Open in new window / Try shogun cloud
--- Log opened Sun Jul 10 00:00:59 2011
* blackburn got insight of various abracadabras DGEMM DPOTRI00:02
CIA-32shogun: Sergey Lisitsyn master * r67826b5 / (src/libshogun/lib/arpack.cpp src/libshogun/lib/arpack.h): Added Cholesky factorization for Arpack wrapper -
CIA-32shogun: Sergey Lisitsyn master * r46aed74 / (2 files): Modified preprocessors using arpack for new parameter -
CIA-32shogun: Sergey Lisitsyn master * r419387b / (2 files in 2 dirs): Removed junk -
blackburnunfortunately lle produces indefinite symmetrical matrix :(00:38
-!- f-x [~user@] has quit [Remote host closed the connection]01:21
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]01:56
-!- in3xes_ [~in3xes@] has joined #shogun05:25
-!- in3xes [~in3xes@] has quit [Ping timeout: 240 seconds]05:28
-!- in3xes_ is now known as in3xes05:34
-!- [1]warpy [] has joined #shogun06:50
[1]warpyhey guys anyone alive06:56
[1]warpyi was wondering how to download 0.1106:56
-!- [1]warpy [] has quit [Quit: HydraIRC -> <- In tests, 0x09 out of 0x0A l33t h4x0rz prefer it :)]07:34
-!- in3xes_ [~in3xes@] has joined #shogun09:02
-!- in3xes_ [~in3xes@] has quit [Remote host closed the connection]09:29
-!- shayan_ [~shayan@] has joined #shogun12:39
-!- shayan_ is now known as shayan13:02
-!- shayan [~shayan@] has quit [Quit: Leaving]13:25
-!- f-x [~user@] has joined #shogun14:32
-!- blackburn [~blackburn@] has joined #shogun18:32
-!- f-x [~user@] has quit [Remote host closed the connection]18:33
blackburngot kindle dx! now can read papers without any hurt ;)18:42
-!- blackburn [~blackburn@] has quit [Read error: No route to host]19:32
-!- blackburn [~blackburn@] has joined #shogun19:32
-!- f-x [~user@] has joined #shogun20:35
-!- f-x [~user@] has left #shogun []20:35
-!- f-x [~user@] has joined #shogun20:35
@sonney2kblackburn, without any hurt?22:22
blackburnsonney2k: yeah reading using laptop suckz, aint it?22:22
@sonney2kblackburn, yeah but is this A4 size / I mean don't you need to zoom the pdf to read anything?22:23
blackburnsonney2k: most papers have large margins - so easy readable22:24
CIA-32shogun: Baozeng Ding master * rb0c3848 / (4 files): add distribution java examples -
CIA-32shogun: Soeren Sonnenburg master * r268bccc / (4 files):22:24
CIA-32shogun: Merge pull request #185 from sploving/master22:24
CIA-32shogun: add remaining distribution java examples -
@sonney2kblackburn, ok...22:24
blackburne.g. elements of statistical learning looks pretty without zoom22:24
blackburnas is22:24
@sonney2khmmhh... looks like things are progressing22:24
blackburnwhich things?22:24
@sonney2kI should probably start test shogun compilation on osx / cygwin22:25
blackburnI'm now trying to get causes of LLE strange results sometimes22:25
-!- blackburn [~blackburn@] has left #shogun []22:28
-!- blackburn [~blackburn@] has joined #shogun22:28
@sonney2kblackburn, that is actually also a good candidate for shogun
@sonney2kall the possible multiclass svms...22:34
blackburnsonney2k: where to get so much consistent developers? ;)22:34
@sonney2kdid you mean where to get enough developers to support all the many features?22:35
blackburnehh sth like that22:36
blackburnwell where to get any developers? :)22:37
@sonney2kblackburn, well year over year there must be another GSoC :)22:37
blackburnyeah I hope to get into it ;)22:38
* sonney2k looks at the mc svm code stuff22:46
@sonney2kthey have sse4.1 tuned dot products in there22:46
@sonney2kblackburn, ^ maybe we should benchmark and borrow their code :)22:47
blackburnwell if it isn't a task for a 1.0 - I could do it some day22:47
blackburnI should be 'vacant' for ~10 hours per week for some shoguning after gsoc22:48
blackburnsonney2k: do we have some up-to-date multiclass svm now?22:50
@sonney2kblackburn, we have GMNPSVM22:50
blackburnit tells me almost nothing :)22:50
@sonney2kand some one-vs-one via LibSVM22:50
@sonney2kbut that's it22:51
blackburnsonney2k: is there any not one-vs-one for SVMs?22:51
@sonney2kthe one in the package were true MC SVMs22:51
blackburnhow it works?22:51
@sonney2kand one-vs-rest one has to do manually currently22:51
@sonney2kblackburn, it is a true multiclass svm22:51
blackburnI wonder how it can be formulated22:52
@sonney2keasy: you want f(x,y) for the true class y to be greater than f(x,y') for y' not the true class22:52
@sonney2kand f is a linear function again - but rest is the same as svm22:53
@sonney2koptimization problem is harder though22:53
@sonney2kblackburn, I mean in SVM you usually just say points should be on the right side of the hyperplane22:57
@sonney2know you say output of the function for the true label should be greater than output of the function for all wrong labels plus some margin22:57
@sonney2kso you have many more constraints22:57
blackburnah seems I understood22:58
blackburn:( got dataset where LLE produces different results for different solvers23:00
blackburndebug session #N23:01
blackburnnot stable problem I guess23:02
blackburnsonney2k: I've finally changed everything to vim :) because of you!23:06
@sonney2kI am sorry23:06
-!- f-x [~user@] has quit [Remote host closed the connection]23:06
blackburnsonney2k: well it is cool, I understand, I guess you (with some experience and skillz) use it almost magically23:07
@sonney2kI don't even use all the features.23:09
@sonney2kbut I got very used to it now...23:09
@sonney2kdo you know whether eclipse has a vim plugin?23:09
blackburnsonney2k: yes, eclipse have one, I know23:10
@sonney2kvery good23:10
@sonney2kif I ever have to use eclipse I can at least feel home :)23:11
blackburnfor java I consider ide ;)23:11
blackburnsonney2k: is it *normal* in ML that some methods can fail if say k is bad?23:23
@sonney2kblackburn, welcome to numerics23:23
blackburne.g. LLE sometimes produces examples lying on diagonal line23:24
blackburnnot that beauty I showed you yesterday23:24
@sonney2ksome methods are very sensitive23:24
@sonney2ksome very robust23:24
blackburnI'll check it for sure - but looks like it is not robust23:24
blackburnsparse matrix eigenproblem doesn't look robust for me, am I right?23:25
blackburnvoices in my head tells me it is not, sonney2k, do you have any?23:26
@sonney2kblackburn, I am not an expert in this subject but I would guess so :)23:28
blackburnme too, but voices.. :D23:28
blackburnscikit-learn started as GSoC project?!23:33
@sonney2kblackburn, I am currently checking java examples23:40
@sonney2kIsomap preprocessor = new Isomap()23:40
@sonney2kgives an error23:40
@sonney2kdoes it not exist or has it a different name?23:40
blackburnsonney2k: isomap is abstract class23:40
blackburnClassicIsomap or LandmarkIsomap23:40
blackburnbetter both :D23:41
@sonney2kyes that works23:41
@sonney2kblackburn, btw, did you touch / polish kernelpca / pca?23:41
blackburnnot yet23:41
blackburnarpack issues..23:42
@sonney2kI see23:42
@sonney2kblackburn, good luck :)23:42
blackburnnow writing to jake vanderplas, he have implemented lle in scikits23:42
blackburnI really warned with some unstability23:43
CIA-32shogun: Soeren Sonnenburg master * ree98378 / (3 files in 2 dirs): fix java examples -
CIA-32shogun: Soeren Sonnenburg Siddharthk-master * r8ed3110 / (4 files in 3 dirs): Merge branch 'master' of into Siddharthk-master -
--- Log closed Mon Jul 11 00:00:05 2011