Open in new window / Try shogun cloud
--- Log opened Wed Oct 17 00:00:17 2012
-!- heiko1 [] has left #shogun []00:35
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]00:36
-!- heiko [] has joined #shogun01:12
shogun-buildbotbuild #118 of nightly_none is complete: Failure [failed compile]  Build details are at
shogun-buildbotbuild #135 of nightly_default is complete: Failure [failed test]  Build details are at
-!- adoniscik [] has joined #shogun05:29
-!- heiko [] has quit [Quit: Leaving.]08:47
-!- Dave [863ab323@gateway/web/freenode/ip.] has joined #shogun09:06
-!- Dave is now known as Guest2307409:06
Guest23074The original LibSVM has support for class weights (to compensate for unbalanced problems). Can Shogun also do that? Couldn't find anything in the documentation though..09:08
Guest23074svm.set_C(c_neg,c_pos)  -- nice :)09:29
-!- Guest23074 [863ab323@gateway/web/freenode/ip.] has quit [Quit: Page closed]10:02
-!- adoniscik [] has quit [Ping timeout: 260 seconds]11:41
wikingsomebody review my PR and possibly apply:
sonne|workwiking: Operators berween?12:51
wikingsonne|work: yep13:07
wikingbut yeah i'll do some extra codes soon13:07
wikingsonne|work: you reckon that SGSparseVectorEntry<T>* is sorted in SGSparsevector by feat_index?13:12
wikingsonne|work: or should we add an option into SGSparseVector to do a sorting....13:16
sonne|workwiking: we should assume it is *always* sorted13:24
sonne|workbut document that we assume this!13:24
wikingsonne|work: oke13:26
-!- Dav_ [863ab324@gateway/web/freenode/ip.] has joined #shogun15:30
-!- Dav_ [863ab324@gateway/web/freenode/ip.] has quit [Client Quit]15:30
-!- blackburn [~blackburn@] has joined #shogun15:50
-!- David [863ab324@gateway/web/freenode/ip.] has joined #shogun16:01
-!- David is now known as Guest8887316:01
Guest88873Is it correct that there's no CEvaluation subclass that allows setting specific losses for TP,TF,FP and FP ?16:05
blackburnGuest88873: hey, could you please elaborate?16:06
blackburnwhat do you mean by specific losses?16:06
Guest88873I' like to parameters of an SVM based on its 'loss', where e.g. a false negative is considered 10 times worse than a false positive16:08
Guest88873That would be: Loss = 10*FN +1*FP16:09
blackburnGuest88873: oh got it, unfortunately that wasn't coded yet16:09
Guest88873ok, thx16:09
blackburnGuest88873: we have class computing all fps,tps,fns,tns so it could be easy to combine16:10
Guest88873related question, could I add such a class on the run-time using Python (I think not, but I hope I'm wrong:) )16:11
Guest88873No indeed, it wouldn't be that in C, but adding it in Python would be handier for experiments16:12
blackburnGuest88873: yeah you could but it would require some C++ code again16:44
blackburnGuest88873: we have such classes for structured risk models and kernels16:47
blackburnit is called 'director' classes16:47
blackburni.e. it has a virtual method in C++ that may be overloaded by python code16:48
blackburnand then used together with all the rest C++ code16:48
Guest88873sounds perfect, I'll check it out16:48
blackburnGuest88873: I'll add such class and example a little later today16:57
Guest88873Thanks blackburn, that would be nice, since - for the moment - I don't get which classes this concerns and how to do it..16:58
wikingblackburn: when u have time check out ;)17:08
blackburnwiking: yeah wanted to17:28
blackburnwill do in 5 mins17:28
wikingokok cool thnx17:33
wikingoh yeah i have a warning supresser fix17:53
wikinganything against it?17:54
blackburnno, looks ok17:54
blackburnwhat is the warning?17:54
wikingwell clang constantly giving me the warning17:55
wikingwhen i compile SpareFeatures.cpp17:55
blackburnthat won't work!17:55
wikingwhy not?17:55
blackburnI mean it is broken already17:55
blackburnone cannot simply copy array of referenced data17:56
wikingwell it's a copy constructor :)17:56
wikingaaah ok17:56
wikingi see17:56
wikingbecause we store reference counter there in the beginning of that array17:57
wikingor why?17:59
blackburnI mean we have to increase ref count of each vector18:02
blackburnhere comes da fix18:03
blackburnCIA passed away again18:04
wikingnew (&sparse_feature_matrix[i]) SGSparseVector<ST>();18:05
wikingi don't get this :D18:05
blackburnwiking: in-place new shit18:06
wikingso this is equivalent of sparse_feature_matrix[i] = new SGSparseVector<ST>();18:06
blackburnwiking: calloc doesn't call constructors18:06
blackburnsparse_feature_matrix[i] is not a pointer it is a memory area already18:07
blackburnwiking: new creates object using memory provided by something, here we provide where to store the object18:08
blackburnanother hardcore option to cut off a leg18:08
blackburnwe like such things here for a reason I do not understand :D18:09
wikingah ok18:09
wikingso you basically have this "placement new"18:09
blackburnsome of shogun's code should be posted in some pornhub or whatever like that18:09
wikingi.e. to just run a ctor on the already allocated memory18:09
blackburncategory 'extreme hardcore'18:09
wiking // Line #3 (see "DANGER" below)18:10
wikinglol "(if you don't know what "alignment" means, please don't use the placement new syntax). You have been warned."18:10
blackburn// here your leg is being cutted off18:10
wikingwell yeah18:11
wikingand is it aligned? :)18:12
wikingon all the possible os/arch variations :)18:12
blackburngood question18:12
wikingas i know that osx malloc will give always give u an aligned memory18:12
blackburndoes sizeof takes alignment into account?18:12
blackburnI hope it does18:12
wikingbut afaik that's only true for osx18:12
wikingi remember when i had troubles with alignment (3 years ago)18:13
wikingaligned memory by malloc was only true for osx18:13
wikinglinux and win did not do that18:13
blackburnwell actually I know it works now18:13
blackburnon linux :)18:13
wikinggood :)18:14
blackburnbut better way is to use new[] here18:14
wikingdo u know why does it only work (clang_complete)18:14
wikingif u do the include <eigen3/Eigen/Dense>18:14
blackburnoh some paths issue18:14
wikingmmm yeah i guess so but why18:14
wikingi cannot get any error out of vim18:14
blackburnthat's the path on my machine IIRC18:15
blackburnnot /usr/include/Eigen/18:15
wikingfor me it's /opt/local/include/eigen318:15
blackburnsee what I mean?18:15
wikingbut if you put18:15
wikingin your clang_complete?18:15
blackburnI didn't try18:15
blackburndid you?18:15
wikingdoes it work for you18:15
wikingbecause i've tried18:15
wikingand i get NO errors18:16
blackburnso works seamless?18:16
wikingbut still clang complete doesn't want to work with SGVector or SGSparseVector for that matter18:16
wikingcould u try plz?18:16
blackburnokay let me try18:16
shogun-buildbotbuild #555 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>, Viktor Gal <>18:16
blackburnwiking: where to put .clang_complete?18:17
wikingblackburn: hahaha have u seen that ? :)18:17
wikingi mean the error18:17
wikingit's some crazy shit18:17
blackburnheiko's thing18:17
wikingah ok18:17
wikingthen i ignore18:17
blackburnhe got into that forest with an axe18:18
wikingi put .clang_complete right in the root of shogun18:18
blackburnin a root or src?18:18
blackburncause I work from src18:18
wikingit looks like this for me18:18
wikingin the root18:18
wikingnot the src18:18
wikingsince this way it works as well18:18
wikingfor examples18:18
wikingas well18:18
blackburnI see18:19
wikingbut u can put it in ./src18:19
wikingthe only difference would be then the first -I18:19
wikingand i guess u don't need the last two flags18:19
wikingas that's for c++11 support in clang18:20
blackburnwiking: works18:20
wikingno fucking way :D18:20
wikingso you can do a this->18:20
wikingin SGVector.cpp18:20
wikingand it gives u the list18:20
wikingw/o errors18:20
blackburnthis-> doesn't work18:20
wikingthen i guess it's the same for u as well18:21
blackburnwiking: some things work18:21
wikingin SGVector.cpp:12318:21
wikingadd a line18:21
wikingand put18:21
blackburnno, doesn't work18:21
wikingand u should be able to get a list for SGSparseVector<T>18:21
wikingah yeah18:21
wikingand what does18:22
blackburnclapack not found18:22
wikingi get18:22
wikingthe usual pattern not found18:22
wikingbut when i do :copen18:22
wikingthere's noooo error at all18:23
wikingdo u have?18:23
blackburnyeah clapack thing18:23
blackburnmay be it is the cause18:23
blackburnlet me fix that18:24
wikinghehehe why don't i get any errors :D18:24
wikingsame here :)18:27
wikingok there's a python scrupt18:27
wikingin clang_complete18:27
wikingit suppose to generate u the right .clang_complete18:27
wikingnow i'm gonna try that18:28
shogun-buildbotbuild #556 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at  blamelist: Viktor Gal <>18:34
wikingok i have to change that18:39
wikingi mean in configure script18:39 was broken before me right?18:39
blackburnyeah I think so18:41
wikingthat python script doesn't help me at all18:45
shogun-buildbotbuild #557 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at  blamelist: Sergey Lisitsyn <>18:54
wikinglet g:clang_use_library = 118:56
wikingthe completion is faster for me18:56
wikingi wonder how can one debug clang_complete18:58
wikingah ok this is why18:58
wiking"Instead of calling the clang/clang++ tool use libclang directly. This gives18:58
wikingaccess to many more clang features. Furthermore it automatically caches all18:59
wikingincludes in memory. Updates after changes in the same file will therefore be a18:59
wikinglot faster."18:59
-!- Guest88873 [863ab324@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]19:05
wikingi'll never find out what's the problem here :D19:25
-!- adoniscik [] has joined #shogun19:30
blackburnwiking: that option is killing my vim :)20:15
wikingblackburn: does it have python support compiled?20:15
wikingw/o that it won't work20:15
blackburnhow can I?20:15
blackburndo tha20:16
wikingi don't know how u can check it if it has python support added20:17
wikingbut for the configure of vim you should add something like this20:17
wiking--enable-pythoninterp --with-python=${prefix}/bin/python2.720:17
wikingor wherever your python interpreter is20:17
wikingcan u do this in your vim20:18
wiking:python print "helo"20:18
wikingif not then pythoninterp is not enabled20:20
wikingthat's why the  g:clang_use_library = 1 does not work20:20
wikingotherwise i have no idea20:20
wikingi had this problem but my vim didn't crash just got an error msg20:20
blackburnI see thanks20:49
blackburnit works20:49
wikingheheh :)20:49
wikingfaster? :)20:49
blackburnnot yet20:49
blackburnI meant python thing20:50
blackburnoh works20:50
blackburnokay let me try20:50
blackburnargh -> doesn't work for me no matter where20:51
wikingmaybe you need to set the library path20:53
wikingwhere's your
wikingmoreover, do you have one? :)20:53
blackburnI think I have to compile shogun now21:01
wikingno21:01 should deal with it21:01
wikingbut try21:01
wikingi might be really wrong ehre :)21:01
blackburnmy installed headers are quite different from actual21:02
blackburnso I have to install I think21:02
-!- mode/#shogun [+o sonney2k] by ChanServ21:59
wikingblackburn: still here?23:50
blackburnwiking: yeao23:55
blackburnwiking: I'm watching my parents' flight via flightradar24 :D23:57
wikingblackburn: ehehe23:58
wikingblackburn: i think the new latent so-svm just overexceeds any previously published solver's performance on one of the benchmark datasets :)23:59
wikingboth in training time and precision :)23:59
--- Log closed Thu Oct 18 00:00:05 2012