Open in new window / Try shogun cloud
--- Log opened Sat Mar 16 00:00:13 2013
-!- FSCV [~FSCV@] has quit [Ping timeout: 276 seconds]00:02
-!- FSCV [~FSCV@] has joined #shogun00:04
-!- zxtx [] has quit [Ping timeout: 264 seconds]00:57
-!- FSCV [~FSCV@] has quit [Quit: Leaving]00:59
-!- n4nd0 [] has quit [Quit: leaving]04:24
-!- hoijui [] has joined #shogun07:34
@sonney2kblackburn, we still have crashes. as soon as make prepare-release works we can go further08:56
@sonney2kblackburn, feel free to run that your own and fix the crashers08:56
@sonney2kblackburn, I don't have time today to work on it so any progress helps09:05
blackburnsonney2k: okay09:09
wikingcheck this plz and if u agree then let's merge it:
-!- hoijui [] has quit [Ping timeout: 256 seconds]10:34
-!- votjak [] has joined #shogun10:56
wikingcheck this plz and if u agree then let's merge it:
wikingoh sorry11:07
-!- n4nd0 [] has joined #shogun11:13
n4nd0hey blackburn11:28
n4nd0where can I find the pictures that compose
blackburnn4nd0: hey11:46
blackburnoh let me find it11:46
blackburnwiking: so -a is more widespread than --append?11:47
wikingonly linux has --apend11:54
blackburnwiking: let it merge then11:55
-!- shogun-notifier- [] has joined #shogun11:55
shogun-notifier-shogun: Viktor Gal :master * c758bf6 / Makefile:
shogun-notifier-shogun: use -a for tee as that's more widespread among OSes11:55
shogun-notifier-shogun: Sergey Lisitsyn :master * 3d7cf1a / Makefile:
shogun-notifier-shogun: Merge pull request #941 from vigsterkr/master11:55
shogun-notifier-shogun: use -a for tee as that's more widespread among OSes11:55
blackburnn4nd0: I found a script that produced that but lost images :D11:56
n4nd0blackburn: hehe ok, I will play around with faces then11:59
shogun-buildbot_build #807 of deb2 - static_interfaces is complete: Success [build successful]  Build details are at
blackburnn4nd0: I will prepare road signs HOG features12:09
blackburnn4nd0: but we'd need your machine to handle that :D12:10
n4nd0blackburn: sure, no problem12:25
n4nd0blackburn: is it to put more tests in tapkee?12:25
blackburnn4nd0: no, it is rather example12:29
blackburnn4nd0: yikes, the data is 1.4 gb12:30
n4nd0blackburn: wow how many feature vectors and dimension?13:27
blackburnn4nd0: 39209 of 3200d14:08
blackburnn4nd0: I'd reduce it to somewhat 400d I think14:20
blackburnor it gets pretty hard to embed in reasonable time14:21
-!- shogun-notifier- [] has quit [Quit: transmission timeout]14:55
-!- n4nd0 [] has quit [Quit: leaving]16:11
wikingsonney2k: around?17:32
-!- wiking [] has quit [Quit: leaving]17:45
-!- hoijui [] has joined #shogun18:36
-!- heiko [] has joined #shogun18:42
heikoblackburn:  hi18:43
blackburnheiko: hey18:43
heikojust reading MulticlassSVM header file18:43
heikowe really should not remove warnings if documentation is missing by putting /** @brief MulticlassSVM */18:43
heikosame for the methods :)18:43
blackburnheiko: heh right18:44
heikothats way too lazy :D18:44
blackburnheiko: I forgot about that bug report18:44
heikoblackburn: no bug here18:44
blackburnrunning prepare-release now18:44
blackburnheiko: what is wrong?18:44
heikothe guy just wrote bad code18:44
heikouninitialised memory etc18:44
heikoeverything works18:44
heikoIll send him a nice example back18:44
heikoand add to libshogun18:44
blackburnwell we are who to blame for that people can write bad code :)18:45
heikoblackburn: you are totally right18:46
heikobut he did not even init his label memory18:46
heikoso there is no way we can avoid that in c++18:46
blackburnheiko: I am not sure about that - may be there is a way18:46
-!- votjak [] has quit []19:00
blackburnheiko: dst.template triangularView<Upper>() = src.template triangularView<Upper>();19:03
blackburnheiko: ^ that's valid C++19:03
blackburnsometimes I wanna cry19:03
heikoblackburn: ? :)19:03
heikowhat happens if you do it19:03
blackburnheiko: dst.template[Space here]<Upper>()19:04
blackburnheiko: crazy C++19:04
heikoblackburn: am I allowed to commit stuff?19:06
blackburnheiko: yes, for a year already :D19:06
heikoblackburn:  you are a scherzkeks19:06
heikoI mean does it interfer with your release work?19:07
blackburnheiko: ah no, I just test if it works19:07
heikoblackburn:  just ran into another multiclass svm bug :(19:08
blackburnheiko: scherzkeks is kind of 'joker'?19:08
heikoblackburn:  yes :)19:08
blackburnheiko: what is da bug?19:09
heikoblackburn: I will add the example19:11
heikothen it will fail19:11
heikoand you are forced to solve it :D19:11
heikoREQUIRE(set_svm(idx, svm),"SVM set failed")19:11
heikoline 194 fails19:11
heikodepending on inpu data19:11
heikoanother really nice bug19:11
blackburnheiko: ahh there was some issue with that19:12
heikoit only occurs with multiple classes19:16
heikoIll add example19:16
heikoand you have to increase number of classes from two to 3 or something to see what fails19:16
blackburnheiko: btw would be nice to finally release asap or it gets forever19:16
heikoblackburn:  agreed19:16
heikothis bug would be nice to fix19:16
heiko(also mention it in NEWS)19:17
blackburnheiko: ok19:17
heikoand the guy from the mailing list would be happy19:17
-!- shogun-notifier- [] has joined #shogun19:19
shogun-notifier-shogun: Heiko Strathmann :master * 51e5322 / examples/undocumented/libshogun/ (2 files):
shogun-notifier-shogun: added a modelselection example for multiclass lib svm19:19
shogun-notifier-shogun: Heiko Strathmann :master * a88be17 / examples/undocumented/libshogun/ (2 files):
shogun-notifier-shogun: Merge pull request #942 from karlnapf/master19:19
shogun-notifier-shogun: new example for model selection multiclass svm19:19
heikoblackburn: set number of dimensions/classes to 3 and you see the problem. If it works, could you permantenly increase it?19:20
heikoblackburn: I gotta go now, will write to the mailing list and see you tomoroow19:20
blackburnheiko: I'll check19:20
heikoblackburn: thanks!19:26
heikoblackburn: and a set_autolock method for multiclass would be great19:26
heikoor even a lock method :)19:26
heikobut thats for later19:26
heikothe set_atuolock would prevent the warning that is outputted in the ex.19:27
heikoblackburn:  ok gotta run now. bye!19:27
-!- heiko [] has left #shogun []19:27
-!- zxtx [~zv@] has joined #shogun20:17
shogun-buildbot_build #354 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>20:26
@sonney2kblackburn, any progress towards release?20:36
blackburnsonney2k: yeah I'll fix a bug pretty soon20:37
blackburnsonney2k: I am not sure what you meant by crashes20:37
blackburnI ran prepare-release (takes \infty) and there were no real crashes20:37
@sonney2kblackburn, really?20:38
@sonney2kit crashed here20:38
blackburnsonney2k: what exactly?20:38
@sonney2kclassifier_perceptron in octave static20:38
blackburnsonney2k: I tried octave_static examples - all ran20:39
blackburnthere are a lot of integration test errors, but we can't fix it yet20:40
@sonney2kyeah sure not a blocker20:40
@sonney2kblackburn, btw make prepare-release takes so long b'coz it compiles shogun for all $LANGs20:41
@sonney2kso now you know why I would love to have lower compile times20:42
blackburnsonney2k: we'd have to speedup swig for that20:43
blackburn[0.993604,1.002475] with alpha=0.050000, mean=0.99803920:43
blackburnsonney2k: minor bug in xval :D20:44
blackburnsonney2k: conf interval is not cut by maximum possible value20:44
blackburnand we tell users accuracy is in between 0.99 and 1.002 hah20:44
@sonney2kblackburn, not swig is the problem...20:48
blackburnsonney2k: compilation of classes is pretty fast20:48
@sonney2klibshogun is fast20:48
blackburnsonney2k: at least getting faster with new versions of compilers20:48
@sonney2kbut we would need split up version of the swig wrapper20:48
@sonney2kblackburn, ohh octave seems to work now - the environment variable needs to be OCTAVE_PATH no longer OCTAVE_LOADPATH20:49
shogun-notifier-shogun: Sergey Lisitsyn :master * a77b5ba / src/ChangeLog:
shogun-notifier-shogun: updated changelog20:51
shogun-notifier-shogun: Sergey Lisitsyn :master * fc7f16e / src/ChangeLog:
shogun-notifier-shogun: updated changelog20:51
shogun-notifier-shogun: Sergey Lisitsyn :master * c77af73 / examples/undocumented/libshogun/ (2 files):
shogun-notifier-shogun: Merge branch 'master' of
shogun-notifier-shogun: Sergey Lisitsyn :master * ec6e149 / / (2 files):
shogun-notifier-shogun: Fixed bug with libsvm 3 classes20:51
blackburnsonney2k: yikes, these two commits were produced by prepare-release20:52
blackburnI am not sure what changelog is used for though20:52
blackburnexcept gpl stuff20:53
@sonney2kblackburn, yeah I know20:54
@sonney2knext time git reset --hard20:54
blackburnsonney2k: git reset HEAD~1 is enough20:55
@sonney2kblackburn, but that is a soft reset then right? and you still have this change lying around21:00
blackburnsonney2k: they are erased but next commits I think21:00
blackburnsonney2k: argh we failed to release on my bday :D21:01
blackburn56 seconds ago21:01
@sonney2kblackburn, when was it?21:01
blackburnsonney2k: today for you, yesterday for me21:01
@sonney2kblackburn, then congrats. now you are approaching the thirties quickly :D21:03
blackburnsonney2k: hah thanks21:03
blackburnsonney2k: 8 more years to approach it21:03
@sonney2kblackburn, so you can still become professor before that :)21:05
blackburnsonney2k: phd in doing useless things :D21:06
@sonney2kargh I need a faster machine21:06
@sonney2kblackburn, for example machine learning ;)21:06
blackburnsonney2k: yeah pretty useless stuff mostly21:06
@sonney2kblackburn, so go to guu and cure cancer!21:06
blackburnguu sounds funny21:07
@sonney2kblackburn, our json serialization is broken :/21:09
@sonney2kserialization_complex_example.lua died21:09
shogun-notifier-shogun: Soeren Sonnenburg :master * 0e282c7 / / (2 files):
shogun-notifier-shogun: release fixes21:26
shogun-notifier-shogun: - disable json serialization for now21:26
shogun-notifier-shogun: - rename OCTAVE_LOADPATH -> OCTAVE_PATH21:26
blackburnsonney2k: hmm21:26
blackburnsonney2k: any reason?21:27
blackburnsonney2k: I mean it worked before right?21:27
@sonney2kno idea21:27
blackburnquite strange then21:27
blackburnI do not remember any chagnes21:27
shogun-buildbot_build #355 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Heiko Strathmann <>21:28
@sonney2kblackburn, I couldn't find any either21:28
@sonney2knice intro to random forests21:35
-!- hoijui [] has quit [Ping timeout: 252 seconds]21:36
blackburnsonney2k: actually I never tried them (shame on me)21:37
@sonney2kblackburn, they work as good as SVMs21:38
blackburnsonney2k: I beat them on GTSRB21:38
@sonney2kI used them to predict whether a molecule is a drug or not back in the days :D21:38
blackburnsonney2k: german traffic sign recognition benchmark21:38
blackburnsonney2k: funny thing happened with GTSDB (detection instead of recognition)21:39
blackburnsonney2k: they all got 1.0 ROCs21:39
@sonney2kblackburn, yeah they cannot provide realvalued outputs21:39
@sonney2kit is a majority vote21:39
@sonney2k100 trees...21:40
@sonney2kbtw, bagged KNN21:40
@sonney2kworks also very very well21:40
@sonney2kno idea if that also has a standard name21:40
blackburnsonney2k: never heard about special name for that21:40
blackburnsonney2k: bagging is effective thing for anything I guess21:41
@sonney2kblackburn, are you in this table?21:41
blackburnsonney2k: I wish I was21:42
@sonney2ksome ELL_SVM2 won?21:42
blackburnsonney2k: I started my research on recognition after GTSRB and I just had no time to do GTSDB21:42
blackburnsonney2k: not sure what ELL means21:43
blackburnsonney2k: second place is integral channel features with multiple scaling I guess21:43
blackburnaccording to its name21:43
@sonney2kblackburn, alright I am off to bed.21:44
blackburnsonney2k: good night21:44
@sonney2kblackburn, get drunk and celebrate!21:44
blackburnsonney2k: I have to finish paper stuff21:45
@sonney2kI hope we get prepare release to work21:45
@sonney2kblackburn, then get drunk and finish paper stuff :)21:45
blackburnsonney2k: 1.5 years of accident writing jmlr paper is crazy21:45
@sonney2kblackburn, well you could have had the paper about 1 year ago but you took the long route21:46
* sonney2k Zzzzz21:46
blackburnsonney2k: good thing is that it is fast now21:46
blackburnand being used for real21:46
blackburnby chemists though :D21:46
blackburnI expected ML/data analysis hah21:46
-!- zxtx [~zv@] has quit [Ping timeout: 240 seconds]22:06
-!- zxtx [] has joined #shogun22:07
shogun-buildbot_build #356 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Sergey Lisitsyn <>22:31
-!- zxtx [] has quit [Ping timeout: 264 seconds]23:01
shogun-buildbot_build #357 of rpm1 - libshogun is complete: Failure [failed test]  Build details are at  blamelist: Soeren Sonnenburg <>23:35
--- Log closed Sun Mar 17 00:00:13 2013