Open in new window / Try shogun cloud
--- Log opened Wed Apr 18 00:00:19 2012
blackburnsonney2k: no idea..00:00
blackburnthere should be a reason though00:00
@sonney2khmmhh maybe because of LIbSVMMultiClass00:00
blackburnSGVector<int32_t> &svm_svs() 142+    { 143+      return svm_proto()->m_svs; 144+    }00:03
blackburnsonney2k: may be that ^?00:03
@sonney2km_machines = SGVector<CMachine *>(num_svms);00:03
@sonney2kthat is sth I don't like00:03
@sonney2kall the rest looks pretty nice to me00:03
blackburnsonney2k: why?00:03
blackburnI did it in the same way00:03
blackburnsonney2k: m_machines = create_machines(num_svms)?00:04
@sonney2kblackburn, SGVector is a numeric thing00:05
@sonney2ke.g. SGVector a,b00:05
@sonney2kwill work at some point00:05
blackburnhmm yes00:05
blackburnbut I prefer even stl vector here00:05
@sonney2kany of the Dyn*Arrays is what you should use here00:05
-!- emrecelikten [~Anubis@] has quit [Ping timeout: 276 seconds]00:06
blackburnsonney2k: have you heard of valarray btw?00:06
blackburnthat's the thing we wouldn't be able to ignore in the future00:06
@sonney2knot sure if this comes too late for us00:10
blackburnsonney2k: we can always change underlying sgvector stuff00:11
blackburnsonney2k: I think it should be as fast if we provide a facade like sgvector00:13
@sonney2kso today will be the anouncement of duplicate students00:13
@sonney2kblackburn, I gave some comments to the multiclasssvm stuff00:13
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]00:14
@sonney2kproblem is really that this is a huge very intrusive patch00:14
blackburnsonney2k: yes I see mails00:14
@sonney2kvery hard to look at00:14
@sonney2kbut I think pluskid did a nice job00:14
@sonney2kmaybe some of the multiclass svm proxy stuff can go away but apart from the minor changes (pending that it all still passes test!) it has my OK00:14
blackburnsonney2k: I would like to merge it soon00:15
@sonney2kIIRC he is away now so would be good if you can give him feedback *before* he returns00:15
@sonney2ksuch that he has a fresh mind and does the final touches and we can quickly merge afterwards00:15
blackburnsonney2k: ok sure00:16
@sonney2kblackburn, btw isn't it 4am or so at your place?00:16
@sonney2ksun rising already?00:16
blackburnsonney2k: hmm 2.1600:16
@sonney2kahh not soo late then00:16
blackburnsonney2k: I am just +2 from you I guess?00:17
blackburnnot so far away actually00:17
blackburnin means of russian distances lol00:17
blackburnif I was in omsk it would be +400:18
@sonney2kblackburn, btw I tried to do some meeting planning - but it is difficult with so many timezones00:18
@sonney2ksomeone will have to stay awake late or wake up early00:18
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun00:19
blackburnsonney2k: after 23?00:19
@sonney2kproblem is we have mentos in NY / melboune00:19
@sonney2kall the rest is inbetween00:19
blackburnsonney2k: actually I am not sure chris in NY right now00:19
blackburnis in *00:19
blackburnsonney2k: he visited tuebingen for some deals so I don't know where he is right now00:20
blackburnbut melbourne is a problem yeah00:20
blackburn(just like cheng soon for me) :D00:20
@sonney2kso only viable time is UTC 11 or 12hrs00:20
@sonney2kblackburn, how can we get rid of the 10 open pull requests?00:21
blackburnsonney2k: we don't have to00:21
blackburnjust live with it :D00:21
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit]00:22
@sonney2kblackburn, you don't loose overview?00:22
blackburnsonney2k: yes that can be hard but I believe it should be as it is00:23
@sonney2kblackburn, for exampel your lib/external stuff could easily be merged if you kept Tron at its place00:23
@sonney2kblackburn, I would rather close things we don't merge within a few days / where there is no obvious progress for some time00:24
@sonney2kit doesn't help to have such requests open00:24
blackburnsonney2k: I agree - my issues are slow because of my exam week00:24
-!- cronor_ [] has joined #shogun00:26
-!- cronor [] has quit [Ping timeout: 276 seconds]00:27
-!- cronor_ is now known as cronor00:27
CIA-64shogun: Victor Sadkov master * r9a4214c / (7 files in 5 dirs): Merge branch 'master' of git:// into refactoring -
CIA-64shogun: Victor Sadkov master * rb886236 / src/README.developer : Fixed outdated file. -
CIA-64shogun: Victor Sadkov master * r5f73ad6 / (2 files): Added new sample code -
CIA-64shogun: Soeren Sonnenburg master * r024fc0a / (3 files in 2 dirs):00:27
CIA-64shogun: Merge pull request #463 from Victor-Sadkov/master00:27
CIA-64shogun: New statistics sample and various whitespace changes. -
@sonney2kblackburn, this week or next - when do you have them?00:27
blackburnsonney2k: *pre-exam sorry00:28
blackburnI have one on 20, and on 27, 4 and 800:28
blackburnsonney2k: however this week is much more hard for me than next will be00:29
blackburnI have to get done all the things I haven't done before :D00:29
@sonney2kgreat - I guess the kickoff meeting then will have only 70% of the participants00:30
@sonney2kwell we have chat logs00:30
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun00:31
@sonney2kI think I should consider that sleep is an option00:31
@sonney2kblackburn, nite00:31
blackburnsonney2k: best option00:31
blackburngood night00:31
blackburnbut I have a couple of things to do still heh00:32
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]00:44
-!- harshit_ [~harshit@] has quit [Quit: Leaving]00:56
-!- av3ngr [av3ngr@nat/redhat/x-hcgxbjfacbxlnpuz] has joined #shogun01:47
shogun-buildbotbuild #489 of python_modular is complete: Success [build successful]  Build details are at
-!- cronor [] has quit [Read error: Connection reset by peer]05:24
shogun-buildbotbuild #490 of python_modular is complete: Failure [failed test_1]  Build details are at  blamelist:, sadkov993@gmail.com05:39
-!- cronor [] has joined #shogun05:43
-!- n4nd0 [] has joined #shogun07:51
n4nd0good morning!07:52
-!- n4nd0 [] has quit [Remote host closed the connection]07:55
-!- n4nd0 [] has joined #shogun08:01
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]08:23
sonne|workn4nd0: nice one too!08:35
-!- uricamic [9320543b@gateway/web/freenode/ip.] has joined #shogun08:46
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun08:46
sonne|workmoin blackburn!08:49
blackburnsonne|work: morgen08:52
-!- wiking [] has joined #shogun08:54
-!- wiking [] has quit [Changing host]08:54
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun08:54
-!- Marty28 [~Marty@] has joined #shogun08:56
n4nd0hey Marty2808:59
Marty28Does have any chance of running, now?09:02
n4nd0Marty28: it looks like sonne|work fixed it yesterday09:04
n4nd0as he said, it should be ready09:04
Marty28Do you know if shogun has a simple random data generator included?09:05
Marty28A data converter for different file formats would also be nice.09:05
Marty28I am currently writing a simple generator for DNA/RNA.09:06
n4nd0Marty28: take a look at random methods in CMath and SGVector09:06
Marty28Some code is in the examples as well.09:08
blackburnsonne|work: I believe it is unfair that they announce on 23th of April09:27
blackburnthat should be friday (not enough room for vodking)09:27
sonne|workMarty28: has mldata-utils converting between various formats09:27
Marty28sonne: thx09:29
-!- n4nd0 [] has quit [Quit: Lost terminal]09:32
-!- n4nd0 [] has joined #shogun09:34
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]09:53
-!- n4nd0 [] has quit [Quit: leaving]09:56
-!- n4nd0 [] has joined #shogun09:59
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:02
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection]10:02
-!- wiking [] has joined #shogun10:02
-!- wiking [] has quit [Changing host]10:02
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:02
-!- av3ngr [av3ngr@nat/redhat/x-hcgxbjfacbxlnpuz] has quit [Quit: That's all folks!]10:19
-!- Marina [55b326f1@gateway/web/freenode/ip.] has joined #shogun10:43
-!- Marina [55b326f1@gateway/web/freenode/ip.] has quit [Client Quit]10:44
-!- Marina [55b326f1@gateway/web/freenode/ip.] has joined #shogun10:44
sonne|workMarina: try again :D10:47
-!- harshit_ [~harshit@] has joined #shogun10:48
wikingsonney2k blackburn any of you here? is there anything more required for defining a new feature type than setting the EFeatureType and EFeatureClass ?10:52
blackburnwiking: nothing more10:53
wikingi wonder why do i get this error then base/class_list.cpp:427:93: error: allocating an object of abstract class type 'CLatentFeatures'10:54
wikingafaik i've implemented all the =0 functions...10:54
blackburnwiking: get_name?10:54
wikingyep i have that as well10:54
blackburnwiking: do you inherit it from dotfeatures?10:54
wikingi have duplicate, get_num_vectors, get_size, get_name, get_feature_type and get_feature_class implemented10:55
wikingnow i've inherited from cfeatures, just for simplicity...10:55
wikingand i still get this10:56
wikingbase/class_list.cpp:427:93: error: allocating an object of abstract class type 'CLatentFeatures'10:56
wikingstatic CSGObject* __new_CLatentFeatures(EPrimitiveType g) { return g == PT_NOT_GENERIC? new CLatentFeatures(): NULL; }10:56
wikingi've tried a make clean && make10:56
wikingbut did not help10:56
wikingi really wonder now what other pure virtual functions are there that i haven't implemented yet in cfeatures10:57
wikingi guess i'll check it now in the source10:57
blackburnthe only virtual of SGObject is get_name10:58
blackburnso you have to implement only get_name + virtual funcs of cfeatures10:58
wikingafaik i did it10:58
wikinghave them :)11:02
wikingmy definition11:02
harshit_hello everyone :)11:07
-!- n4nd0 [] has quit [Quit: leaving]11:12
wikingblackburn: any ideas?11:17
wikingi really couldn't find more pure virtual functions that i'm missing and no other things i would miss... but for sure it's something with my implementation11:18
blackburnwiking: what is .cpp?11:24
wikingit's most def thinking that it's still an abstract class11:25
wikingfeatures/LatentFeatures.cpp:26:13: error: allocating an object of abstract class type 'CLatentFeatures'11:25
wiking        return new CLatentFeatures (*this);11:25
blackburnwiking: but there is CLatentFeatures(CLatentFeatures) constructor?11:26
wikingi haven't defined a copy constructor11:26
wikingi think i see the problem11:27
wiking../shogun/features/Features.h:100:25: note: unimplemented pure virtual method 'get_feature_class' in 'CLatentFeatures'11:27
wiking                virtual EFeatureClass get_feature_class()=0;11:27
wikingusing inline is not good11:28
blackburnwiking: ah11:28
blackburndon't use inline on virtuals11:28
blackburnmakes no sense11:28
blackburnas calling the virtual function leads to look up in vtable it can't be inlined afaik11:30
wikingblackburn: have you tried this: git grep "inline virtual"11:30
blackburnwiking: heh you are right11:30
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has joined #shogun11:31
harshit_blackburn, : having problems with tsvm, please help !11:31
wikingok this is getting funnier and funnier11:31
wikingi still get the same error11:31
blackburnwiking: still?11:31
wikingeven though i've moved the thing into .cpp11:31
wikingfor get_feature_class11:31
blackburnharshit_: what is the problem?11:32
wikingi might know what's the problem11:32
wikingahahah yeah11:32
wikingso can i have a question?11:32
wikingwhy virtual EFeatureClass get_feature_class()=0; and not virtual EFeatureClass get_feature_class() const=0;11:32
harshit_blackburn: I am submitting correct parameters to ssl.cpp but getting wrong weights as output11:32
-!- xiangwang [~chatzilla@] has joined #shogun11:32
wikingin CFeatures...11:33
xiangwanghi wiking11:33
blackburnwiking: const? yes this should be fixed11:33
wikingxiangwang: heey11:33
wikingblackburn: it's not const11:33
blackburnwiking: I meant that11:33
wikingi naturally assumed it should be a const as it does not change anything in the class..11:33
blackburnwiking: yes it should11:33
wikingand that was the problem... since really it was a different func11:33
wikingok i'll write up a sed script for fixing this11:34
blackburnhmm it seems I have a task for gsomix11:34
blackburngsomix: !11:34
-!- gsomix [~gsomix@] has quit [Quit: ????? ? ?? ??? (xchat 2.4.5 ??? ??????)]11:35
-!- gsomix [~gsomix@] has joined #shogun11:35
gsomixblackburn, hey11:35
xiangwang in SVMLight.h not all the functions of SVMLight realized ?  like svm_learn_optimization?11:35
harshit_blackburn: please have a look at the comment I posted on my pull request !11:36
blackburngsomix: you may fix the const issue11:36
blackburni.e. all the methods that should be const should be const11:36
gsomixblackburn, ok/11:37
blackburnharshit_: checking11:37
blackburnharshit_: ok I can't help with some code right now but you may compare our ssl.cpp and the one you assume as ground truth11:39
harshit_blackburn: ok no worries :)11:39
harshit_blackburn: and yeah I did compare that only .. but the problem was that there were a few functions which where not ported to shogun11:40
harshit_so I did that and when now I am comparing the 2 results it is not the same !!!11:40
blackburnhmm that's strange11:41
blackburnare you sure all the parameters are the same then?11:41
n4nd0harshit_: hey! did you compare the results of those funcitons you ported too?11:41
wikingxiangwang: u wanted to msg me?11:41
blackburnwiking: ?11:42
harshit_n4nd0: hey ! yeah ,Actually they are only 2 functions for creating subset of labeled features from mix of labeled and unlabeled11:42
xiangwangsorry, wrong button:)11:42
harshit_that I ported11:42
harshit_n4nd0, and I think they are working fine11:42
wikingxiangwang: ah ok :)11:42
wikingblackburn: ehhehe of course now it compiles... ;)11:43
n4nd0harshit_: uhm, any clue where the error can be then?11:43
xiangwangi'm trying to realize Latent SVM, but the code is not fully written.11:43
blackburnwiking: I kind of delegated this task to gsomix so we will ensure that const is in place11:44
wikingblackburn: actually virtual int32_t get_size (); should be the same11:44
wikingblackburn: ahhaha :)11:44
harshit_n4nd0, somewhere in between ssl.cpp( file where svmlin code for shogun is )11:44
blackburnI have to go now guys :) n4nd0 thanks I hope you would help harshit_ ;)11:44
wikingxiangwang: which code?11:45
xiangwangshould this part of code push to the shogun svm for gsoc qualification patch?11:45
n4nd0blackburn: I will try :)11:45
n4nd0harshit_: between ssl.cpp and?11:45
xiangwangIntergration of Latent SVM11:45
xiangwangto shogun11:45
harshit_n4nd0, I meant inside ssl.cpp11:45
wikingxiangwang: i think you should be able to do a pull request...11:46
wikingwith some patch11:46
harshit_hey maybe i should test svm of svmlin11:46
xiangwangbut the code is not fully written...11:46
n4nd0harshit_: yeah, sure11:46
harshit_that would give me idea whether svmlin is correctly ported or not11:46
* blackburn is looking how one candidate student is asking his competitor how to apply to the task11:46
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]11:47
xiangwangok I'll push for several times with the code being written.11:47
-!- xiangwang [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120312181643]]11:48
harshit_n4nd0, btw is your covertree done ??11:49
wikingblackburn :DDD11:49
n4nd0harshit_: it is working and it is faster than what we used to have if the number of queries is more or less large11:49
n4nd0harshit_: but it is not yet as fast as the standalone version11:50
n4nd0harshit_: they use a computation of distance with quite a bit of speedups, I have still to work on those11:50
harshit_n4nd0, oh great , whats standalone version ?11:50
n4nd0harshit_: the original version from John Langford11:50
harshit_you mean actual implementation11:50
harshit_oh okay11:51
harshit_great dude :)11:51
harshit_Inform me when you'll get this merged, I can test it on my project (I did one project in which I used knn)11:52
n4nd0that could help :)11:52
n4nd0what was the number of training vectors?11:52
n4nd0how many queries?11:52
n4nd0and what dimension?11:53
n4nd0lot of questions :))11:53
harshit_really huge :) I used subset of 20 Newsgroup dataset11:53
harshit_i dont remember exactly but it was about 5000 training vectors and 4000 features/words11:53
harshit_That was a stupid project actually, I did that when I was first learning machine learning.11:54
harshit_n4nd0, And It used to take 5-6 hrs to get results on matlab ( I used my own implementation of knn)11:55
n4nd0harshit_: ok11:56
n4nd0harshit_: so feature vectors of dimension equal to 4000? that's a big one11:56
harshit_n4nd0, nlp tasks normally have bow representation of documents which make feature dimensions really large11:57
harshit_n4nd0, I dont know if shogun has any support for sparse features ?11:58
harshit_do you knw if it has11:58
harshit_but that doesn't work in octave. right ?12:03
harshit_oh thats really good12:05
n4nd0I remember I heard something about that the swig typemaps were not ready for that12:09
n4nd0gsomix: I think you know something else about that ^, don't you?12:09
gsomixn4nd0, mmm, it work in octave now.12:12
n4nd0harshit_: then it looks like it is ready12:12
n4nd0gsomix: it was something you did recently?12:13
gsomixn4nd0, yep.12:14
gsomixharshit_, examples/undocumented/octave_modular/features_sparse_modular.m12:15
-!- abhishek [dce1fbc1@gateway/web/freenode/ip.] has joined #shogun12:20
-!- abhishek is now known as Guest5879112:21
-!- Guest58791 [dce1fbc1@gateway/web/freenode/ip.] has left #shogun []12:22
-!- abhi1234 [dce1fbc1@gateway/web/freenode/ip.] has joined #shogun12:22
-!- cronor [] has quit [Quit: cronor]12:28
harshit_gsomix, n4nd0 : great,thats really helpful :)12:28
n4nd0no problem12:28
n4nd0I have to go now12:28
n4nd0see you guys!12:28
harshit_me too :) bye12:28
harshit_adios !12:29
-!- harshit_ [~harshit@] has quit [Quit: Leaving]12:29
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has quit [Quit: Page closed]12:29
wikingabhi1234: yes?12:30
wikingabhi1234: here12:30
abhi1234i have been using12:30
abhi1234hi wiking12:30
abhi1234i am sorry i was sending you personal messages12:31
abhi1234this cygwin thing is giving me a hard time12:31
wikingoh i'm sorry12:31
abhi1234building modshogun.dll12:31
wikingbut i think i'm not the best person to ask about compiling shogun on windows :(12:31
abhi1234can you tell me one thing12:31
abhi1234cant we get a dll directly from the website12:32
wikingabhi1234: it's better to ask here about it since i think somebody here might have better experience with shogun and windows12:32
abhi1234can some one please tell me how to use shogun within visual studio environment12:33
abhi1234i am really having a hard time using libsvm dot net version12:34
wikingabhi1234: and u could also try on the mailing list as some people from the shogun project are only checking the mailing list and are not online here12:38
abhi1234oh thanks12:38
abhi1234i will do that12:38
-!- xiangwang [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has joined #shogun12:41
wikingxiangwang: please here13:04
* wiking thinks that he is becoming a git support line on #shogun13:06
wikingsonney2k: can u add this to the channel's topic:
wikingxiangwang: u should have your own forked git repo of shogun13:09
wikingdo the development there13:09
wikingand commit/push all your changes there13:09
wikingand once u have that13:09
wikingu can do a pull request13:09
wikingof course this is only true if u maintain your git repo on github13:10
CIA-64shogun: Michal Uricar master * r0f16159 / (2 files): Gauss-Seidel iterative method for Kernel Ridge Regression learning -
CIA-64shogun: Michal Uricar master * r7a4b0cc / (2 files): GS KRR work on comments -
CIA-64shogun: Michal Uricar master * r4b4bde8 / (2 files): Indentation fixed. -
CIA-64shogun: Soeren Sonnenburg master * r6653329 / (2 files):13:10
CIA-64shogun: Merge pull request #460 from uricamic/GSKRR13:10
CIA-64shogun: Gauss-Seidel iterative method for Kernel Ridge Regression learning -
wikinguricamic: here?13:11
uricamicwiking: yep13:15
wikinguricamic: sorry for yesterday... :(13:18
uricamicwiking: its ok13:18
wikingdo u have some time now?13:19
uricamicwiking: yes, for now, but soon I will have to talk with my adviser13:20
xiangwangthanks wiking, but i can't find my pull now in
uricamicwiking: but he is not here right now, so I have some time :)13:21
wikinguricamic: ok so the thing is that instead of simply just adding 2 headers of libqp13:22
wikingwe should do some c++ wrapping of libqp13:22
uricamicwiking: ok13:23
wikingjust like liblinear/libsvm or svmlight13:23
wikingbtw the libqp_splx_solver function in shogun case13:24
wikingit's copied from libqp?13:24
wikingah yeah it is13:25
-!- PhilTillet [~Philippe@] has joined #shogun13:25
uricamicwiking: I think that lipqp.h is the header file of libqp_splx.cpp, where is this solver implemented13:26
wikingi was just wondering13:26
wikingwhether these are like copied from the libqp13:26
wikingand actually it's just copied from libqp into shogun's repo13:27
wikingand that's it basically13:27
wikingand actually at this moment only libqp_splx_solver is being copied13:27
wikingso i guess we should follow the style of shogun, i.e. copy the source from libqp and not 'depend' on it ;)13:28
wikingi have objections about this but then again i guess sonney2k and blackburn has their reasons to do this this way...13:28
wikingbut still we should do a c++ wrapper for both of the functions13:29
uricamicby libqp you mean the original library ?13:29
wikingi don't know though how often vojtech is planning to do upgrades on libqp ;)13:30
wikingfrom an email of his he seems to be quite busy lately13:30
wikingso i guess not too soon13:30
uricamicwiking: yes he is, I can ask him13:31
uricamicbut he is not here right now :)13:31
wikingheheh his your advisor?13:31
uricamicyep :)13:32
wikingah cool! you are lucky! ;)13:32
uricamicbut I don't think, that he plans to do some chenges13:33
wikingso anyhow i'm just checking now where could we fit nicely the whole thing13:35
wikingso whether we could fit it into one abstract class13:37
wikingor simply just create an own standalone class for it13:38
wikingsonney2k: here?13:41
wikingafais we should just do a standalone cpp wrapper for it but i need to see what sonney2k thinks about it13:43
wikinguricamic: r u going to continue bmrm?13:44
uricamicwiking: I hope so13:45
uricamicnow it is basically in the state, that it also requires a wrapper13:46
-!- gsomix [~gsomix@] has quit [Read error: Operation timed out]13:46
sonne|workwiking: whats up?13:46
wikingsonne|work: it's just a quick question about libqp integration13:49
sonne|workwiking: so are uricamic and you intending to do sth about it?13:50
wikingsonne|work: yep hoping so13:50
sonne|workwould be great13:50
sonne|workI assume you both need it for your projects anyways...13:50
uricamicsonne|work: yep, I will definitely need it13:51
wikingi was just wondering if u have any opinion about whether the libqp wrapper should be a standalone class (as afaik there's no abstract class where we could plug this into) or shall we start thinking about making an abstract qp solver class ...? i.e. maybe the latter one would be wiser as i can see there's the QPBSVMLib13:51
wikingthat i *guess* it's another qp solver13:52
-!- Marina [55b326f1@gateway/web/freenode/ip.] has quit [Quit: Page closed]13:53
wikingso maybe having like a QP solver abstract class and doing the various implementations inherited from that would be good to have...13:54
wikinguricamic: me too ;)13:54
sonne|workwiking: I think it should be similar to what libocas is now - basically a C library we have in shogun and use with all the necessary callbacks to enable sufficient flexibility13:54
sonne|workthen a wrapper on top13:54
abhi1234hi sonne13:55
abhi1234i have a quick question13:55
abhi1234if it is a good time to ask13:55
sonne|workthis way vojtech can continue to maintain this library outside and we only have to adjust the wrapper13:55
abhi1234thanks in advance13:55
sonne|workabhi1234: ask13:55
sonne|work(don't ask to ask)13:55
wikingsonne|work: mmmm13:55
abhi1234I am planning to run shogun in visual studio13:56
abhi1234is it possible13:56
abhi1234i have window 7 and cygwin13:56
abhi1234and it is really difficult to compile the modshogun.dll13:56
abhi1234in cygwin13:56
abhi1234so if it is possible can I get a modshogun.dll for a quick run13:57
sonne|workwiking: what does that mean?13:57
wikingsonne|work: just thinking loudly ;)13:57
sonne|workabhi1234: I am afraid that this is not so easy13:58
sonne|workno one has succeeded in a pure windows port of libshogun13:58
-!- xiangwang [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has quit [Remote host closed the connection]13:58
wikingsonne|work: so let's keep the libqp as it is now (i.e. an easy way to add the future changes), but still aim for not only a nice c++ wrapper on top of the qp solver but start defining an abstract qp solver as well...?13:59
abhi1234ohh terrible then13:59
abhi1234i am really impressed by the kind of algorithm available in shogun14:00
abhi1234you people have done an amazing work14:00
wikingabhi1234: maybe u could be the one who does the work ;)14:00
abhi1234yeah i would like to be14:00
wikingabhi1234: what's your current error?: )14:00
sonne|workwiking: that is what I meant - some nice C++ interface to enable using it within shogun and all its interfaces14:00
sonne|workwiking: but using libqp internally14:01
abhi1234my current error is related to option -pthread14:01
wikingsonne|work: ok... and try to do something about QPBSVMLib as well... ;)14:01
wikingsonne|work: ooh i think we are on the same page then... i saw blackburn's possible patch to move out to an 'external' directory the external libraries... i totally second that idea14:02
abhi1234but cant i get the modshogun.dll from someone  who built it in linux environment only14:02
wikingabhi1234: well it's not a dll14:02
wikingi can give u a .so14:02
wikingbut i doubt that it'll work for u14:02
sonne|workwiking: well if we manage that kind of integration, the files cplex / pr_loqo etc should all be in there too14:04
sonne|workand e.g. SVMlight could work with any of the solvers then14:04 i am afraid wont work for me14:04
wikingsonne|work: cool14:04
wikingi see my competitor did a pull request i should get onto giving a patch today/tomorrow ;P14:04
wikingabhi1234: ok so you die with -pthreads14:05
wikingu use cygwin i assume14:05
wikingabhi1234: have u actually 'made' cygwin to install the pthread library?14:06
wikingas it is part of cygwin but you need to tell the installer that u want it14:06
wikingif not then try to install it now14:06
abhi1234let me try it14:06
wikingcan u do this14:06
wikingls -la /usr/lib/libpth*14:06
abhi1234one moment14:07
wikingand what's the result for it14:07
wikinguricamic: u still here or your advisor caught you?14:08
abhi1234-rw-r--r-- 114:08
wikingabhi1234: the whole lien ;)14:08
abhi1234is the result i get14:08
uricamicwiking: still here :)14:08
abhi1234one moment14:08
wikinguricamic: so i think we should then start defining an abstract qp solver class14:08
uricamicwiking: yep14:09
abhi1234-rw-r--r-- 1 TTG None 30678 Apr  5 20:35 /usr/lib/libpthread.a14:09
wikingabhi1234: so u have then actually pthread14:10
wikingabhi1234: why don't you pastebin your full error and send it to us14:10
wikingi mean compilation error14:10
abhi1234sure I will do that14:11
abhi1234one moment14:11
wikinguricamic: ok i'll sketch up something soon and put it into one branch of mine so that everybody can comment on it via github, but if you have already some ideas please don't be shy ;)14:11
uricamicwiking: ok, I will look at it14:12
abhi1234earlier when i tried to compile it was throwing error "unrecognized option -pthread"14:15
abhi1234now it is showing the same thing as warning14:16
-!- pluskid [~pluskid@] has joined #shogun14:30
-!- gsomix [~gsomix@] has joined #shogun14:33
-!- abhi1234 [dce1fbc1@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]14:37
PhilTillethello everybody :)14:41
Marty28Does anyone know the ERROR "ImportError: cannot open shared object file: No such file or directory"? I just installed a shogun version locally and this comes upon importing shogun.14:41
Marty28I mean import modshogun.14:41
Marty28(trying to use another destdir)14:44
-!- cronor [] has joined #shogun14:56
pluskidwhich distribution are you using? and how did you install Shogun?15:00
Marty28might be NTFS15:09
Marty28cannot change rights on this drive15:09
Marty28ok, moving to /HOME/somewhere15:09
Marty28the shogun_modular sonne buildt yesterday15:11
-!- pluskid [~pluskid@] has quit [Ping timeout: 265 seconds]15:22
Marty28nah, does not work15:33
Marty28I did the following:15:57
Marty28./configure --enable-svm-light --python=python2.7 --destdir=/home/ralf/shogun-builds --prefix=/ | tee configure_log.txt15:57
Marty28make | tee makelog.txt15:57
Marty28sudo make install | tee makeinst_log.txt15:57
Marty28THEN: ~/shogun-builds/shogun0417/bin$ shogun15:58
Marty28shogun: error while loading shared libraries: cannot open shared object file: No such file or director15:58
Marty28Why cannot the be found?15:58
Marty28Is it because of <configure_log.txt> "Checking for libshogun ... yes (internal in source tree)" ?15:59
Marty28I already installed shogun 1.10 and now I am installing the current version from github16:00
sonne|workMarty28: you need to set you LD_LIBRARY_PATH16:01
sonne|workto the dir where libshogun*.so* is16:01
Marty28shell variable?16:02
Marty28BTW: contains outdated interfaces, right?16:03
-!- PhilTillet [~Philippe@] has quit [Read error: Operation timed out]16:08
Marty28whew, works16:10
Marty28sonne: git clone git://
Marty28sonne: Do I get the latest shogun version with this including the changes from yesterday?16:17
Marty28Did you change WeightedDegreePositionStringKernel.prepare_POIM2 to SGVector/Matrix?16:22
sonne|workwill do it on the train ride then16:24
Marty28super :)16:24
-!- vikram360 [~vikram360@] has quit [Ping timeout: 260 seconds]16:27
-!- vikram360 [~vikram360@] has joined #shogun16:28
@sonney2kwiking, btw have you tried to create dotfeatures from your homo kernel map?16:44
-!- Marina__ [8d175a52@gateway/web/freenode/ip.] has joined #shogun16:45
Marina__sonney2k: i tried again... but it doesn't work16:46
Marina__i get the same error16:46
Marina__i have install the latest version of shogun16:46
Marina__and i still get the error:16:46
Marina__   File "/home/mvidovic/POIM/", line 37, in compute_poims     kernel.prepare_POIM2(distr,int(poimdegree), int(max_len)) TypeError: in method 'WeightedDegreePositionStringKernel_prepare_POIM2', argument 2 of type 'float64_t *'16:46
CIA-64shogun: Soeren Sonnenburg master * r9067389 / (2 files): convert prepare_POIM2 to SGMatrix -
@sonney2kMarina__, voila16:47
Marina__so i have to go to src/shogun/kernel/WeightedDegreePositionStringKernel.cpp and delete and add the lines??16:49
CIA-64shogun: Evgeniy Andreev master * r46fd9e4 / (67 files in 2 dirs): fixed string list functions -
CIA-64shogun: Soeren Sonnenburg master * ra8231f2 / (67 files in 2 dirs):16:49
CIA-64shogun: Merge pull request #466 from gsomix/python3_interface16:49
CIA-64shogun: Python 3 support for static python interface -
@sonney2kMarina__, or checkout the whole beast16:49
@sonney2kgit pull16:50
@sonney2kMarty28, done16:50
@sonney2kneeds someone doing the testing though16:50
Marina__@sonney2k: I checkout the whole src - isn't that enough?16:52
@sonney2kMarina__, you do git clone once and then only update via git pull16:54
@sonney2kno need to checkout the whole source16:54
@sonney2kagain and again16:54
* sonney2k leaves the train16:59
Marty28Is the git pull syntax "git pull git://"?17:06
Marina__@sonney: it works :D17:10
Marty28sonne: Works here as well17:14
Marty28Now the result has to make sense as well - lets see...17:15
gsomixsonney2k, thanks. :)17:16
-!- blackburn [~qdrgsm@] has joined #shogun17:18
-!- Marty28 [~Marty@] has quit [Quit: ChatZilla [Firefox 11.0/20120310010446]]17:22
blackburnwiking: I just read logs and :D :D17:23
-!- uricamic [9320543b@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:44
-!- cronor_ [~cronor@] has joined #shogun18:20
-!- cronor [] has quit [Read error: Operation timed out]18:22
-!- cronor_ [~cronor@] has quit [Ping timeout: 276 seconds]18:25
-!- cronor [~cronor@] has joined #shogun18:26
-!- cronor [~cronor@] has quit [Remote host closed the connection]18:26
-!- cronor [] has joined #shogun18:27
-!- cronor [] has quit [Client Quit]18:30
@sonney2kMarina__, let us know if the results are OK18:57
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]18:59
-!- n4nd0 [] has joined #shogun19:08
@sonney2kgsomix, keep them coming19:10
@sonney2kgsomix, which reminds me are you progressing with sgvector stuff?19:10
n4nd0sonney2k: I apologize that I didn't change the time thing in KNN, I forgot about it :S19:11
n4nd0sonney2k: I will apply the change now19:11
@sonney2kn4nd0, great to hear19:12
@sonney2kthis thing should finally be merged19:12
n4nd0sonney2k: yeah, my pace has been reduced a bit, exams coming next week! :O19:13
@sonney2kn4nd0, no problem19:13
@sonney2kbtw did anyone apply at some other org?19:13
n4nd0I didn't19:14
n4nd0sonney2k: should I have done it? :P19:16
@sonney2kapr 23 is the day you will know19:17
n4nd0sonney2k: by the way, about that meeting you were talking about, the one where the timezone was an issue19:18
n4nd0that is something just for mentors right?19:18
Marina__@sonney2k: my results are ok - they are like the results half a year ago!19:18
@sonney2kn4nd0, no students + mentors19:18
@sonney2kMarina__, good to hear19:19
n4nd0Marina__: awesome!19:19
@sonney2kso then happy poim'ing19:19
wikingblackburn what? :)19:19
wikingsonney2k: noun not yet... i think first the homo kernel map should be extended with the RBF approximation at least19:20
@sonney2kwiking, but that is just an addon19:32
@sonney2kn4nd0, btw I guess you didn't have time to compare things with the dnrm2?19:32
@sonney2kcblas function19:32
@sonney2kfor computing the euclidean distance?19:32
n4nd0no, I have not done that yet19:33
n4nd0I will start looking into that once the current patch is merged19:33
* sonney2k thinks we should add a norm function to DotFeatures19:33
@sonney2kn4nd0, would be nice - but first things first19:34
n4nd0sonney2k: your idea was that it should be faster with that than doing our own version of JL's distance function19:34
n4nd0with unrolled loops and so on19:34
@sonney2kn4nd0, yeah19:35
@sonney2kand screw the (small) speedup that the upper bound to distance gives us19:36
-!- Marina__ [8d175a52@gateway/web/freenode/ip.] has quit [Quit: Page closed]19:46
-!- Marty28_ [~marty@] has joined #shogun19:52
Marty28_New irc client on ipad :)20:06
Marty28_My testing of poim with artificial data is set to tomorrow.20:08
-!- Marty28_ [~marty@] has quit [Quit: Marty28_]20:09
@sonney2kn4nd0, do you have a ptr to JL's distance code20:20
@sonney2kthe snipped you pasted?20:20
n4nd0sonney2k: yeah, give me a sec20:22
* gsomix have returned from dance floor.20:24
n4nd0sonney2k: there it goes20:25
n4nd0sonney2k: do we need the assumption of length multiple of 8 in order to use cblas?20:25
n4nd0I guess not20:25
gsomixsonney2k, I'm doing SGVector -> SGVector& conversion for the arguments.20:27
gsomixsonney2k, slowly at this days. trying to deal with study in university and with choosing supervisor.20:27
-!- vikram360 [~vikram360@] has quit [Read error: Connection reset by peer]20:31
-!- vikram360 [~vikram360@] has joined #shogun20:32
@sonney2kgsomix, ok just keep me updated...20:35
@sonney2kn4nd0, not necessary20:35
-!- harshit_ [~harshit@] has joined #shogun20:37
gsomixsonney2k, ok.20:43
@sonney2kn4nd0, I think that JL's code can be twice as fast if he uses d1 ...d420:51
@sonney2kthen it would really be just two sse instructions20:51
@sonney2kbut I guess we need a separate benmark for that20:51
n4nd0sonney2k: yeah that makes actually sense, I thought of it, if they unrolled 2, why not 4? but I thought that maybe it didn't pay off or sth20:52
@sonney2kmaybe when he wrote it it didn't make any difference20:52
@sonney2kor he wrote it for double20:52
@sonney2kn4nd0, still I don't understand why you measured that in JL's implementation the thing is *much* fuster20:53
@sonney2k(like 10x or so)20:53
n4nd0do you think it shouldn't?20:53
@sonney2kthis thing here looks like it can be twice as fast due to the d1,d2 idea and another factor of 2 due to using floats20:54
@sonney2kso 4x20:54
@sonney2kbut no more20:54
@sonney2k(if the upper bound doesn't help a lot)20:54
n4nd0some overhead in shogun somewhere?20:55
n4nd0I will profile this in any case, I find it interesting20:55
@sonney2kn4nd0, I would start with some simple example just computing the distance matrix for say 1000 vectors once with the shogun and once with the jl code20:55
@sonney2kI would predict 4 times faster :D20:56
n4nd0I write that down, we'll see if your prediction is correct ;)20:56
@sonney2kn4nd0, then you could use the *slow* shogun distance version in Jl's tree and see if speeds are the same20:59
@sonney2kif not then we have a timesink somewhere :)20:59
-!- PhilTillet [] has joined #shogun21:02
gsomixit's time to reading the philosophy.21:13
gsomixcu, guys21:13
@sonney2kn4nd0, actually I bet on the time sink - the distance function will only be marginally faster21:17
gsomix"Machine Learning Andrew Ng, Associate Professor 23 April 2012" wow, most interesting things are after 23 :)21:34
n4nd0sonney2k: where the time sink would be then?21:35
@sonney2kn4nd0, that is the hard part to find out :)21:37
@sonney2kn4nd0, but if this distance function is soo fast it would speed up the gaussian kernel computation and lots of stuff *a lot*21:37
@sonney2kn4nd0, which reminds me that we can speed things up anyways by using ||x-y||^2 == x^T x - 2xy + y^Ty21:39
@sonney2kx^Tx can be precomputed21:39
n4nd0sonney2k: would that entail a big difference in general?21:40
@sonney2kand so we only need dot(x,y)21:40
@sonney2kthat is what we use for gaussian kernel at least for sparse features21:40
@sonney2kactually for all features21:41
@sonney2kthat certainly is *a lot* faster even than what JL uses21:42
@sonney2kn4nd0, so yes :D21:43
@sonney2kn4nd0, I hope you pass exams soon and have some time to investigate21:43
n4nd0sonney2k: there was also another speedup suggested in a comment within his code but not implemented21:48
n4nd0line 59421:49
@sonney2kno idea about that (no covertree guru)21:51
n4nd0sonney2k: I have updated the patch with time functions used from shogun22:03
@sonney2kthen let me merge it22:03
n4nd0sonney2k: can you please take it a look and tell me if I have done the git part correctly? I see that this time there are no duplicates commits by me22:03
n4nd0but commits from other people appear in the PR :S22:04
@sonney2kn4nd0, you did it again22:04
@sonney2kno git pull --rebase22:04
n4nd0I did that this time!22:04
@sonney2kwhy am I seeing
@sonney2k uricamic and others added some commits 2 days ago22:04
@sonney2kGauss-Seidel iterative method for Kernel Ridge Regression learning22:04
n4nd0yeah I know but I did it that way this time22:05
@sonney2kn4nd0, can you tell me exactly what you did?22:05
n4nd0that's my history22:05
n4nd0grepped to git commands22:05
n4nd0let's find out what I am doing wrong, I don't really understand :(22:06
n4nd0the experimental stuff is just something I did to try out if I got in that case the non-fast-forward allowed and how I can delete a remote branch22:07
n4nd02058 gave the familiar error about non-fast-forward not allowed22:08
n4nd0so then I git pull --rebase22:08
@sonney2kmight be related to 2060  git pull --rebase22:10
@sonney2k 2061  git pull --rebase origin patch-jlcovertree22:10
@sonney2kI mean *I* would have git checkout master ; then git pull --rebase22:10
@sonney2kthen git checkout patch-jlcovertree22:10
@sonney2kgit rebase master22:10
@sonney2kgit push origin patch-jlcovertree22:11
n4nd0even if the problem is to push in the branch, do I have to pull in my master?22:11
n4nd0as you see in 2040, the push in master worked fine22:11
@sonney2kdid you do work inbetween 2040 and 2060?22:12
@sonney2kanyway I think the problem is the call to git pull --rebase origin patch-jlcovertree22:13
@sonney2kI think this rebases everything new on top of what is in origin patch-jlcovertree22:14
n4nd0between 2040 and 2060 I didn't work22:14
n4nd0the whole history is there22:14
@sonney2kI think you don't ever need to pull from that and if you do not with rebase22:15
n4nd0I thought you told me that last time22:16
n4nd0so then, do you have any clue about how should I do it?22:17
@sonney2kno I said you shoudl do git checkout master; git pull --rebase ; git checkout patch-jlcovertree ; git rebase master ; git push origin patch-jlcovertree22:17
@sonney2kn4nd0, do you still have the patch lying around isolated?22:18
@sonney2kor can you follow what is described in git help reset22:19
@sonney2kand in there 'Undo a merge or pull inside a dirty working tree'22:19
n4nd0I can do as I did for this pull request, create a new branch and apply the patches for each of the commits22:19
-!- ckwidmer [] has joined #shogun22:32
wikingsonney2k: around?22:32
wikingor anybody has been doing some hierarchical classification lately?22:32
@sonney2kwiking, topic hopping again?22:37
n4nd0sonney2k: there it goes the new PR and the other closed22:37
n4nd0sonney2k: I am sorry for the inconveniences due to my git problems :S22:38
CIA-64shogun: Soeren Sonnenburg master * r3d4d21f / (8 files in 3 dirs):22:38
CIA-64shogun: Merge pull request #467 from iglesias/repath-jlcovertree22:38
CIA-64shogun: John Langford's Cover Tree Integration (+8 more commits...) -
n4nd0it looks that I'll learn it the hard way22:38
@sonney2kn4nd0, don't worry22:38
wikingsonney2k: nah ... just i'll need it soonish22:39
@sonney2klast year we ourselves had lots of commits like this22:39
@sonney2kand yes me and  blackburn producing them too22:39
wikingsonney2k: since now i'm doing a classification challenge and today they've released the training set... and when i saw it i've realized that there's no fucking way that the task can be done without some hierarchy22:39
wikingso i'm just 'preparing' myself for the future :)))22:40
wikinghahahah today new spam day... getting 'part time job offers' :D22:41
-!- Marty28 [~chatzilla@] has joined #shogun22:44
@sonney2kwiking, I see - and well it is part of the multiclass task we have this GSoC - some tree based ecoc...22:45
@sonney2knite everyone22:46
gsomixsonney2k, nite22:46
n4nd0sonney2k: good night22:46
wikingn4nd0: yo22:46
n4nd0wiking: hey!22:46
wikingmmm i think it's pluskid who's doing ^that22:47
wiking"well it is part of the multiclass task we have this GSoC - some tree based ecoc"22:48
n4nd0aham, who knows!22:48
wikingi guess soeren knows :>22:48
-!- Marty28 [~chatzilla@] has quit [Quit: ChatZilla [Firefox 11.0/20120312181643]]22:51
gsomixgood night22:52
n4nd0good night gsomix22:52
gsomixphilosophy puts to sleep =____=22:53
n4nd0gsomix: philosophy exam?22:55
gsomixn4nd0, just seminar. exam will be in june22:56
n4nd0for the philosophy exams tend to be like ... no idea what I am doing, sometimes I even laugh about it :D22:58
-!- harshit_ [~harshit@] has quit [Remote host closed the connection]23:19
-!- ckwidmer [] has quit [Remote host closed the connection]23:28
-!- blackburn [~qdrgsm@] has joined #shogun23:29
blackburnhey there23:30
n4nd0blackburn: hey!23:31
blackburndamn I missed chris23:36
blackburnsonney2k: still there?23:39
blackburnn4nd0: are you finished with that stuff?23:40
n4nd0blackburn: with the cover tree pr?23:41
n4nd0the first version at least :)23:41
n4nd0now I am fighting agains my databases course23:41
blackburnuh I see23:42
n4nd0literally, fighting hehe23:42
blackburnwhen are you going to pass it?23:42
blackburngsomix: still here?23:42
n4nd0I am going to take the exam next Thursday23:42
n4nd0not tomorrow23:42
n4nd0in the next week23:42
n4nd0I am going to Spain on Tuesday :S23:42
blackburnare you going to take this exam in spain?23:43
n4nd0this one, and other two23:43
blackburnwill you go back to sweden then?23:44
n4nd0of course!23:44
n4nd0hehe, as soon as I am done with my last exam this period23:44
blackburnI have to take exam today but I'm going to take it tomorrow as it also possible23:46
blackburnand yeah I don't know anything yet23:47
blackburnn4nd0: when will you have last exam then?23:52
n4nd0blackburn: May 7th23:52
blackburnjust like me23:52
n4nd0blackburn: I've to reboot to windows :O brb23:52
-!- n4nd0 [] has quit [Quit: leaving]23:52
blackburn(but 8)23:52
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.] has joined #shogun23:55
n4nd0blackburn: so what is your exam about?23:55
blackburnn4nd0: tomorrow's is called as 'optical informatics'23:56
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]23:56
blackburna lot of shit there23:56
n4nd0blackburn: what is that about?23:57
n4nd0optics or informatics :P?23:57
blackburnn4nd0: yes exactly they are all crazy23:57
blackburnoptics mainly23:57
blackburnI bet they put some info* word there just to be it more official23:57
n4nd0stuff like Fresnel coefficients, Snell law, reflexion and so on?23:58
blackburnyes but more advanced23:58
blackburnsome photo stuff also23:58
n4nd0I don't know about that23:58
blackburnccd matrices23:58
blackburnme neither23:58
blackburnnext exams are on some software engineering, digital signal processing and (sic) politology23:59
--- Log closed Thu Apr 19 00:00:19 2012