Open in new window / Try shogun cloud
--- Log opened Tue Jun 04 00:00:03 2013
--- Day changed Tue Jun 04 2013
@iglesiasggsomix: what is it about? the one at the end00:00
gsomixiglesiasg, big state exam in general physics00:01
@iglesiasgit sounds hard indeed!00:01
gsomixiglesiasg, I think everything will be fine. this is hard year in uni, next will be more easy: just some interesting classes and getting bachelor's degree00:05
@iglesiasggood night people00:17
-!- iglesiasg [d58f3279@gateway/web/freenode/ip.] has quit [Quit: Page closed]00:18
-!- shogun-notifier- [] has quit [Quit: transmission timeout]00:56
-!- lambday [67157f4e@gateway/web/cgi-irc/] has joined #shogun01:44
lambdaymore failures for unit-tests.. gotta make all of those 1e-15 it seems01:50
-!- foulwa___ [~foulwall@2001:da8:215:6902:802c:e72b:d87a:bde8] has joined #shogun02:38
van51sonney2k: I 've modified the PR03:19
van51I ran into an issue03:19
van51please have a look and let me know03:20
-!- van51 [] has left #shogun ["QUIT :Leaving."]03:25
-!- nube [~rho@] has quit [Quit: Leaving.]04:49
-!- shogun-notifier- [] has joined #shogun05:33
shogun-notifier-shogun-demo: foulwall :master * 652dfc1 / / (10 files):
shogun-notifier-shogun-demo: unify the css.05:33
shogun-notifier-shogun-demo: Soeren Sonnenburg :master * 9144261 / / (10 files):
shogun-notifier-shogun-demo: Merge pull request #6 from foulwall/refractor05:33
shogun-notifier-shogun-demo: unify the css.05:33
-!- foulwa___ [~foulwall@2001:da8:215:6902:802c:e72b:d87a:bde8] has quit [Remote host closed the connection]05:34
-!- foulwall [~foulwall@2001:da8:215:503:7961:ad3a:cae1:e662] has joined #shogun06:59
foulwallsonney2k: ping07:00
-!- gsomix [~gsomix@] has quit [Ping timeout: 264 seconds]07:01
* foulwall tired07:20
-!- nube [~rho@] has joined #shogun07:23
-!- fw2k [~androirc@2001:da8:215:503:c874:6a48:8829:6aae] has joined #shogun07:48
-!- fw2k [~androirc@2001:da8:215:503:c874:6a48:8829:6aae] has quit [Client Quit]07:51
-!- foulwall|mobile [~foulwall@2001:da8:215:503:c874:6a48:8829:6aae] has joined #shogun07:52
-!- foulwall|mobile [~foulwall@2001:da8:215:503:c874:6a48:8829:6aae] has quit [Client Quit]07:53
-!- foulwall|mobile [~foulwall@2001:da8:215:503:c874:6a48:8829:6aae] has joined #shogun07:53
-!- foulwall|mobile [~foulwall@2001:da8:215:503:c874:6a48:8829:6aae] has quit [Remote host closed the connection]07:57
@sonney2kfoulwall, hey there!08:30
foulwallsonney2k: I'm trying to compress the messy js in a single file, almost finished08:31
@sonney2kfoulwall, yeah I've seen that complexity went down quite a bit!08:32
@sonney2kfoulwall, happy to see that :)08:32
foulwallsonney2k: haha08:32
@sonney2kfoulwall, you can assume to have the +/-1 coordinate system in all the examples08:32
@sonney2kalso in clustering I mean08:32
-!- shogun-notifier- [] has quit [Quit: transmission timeout]08:32
@sonney2kand classification08:33
foulwallsonney2k: but when we want to import custom data, I'd better make it scalable?08:34
@sonney2kfoulwall, yes - but you did that already. I mean by default -1/+1 for everything. For loading data we do sth different08:35
foulwallsonney2k: ok~08:35
@sonney2kfoulwall, btw I will reply to cheng's email08:36
foulwallsonney2k: ok~08:36
@sonney2kfoulwall, ohh and I never know what your real name is or how I should call you in emails08:36
foulwallZhengyang Liu, or just foulwall08:36
lambdaysonney2k: good morning :)08:40
lambdayfoulwall: good morning :)08:40
foulwallmorning lambday08:40
lambdaysonney2k: I fixed the unit-tests... edited JacobiEllipticFunctions to use complex64_t and related tests08:43
-!- lisitsyn [~blackburn@] has joined #shogun08:48
sonne|workfoulwall: hey08:53
foulwallsonne|work: ;)08:53
sonne|workI am back he - OK so Liu or Zhengyang? or both?08:55
sonne|worklambday: excellent!08:55
lambdaysonne|work: sending the PR08:55
foulwallZhengyang is better.08:55
sonne|workfoulwall: ok.08:56
sonne|worklambday: not there yet?08:57
lambdaysonne|work: sent08:58
lambdaysonne|work: just a sec.. I forgot something08:59
lambdaysonne|work: done! :)09:01
-!- hushell [] has quit [Ping timeout: 260 seconds]09:07
-!- nube1 [~rho@] has joined #shogun09:10
-!- nube [~rho@] has quit [Ping timeout: 256 seconds]09:13
-!- shogun-notifier- [] has joined #shogun09:14
shogun-notifier-shogun: lambday :develop * 75a8272 / / (5 files):
shogun-notifier-shogun:  complex unit-tests fixed, JacobiEllipticFunctions updated09:14
shogun-notifier-shogun: Soeren Sonnenburg :develop * 196a0a3 / / (5 files):
shogun-notifier-shogun: Merge pull request #1142 from lambday/develop09:14
shogun-notifier-shogun:  complex unit-tests fixed, JacobiEllipticFunctions updated09:14
sonne|worklambday: btw I've seen one reason why results are different: You permute the vector internally before summing09:14
sonne|worklambday: please seed the rng for that to get reproducible results09:14
sonne|workotherwise this ~1e-15 deviation makes sense with doubles09:14
sonne|worka+b+c != c+b+a with floating point numbers!09:15
lambdaysonne|work: okay..09:15
lambdayyes you're right!09:15
lambdaysonne|work: I will fix it09:17
shogun-buildbotbuild #898 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at  blamelist: lambday <>09:17
lisitsynsonne|work: do you live at your job or what? :D09:17
sonne|worklisitsyn: sure or do you have some other live?09:18
lisitsynsonne|work: I also have exams and I do not sleep at job!09:19
lisitsynso yes09:19
lisitsynI have to pass exam on some useless shit again09:20
lisitsynCORBA and ODBC haha09:20
lambdaysonne|work: lisitsyn: I'll be back in a few hours.. there is this school for orphan kids, their computer is broken or something.. our prof is the in-charge there..asked us sysads to check it :)09:21
sonne|worklisitsyn: enjoy!09:21
sonne|worklambday: yes go do some good deeds :)09:21
lambdaysonne|work: will fix it as soon as I come back09:22
lambdaysee you09:22
lambdaysee you lisitsyn :)09:22
lisitsynsee you lambday09:22
-!- lambday [67157f4e@gateway/web/cgi-irc/] has quit [Quit: lambday]09:23
lisitsynokay heading to give an imitation I have learned all these 1956' techs09:23
shogun-buildbotbuild #1066 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: lambday <>09:25
-!- foulwall|mobile [~foulwall@] has joined #shogun09:25
-!- lisitsyn [~blackburn@] has quit [Quit: Leaving.]09:26
shogun-buildbotbuild #899 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at  blamelist: Soeren Sonnenburg <>09:27
shogun-buildbotbuild #1067 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Soeren Sonnenburg <>09:34
-!- foulwall|mobile [~foulwall@] has quit [Read error: Connection reset by peer]09:39
-!- nube1 [~rho@] has quit [Quit: Leaving.]09:46
-!- nube [~rho@] has joined #shogun09:46
-!- foulwall|mobile [~foulwall@] has joined #shogun10:00
-!- fw2k [~foulwall@] has joined #shogun10:29
-!- foulwall|mobile [~foulwall@] has quit [Read error: Connection reset by peer]10:29
-!- iglesiasg [~iglesiasg@] has joined #shogun10:37
-!- mode/#shogun [+o iglesiasg] by ChanServ10:37
-!- foulwall [~foulwall@2001:da8:215:503:7961:ad3a:cae1:e662] has quit [Remote host closed the connection]10:43
-!- vgorbati [~vgorbati@] has joined #shogun10:55
-!- fw2k [~foulwall@] has quit [Remote host closed the connection]11:02
thoralfHey guys.11:19
thoralfI just sent a small patch for a segfault in CMath::qsort(), which also affects SGVector.qsort().11:20
@iglesiasghey thoralf, I just commented your patch11:31
@iglesiasgthanks for it!11:31
@iglesiasgjust send it against develop please11:31
thoralfiglesiasg: Okay, no problem.  But you should update your documentation. ;)) See:
thoralfSection VERSIONING SCHEME:11:33
@iglesiasgwe should indeed11:33
@iglesiasgsorry about that11:33
@iglesiasgfeel free to send another patch with the doc update :)11:33
thoralfiglesiasg: I just found out github allows to send pull request against arbitrary upstream branches.11:38
@iglesiasgyou should actually be using another branch too btw11:39
-!- gsomix [~gsomix@] has joined #shogun11:39
gsomixlol, fail11:40
thoralfiglesiasg: That's right.11:40
@iglesiasgthe idea is that your master in synced with upstream master, your develop with upstream develop and you submit patches from new feature branches11:40
@iglesiasghey gsomix11:40
@iglesiasgwhy fail? :)11:40
@iglesiasgthoralf, but I think it should be fine to merge it like it is this time11:40
thoralfiglesiasg: Btw, my orignal PR was against shogun/develop.11:40
@iglesiasgiglesiasg, oh fuck yes11:40
@iglesiasgit was the same11:41
@iglesiasgepic fail11:41
thoralfiglesiasg: But I will switch to develop locally, anyway.  Thanks for telling. ;)11:41
@iglesiasgthoralf, ok, sorry for my misunderstanding11:41
gsomixiglesiasg, I'm just stupid, huh. will try tomorrow again. :)11:41
shogun-notifier-shogun: Thoralf Klein :develop * d6b0fbb / src/shogun/mathematics/Math.h:
shogun-notifier-shogun: CMath::qsort() - Prevent segfault when trying to sort zero-size-vectors.11:51
shogun-notifier-shogun: Fernando Iglesias :develop * 45d7f76 / src/shogun/mathematics/Math.h:
shogun-notifier-shogun: Merge pull request #1144 from tklein23/master11:51
shogun-notifier-shogun: Fixed: Segfault when sorting empty lists using CMath::qsort() (re-sent to develop)11:51
shogun-buildbotbuild #1068 of bsd1 - libshogun is complete: Failure [failed configure]  Build details are at  blamelist: Thoralf Klein <>12:00
thoralfError: Unknown parameter: --swig=swig2.012:05
thoralfI think I'm not responsible for this. ;)12:05
@iglesiasgthoralf, the configure script doesn't recognize your swig automatically?12:07
thoralfiglesiasg: No, it's the log from the failed build above.12:11
shogun-buildbotbuild #900 of cyg1 - libshogun is complete: Success [build successful]  Build details are at
@iglesiasgah ok12:12
-!- nube [~rho@] has quit [Quit: Leaving.]12:12
@iglesiasgthoralf, I am not sure why that happened12:12
@iglesiasgbut it can't be related to the patch12:13
thoralfiglesiasg: Sure not. :)12:13
shogun-buildbotbuild #1069 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Fernando Iglesias <>12:14
thoralfiglesiasg: I just wanted to know if I have to wear the tin-foil-head now.12:14
@iglesiasghehe no I don't think so12:15
@iglesiasgthoralf, SGVector test failed though12:16
@iglesiasgbut I think it is unrelated as well12:17
shogun-buildbotbuild #1193 of deb3 - modular_interfaces is complete: Failure [failed compile lua_modular]  Build details are at  blamelist: Thoralf Klein <>12:24
shogun-buildbotbuild #901 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at  blamelist: Fernando Iglesias <>12:24
@iglesiasgthoralf, we have killed shogun :P12:25
@iglesiasgshogun-buildbot force bsd1-libshogun12:36
@iglesiasgshogun-buildbot, force "bsd1 - libshogun"12:37
shogun-buildbottry 'force build [--branch=BRANCH] [--revision=REVISION] [--props=PROP1=VAL1,PROP2=VAL2...]  <WHICH> <REASON>'12:37
@iglesiasgshogun-buildbot, force "bsd1-libshogun"12:37
shogun-buildbottry 'force build [--branch=BRANCH] [--revision=REVISION] [--props=PROP1=VAL1,PROP2=VAL2...]  <WHICH> <REASON>'12:37
@iglesiasgshogun-buildbot, force build "bsd1-libshogun"12:37
shogun-buildbotno such builder 'bsd1-libshogun'12:37
@iglesiasgshogun-buildbot, force build "bsd1 - libshogun"12:37
shogun-buildbotbuild forced [ETA 7m02s]12:37
shogun-buildbotI'll give a shout when the build finishes12:37
shogun-buildbotbuild #1070 of bsd1 - libshogun is complete: Failure [failed configure]  Build details are at
@iglesiasgit seems that option stopped working for some reason12:38
@iglesiasgshogun-buildbot, force build "cyg1 - libshogun"12:39
shogun-buildbotbuild forced [ETA 13m33s]12:39
shogun-buildbotI'll give a shout when the build finishes12:39
@iglesiasgshogun-buildbot, force build "deb3 - modular_interfaces"12:39
shogun-buildbotbuild forced [ETA 44m14s]12:39
shogun-buildbotI'll give a shout when the build finishes12:39
shogun-buildbotbuild #902 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at
@iglesiasgwiking, any idea what's wrong with swig2.0 option in bsd?12:42
shogun-buildbotbuild #1195 of deb3 - modular_interfaces is complete: Failure [failed compile lua_modular]  Build details are at
thoralfiglesiasg: "we have killed shogun :P" <-- That was easy. ;)13:20
thoralfiglesiasg: Anything we can do about it?13:20
@iglesiasgthoralf, as far as I can tell it doesn't seem to be our fault13:20
@iglesiasgthoralf, bsd1 seems to fail sometimes because of the swig2.0 option13:21
@iglesiasgand also there is a test related to complex numbers13:21
-!- vgorbati_ [~vgorbati@] has joined #shogun13:21
-!- vgorbati [~vgorbati@] has quit [Read error: Operation timed out]13:21
-!- vgorbati_ is now known as vgorbati13:21
@iglesiasgand the other two seem unrelated tome as well13:21
shogun-buildbotbuild #1194 of deb3 - modular_interfaces is complete: Failure [failed test ruby_modular]  Build details are at  blamelist: Fernando Iglesias <>13:25
sonne|workthoralf: it is totally weird. this ruby test is failing every now and then but no valgrind issues...13:28
@iglesiasgsonne|work, any idea about the others?13:30
sonne|workiglesiasg: yes the cyg one is wtf too13:31
sonne|workand the bsd one was about to be fixed by lambday13:31
sonne|workbut he has to do some charity work currently13:31
sonne|workright after that13:31
@iglesiasgsonne|work, but it seems that in the bsd not only the test fails13:32
@iglesiasgsometimes the option swig=swig2.0 is not recognized13:32
sonne|workiglesiasg: errm did somebody commit to master?13:33
sonne|workiglesiasg: only in master this option is unknown...13:33
@iglesiasgsonne|work, thoralf patch was from his master to upstream develop13:34
@iglesiasgI thought that shouldn't be trouble13:34
@iglesiasgbut maybe I was wrong13:34
sonne|workno that is OK13:40
-!- iglesiasg [~iglesiasg@] has quit [Quit: Leaving]13:59
gsomixsonne|work, hey. how are you? I'll be available at late evening - need to prepare for next try of exam. :)14:00
sonne|workgsomix: working as usual :D14:05
sonne|workgood luck!14:05
-!- lisitsyn [] has joined #shogun14:17
lisitsynha-ha that was funny14:18
sonne|worklisitsyn: what did you do?14:21
lisitsynsonne|work: pretended I know sth on databases14:22
sonne|workohh I did that too14:24
sonne|workthough I never attended the lecture14:24
lisitsynsonne|work: do you think I attended any lecture?14:34
sonne|workpretending is attending14:34
lisitsynsonne|work: well I attended exam14:35
-!- mode/#shogun [+o lisitsyn] by ChanServ14:45
-!- shogun-notifier- [] has quit [Quit: transmission timeout]14:51
-!- HeikoS [] has joined #shogun15:18
-!- mode/#shogun [+o HeikoS] by ChanServ15:18
-!- iglesiasg [c1934d18@gateway/web/freenode/ip.] has joined #shogun15:33
-!- mode/#shogun [+o iglesiasg] by ChanServ15:33
@lisitsyniglesiasg: are you following game of thrones?15:33
@lisitsynI heard about some tragedy in the last episode :D15:34
@lisitsynpeople are out of rage15:34
@iglesiasglisitsyn: yeah man15:37
@iglesiasglisitsyn: it was so epic15:37
@iglesiasgI was shocked15:38
@iglesiasgI watched it twice in a row to recover actually15:38
@iglesiasgas weird as it must sound haha15:38
-!- nube [~rho@] has joined #shogun15:44
thoralfSome examples in examples/undocumented/libshogun are expecting -lpthread, may I send a PR with fixed Makefile (LIBS_ADD=... -lpthread)?15:48
@iglesiasgthe -lpthread issue seems to be always around15:49
@iglesiasgI honestly  don't really get when it is needed and when not15:49
thoralfiglesiasg: There are two or three tests, which are depending on -lpthread.15:50
sonne|workthoralf: no15:50
thoralfsonne|work: So why are they failing?15:51
thoralfsonne|work: Okay, one step back - which part exactly are you rejecting? ;)15:52
sonne|workthoralf: no to add this to the Makefile15:52
sonne|workthoralf: when you run make check-examples from the src directory15:52
sonne|workthey will be run with additional link flags etc taken from .config15:52
thoralfsonne|work: Yes, .config contains come definitions with -lpthread.  But no, when I run make check-examples from src, library_map and classifier_svmlight_string_features_precomputed_kernel always fail.16:00
thoralfsonne|work: Just double-checked it on clean upstream/develop16:01
sonne|workthoralf: which OS is this?16:01
thoralfsonne|work: Ubuntu 13.0416:01
sonne|workwe do exactly this on all build bots / travis16:01
thoralfsonne|work: But Makefile in undocumented/libshogun does not include .config, so how do the options get to the gcc?16:02
thoralferror in make classifier_svmlight_string_features_precomputed_kernel CC=c++ LIBRARY_PATH=LD_LIBRARY_PATH LIB_PATH=~/sg/lib INC_PATH=~/sg/include LIBS=-Xlinker --no-undefined  -lshogun16:03
sonne|workthoralf: which examples fail again?16:03
thoralflibrary_map and classifier_svmlight_string_features_precomputed_kernel16:03
thoralfclassifier_svmlight_string_features_precomputed_kernel: undefined reference to symbol 'pthread_create@@GLIBC_2.2.5'16:04
thoralflibrary_map: undefined reference to symbol 'pthread_spin_init@@GLIBC_2.2.5'16:04
sonne|workI don't understand why linking against libshogun doesn't help16:07
thoralfsonne|work: But why should -lshogun export "foreign" symbols, i.e. symbols of libpthread?16:09
thoralfsonne|work: ldd in shows: => /lib/x86_64-linux-gnu/ (0x00007f1d81084000)16:12
thoralfnm -C -D shows pthread_* functions, but without address.16:15
-!- vgorbati [~vgorbati@] has quit [Quit: vgorbati]16:16
-!- Rasmi [~Rasmi@] has joined #shogun16:24
sonne|workthoralf: erm, you don't have to link against blas/lapack when you use libshogun that internally uses it16:25
sonne|workso we should somehow get to the same state with libpthread16:26
thoralfsonne|work: You're right.  I was expecting library_map using it - but it does not.16:27
sonne|workthoralf: hmmhh maybe there is something subtle here16:28
sonne|workit uses Map.h which includes pthread16:28
-!- Rasmi [~Rasmi@] has quit [Ping timeout: 252 seconds]16:29
sonne|workI guess for locking / thread safety16:29
thoralfsonne|work: Yes, I'm already following this trail... :)16:29
thoralfsonne|work: If some of the header files depends on pthread_* functions directly, then we're doomed. ;)16:30
sonne|workif so we should somehow try to move everything over to a .cpp file16:30
-!- Rasmi [~Rasmi@] has joined #shogun16:32
-!- Rasmi [~Rasmi@] has left #shogun []16:35
-!- vgorbati [~vgorbati@] has joined #shogun16:36
thoralfSo, what did we found?  Map.h uses pthread_*() directly.  So every use of Map.h introduces a compile-time dependency on pthread-symbols, e.g. -(l)pthread16:37
thoralfBut this does not (yet) explain classifier_svmlight_string_features_precomputed_kernel16:37
thoralfMath.h also depends on pthreads.16:41
thoralfWhich is used by SVMLight.h16:41
thoralfwhich is used in the example.16:41
thoralfq.e.d. ;)16:41
-!- vgorbati [~vgorbati@] has quit [Quit: vgorbati]16:50
-!- Rasmi [~Rasmi@] has joined #shogun17:00
@iglesiasghey Rasmi17:01
Rasmiya I want to know when I shall know that I have configured shogun17:02
@iglesiasgwhen you run the configure script and no error fires17:03
RasmiIs it that when I write shogun it will show the shogun prompt17:03
@iglesiasgRasmi: mmm no, not really17:04
@iglesiasgyou don't get something like Matlab or octave if you are waiting for that17:04
@iglesiasgRasmi: have you tried running any of the examples? That should give you a better feeling of it17:05
Rasmino, I think your are not getting what I want to say17:05
RasmiI want to say that when I will write shogun in the terminal it will show me the shogun prompt17:06
@iglesiasgoh then I got what you said :)17:07
RasmiI have my laptop if you can come to web cam then it will be better17:08
@sonney2kthoralf, hmmhh but lots more uses CMath17:08
@sonney2kthere is something fishy17:09
@iglesiasgRasmi: I think there is no need for that17:09
RasmiI am not getting what to do with this shogun, please help me17:10
Rasmihow to configure nad how to use17:10
@iglesiasgRasmi: from your other mails I think you have already configured and installed17:11
@lisitsynwhat do you to use it for?17:11
@iglesiasgnow try out some of the examples and you will get a feeling about "what to do with this shogun"17:11
RasmiI am doing my project using SVM17:12
thoralfsonney2k: So I think pthreads are not "optional" any more. ;)17:12
-!- vgorbati [~vgorbati@] has joined #shogun17:12
@sonney2kthoralf, they are optional (otherwise we could not compile under cyg)17:13
Rasmiand now my professor told me to solve a problem using shogun17:13
@iglesiasgwow you have a cool professor!17:14
@sonney2kiglesiasg, I've recently seen in a job ad: "requirements: experience in one of R, weka, shogun" :D17:14
thoralfsonney2k: You're right.  The problems only arise when pthreads are enabled.17:14
Rasmiok let me try some examples17:15
@iglesiasgRasmi: if you want to use it for SVMs and the C++ interface try for instance the example classifier_minimal_svm.cpp17:15
@iglesiasgsonney2k: nice!17:15
thoralfsonney2k: as far as i can see, all the pthread calls are enclosed by "#ifdef HAVEPTHREADS"17:15
@iglesiasgRasmi: just in case, you can find it in <path to your shogun directory>/examples/undocumented/libshogun/classifier_minimal_svm.cpp17:16
@lisitsynsonney2k: no way, rly?17:16
Rasmiya I got it17:16
@iglesiasgok I've got to go now, I will be back later in a few hours17:17
Rasmithat exmple17:17
-!- Rasmi [~Rasmi@] has quit [Quit: irc2go]17:18
-!- zxtx [] has quit [Ping timeout: 240 seconds]17:18
@sonney2klisitsyn, yup17:19
-!- iglesiasg [c1934d18@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:19
@lisitsynsonney2k: what are your working hours?17:24
@sonney2klisitsyn, huh?17:26
@sonney2kwhat's wrong?17:27
@lisitsynsonney2k: just curious17:27
* sonney2k is suspicious now17:27
@lisitsynsonney2k: haha no I just want to know when you are at job17:28
@sonney2kwell you talked to me in the morning when I was there and now that I just left :D17:28
@lisitsynit floats a bit so I am unsure now17:28
@lisitsynlike 8 - 17 or what?17:29
@sonney2kno it doesn't really float17:29
@sonney2kin this range but my only 8hrs is the standard contract here :D17:29
@sonney2kbut you know when you are just about to finish $FEATURE17:29
@lisitsynsonney2k: what about lunch?17:30
@sonney2kalrighty gtg17:30
@lisitsynI mean 8-17 is 8+117:30
@lisitsynI mean speaking of official stuff I am working 10-1917:30
@lisitsynthat's 8 hours17:30
thoralfJust sent another PR for qsort() in Math.h - this time with unit tests.17:45
thoralf(Unit test without assertions, because they're meant to trigger segfaults.)17:45
thoralfDamnit.  Found another two potential segfaults in CMath::qsort*().18:04
thoralfBefore fix, make valgrind gives: ERROR SUMMARY: 42150 errors from 79 contexts (suppressed: 2 from 2)18:04
thoralfAfter fix: ERROR SUMMARY: 8076 errors from 75 contexts (suppressed: 2 from 2)18:04
-!- lisitsyn [] has quit [Quit: Leaving.]19:37
-!- van51 [] has joined #shogun19:48
@sonney2kvan51, hey!19:48
@sonney2kvan51, good evening19:48
@sonney2kvan51, I've lost your email somehow19:48
van51sonney2k: hello19:48
van51i'll send it to you then19:48
van51sonney2k: at your shogun address, right?19:49
@sonney2kvan51, yes please19:52
@sonney2kvan51, I only recall that sparsefeatures gave you some trouble19:53
@sonney2kand the feature iterator19:53
@sonney2kwhich is not implemented for everything19:53
@sonney2kvan51, let me just explain a bit the idea of dotfeatures19:53
@sonney2kwhen you usually do learning you get your features from some domain, say R^n19:53
@sonney2kso you have all fixed size n-dim vectors19:54
@sonney2kwith dot features you can have features of arbitrary domain19:54
@sonney2kyou can mix dense and sparse features and string features19:54
@sonney2kall you need is to provide a certain number of operations like add_to_dense_vec19:55
@sonney2ketc that are defined in dotfeatures19:56
van51sonney2k: ok so far19:56
@sonney2kwith these operations one can use quite a number of methods already19:57
@sonney2kfor example linear SVMs like liblinear can be expressed like using these dense_dot, add_to_dense_vec etc19:57
@sonney2kvan51, the good thing about providing these functions is that you only need to operate on non-zero features19:58
@sonney2kso if you have some hash(word) and you would need to implement add_to_dense_vec19:58
@sonney2kyou would only need to do w[hash(word)]+=some_value19:59
@sonney2kfor all words in a text document19:59
@sonney2keven though w (the dense vector) might be billion-dimensional20:00
van51sonney2k: yeah ok, I get that20:00
van51my question was, you usually create a Features object and pass it to a classifier20:01
van51doesn't this require the dataset to be in memory?20:01
@sonney2know about these feature iterators that failed20:01
@sonney2kthey are there because not all functions can be expressed with dense_dot etc20:01
@sonney2kidea is that you iterate over the non-zero features.20:01
@sonney2kit certainly is not that fast to use them but maybe it is worth to make some algorithms work with them...20:01
@sonney2kvan51, that's it20:01
@sonney2kvan51, yes20:02
@sonney2kvan51, of course we have the same streaming*dotfeatures20:03
@sonney2kso you can do online learning20:03
thoralfvan51: I think this topic is related to my work, too. ;)20:04
thoralfvan51: Can you please talk in public?20:04
@sonney2kunfortunately we didn't have the shiny unit/integration tests (activated) so we broke quite a bit of the streaming* code20:05
@sonney2kwhen we introduced SGVector & friends20:05
@sonney2kvan51, anyway my suggestion would be you start with some very simple function that gets a char[256] array as input20:07
@sonney2kto tokenize a string20:07
thoralfWhat topic is van51 working on?20:07
@sonney2kyou do that once as shogun preprocessor20:07
van51sonney2k: ok w820:08
-!- nube [~rho@] has quit [Ping timeout: 252 seconds]20:08
van51sonney2k: you are referring to the iterator issue, right?20:08
van51sonney2k: I thought you meant my email from yesterday in the beginning20:08
van51van51: about the iterator, I pretty much figured its usage from the comments20:08
van51sonney2k: but it is implemented in SparseFeatures20:08
van51and I am getting a seg fault when calling it20:08
@sonney2kand once inside dot features20:08
-!- nube [~rho@] has joined #shogun20:09
van51I think I have a problem with my connection because I'm getting all the messages at once :S20:09
van51thoralf: are you referring to the online learning part that sonney2k mentioned?20:10
@sonney2kand train once (with preprocessor) just with liblinear and once with the dotfeatures and liblinear20:10
-!- Netsplit *.net <-> *.split quits: @wiking, shogun-buildbot20:11
@sonney2kyou should get *exactly* the same result20:11
@sonney2kjust try to avoid code duplication20:11
-!- wiking [] has joined #shogun20:13
thoralfsonney2k: did you refer to me? ;)20:13
@sonney2kvan51, did you run away screaming?20:13
@sonney2kthoralf, all kinds of large scale stuff - hashing and random kitchen sinks / fast food20:14
thoralfsonney2k: Yes, when you told him to avoid code duplication :)20:14
-!- Netsplit over, joins: shogun-buildbot20:14
@sonney2kvan51, wow all messages in one go20:14
@sonney2kvan51, heh you send me an email but not the one with the iterator stuff20:15
@sonney2kvan51, I need to see the code20:15
@sonney2kI suspect you are not creating the sparse features correctly20:15
@sonney2kthoralf, sry was that for van51?20:16
thoralfsonney2k: No, just a joke.  "van51, did you run away screaming?" -> "Yes, when you told him to avoid code duplication :)"20:17
thoralfJoke was killed by lag.20:17
@sonney2kahh :)20:18
-!- van511 [] has joined #shogun20:18
@sonney2kvan511, you missed that we were making fun of you :P20:18
van511I was reading the logs from the site :p20:18
van511sonney2k: all the messages I received were in one go and it was a bit confusing20:19
@sonney2kvan511, so what do we do now? shall we merge your list ->dynarray for combined dot features?20:20
van511sonney2k: ok, first things first20:21
van511sonney2k: the iterator code that causes this issue20:21
-!- lisitsyn [~blackburn@] has joined #shogun20:21
-!- mode/#shogun [+o lisitsyn] by ChanServ20:21
van511sonney2k: is in the last test in the unit test I created20:21
van511sonney2k: if I remove it for now, you can probably merge20:21
@sonney2kand you send me some example on how to crash sparsefeatures?20:22
van511sonney2k: but as I said in the comments there are a couple functions that are not tested20:22
-!- Netsplit *.net <-> *.split quits: van5120:22
van511sonney2k: here :
van511sonney2k: end of page20:23
-!- van511 is now known as van5120:24
-!- HeikoS [] has quit [Quit: Leaving.]20:25
@lisitsynsonney2k: where are students :D20:29
-!- sonney2k [] has quit [Excess Flood]20:33
-!- sonney2k [] has joined #shogun20:38
-!- mode/#shogun [+o sonney2k] by ChanServ20:38
-!- van511 [] has joined #shogun20:40
-!- lisitsyn1 [~blackburn@] has joined #shogun20:40
van511sonney2k: I have to go for now20:40
van511sonney2k: I'll remove it as a test when I get back so you can merge20:41
van511sonney2k: and let me know if it was my fault so I can create it correctly afterwards20:41
van511bye for now20:41
-!- van511 [] has left #shogun []20:43
@sonney2klisitsyn, have you heard anything from pickle27?20:43
@sonney2khardcore conference hopping?20:43
@sonney2klooks like freenode is dying today20:43
@sonney2kvan51, not your fault20:43
@sonney2kkeep it (but comment it)20:43
@sonney2kI will investigate then20:43
@sonney2kvan51, cu20:43
-!- iglesiasg [d58f327d@gateway/web/freenode/ip.] has joined #shogun20:43
-!- mode/#shogun [+o iglesiasg] by ChanServ20:44
lisitsyn1sonney2k: well he said he will be at conferences20:45
lisitsyn1I'll mail him20:46
-!- Netsplit *.net <-> *.split quits: @lisitsyn, van5120:49
lisitsyn1ok kick ass time :D20:49
@sonney2kscrew freenode :/20:51
-!- wiking_ [] has joined #shogun20:59
-!- wiking [] has quit [Ping timeout: 245 seconds]21:05
-!- iglesiasg [d58f327d@gateway/web/freenode/ip.] has quit [Quit: Page closed]21:15
-!- zxtx [] has joined #shogun21:54
-!- vgorbati [~vgorbati@] has quit [Quit: vgorbati]22:09
-!- Netsplit *.net <-> *.split quits: wiking_22:18
-!- wiking [] has joined #shogun22:24
-!- iglesiasg [d58f3239@gateway/web/freenode/ip.] has joined #shogun23:11
-!- mode/#shogun [+o iglesiasg] by ChanServ23:14
-!- lisitsyn1 [~blackburn@] has quit [Ping timeout: 256 seconds]23:56
-!- iglesiasg [d58f3239@gateway/web/freenode/ip.] has quit [Quit: Page closed]23:57
--- Log closed Wed Jun 05 00:00:03 2013