@sonney2k @sonney2k @sonney2k --- Log opened Sat Oct 27 00:00:17 2012 blackburn, look may 7 2012 seems to have been the most active at #shogun 00:10 http://shogun-toolbox.org/page/contact/irclogs/ 00:10 is that due to us enabling the bots maybe? 00:11 sonney2k: gott! we were talking all day long 00:12 yeah crazy everyone was active that day 00:12 blackburn, I just got a confirmation from Carol - our application got submitted 00:13 sonney2k: nice 00:13 sonney2k: I am in MLdepression :( 00:14 we have to extend shogun somehow 00:14 Bayesian learning! 00:16 n4nd0: I am too stupid to understand bayesian stuff 00:17 :D 00:17 c'mon you are not 00:18 but seriously, I think Bayesian stuff can be really cool 00:18 no idea 00:18 although I must admit I have just seen a few fundamentals of it so far 00:18 I fucked up my brain with alexandrescu 00:19 can't see shogun code anymore 00:19 haha 00:19 what do you want to do about it? 00:19 may be lay somewhere and cry, may be travel around siberia 00:20 not sure what can i do with it 00:20 hehe 00:21 blackburn, write some nice example! 00:24 blackburn, I have something concrete in mind 00:24 sonney2k: they are all useless :( 00:24 what is your idea? 00:24 somthing that is live on the website basically 00:24 so just some 2-class svm but writting in java script / python django 00:25 + utilizing shogun underneath 00:25 and some fancy visualization via d3js 00:25 is it possible to call pythno from java script? 00:26 -!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Quit: Page closed] 00:28 blackburn, sure on our website w/ dango 00:30 django 00:30 I think such kind of demos would make it much easier for people to grasp what shogun is about 00:31 and we could have a toy demo for each class of shogun's algorithms 00:31 blackburn, for other serious demos we really need concrete examples. I think the gender detector based on opencv would also be cool 00:32 or some of wiking's live image segmentation 00:32 sonney2k: there is no machine learning in gender detection nor segmentation nor anything machine vision 00:33 I mean it is just learn a model and apply 00:33 sonney2k: in opencv they even include "w" vector for pedestrian detection 00:34 no machine learning anymore at all 00:34 sonney2k: are your kids better? 00:35 blackburn, huh? What do you mean? The w vector needs to be learned so why is it not ML? 00:36 sleeping now 00:36 sonney2k: one don't need a huge library to learn just w 00:36 and in the end nothing matters in ml sense, it is rather what you do with features 00:37 yeah I am in marvin state again 00:37 so marvin tell me which field do you want to kill next? 00:41 and actually sure you do - for the gender detection you need coffin or you won't be able to learn 00:42 sonney2k: liblinear doesn't use coffin 00:45 blackburn, it does 00:46 in shogun? 00:46 sure 00:46 but in liblinear? 00:46 btw we totally didn't mention in NEWS that the gsoc's student's projects are in 00:46 in liblinear nope - that is the problem 00:47 yeah we totally didn't mention 50% of things we should better 00:47 blackburn, what else? 00:47 sonney2k: but it is not a problem for other people - that is the problem I am depressed with 00:47 sonney2k: SLEP fuckup at least 00:47 to remember :) 00:47 we forgot heikos stuff too 00:47 blackburn, sure it is - only very few people have 512G mem and hundreds of cores 00:48 sonney2k: btw about coffin 00:48 I think it could be implemented using generic programming 00:48 and with shogun and that amount of mem they could have used orders of magnitude more examples and get much higher accuracy 00:48 dense and add are functors 00:48 didn't try but conceptually it could simplify code 00:49 no idea about that 00:49 blackburn, I think you should focus on submitting your edrt to jmlr 00:49 ha 00:50 yeah 00:50 I am working on it 00:50 just now 00:50 you have too many things going on 00:50 sonney2k: btw n4nd0 will be in (I hope so) 00:50 for what? 00:50 sonney2k: edrt 00:50 what did he do? 00:50 sonney2k: nothing yet - he will help me with porting to eigen3 and new algorithms 00:51 hmmhh, don't make it too big otherwise you will never finish 00:51 I recall that reviewers were happy already last time 00:52 blackburn, look at this http://mbostock.github.com/d3/ex/stack.html 00:52 and then click group 00:53 whoop! 00:53 isn't that cool 00:53 ? 00:53 that is d3js :) 00:53 yeah it is 00:53 blackburn, or http://mbostock.github.com/d3/ex/force.html 00:53 drag around some node 00:53 omfg 00:53 so that is what I want for shogun demo 00:54 blackburn, that is the mode you are in right https://www.youtube.com/watch?v=ibEdgQJEdTA&feature=related 01:17 om nom nom 01:18 sonney2k: exactly 01:18 blackburn, hmmhh well your edrt spin off - well do whatever you want 01:48 sonney2k: ? 01:49 sonney2k: I want to develop it standalone and then bind shogun to newest version 01:49 sonney2k: what did you mean? 01:52 I know. 01:54 exactly that 01:54 I still dont' get why but hey do whatever you want 01:54 sonney2k: do you see any other way for that paper? 01:54 I can't demarkate where shogun ends 01:54 sure. they didn't care about that 01:55 all you have to do is explain just edrt's features and not so much shogun's 01:55 I think cheng meant that 01:55 no cheng complained about that you described shogun's features 01:56 sonney2k: it is just like libqp 01:57 :) 01:57 blackburn, I don't understand your gsoc summary 02:13 " During the summer Sergey has ported a few algorithms from the SLEP and 02:14 MALSAR libraries with further extensions and improvements. 02:14 " 02:14 ehmm what is wrong 02:14 ? 02:14 didn't you implement some multitask stuff your own? 02:14 sonney2k: from matlab code mostly 02:15 blackburn, can you give me the links to slep/malsar please? 02:15 blackburn, from chris or whom? 02:15 sonney2k: SLEP 02:15 and MALSAR 02:15 http://www.public.asu.edu/~jye02/Software/SLEP/ 02:15 http://www.public.asu.edu/~jye02/Software/MALSAR/ 02:15 sonney2k: they have a lot of matlab code - I adopted their C code and developed C++ code using the matlab code 02:15 but with coffin and all the stuff 02:16 latest are using eigen3 as well with big impact on performance 02:16 currently your summary doesn't sound impressive 02:16 heh 02:16 though what you say here is cool so please tweak. 02:17 name the algorithms you ported explicitly and what you did to them 02:17 well I didn't do anything special this summer 02:17 I think that will make it sound better and gets closer to the truth 02:17 pretty bad work actually 02:18 as usual 02:18 marvin!!! 02:18 blackburn, name them please 02:18 sonney2k: there will be too many algorithms named 02:18 that is good 02:19 and needed 02:19 and please mention matlab -> c++ / coffin eigen3 too 02:19 Sergey Lisitsyn was our GSoC student for the second time and he focused on multitask learning. Multitask learning is a modern approach to machine learning that learns a problem together with other related problems at the same time using a shared representation. This approach often leads to a better model for the main task, because it allows the learner to use the commonality among the tasks. During the summer Sergey has developed efficient implementatio 02:24 sonney2k: ^ ? 02:24 I'd actually prefer 'He's spent some time on useless things nobody will ever touch' 02:24 :D 02:24 blackburn, your project description is currently worst of all :/ 02:28 and that is really is because of your stupid marvin mood 02:29 no idea what to do with it 02:29 blackburn, *please give me the list of algoritms you ported* 02:29 *now* 02:29 it is already in text I  sent 02:29 when? 02:30 Namely, they are L12 group tree and L1q group multitask logistic regression and least squares regression, trace-norm multitask logistic regression, clustured multitask logistic regression, basic group and group tree lasso logistic regression. 02:30 sonney2k: ^? 02:30 when did you sent that where? 02:30 ehmm 02:30 irc filtered it out probably 02:30 Sergey Lisitsyn was our GSoC student for the second time and he focused on multitask learning. Multitask learning is a modern approach to machine learning that learns a problem together with other related problems at the same time using a shared representation. This approach often leads to a better model for the main task, because it allows the learner to use the commonality among the tasks. 02:31 During the summer Sergey has developed efficient implementations of various multitaks algorithms using the SLEP and MALSAR libraries. Namely, they are L12 group tree and L1q group multitask logistic regression and least squares regression, trace-norm multitask logistic regression, clustured multitask logistic regression, basic group and group tree lasso logistic regression. All the implemented algorithms use COFFIN framework for flexible and efficient l 02:31 sonney2k: works this time? 02:31 yes now I received it 02:31 blackburn, but cut off 02:32 what is following efficient I 02:32 ? 02:32 2nd message? 02:32 yes the "All the implemented..." is not complete 02:32 sentence I mean 02:33 All the implemented algorithms use COFFIN framework for flexible and efficient learning and some of the algorithms were implemented using efficient Eigen3 library. 02:33 that's it then right? 02:33 stupid irc 02:33 yeah 02:33 now it sounds much better 02:33 blackburn, what is clustured multitask ? 02:34 sonney2k: tasks are clustered 02:34 clustered? 02:34 automagically 02:34 or structured 02:34 clustred 02:34 k 02:34 sonney2k: you just say you have 81 tasks and 9 clusters 02:35 and it finds you the clusters 02:35 and solution as usual 02:35 relaxed k-means 02:35 stupid idiot thing no idea why ever use it 02:35 yes marvin 02:37 desperation 02:39 * sonney2k hands blackburn a cookie 02:41 cookie++ 02:41 cookies are useless 02:42 my gf got bored and wrote a QNX problem, why on the earth one should write programs for QNX 02:42 program* 02:43 blackburn, what other big changes did we have in shogun 02:44 I recall SGVector etc + refcounting 02:44 and utilizing eigen3 02:44 sonney2k: hmm 02:45 let me recall 02:45 blackburn, I am reading what we have achieved this summer and I cannot *not at all* see the slightest reason to be depressed 02:48 to the contrary 02:48 looks like a heroic effort 02:48 sonney2k: yeah in number of lines probably 02:49 sonney2k: a few useless refactoring operations were carried out as well 02:51 we now have even more folders 02:51 haha :D 02:52 okay time to sleep 02:53 sonney2k: have a nice weekend and good night 02:54 blackburn, sweet as in sugar dreams! 02:54 -!- blackburn [~blackburn@188.168.14.62] has quit [Quit: Leaving.] 02:56 -!- ptizoom [~christian@79-71-87-88.dynamic.dsl.as9105.com] has quit [Quit: Ex-Chat] 04:32 -!- ptizoom_ [~christian@79-71-87-88.dynamic.dsl.as9105.com] has joined #shogun 04:32 -!- ptizoom_ [~christian@79-71-87-88.dynamic.dsl.as9105.com] has quit [Quit: Ex-Chat] 08:22 -!- ptizoom__ [~christian@79-71-87-88.dynamic.dsl.as9105.com] has joined #shogun 08:22 -!- Marty28 [~marty@5.28.65.203] has joined #shogun 08:25 -!- Marty28 [~marty@5.28.65.203] has quit [Quit: Colloquy for iPad - http://colloquy.mobi] 08:35 -!- gsomix [~gsomix@109.169.186.78] has joined #shogun 08:36 good morning 08:37 -!- ptizoom__ is now known as ptizoom 09:17 hey 10:44 gsomix: here? 10:50 actually you might be the person i need now :D 10:50 sonney2k: you awake? :) 11:07 build #147 of nightly_default is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/147 11:25 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun 13:23 -!- blackburn [~blackburn@188.168.4.221] has joined #shogun 13:41 blackburn: ping 14:06 pong 14:07 ok so i've did like almost everything 14:07 but still cannot get python to be debugged 14:08 i get a segfault if i start using Director class 14:08 that's all i know :) 14:08 did you try that gdb attach or anything like thaT? 14:08 wiking: did you try valgrind? 14:09 i cannot attach 14:09 since it's very fast 14:09 it works for such debugging as well 14:09 valgrind mmm 14:09 ok 14:09 let's see 14:09 valgrind catches segfaults as well 14:09 trying :) 14:10 didn't work :) 14:10 ?? 14:11 only get segfault 14:11 and i can dig out this from apple's crash reporter 14:11 Thread 0 Crashed:: Dispatch queue: com.apple.main-thread 14:11 0   org.python.python             0x000000010a6dd6b5 _PyType_Lookup + 149 14:11 1   org.python.python             0x000000010a6c9f6a _PyObject_GenericGetAttrWithDict + 170 14:11 2   org.python.python             0x000000010a68db18 PyObject_CallMethodObjArgs + 168 14:11 3   _modshogun.so                 0x000000010b12d10d SwigDirector_DirectorData::get_name() const + 173 (modshogun_wrap.cxx:10315) 14:12 4   libshogun.13.dylib            0x000000010c9f56fc shogun::CSGObje 14:12 oh getname 14:12 interesting 14:12 did you declare get_name too? 14:12 i'll kill nerdtree NOW 14:14 virtual const char* get_name() const { return "DirectorData"; } 14:14 I think I had that before 14:18 wiking: can you show your python code? 14:20 yes 14:21 https://gist.github.com/3964468 14:22 blackburn: if i use Data as base class for DNA 14:22 everything is ok 14:23 wiking: ok the problem is that somewhere in C++ code get_name is called 14:25 and swig got crazy with it somehow 14:25 hehehe 14:25 can't remembed what did I do with it last time 14:27 wiking: ok I am near 14:29 -!- wiking_ [~wiking@info2k1.hu] has joined #shogun 14:31 -!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: Connection reset by peer] 14:31 wiking_: 14:33 first, we have to be sure only one thread is used 14:34 ok how? 14:34 anyshogunobject.parallel.set_num_threads(1) 14:35 tried 14:39 the same 14:39 Thread 0 Crashed:: Dispatch queue: com.apple.main-thread 14:40 0   org.python.python             0x00000001019a96b5 _PyType_Lookup + 149 14:40 1   org.python.python             0x0000000101995f6a _PyObject_GenericGetAttrWithDict + 170 14:40 2   org.python.python             0x0000000101959b18 PyObject_CallMethodObjArgs + 168 14:40 3   _modshogun.so                 0x00000001023fc10d SwigDirector_DirectorData::get_name() const + 173 (modshogun_wrap.cxx:10315) 14:40 4   libshogun.13.dylib            0x0000000103cc46fc shogun::CSGObject::unref() + 156 (SGObject.cpp:174) 14:40 5   _modshogun.so                 0x0000000102c804f5 _wrap_delete_DirectorData + 197 (modshogun_wrap.cxx:263538) 14:40 6   _modshogun.so                 0x0000000102608028 _wrap_delete_DirectorData_closure + 56 (modshogun_w 14:40 ok this i'll pastebin 14:40 http://pastebin.com/QNtJGSpu 14:40 -!- wiking [~wiking@info2k1.hu] has joined #shogun 14:42 -!- wiking_ [~wiking@info2k1.hu] has quit [Read error: Connection reset by peer] 14:42 wiking: see the problem? 14:42 where? 14:43 we call get_name of object that was deleted 14:43 sorry i'm trying to put together an osx buildbot 14:43 ok but why it get's deleted i really wonder 14:43 and why it's not happening 14:44 when i don't use the director class? 14:45 wiking: we have to put some flag or call some function or so 14:45 I recall that 14:45 but what to call/set I do not recall 14:45 trying to find 14:45 !!!!!!!!! 14:46 wiking: call data.__disown__ after creation 14:46 ok 14:47 where? :) 14:47 wiking: so you create MotifData 14:47 somewhere in code 14:47 and after instantiation 14:47 call __disown__() method 14:47 ok 14:47 tyring 14:47 ok cool 14:47 now it's not crashing 14:48 :DD 14:48 wtf? 14:48 wiking: some reference counting issue.. 14:48 asdf 14:48 :> 14:48 ok but i have the same shiftfuck now 14:48 ?? 14:48 that i get back a CData from features.get_sample(i) 14:49 but i would need DNA or MotifPos 14:49 wiking: oh you finally found what is obtain_from_generic for 14:50 :D 14:50 and it seems there's no way around this 14:50 yeah but what if i specify a class 14:50 in python 14:50 or in this case this is not possible 14:50 and i should just not even try it :D 14:50 because if this is the case 14:51 the whole concept of having latent svm in python is fucekd :) 14:51 no, why? 14:51 so here's the deal 14:51 i need to define 3 things for a latent svm 14:51 latent model 14:51 you get Data but you need a DNA? 14:51 yes 14:51 i need DNA 14:51 since it has some fields that i need in LatentModel 14:52 but obviously Data doesn't have it 14:52 in similar fashion it would be DNA.obtain_from_generic(m_features.get_sample(idx)) 14:52 don't you like it? 14:52 blackburn: but how do i do the typecasting in python/ :) 14:52 i mean i'll have to write the c++ for DNA class then 14:52 ah 14:53 i mean i'll have to define DNA class and implement in c++ 14:53 right 14:53 that's what you are saying now 14:53 fucking python 14:53 hehe hyeah 14:53 exactly 14:53 :)) 14:53 why not to have casting? 14:53 they denied it allthrough 14:53 wiking: so what to do with it.. 14:55 well 14:57 i guess there's no way this will work then 14:57 wiking: here 14:58 gsomix: hehehe we've just realised that it's not going to work out with blackburn 15:00 ok so i have to reimplement now this in c P) 15:00 :> 15:00 nice 15:00 i was hoping that i can use soon python interface 15:00 wiking: ok. it's comfortably, that there is blackburn. 15:01 build #596 of deb3 - modular_interfaces is complete: Failure [failed test libshogun]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/596  blamelist: Sergey Lisitsyn 15:02 -!- heiko [~heiko@host86-176-2-92.range86-176.btcentralplus.com] has joined #shogun 15:27 wiking: I've no idea what to do with it still 15:43 -!- K0stIa [~kostia@alt2.hk.cvut.cz] has joined #shogun 15:46 -!- K0stIa [~kostia@alt2.hk.cvut.cz] has left #shogun [] 15:46 blackburn: afaik it's not possible and that's it :( 16:35 wiking: we should ask somewhere 16:36 may be guido 16:36 :D 16:36 :D 16:36 why the heck you do it with us, guido 16:36 :> 16:37 wiking: https://github.com/lisitsyn/edrt/blob/master/src/edrt/libedrt.cpp having STL party in code 16:38 :D 16:38 blackburn: hehehe well the new sosvm has stl as well 16:39 std::vector 16:39 i still haven't got around changing it 16:39 :> 16:39 wiking: IIRC sonney2k has nothing against STD in .cpp 16:39 then? 16:39 where? :) 16:39 haha 16:39 :D 16:40 well I think so 16:40 today might be the day that i can give an osx buildbot 16:41 and then we'll see the bug in StreamingFeatures as well 16:41 nice 16:41 heheheheh 16:41 so somebody should fix that :) 16:41 idnotwanto 16:41 we have to solve your problem somehow 16:41 it's funny that it's only osx who's memory manager fails on that bug 16:41 blackburn: yeah would be nice to be able to lsvm in python 16:42 wiking: I started to fix that bug but had to change 100+ lines and gave up 16:42 : 16:42 :D 16:42 so you acted like this: http://webtrax.hu/myfacewhen/faces/lineart-memes/fuck-this-shit-table-flip.jpg 16:42 and shouted fuckthis shit :) 16:43 exactly 16:45 mmm 16:49 i'll join now #swig 16:49 if there's one and ask around 16:49 joined 16:50 title: "Semantic Web Interest Group 24x7 publicly logged hack-'n'-chat." 16:50 i guess i'm on the wrong place :) 16:50 hehehe nice 16:50 in the topic: "ften confused with the Interface Generator, http://www.swig.org/ which is on irc.oftc.net channel #swig" 16:50 nice there are exactly 2 ppl in the #swig channel 16:53 hah 16:54 wiking: consider mailing list then :) 16:57 btw 17:01 is it now intenional that MSG_ERROR will not cause and exit(1) 17:01 ? 17:01 i.e. SG_ERROR("asdf"); 17:01 -!- heiko [~heiko@host86-176-2-92.range86-176.btcentralplus.com] has quit [Quit: Leaving.] 17:02 wiking: no idea 17:14 currently it's not killing the app 17:24 so basically we have a some stupid warnings because of this 17:24 in the static analyzer 17:24 -!- heiko [~heiko@host86-176-2-92.range86-176.btcentralplus.com] has joined #shogun 18:42 -!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Read error: Operation timed out] 19:08 wiking, no it should kill the app if we are not on a python etc interface 19:13 mmm weird thing 19:24 wiking, ? 19:24 as scan-build detects it as if it's not killed 19:24 sonney2k: do you maybe have any solution ideas how one can typecast back to a derived class in python? 19:34 if i defined the derived class in python 19:34 sonney2k: yeah we got catch 22 situation 19:49 -!- gsomix [~gsomix@109.169.186.78] has quit [Quit: Leaving] 20:22 wiking, not possible 20:51 -!- sonney2k [~shogun@7nn.de] has left #shogun ["Ex-Chat"] 20:51 blackburn: here's our answer :) 20:54 wiking: amazing 20:54 -!- heiko [~heiko@host86-176-2-92.range86-176.btcentralplus.com] has quit [Quit: Leaving.] 20:57 -!- heiko [~heiko@host86-176-2-92.range86-176.btcentralplus.com] has joined #shogun 21:04 -!- heiko [~heiko@host86-176-2-92.range86-176.btcentralplus.com] has quit [Quit: Leaving.] 21:31 -!- sonney2k [~shogun@7nn.de] has joined #shogun 21:43 -!- mode/#shogun [+o sonney2k] by ChanServ 21:43 sonney2k: ok to add iterators to features? 21:48 sonney2k: with all you hate to stl do you realize stl-like for or for_each will work for any possible features with no changes at all (including streaming) 21:49 blackburn, I have no idea about iterators in C++ 21:50 sonney2k: I believe it is a good pattern actually 21:51 speed trade-off? 21:52 sonney2k: it is compile-time abstraction 21:52 sonney2k: just like eigen3 stuff it goes away on compilation 21:53 sonney2k: iterator is a pointer but it is generic 21:54 :> 22:02 wiking: agree? :D 22:03 blackburn, did you measure? 22:03 sonney2k: if u find me a good way to convert dmg to iso then we can have an osx buildbot 22:03 sonney2k: measure what exactly? 22:03 speed 22:03 sonney2k: i've googled and tried everything in the afternoon but couldn't get working any of it 22:03 sonney2k: okay let me show you assembly code in 5 minutes :D 22:03 blackburn: -Os the code :) 22:04 wiking, what did you not get working? 22:04 sonney2k: couldn't get a dmg converted to iso 22:04 sonney2k: and i need an iso for VB 22:04 vb = virtualbox 22:05 wiking, doesn't dmg2img work? http://vu1tur.eu.org/tools/ 22:09 sonney2k: tried 22:09 sonney2k: no luck 22:09 sonney2k: i've tried doing dmg2img && mount as hfsplus 22:09 but couldn't mount the img 22:09 had some problems... couldn't find hfsplus superblock 22:10 wiking, well worst case burn the dmg to a dvd and then copy / create an iso from it 22:10 sonney2k: heheh no bw for upping now an iso from my place 22:11 build #597 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/597 22:42 -!- heiko [~heiko@host86-176-2-92.range86-176.btcentralplus.com] has joined #shogun 22:42 heiko, around? 23:00 sonney2k, yes! hi 23:00 heiko, I'd like to have your blog on shogun planet 23:00 not the gsoc blog 23:00 the shogun one? 23:00 sure go ahead! 23:00 your last entry didn't appear on planet 23:01 yes, I started a new one 23:01 shogun related 23:01 blackburn: where's asm? :D 23:01 sonney2k, where I want to write stuff I do for shogun 23:01 heiko, well what is the url for all? 23:01 wiking: it looked similar and I gave up :D 23:01 Ehm, I dont have one I think, let me construct one 23:01 with -O3 it is nearly the same 23:01 Ill send it to you 23:01 well shogun planet doesn't necessarily mean only shogun stuff 23:02 anything authors (or related people) write is totally fine 23:02 can be about weather or whatever 23:02 sonney2k, ok 23:02 but missing your last blog post on planet is stupid 23:02 sonney2k, indeed 23:03 sonney2k, can you just add the other blog url feed? 23:03 sonney2k, since I am pretty lame in webprogramming and dont know how to create a feed of the featured articles from the main site (using joomla) 23:04 sonney2k, this is the url 23:04 http://herrstrathmann.de/shogun-blog 23:04 the feed is on the bottom 23:04 http://herrstrathmann.de/shogun-blog.feed?type=rss 23:04 matrix experts anyone? 23:05 heiko, http://shogun-toolbox.org/page/planet/shogun 23:07 much better 23:07 heiko, how do you get the tex stuff translated into formulas? 23:07 sonney2k, yes its annoying that the graphics dont appear 23:07 its a joomla plugin 23:07 Y' * (W^-1) * Y 23:08 I think they are loaded from an external page 23:08 by the plugin 23:08 can somebody translate that into equation to avoid inverse? 23:08 heiko, too bad that this is not preserved in the feed 23:08 sonney2k, indeed, Ill check for a solution to generate graphics instead 23:09 sonney2k, http://extensions.joomla.org/extensions/living/education-a-culture/tex/12357 23:09 heiko, there is this http://www.mathjax.org/ 23:11 heiko, you basically just write latex code and it is translated via javascript 23:11 yooohoooo 23:11 ocas_time 3082.422861 s 23:11 let's see how it is :> 23:12 sonney2k, ok I will take a look 23:12 heiko, let me just add it to the website 23:14 sonney2k, how to do that? 23:14 I actually have no idea about this stuff ;) 23:14 heiko, well the mathjax doc says just to include a .js script 23:14 that's all 23:15 shouldwork 23:15 so then as soon as you write latex math code it will be translated 23:15 chrome doesnotlike mathml 23:15 heiko, and this would work on your joomla etc too 23:15 blackburn: ping 23:15 wiking: pong 23:15 blackburn: whatyaneed? 23:16 wiking: B = Y' inv(W) Y 23:16 want to have W without inv 23:16 costs too much for me 23:16 blackburn: btw check out this one http://maeth.com/pres.test 23:16 blackburn: press 'f' for fullscreen 23:16 blackburn: and of course you can do all the fancy css3 transformations ;) 23:16 wiking: your work? 23:17 blackburn: noup of course not :) 23:17 blackburn: just put together 2 jquery plugins 23:17 :) 23:17 blackburn: http://shama.github.com/jmpress.js/ 23:17 I see 23:18 blackburn: wekk i guess u r looking for B 23:18 or? 23:18 wiking: yes I need to translate it to WB = .. 23:19 hmm may be it is possible in two steps? 23:19 /*W 23:19 wiking: what does that mean? 23:19 wiking, I've seen someone use that + d3js and mathjax 23:19 very fancy talk that you can afterwards easily put on the web 23:20 sonney2k: yep 23:21 heiko, that is how you can use it in joomla http://docs.mathjax.org/en/latest/platforms/index.html 23:22 blackburn: does Y have any special property? 23:23 wiking: orthogonal 23:23 hmm 23:24 wait wait so 23:24 B = Y' inv(W) Y 23:24 Y B = inv(W) Y 23:24 wiking: looks correct for you? 23:25 how? 23:25 no 23:25 y' = transpose of y or? 23:25 left multiply by Y? 23:25 Y' is transpose 23:25 Y B = Y Y' inv(W) Y 23:25 can we do that? 23:25 I haven't been working with matrices for a while :D lol 23:26 well 23:26 matrix mult is does not commute 23:26 yes it is not 23:26 sonney2k, I now changed the latex to cache the images locally, so it should be fine, how can I tell shogun planet to update itself? 23:26 unless they are diagonal and are the same dim 23:26 you mean it will be 23:26 Y B = Y (Y' inv(W) Y) 23:26 yes? 23:27 BY' = Y' inv(w) Y Y' 23:27 yeah it's almost the same :) 23:27 shouldwork 23:27 okay BY' = Y' inv(W) 23:27 but it is not a linear equation system yet 23:27 heiko, well even better just use mathjax 23:27 wiking: do you remember whether we can transpose both left and right side? 23:28 sonney2k, then I have to install it, I really would like to avoid that if it works otherwise 23:28 heiko, look I just did this http://shogun-toolbox.org/page/planet/shogun 23:28 we can 23:28 sonney2k, how worked that? 23:29 heiko, no only write this one line in your doc 23:29 but dont I have to install the software before? 23:29 then you can immediately have latex code in there 23:29 blackburn: but A*A' != I 23:29 really, just this one liner? 23:29 cool thing 23:29 heiko, no just followhttp://docs.mathjax.org/en/latest/platforms/index.html 23:29 sorry I got that wrong 23:29 http://docs.mathjax.org/en/latest/platforms/index.html 23:29 wiking: what is A? 23:29 blackburn: any matrix 23:29 looks cool right 23:29 wiking: A*A' = I for an orthonormal matrix 23:29 heiko, for that to work I replaced {tex} with $and {/tex} with$ on all pages ... 23:30 ah yeah u said that it's orthonormal 23:30 when n4nd0 is around I will tell him too 23:30 sonney2k, I think n4nd0 is on wordpress where things work automatically 23:30 wiking: hmm when we have BY' = Y' inv(W), we transpose both parts YB' = inv(W)' Y 23:31 heiko, well look at how awful it looks on his blog 23:32 so he should convert to using mathjax too 23:32 well then altogether u have YBY' = inv(w) 23:32 :P 23:32 or? 23:32 :> 23:32 sonney2k, youre right, this mathjax is an awesome tool 23:32 mathjax is a shitty replacement of mathml 23:32 imho 23:32 :) 23:32 wiking: inv(W)' Y can be easily computed 23:32 denote it by K then we have YB' = K 23:32 heiko, now imagine this used with http://bartaz.github.com/impress.js/#/bored 23:32 it's just helping chrome and other shitty browsers who doesn't support mathml 23:32 impress js 23:32 heiko, to give talks 23:33 sonney2k: jmpress.js is better 23:33 sonney2k: imho 23:33 sonney2k: as it's jquery based so u really can do some fancy shit 23:33 i've started with impress.js 23:33 but for some reason bartaz doesn't want jquery in it 23:33 :( 23:33 so that's why there's jmpress.js 23:33 sonney2k, that one is pretty cool! 23:34 sonney2k: impressive, it is a next step after beamer ones 23:34 so evolution is god damn shit powerpoint -> beamer -> js :D 23:34 wiking: thanks I think it should work 23:34 heiko: http://shama.github.com/jmpress.js/docs/#docs-examples 23:35 wiking: I am trying to leverage redsvd for my dirty purposes 23:35 heiko, then combine this with d3js for figures http://d3js.org/ 23:35 heiko, e.g. http://mbostock.github.com/d3/ex/force.html 23:36 heiko, or http://mbostock.github.com/d3/ex/stack.html 23:36 battlestation down :( 23:36 sonney2k: !!! 23:37 sonney2k: irc hook for github 23:37 sonney2k: can u fire it up plz? 23:37 wiking, cia is dead 23:37 sonney2k: but it's not based on cia 23:37 or? 23:37 and I tried to configure the standard irc hook for github but it doesn't work 23:37 wiking, no idea how to get it to work 23:38 sonney2k: r u sure u used shogun-toolbox account and not yours? 23:38 like in case of travis 23:38 wiking, sure 23:38 wiking, try to add notifications to your local fork 23:38 ok 23:38 if it works we get this to work 23:39 sonney2k: 23:39 notice - Enables notice support. Make sure you configure channel to support notices ("/mode #yourchannelname -n" or "/msg chanserv set mlock -n") 23:39 could u plz? 23:39 wiking, long done 23:40 Configure your IRC channel to allow external messages (/mode #yourchannelname -n). 23:40 ? 23:40 wiking, that is the same thing 23:40 only chanserv is doing this kind of stuff on freenode 23:41 -GitHub153:#shogun- [shogun] none pushed 3 new commits to master: https://github.com/vigsterkr/shogun/compare/21d55d24675b...4e01cd072e2b 23:41 -GitHub153:#shogun- shogun/master 046a89c Sergey Lisitsyn: Fixed ifdefs in SGBase to avoid failures with all but perl interfaces 23:41 -GitHub153:#shogun- shogun/master 5f36e99 Sergey Lisitsyn: Removed wrong free of sgvector field of LinearKernel 23:41 -GitHub153:#shogun- shogun/master 4e01cd0 Viktor Gal: Use SG_CALLOC instead of calloc Just to fix some of the static analyzer reports 23:41 ;) 23:41 lalala 23:41 :D 23:41 wtf 23:41 hahahaha 23:41 tolda 23:41 toldya 23:41 :D 23:41 alright 23:41 I told you sooooo 23:41 :D 23:41 hhehehehe 23:41 FUCKTHISSHIIIIT 23:41 :D 23:41 sonney2k: i guess it was your other account or? 23:41 wiking: and then table is flying? 23:41 wiking, show me your config 23:42 wiking, no 23:42 sonney2k: just followed the 23:42 howto 23:42 For freenode, try the following: 23:42 # server: irc.freenode.net 23:42 # port: 6667 23:42 # room: #yourroom 23:42 # message_without_join: checked 23:42 # long_url: checked 23:42 # notice: checked 23:42 # active: checked 23:42 # NOTE: Ensure you enable notice support (see above) 23:42 -shoGitter:#shogun- [shogun] none pushed 3 new commits to master: https://github.com/vigsterkr/shogun/compare/21d55d24675b...4e01cd072e2b 23:43 -shoGitter:#shogun- shogun/master 046a89c Sergey Lisitsyn: Fixed ifdefs in SGBase to avoid failures with all but perl interfaces 23:43 -shoGitter:#shogun- shogun/master 5f36e99 Sergey Lisitsyn: Removed wrong free of sgvector field of LinearKernel 23:43 -shoGitter:#shogun- shogun/master 4e01cd0 Viktor Gal: Use SG_CALLOC instead of calloc Just to fix some of the static analyzer reports 23:43 heheh even the nick part works 23:43 cool 23:43 ;) 23:43 wiking, no it is not my account because email notifications, cia, travis etc stuff is in there 23:43 ah ok 23:43 try it then with that 23:43 as u see it works 23:43 let me turn it off 23:43 wiking: where to set it up? 23:43 blackburn: admin -> service hooks -> IRC 23:44 found thanks 23:44 sonney2k: and then just push the test hook 23:44 it takes a about 10-20 seconds 23:44 lets try 23:44 to get on irc 23:44 wiking: how do you call it to announce again? 23:45 blackburn: test hook 23:45 after you've updated the settings 23:45 you'll have a test hook button 23:45 -!- GitHub96 [~GitHub96@sh2.rs.github.com] has joined #shogun 23:45 -GitHub96:#shogun- [shogun] none pushed 3 new commits to master: https://github.com/shogun-toolbox/shogun/compare/cc110627342e...bbc4a3bbb433 23:45 -GitHub96:#shogun- shogun/master 0d5f851 ptizoom: forced local shogun library and module path in tests/integration/perl_modular. 23:45 -GitHub96:#shogun- shogun/master 6b6d83b Sergey Lisitsyn: Marked methods in director structured model as virtual 23:45 -GitHub96:#shogun- shogun/master bbc4a3b Soeren Sonnenburg: Merge pull request #805 from ptizoom/perl_121026... 23:45 -!- GitHub96 [~GitHub96@sh2.rs.github.com] has left #shogun [] 23:45 -EDRTbot:#shogun- [edrt] none pushed 3 new commits to master: https://github.com/lisitsyn/edrt/compare/e46aa1007b1b...f3daae107f61 23:45 -EDRTbot:#shogun- edrt/master d7797e7 Sergey Lisitsyn: Big bang 23:45 -EDRTbot:#shogun- edrt/master ca8ca4a Sergey Lisitsyn: Towards proof of concept 23:45 -EDRTbot:#shogun- edrt/master f3daae1 Sergey Lisitsyn: One more step towards prototype of KLLE 23:45 wiking: haha 23:45 blackburn: :DDD 23:45 sonney2k, amazing stuff :), but lots of programming work :D 23:46 sonney2k: works? :) 23:46 sonney2k how to the \$ -equivalent in mathjax? 23:46 sonney2k: let's call the bot Bartender :) 23:46 -shogungit:#shogun- [shogun] none pushed 3 new commits to master: https://github.com/shogun-toolbox/shogun/compare/cc110627342e...bbc4a3bbb433 23:47 -shogungit:#shogun- shogun/master 0d5f851 ptizoom: forced local shogun library and module path in tests/integration/perl_modular. 23:47 -shogungit:#shogun- shogun/master 6b6d83b Sergey Lisitsyn: Marked methods in director structured model as virtual 23:47 -shogungit:#shogun- shogun/master bbc4a3b Soeren Sonnenburg: Merge pull request #805 from ptizoom/perl_121026... 23:47 baaartendeeer! 23:48 travis already knows the drill :P 23:48 he always serves another round on blackburn's tap :D 23:48 wiking, alright it works now. the reason why it did not work was that the length of the nick has strict limitations 23:49 sonney2k: hehehe u gave like a 1024 chars long nick?:) 23:49 ah got it 23:49 wiking, nope around 10 23:49 sonney2k: :> 23:49 none the less 23:49 thanx a lot 23:49 sonney2k: was it DieShogunAlarmgeber? 23:51 blackburn: :D 23:51 blackburn, of course sth like that - you know german have these long words like eierlegendewollmilchsau 23:51 sonney2k: the password you gave me @shogun-toolbox.org was like that as well 23:52 sonney2k: this d3 is pretty crazy 23:53 blackburn, like HosenRunterUnterHosen :D 23:54 wiking, yeah I love it. We need to craft all demos utilizing d3js - so we can have them live on the website 23:54 sonney2k: i suppose u can feed it with json data or? 23:55 wiking, csv or cmdline or table or ... 23:55 cmdline == as arguments 23:55 sweet 23:55 seems like one can do anything with it 23:55 we can generate json right? 23:55 we are python 23:56 never tried it but i know that shogun depends on libjson :) 23:56 sonney2k: no i'm not  :( 23:56 sonney2k: fuckshit class inheritance doesn't let me python 23:56 wiking, what are you trying to do? 23:56 well as said earlier 23:56 if i want to be able to do latent svm fully from python 23:57 but? 23:57 i would need to be able to define my custom classes (derived from CData) 23:57 -EDRTbot:#shogun- [edrt] lisitsyn pushed 1 new commit to master: https://github.com/lisitsyn/edrt/commit/92c4ea0db10d1cbf7a323cd007f036ecf272aa52 23:57 -EDRTbot:#shogun- edrt/master 92c4ea0 Sergey Lisitsyn: Added REDSVD-like eigensolver routine (not tested yet) 23:57 and then i set these in my custom LatentModel (defined in python as well) 23:57 this works fine 23:57 but then 23:57 blackburn, could you please stop the edrtbot logging? 23:57 I will kill myself if it works faster than arpack 23:58 :D 23:58 blackburn: :DDD 23:58 sonney2k: why? I can 23:58 sonney2k: so but then when i try to get back the CData in LatentModel 23:58 i only get CData* 23:58 i cannot get back the class that i've defined in python 23:58 and there's no way i can 'cast' it to the right format 23:58 or at least i couldn't find a way 23:59 blackburn, because of too much noise on the channel 23:59 (latentmodel stores CData* in a DynamicArray) 23:59 sonney2k: I hardly commit once per day :D 23:59 blackburn, this IRC thing is even more verbose than cia - not too nice :/ 23:59 --- Log closed Sun Oct 28 00:00:04 2012