Open in new window / Try shogun cloud
--- Log opened Wed Feb 01 00:00:19 2012
-!- blackburn [~qdrgsm@] has left #shogun []00:27
-!- naywhayare [] has quit [Ping timeout: 240 seconds]04:13
-!- naywhayare [] has joined #shogun04:14
-!- dfrx [] has joined #shogun06:04
CIA-11shogun: Soeren Sonnenburg master * raaf479e / (src/shogun/features/Labels.h src/shogun/regression/KRR.h): fix minor documentation typos -
-!- sonne|work [~sonnenbu@] has left #shogun []10:51
-!- dfrx [] has left #shogun []11:37
-!- blackburn [~qdrgsm@] has joined #shogun13:58
-!- Ram108 [~amma@] has joined #shogun15:25
-!- Ram108 [~amma@] has quit [Ping timeout: 252 seconds]15:40
-!- Ram108 [~amma@] has joined #shogun15:50
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]16:09
-!- blackburn [~qdrgsm@] has joined #shogun16:19
blackburnsonney2k: is it true submission could take an year? :)17:20
-!- Ram108 [~amma@] has quit [Ping timeout: 252 seconds]17:32
-!- Ram108 [~amma@] has joined #shogun17:45
-!- blackburn [~qdrgsm@] has quit [Ping timeout: 240 seconds]18:08
-!- blackburn [~qdrgsm@] has joined #shogun18:16
-!- n4nd0 [] has joined #shogun19:09
n4nd0I am interested in gsoc 2012, I know that the organizations have not even been selected yet19:11
n4nd0but I would like to start getting a bit in shogun in case it is selected I can make a good proposal19:11
blackburnn4nd0: hi19:13
blackburnn4nd0: well gsoc was not announced too (yet)19:13
Ram108blackburn: hi do u mind if i seek some help?19:14
n4nd0blackburn, yeah, that's right :)19:15
blackburnRam108: sure19:15
n4nd0blackburn, but I guess it will run again this year19:15
blackburnn4nd0: yeah, some evidences are here already:
n4nd0blackburn, and anyway, if it doesn't, I would like to learn more about machine learning and this can be a good place to do it :)19:16
blackburnn4nd0: how familiar are you with ml?19:16
Ram108i ll catch u tomorrow......... my laptop ll run out of charge in a min and i dont have an outlet to charge19:16
Ram108lol bye have a great day19:17
n4nd0blackburn, I have taken some courses at the university about ml and applied it for the projects in the courses19:17
blackburnRam108: sure, feel free to write emails19:17
blackburnRam108: see you19:17
Ram108thanks :)19:17
Ram108bye :)19:17
blackburnn4nd0: ok so what is the most interesting subarea in ml for you? I'm just curious what to suggest19:18
-!- Ram108 [~amma@] has quit [Quit: Ex-Chat]19:18
n4nd0blackburn, so in one course we talked about kernel functions I found them really cool, I have not applied them to anything though19:19
n4nd0blackburn, I would like to learn more about them19:19
blackburnn4nd0: kernels, like in svms?19:19
n4nd0blackburn, yes19:20
blackburnhmm we have a lot of them :)19:20
n4nd0blackburn, but are they just for svms or also for more stuff?19:20
blackburnn4nd0: well kernels are used in svms very often19:20
n4nd0blackburn, all right, I believed they were also applied to other stuff19:21
blackburnthey are19:21
n4nd0blackburn, I like svms anyway so it is cool19:21
blackburnfor example I 'specialize' in dimensionality reduction19:21
blackburnand there are some kernel methods19:21
n4nd0ok, I see19:22
n4nd0and apart from the subarea in ml I am interested, what is interesting for the toolbox to have?19:22
n4nd0what are you working on?19:23
blackburnn4nd0: I'm author of all dimensionality reduction methods in shogun19:23
blackburnan *19:23
blackburnwell this year I'll probably be working on domain adaptation or multitask learning19:23
blackburndo you know anything about structured output learning?19:24
n4nd0sincerely not19:25
blackburnokk what about langs you are familiar with?19:25
blackburnn4nd0: that's the thing I'm mainly working at :)19:26
n4nd0blackburn, I work the most with C++ and matlab/octave19:26
n4nd0blackburn, but I can work in java and python also19:26
blackburnI see19:26
blackburnwell how do you want to start 'shoguning'? :)19:27
blackburnyou could write some examples or extend with new kernels or distances19:27
n4nd0blackburn, haha that's the big question19:27
blackburnor anything similar19:28
n4nd0blackburn, I don't really know, I guess almost anything could work19:28
n4nd0blackburn, that sounds good19:28
n4nd0maybe I can start with examples and later go more into extending some functionality / creating new stuff19:29
blackburnwith which tasks you was involved?19:29
blackburnin your studies I mean19:29
blackburnI would suggest you to reproduce it with shogun19:30
n4nd0I worked on face recognition19:30
blackburnoh nice19:30
blackburnon which datasets?19:30
n4nd0digit recognition and stuff like that19:30
blackburnok so you know mnist and ORL at least, right?19:30
n4nd0we implemented the paper of Viola and Jones19:31
n4nd0boosting with haar like features19:31
blackburnhmm I see19:31
blackburndo you have your own code for haar-like features you can integrate?19:31
n4nd0we did the full implementation in matlab19:32
n4nd0so I would have to look for it but yes, I must have that code19:32
blackburnok then my suggestion is:19:33
blackburnthere could be some script19:33
blackburnthat 1) downloads some dataset like olivetti or so19:33
blackburn2) constructs these features19:33
blackburn3) classifies with shogun's classifiers19:33
n4nd0compare different classifiers?19:35
blackburnwell could be nice too19:35
blackburnbut even with only one classifier it works well too19:35
n4nd0all right19:35
blackburnsounds like a plan?19:35
n4nd0sounds like a nice plan :)19:35
blackburnyou may contact me via blackburn91@gmail.com19:36
blackburnor here, in irc19:36
n4nd0ok, thank you19:36
n4nd0I will start then refreshing what I did in my lab to compute the features and later decide the classifier to use19:37
n4nd0is it ok to start out with my matlab code? or would it be better in C++ or any other?19:42
blackburnn4nd0: does it work with octave?19:43
blackburnpython is preferrable but no need to rewrite it if it is already working with octave19:43
n4nd0blackburn, I think it should work with octave, I don't think I used anything fancy, but I should try it out first19:44
n4nd0blackburn, so if the part that I have already written to compute the Haar features works well in octave, can it be used from python later?19:45
blackburnn4nd0: sure just decompose it to features.m and :)19:46
n4nd0blackburn,  cool19:46
blackburnn4nd0: are you from sweden? :)19:50
n4nd0blackburn, no, but I live in Sweden19:50
blackburnI see19:50
n4nd0blackburn, where are you from? Sweden or Scandinavia maybe?19:50
blackburnwhere are you from then (if it is not a secret)? :)19:50
blackburnn4nd0: well my name is Sergey :)19:51
blackburnI see19:51
blackburnso I'm russian19:51
n4nd0blackburn, was it just a good guess :P? or did you find something related to my nickname and Sweden?19:52
blackburnfrom cold country too hah19:52
n4nd0haha yes19:52
blackburnn4nd0: Username: :)19:52
n4nd0haha I see19:52
-!- f-x [~user@] has joined #shogun19:57
-!- axitkhurana [~akshit@] has joined #shogun19:57
-!- f-x [~user@] has quit [Remote host closed the connection]20:00
-!- axitkhurana [~akshit@] has quit [Quit: Leaving.]20:04
CIA-11shogun: Sergey Lisitsyn master * r0be4076 / src/shogun/classifier/ConjugateIndex.cpp : Conjugate index improvement -
-!- n4nd0 [] has quit [Read error: Operation timed out]20:51
-!- n4nd0 [] has joined #shogun20:52
n4nd0blackburn, so did you have any classifier in particular that could work well with the haar features?20:55
blackburnn4nd0: unfortunately I'm not very familiar with haar features20:55
n4nd0blackburn, in the implementation I have I use weak classifiers with boosting20:56
blackburnwell it is easy to try different ones20:56
blackburnwith shogun I mean20:56
n4nd0does the toolbox include boosting then?20:57
blackburnwe only have LPBoost currently20:58
n4nd0I am not familiare with it20:58
blackburnme neither20:59
CIA-11shogun: Soeren Sonnenburg master * r013c234 / (32 files in 9 dirs):21:02
CIA-11shogun: Merge pull request #361 from karlnapf/master21:02
CIA-11shogun: KRR/SVR and cross-validation/model-selection -
-!- n4nd0 [] has quit [Ping timeout: 272 seconds]21:04
-!- n4nd0 [] has joined #shogun21:16
CIA-11shogun: Sergey Lisitsyn master * re251619 / (2 files): Added confusion matrix routine -
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]22:20
shogun-buildbotbuild #122 of nightly_default is complete: Failure [failed test_1]  Build details are at
-!- n4nd0 [] has quit [Ping timeout: 240 seconds]23:59
--- Log closed Thu Feb 02 00:00:19 2012