun/CMakeLists.txt13:44
n4nd0I am going to try13:44
n4nd0but actually I am not using cmake so I don' have much hope here13:45
ckwidmern4nd0, have you checked whether your new module (or files therein) is included in some "intermediate" file, such as python_modular/modshogun_wrap.cxx13:47
n4nd0ckwidmer: so I have several lines like this one13:48
n4nd0#define SWIGTYPE_p_shogun__CStructuredModel swig_types[947]13:48
n4nd0CStructuredModel is something I have defined inside shogun/so/13:48
ckwidmerok13:49
ckwidmeralso the includes?13:49
ckwidmere.g.  #include <shogun/classifier/LPBoost.h>13:49
n4nd0ckwidmer: yes13:51
n4nd0although I have found that python_modular/Makefile didn't contain SO in a for loop it does in what seems to me each of the modules13:51
n4nd0so I have added SO and I am building again13:51
n4nd0nice it worked!!13:52
ckwidmerawesome13:52
n4nd0yeah13:52
ckwidmeryeah, just saw that loop13:52
n4nd0but I don't think it is the good solution though13:52
n4nd0isn't this makefile automatically generated?13:53
ckwidmeris that makefile autogenerated?13:53
ckwidmeri guess not13:53
n4nd0:)13:53
n4nd0no idea13:53
n4nd0I will see next time I build13:53
ckwidmerwell13:53
n4nd0it is not in the repo so probably it is autogenerated13:53
ckwidmerif it isn't13:53
ckwidmerthis should be documented13:54
ckwidmeryeah, I would think os13:54
ckwidmerso13:54
n4nd0agree13:54
ckwidmerhave you tried configure again13:54
n4nd0I read README.developer but there is nothing about creating new modules13:54
n4nd0no, I have not13:54
ckwidmeryeah13:55
ckwidmerI think just quickly figure out if its autogen13:55
n4nd0I hope you don't mind but I will do this later since I don't want to re-build everything again now :S13:57
ckwidmerjust don't forget about it, as you've figured it out now - if we don't get this right, now someone else will have to retrace your steps in the future13:57
n4nd0yes13:58
ckwidmergreat :)13:58
n4nd0ckwidmer, gsomix : thank your very much for your help guys!13:59
ckwidmeryou're welcome14:00
n4nd0I am very excited about doing this test :D14:02
* n4nd0 has changed to debug mode after weeks of untested coding14:06
n4nd0ckwidmer: do you have another moment?14:56
-!- uricamic1 [~uricamic@2001:718:2:1634:9d6b:332c:ffea:54e8] has joined #shogun14:56
-!- uricamic [~uricamic@2001:718:2:1634:284e:b30f:f03f:5117] has quit [Ping timeout: 248 seconds]14:58
ckwidmern4nd0 sure15:10
ckwidmerwhat's up?15:10
n4nd0yeah, I have a python/swig problem that I don't manage to fix15:10
n4nd0I am sorry, I think I am kind of abusing consulting today :S15:11
ckwidmerno worries, go ahead15:11
n4nd0it looks like this15:11
n4nd0NotImplementedError: Wrong number or type of arguments for overloaded function 'new_PrimalMosekSOSVM'.15:11
n4nd0  Possible C/C++ prototypes are:15:11
n4nd0    shogun::CPrimalMosekSOSVM::CPrimalMosekSOSVM()15:11
n4nd0    shogun::CPrimalMosekSOSVM::CPrimalMosekSOSVM(shogun::CStructuredModel *,CLossFunction *,shogun::CStructuredLabels *,shogun::CFeatures *)15:11
n4nd0I have checked that the arguments are correct in order15:11
ckwidmerhmm15:12
n4nd0I find weird that the second argument doesn't have shogun::15:12
ckwidmermaybe a good place to start is trying methods that take only CStructuredModel or CStructuredLabels as arguments15:12
ckwidmertrue15:12
ckwidmerare you sure the individual ones work elsewhere?15:13
ckwidmerpossibly your new classes aren't registered correctly15:13
n4nd0do you mean like for example using setters to try them out?15:13
ckwidmeryes15:13
ckwidmerexactly15:13
ckwidmerso you can narrow down the problem15:13
n4nd0I am going to create those setters then and try it15:14
ckwidmeralright15:14
ckwidmerif you discover that it doesn't work with one particular class15:14
ckwidmerretrace all the occurences of another class15:15
ckwidmerand make sure you have it registered in all right places15:15
n4nd0I don't think I understand what do you mean with retrace all the occurrences of another class15:16
ckwidmerwell15:20
ckwidmergrepping through all files15:20
ckwidmerlooking for e.g. LPBoost15:20
n4nd0ok, I think I got you now15:21
gsomixgedit works with modshogun_wrap.cxx too slow :(15:31
n4nd0too many lines there15:32
n4nd0gsomix: join the vim club!15:33
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has joined #shogun15:34
n4nd0ckwidmer: I have found the arguments that fire the error, I am passing a child class and the method is declared to accept the parent15:35
n4nd0ckwidmer: maybe I have to do something in swig to indicate inheritance?15:35
n4nd0hey nicococo15:35
nicococohi n4nd0 & chrisssss15:36
n4nd0nicococo: I finished the multiclass model but I am still stuck with the example15:37
n4nd0nicococo: quite a few errors appearing :(15:37
n4nd0nicococo: please, tell me what do you think https://github.com/iglesias/shogun/blob/so/src/shogun/so/MulticlassModel.cpp15:38
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun15:38
nicococon4and0: very good! i don't see any logical mistakes :)15:41
n4nd0nicococo: nice then!15:41
n4nd0nicococo: let's see if I manage to solve the problems I am finding15:42
n4nd0they are turning out to be tough ones :S15:42
nicococoohh. why?15:42
n4nd0I don't manage to solve them :P15:42
nicococoi see15:43
n4nd0I don't know if they are just swig/python related or of the framework in general15:43
nicococookay.. i am sorry i can not help with that :(15:44
n4nd0it is like if the hierarchy is not detected  when I want to make use of it from python15:44
n4nd0e.g. I do in an object called sosvm that is of type PrimalMosekSOSVM15:44
n4nd0sosvm.set_model(model)15:44
n4nd0model is of type MulticlassModel15:44
n4nd0MulticlassModel inherits from StructuredModel15:44
n4nd0but it tells me that I cannot do that set_model15:45
n4nd0that the argument needs to be StructuredModel15:45
n4nd0:S15:45
nicococooha15:46
n4nd0nicococo: just tell me if you have any guess, I am already witout ideas ...15:47
nicococowell its rather technical .. i wouldn't guess, instead i would call soeren :)  (or someone else like ckwidmer)15:48
n4nd0blackburn: hey! do you have any idea about it ^?15:49
blackburna min15:49
n4nd0thank you15:50
blackburnn4nd0: which problem exactly?15:50
n4nd0when I do set_model(model) I get15:51
n4nd0TypeError: in method 'StructuredOutputMachine_set_model', argument 2 of type 'CStructuredModel *'15:51
n4nd0the object model I am using is from a child class of CStructuredModel15:51
-!- ckwidmer [~chris@HSI-KBW-046-005-237-106.hsi8.kabel-badenwuerttemberg.de] has quit [Remote host closed the connection]15:51
blackburnwhat is header?15:51
n4nd0it is like it doesn't recognize the inheritance15:52
n4nd0https://github.com/iglesias/shogun/blob/so/src/shogun/so/MulticlassModel.h15:52
n4nd0that is the child class15:52
n4nd0here is the parent15:52
n4nd0https://github.com/iglesias/shogun/blob/so/src/shogun/so/StructuredModel.h15:52
blackburnno swig works perfectly with inheritance :)15:52
n4nd0yeah sure15:52
n4nd0I meant it more like I am doing something wrong somewhere about it :D15:52
n4nd0I have the same problem with another member CLossFunction15:53
blackburnjust some quick guess while I'm thinking15:53
blackburntry to add "class CStructuredModel;" in multiclass model15:53
blackburnand check whether they are included in SO.i in a proper order15:54
n4nd0all right15:55
n4nd0first parent, then child right?15:55
blackburnyes15:55
n4nd0they were in the right order15:56
n4nd0I am compiling after including class CStructuredModel15:56
blackburnaham15:57
n4nd0it didn't make the trick :(16:00
blackburnhmm16:05
blackburnbad16:05
n4nd0I think I will have to wait and ask Soeren about this16:07
n4nd0nicococo: I will try later to do the example in C++16:13
n4nd0I don't think that this error will appear there16:14
nicococook16:14
n4nd0nicococo: I can let you know by mail once I know something more about it ok?16:14
nicococothat would be fine16:14
n4nd0nicococo: did you take a look to the other parts of the code as well?16:15
n4nd0MulticlassSOLabels and PrimalMosekSOSVM16:15
nicococoyep16:15
n4nd0anything about it?16:16
nicococolooks pretty good :)16:16
nicococoi need to have a look again on PrimalMosekSOSVM16:17
n4nd0ok16:17
nicococolooks okay to me too.. so, i am looking forward to hear some exciting news about working multiclass example ;)16:19
n4nd0ok16:20
n4nd0I will let you know :)16:20
n4nd0I am off now for a while so bye!16:20
nicococohappy coding.. see you16:21
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]16:21
-!- nicococo [~nico@lacedcoffee.ml.tu-berlin.de] has left #shogun []16:21
-!- hoijui [~hoijui@dslb-088-074-106-085.pools.arcor-ip.net] has quit [Quit: Leaving]16:23
gsomixvim .__.16:29
-!- uricamic1 [~uricamic@2001:718:2:1634:9d6b:332c:ffea:54e8] has quit [Quit: Leaving.]16:33
gsomixmidnight commander and nano! :)16:34
blackburnnano sucks16:38
gsomix:(16:39
blackburn145             //dear god, this is super verbose, isn't there a way to shorten this?16:48
blackburn:D16:48
-!- gsomix [~gsomix@109.169.230.79] has quit [Ping timeout: 252 seconds]17:04
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has joined #shogun17:05
KMcQuistenHello17:05
blackburnhey17:05
KMcQuistenHow's things?17:06
blackburngoing on :)17:07
-!- gsomix [~gsomix@109.169.230.79] has joined #shogun17:07
blackburnKMcQuisten: is it ok now with mkl?17:07
KMcQuistenYeah, the MKL is working.  Avoiding the LINADD seems to help.  I'm working on something else now.17:08
KMcQuistenI'm really excited about using DirectorKernels, but the examples acess data from a global.  to access data from withing one, using get_rhs() and get_lhs(), i'm working on a version of obtain_from_generic for DenseFeatures and StringFeatures17:09
blackburnah that's nice17:10
gsomixKMcQuisten, hello. Sorry, I don't replied to your mail. I'm working on this problem now.17:11
KMcQuistenWell, I wrote up the code, it compiles properly, wraps properly in swig, and when I test it I can extract the data.  However, I get the following code when I actually try and use it in a kenrel in LibSVR:17:11
blackburnKMcQuisten: do you have a fork?17:12
KMcQuistenYes17:12
blackburnso will you do a pull request with it?17:12
KMcQuisten(This is the output from training with MSG_DEBUG on17:12
KMcQuisten:17:12
KMcQuisten[INFO] 522 trainlabels17:12
KMcQuisten[INFO] training one class svm or doing epsilon sv regression17:12
KMcQuisten[INFO] Computing gradient for initial set of non-zero alphas17:13
KMcQuistenFatal Python error: GC object already tracked17:13
KMcQuistenAborted17:13
blackburnwhoops17:13
KMcQuistenIf I can get it to work, yes.17:13
KMcQuistenWhat kind of approach have you taken with the, gsomix17:13
KMcQuisten?17:13
KMcQuistenI know you're working on Director stuff, so I don't want to step on your solution if you have one17:14
KMcQuistenAll i've done is modify the obtain_from_generic used in BinaryLabels, and adapeted it to the features17:15
blackburnKMcQuisten: what is code of that compute method?17:15
KMcQuistendef kernel_function(self, idx_a, idx_b):17:16
KMcQuistengenf1 = self.get_lhs()17:16
KMcQuistengenf2 = self.get_rhs()17:16
KMcQuistenf1 = StringCharFeatures.obtain_from_generic(genf1)17:17
KMcQuistenf2 = StringCharFeatures.obtain_from_generic(genf2)17:17
KMcQuistens1 = f1.get_feature_vector(idx_a).tostring()17:17
gsomixKMcQuisten, I'm trying to make automagically converting in swig typemaps.17:17
KMcQuistens2 = f2.get_feature_vector(idx_b).tostring()17:17
KMcQuistenreturn exp(-self.gamma * editdist.distance(s1,s2))17:18
blackburnaham17:18
KMcQuistengamma is a parameter of the object I descended from DirectorKernel17:18
blackburncould you please cut after genf2 and return 0.017:18
KMcQuistengsomix:  That sounds like the best way17:19
KMcQuistenblackburn: Sure17:19
blackburnjust to check whether obtain does something wrong17:19
KMcQuistenit gets to [INFO] Computing gradient for initial set of non-zero alphas, and then segfaults17:20
blackburnwow17:20
-!- foo__ [2e1fd566@gateway/web/freenode/ip.46.31.213.102] has quit [Ping timeout: 245 seconds]17:20
blackburnand without anything? just return 0.017:20
KMcQuistenlets see17:21
KMcQuistenFatal Python error: GC object already tracked17:22
blackburnhmm that sounds bad17:22
KMcQuistenYeah, something's getting garbage collected before it should17:22
KMcQuistenI looked around to find a solution, and the only seemingly informative thing I've found is this:17:23
KMcQuistenhttp://pyrit.wordpress.com/2010/02/18/385/17:23
KMcQuistenIt has an explanation of a similar problem, along with a specific link to some python documentation17:24
KMcQuistenBut I don't know enough about SWIG to do anythign useful with this info17:24
KMcQuistengsomix:  What do you think?  How's the typemapping work going so far?17:24
blackburnhmm let me check17:25
blackburnKMcQuisten: I'll check typemaps17:25
KMcQuistenOk, if I strip out everything but return 0 in the kernel_function, and use only the DirectorKernel default __init__(), I get this:17:28
KMcQuistenterminate called after throwing an instance of 'Swig::DirectorMethodException'17:29
blackburnhmm17:29
CIA-9shogun: Sergey Lisitsyn master * rf54eb89 / (8 files in 3 dirs): Merged code of LibLinearMTL by Chris Widmer - http://git.io/UwYHNg17:33
gsomixKMcQuisten, 'Swig::DirectorMethodException' | it's seems that I forget about some virtual methods.17:34
gsomixsorry, need some time to check .____.17:34
KMcQuistenNo sweat :)17:35
gsomixKMcQuisten, aha, forgot17:43
KMcQuisten?17:43
gsomixKMcQuisten, forgot to add some virtuals methods from CKernel to CDirectorKernel. it's might be a reason of 'Swig::DirectorMethodException'.17:45
KMcQuistenAh, I see17:46
KMcQuistenLet me know how it goes so I can test it18:09
KMcQuisten:)18:09
gsomixKMcQuisten, ok.18:10
gsomixKMcQuisten, little experiment with automagically converting (CFeatures* to CDenseFeatures*) is works18:12
KMcQuistenSweet!18:13
KMcQuistenThat and CStringKernels are the two main types I use18:13
gsomixI need time to write complete converting.18:13
KMcQuistenRight18:13
-!- KMcQuisten_ [d8338942@gateway/web/freenode/ip.216.51.137.66] has joined #shogun18:18
KMcQuisten_Can someome please kill ym clone18:19
KMcQuisten_my18:19
blackburnwhat?18:19
blackburnahh18:19
blackburnI can't18:19
KMcQuisten_I have a duplicate in the chat room18:19
blackburnsonney2k could but I guess he is afk18:19
KMcQuisten_Well, shucks.  it'll time out eventually18:19
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has quit [Ping timeout: 245 seconds]18:20
-!- KMcQuisten_ is now known as KmcQuisten18:20
-!- KmcQuisten is now known as KMcQuisten18:20
KMcQuistenbbiab18:21
-!- KMcQuisten is now known as KMcQuisten-afk18:21
-!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: Connection reset by peer]18:54
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun18:54
-!- KMcQuisten-afk [d8338942@gateway/web/freenode/ip.216.51.137.66] has quit [Quit: Page closed]19:18
emreceliktenOnly he can kill clones as far as I know19:27
emreceliktenusing /msg nickserv ghost19:28
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has joined #shogun19:44
KMcQuistenback19:55
CIA-9shogun: Soeren Sonnenburg master * rabe2a85 / (81 files in 26 dirs): Clean up CMath: move respective functions to CStatistics, SGVector and SGMatrix - http://git.io/T9L6bA20:56
blackburnthe commit from island20:56
CIA-9shogun: Soeren Sonnenburg master * r25d7549 / (2 files): hopefully fix b0rken merge attempt in statistics - http://git.io/a1kTSA21:10
-!- wiking_ [~wiking@huwico/staff/wiking] has quit [Quit: wiking_]22:40
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.216.51.137.66] has quit [Quit: Page closed]22:40
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 260 seconds]23:20
-!- gsomix [~gsomix@109.169.230.79] has quit [Ping timeout: 256 seconds]23:24
--- Log closed Sat Jun 09 00:00:41 2012