Open in new window / Try shogun cloud
--- Log opened Fri May 16 00:00:41 2014
--- Day changed Fri May 16 2014
-!- pickle27f [~pickle27@] has quit [Quit: This computer has gone to sleep]00:00
-!- juliohm [~julio@unaffiliated/juliohm] has quit [Quit: Leaving]02:18
shogun-buildbotbuild #5 of nightly_fedora is complete: Failure [failed git]  Build details are at  blamelist: lambday <>, Soumyajit De <>03:00
-!- pickle27 [] has joined #shogun03:23
-!- kislay [~Abhijeet@] has quit [Read error: Connection reset by peer]03:30
-!- shogun-notifier- [] has joined #shogun03:44
shogun-notifier-shogun-web: Kevin Hughes :master * b7e6feb / static/css/styles.css:
shogun-notifier-shogun-web: comment out gsoc image in css while the box is about europython03:44
-!- PirosB3 [] has quit [Quit: PirosB3]04:01
-!- soumyaC [uid15286@gateway/web/] has quit [Quit: Connection closed for inactivity]04:16
shogun-buildbotbuild #803 of nightly_default is complete: Failure [failed notebooks]  Build details are at  blamelist: lambday <>, Soumyajit De <>04:34
-!- pickle27 [] has quit [Remote host closed the connection]06:30
-!- shogun-notifier- [] has quit [Quit: transmission timeout]06:44
-!- kislay [~Abhijeet@] has joined #shogun07:19
-!- kislay [~Abhijeet@] has quit [Read error: Connection reset by peer]08:00
-!- jiaolong [9e6d09e4@gateway/web/freenode/ip.] has joined #shogun09:21
-!- soumyaC [uid15286@gateway/web/] has joined #shogun10:29
-!- zxtx [] has quit [Ping timeout: 258 seconds]11:25
-!- zxtx [] has joined #shogun11:43
-!- PirosB3 [] has joined #shogun11:59
-!- zxtx [] has quit [Ping timeout: 258 seconds]12:01
-!- kislay [~Abhijeet@] has joined #shogun12:15
-!- zxtx [] has joined #shogun12:17
-!- PirosB3 [] has quit [Quit: PirosB3]12:18
-!- PirosB3 [] has joined #shogun12:48
-!- kislay [~Abhijeet@] has quit [Quit: Leaving]13:32
-!- pickle27 [] has joined #shogun14:44
-!- pickle27 [] has quit [Remote host closed the connection]14:49
-!- khalednasr [~k.nasr92@] has joined #shogun15:14
@besser82thoralf, hey!15:42
-!- khalednasr [~k.nasr92@] has quit [Quit: Leaving]15:43
--- Log closed Fri May 16 15:46:00 2014
--- Log opened Fri May 16 15:46:09 2014
-!- shogun-toolbox [] has joined #shogun15:46
-!- Irssi: #shogun: Total of 14 nicks [3 ops, 0 halfops, 0 voices, 11 normal]15:46
-!- Irssi: Join to #shogun was synced in 6 secs15:46
-!- naywhayare [] has joined #shogun15:46
-!- pickle27f [~pickle27@] has joined #shogun15:53
-!- kislay [~Abhijeet@] has joined #shogun16:28
sonne|workpickle27f: can you upload a fixed website?16:33
pickle27fI pushed the fix for the css16:34
pickle27fI don't know what text is supposed to be there16:34
pickle27fbesser82, you around?16:36
pickle27fkislay, has a PR that adds some stuff to cmake16:36
@besser82pickle27f, yo!16:37
@besser82pickle27f, which one?16:37
@besser82pickle27f, k, thx! ^^16:37
pickle27fnp thanks for taking a look!16:37
@besser82pickle27f, np  =)16:37
pickle27fsonne|work, what should the text be?16:40
sonne|workpickle27f: there is some text in the .html16:40
sonne|workjust flickering and going away16:40
sonne|workI guess that but fernando would know16:40
pickle27fits loading fine for me16:41
pickle27fsonne|work, here is what I see16:43
@besser82pickle27f, just commented on the PR16:44
pickle27fawesome thanks!16:46
pickle27fI'll have kislay address your comments and then anyone can merge it16:46
@besser82pickle27f, allrighty! thx!  Lemme know, when it's ready.  I'll have another look then && merge16:47
@besser82pickle27f, btw. you did do anything related to NLP?16:47
pickle27fcan't say I have16:48
@besser82pickle27f, kk16:48
@besser82pickle27f, I'm just looking for someone, who can show me how to conv LDA-model into lLDA / sLDA...16:48
@besser82and clarify the most signficant differences...16:49
@besser82read lots of papers but I'm not smarter than before  :(16:49
@besser82just eye-sore :(16:49
sonne|workpickle27f: c'mon look at the source code16:49
@besser82sonne|work, hey!16:50
@besser82sonne|work, can you tell me the significant difference between lLDA and sLDA?16:50
pickle27fsonne|work, pooh how did I mess that before ...16:50
@besser82and possibly MedLDA?16:50
pickle27fI'll have to fix that when I get home later slash maybe tomorrow but yeah I'll push something up16:51
kislaypickle27f, see if it's okay. I have removed the system include for opencv from CMakeLists.txt17:22
pickle27fkislay, looks good and it still works right? - did you try a clean build?17:23
kislaypickle27f, yeah! np with that!17:24
pickle27fbesser82, I think we are good now!17:25
pickle27fjust waiting on travis17:25
@besser82pickle27f, lgtm  ;)  will merge after travis run is clean17:34
@besser82kislay, could you remove the mentioned cluttering whitespace, too?17:36
@besser82kislay, plz...17:36
kislaybesser82, sure sure.17:36
@besser82kislay, fine  =)  let's w8 on travis then17:36
-!- thoralf [] has quit [Quit: Konversation terminated!]17:38
-!- pickle27f [~pickle27@] has quit [Quit: This computer has gone to sleep]18:07
-!- pickle27f [~pickle27@] has joined #shogun18:23
-!- pickle27f [~pickle27@] has quit [Quit: This computer has gone to sleep]18:51
-!- shogun-notifier- [] has joined #shogun18:58
shogun-notifier-shogun: Abhijeet :develop * ece3607 / CMakeLists.txt,tests/integration/opencv/opencv_test.cpp:
shogun-notifier-shogun: added opencv as the optional dependency and a basic test in tests/integration/opencv.18:58
shogun-notifier-shogun: Abhijeet :develop * 7968225 / CMakeLists.txt:
shogun-notifier-shogun: added flags for opencv integration in CMakeLists.txt.18:58
shogun-notifier-shogun: Abhijeet :develop * 2dc8e4d / CMakeLists.txt:
shogun-notifier-shogun: shortened the flag name for OpenCV integration18:58
shogun-notifier-shogun: Abhijeet :develop * 1210365 / tests/integration/opencv/opencv_test.cpp:
shogun-notifier-shogun: opencv integration test cleaned_.18:58
shogun-notifier-shogun: Abhijeet :develop * 177c519 / CMakeLists.txt:
shogun-notifier-shogun: removed system include for OpenCV from CMakeLists.txt18:58
shogun-notifier-shogun: Abhijeet :develop * 5aa2b89 / CMakeLists.txt:
shogun-notifier-shogun: cleaned whitespaces.18:58
shogun-notifier-shogun: Bj?rn Esser :develop * 96b815f / CMakeLists.txt,tests/integration/opencv/opencv_test.cpp:
shogun-notifier-shogun: Merge pull request #2221 from kislayabhi/develop18:58
@besser82kislay, merged ^^18:59
kislaybesser82, yeah! :)18:59
shogun-buildbotbuild #2946 of deb1 - libshogun is complete: Success [build successful]  Build details are at
-!- pickle27f [~pickle27@] has joined #shogun19:34
pickle27fthanks for the merge!19:34
pickle27fwho do kislay and I talk to about getting an OpenCV build bot? slash can we just add to Travis?19:35
kislayyeah. wo do we talk :)19:35
pickle27fI think we can add to Travis our selves if you want to look into that - you'll need to edit the travis.yml file in the repo19:40
pickle27fI don't know about the other bots19:41
pickle27fslash I don't even know how or who configures those19:41
-!- thoralf [~thoralf@] has joined #shogun20:19
-!- PirosB3 [] has quit [Quit: PirosB3]20:45
-!- thoralf [~thoralf@] has quit [Read error: Connection reset by peer]20:45
-!- besser82 [quassel@fedora/besser82] has left #shogun ["Verlassend"]20:45
-!- besser82 [quassel@fedora/besser82] has joined #shogun20:46
-!- mode/#shogun [+o besser82] by ChanServ20:46
-!- thoralf [~thoralf@] has joined #shogun20:46
-!- kislay [~Abhijeet@] has quit [Ping timeout: 265 seconds]20:52
-!- thoralf [~thoralf@] has quit [Quit: Konversation terminated!]20:54
shogun-buildbotbuild #301 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at  blamelist: Abhijeet <>, Bj?rn Esser <>21:35
-!- PirosB3 [] has joined #shogun21:57
-!- shogun-notifier- [] has quit [Quit: transmission timeout]21:58
-!- pickle27f [~pickle27@] has quit [Quit: This computer has gone to sleep]22:04
-!- pickle27f [~pickle27@] has joined #shogun22:05
-!- PirosB3 [] has quit [Quit: PirosB3]22:32
-!- abinash_panda [0e8b2996@gateway/web/freenode/ip.] has joined #shogun23:05
-!- pickle27f [~pickle27@] has quit [Quit: This computer has gone to sleep]23:07
-!- pickle27f [~pickle27@] has joined #shogun23:10
-!- thoralf [] has joined #shogun23:11
abinash_pandathoralf: Hi !23:11
thoralfHey :)23:11
abinash_pandathoralf: So, I was going through the implementation of CMulticlassModel for getting to know more about the implementation of get_joint_feature_vector method.23:14
abinash_pandathoralf: Do the implementation of get_joint_feature_vector would be somewhat like this is CMultilabelModel ?23:16
thoralfabinash_panda: Of course - because get_joint_feature_vector() is very central in defining a structured model.23:17
thoralfWe're interested in very efficient implementation, so suggestions are welcome.23:18
abinash_pandathoralf: Ok.23:18
thoralfabinash_panda: Any ideas?23:19
abinash_pandathoralf: I was going through the paper for getting to know more about joint feature map.23:20
abinash_pandathoralf: It has some special cases of structured output learning as examples.23:21
thoralfYeah, this paper is great.23:21
abinash_pandathoralf: So, I what I think that the implementation of get_joint_feature_vector() would be the tensor product of input and output labels. Right ?23:22
thoralftensor product of label-vector and feature-vector, but yes.23:23
abinash_pandathoralf: Yeah. I meant the same.23:23
thoralfOkay :)23:24
abinash_pandathoralf: Now, regarding the argmax function.23:24
abinash_pandathoralf: I am still confused. Could you please help me out23:24
thoralfShoot. :)23:25
abinash_pandathoralf: You have mentioned in the mail, for the initial implementation we can treat the labels independently.23:28
abinash_pandathoralf: I am still unable to understand its implementation.23:30
thoralfHmm.  Okay.23:34
thoralfImagine you have one input vector and want to predict a binary label vector.23:34
abinash_pandathoralf: Ok23:34
thoralfLet's say 3 labels, like (0,0,1)23:34
thoralfFor one input x you have 8 possible outputs (000), (001), (010), ..., (111).  You have a scoring function f(x,y) which tells you a score for (input, possible outputs)... and you want to find the "best" output labeling.23:36
thoralfbest means, highest score.23:36
thoralfBut you don't want to "brute force" them.23:36
thoralfYou want some "clever" decoding.23:37
thoralfSomething that is cheaper than trying all 2^n combinations.23:37
thoralfAnd that has to be implemented in argmax.23:37
thoralf"Find the best labeling"23:37
thoralfDo you have an idea how to implement this multilabel stuff with binary classifiers?23:38
abinash_pandaWe can use binary relevance method, training one binary classifier for each label. Right?23:40
thoralfAnd then we ask the i-th classifier for the i-th label.23:42
thoralfThe cost is linear in the number of labels, which is totally fine. :)23:43
abinash_pandaOk. So what I think that we can implement the same in the argmax function.23:43
thoralfBut instead of training n classifiers, we train one classifier over the joint-feature-space.23:46
thoralfYou know what I mean?23:46
-!- pickle27f [~pickle27@] has quit [Quit: This computer has gone to sleep]23:46
abinash_pandaYeah. We would be training only one classifier.23:47
thoralfThen you probably know how to transform label-vector and feature-vector to form the joint-feature-vector?23:48
thoralfSorry for asking dumb questions.  I'm trying cover all possible problems. ;)23:49
abinash_pandaIn the argmax method, for N labels we can go for computing the argmax N number of times and would return only those labels for whose the output is 1.23:51
abinash_pandaAm I able to make myself clear23:52
thoralfYeah, I think so.23:53
abinash_pandaOk. :)23:53
thoralfIt's even easier: Output only labels which decision value is positive.23:53
abinash_pandaYeah, thats better :)23:53
thoralf">= 0" for true, "< 0" for false.23:53
thoralfBtw., if you like to, you can also implement a multilabel class which uses N independent binary SVMs.  This sounds strange, but you can learn them all in parallel. :)23:55
abinash_pandaYeah, sure. I would try to implement it.23:55
thoralfYou get parallel training on many cores for free.23:56
thoralfWho many cores does your biggest computer have? ;)23:56
* thoralf was able to train multilabel with 1000 classes on a weekend.23:57
abinash_pandaCurrently, my computer have 4 cores ;)23:57
abinash_pandathoralf: On how many cores you have implemented this multilabel classification?23:58
thoralfWell, the computer had 8 cores (with hyperthreading it was 16 threads)23:59
thoralfSo scaling by factor 16 for free.23:59
thoralfNo worrying about how to parallelize SVM solvers.23:59
--- Log closed Sat May 17 00:00:02 2014