Open in new window / Try shogun cloud
--- Log opened Sun Jul 24 00:00:05 2011
@sonney2kgtg need to catch the train ;-)00:19
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]00:23
-!- in3xes1 [~in3xes@] has quit [Ping timeout: 240 seconds]00:39
-!- in3xes [~in3xes@] has joined #shogun00:57
-!- in3xes [~in3xes@] has quit [Ping timeout: 240 seconds]01:44
-!- in3xes [~in3xes@] has joined #shogun01:44
-!- f-x [~user@] has joined #shogun02:09
-!- in3xes [~in3xes@] has quit [Ping timeout: 252 seconds]02:34
-!- in3xes [~in3xes@] has joined #shogun02:34
-!- in3xes [~in3xes@] has quit [Ping timeout: 258 seconds]02:57
-!- in3xes [~in3xes@] has joined #shogun03:02
-!- in3xes [~in3xes@] has quit [Ping timeout: 240 seconds]03:06
-!- in3xes [~in3xes@] has joined #shogun03:07
-!- f-x [~user@] has quit [Ping timeout: 260 seconds]04:24
-!- in3xes [~in3xes@] has quit [Ping timeout: 240 seconds]05:14
-!- in3xes [~in3xes@] has joined #shogun05:21
-!- in3xes_ [~in3xes@] has joined #shogun05:42
-!- in3xes [~in3xes@] has quit [Read error: Operation timed out]05:44
-!- in3xes_ [~in3xes@] has quit [Ping timeout: 258 seconds]05:48
-!- in3xes_ [~in3xes@] has joined #shogun05:49
-!- in3xes_ [~in3xes@] has quit [Ping timeout: 276 seconds]05:57
-!- in3xes_ [~in3xes@] has joined #shogun05:57
-!- f-x [~user@] has joined #shogun06:12
-!- f-x [~user@] has quit [Ping timeout: 260 seconds]06:17
-!- in3xes_ [~in3xes@] has quit [Ping timeout: 258 seconds]06:36
-!- in3xes [~in3xes@] has joined #shogun06:53
-!- in3xes [~in3xes@] has quit [Ping timeout: 264 seconds]07:25
-!- in3xes [~in3xes@] has joined #shogun08:00
-!- gsomix [~gsomix@] has joined #shogun08:42
-!- f-x [~user@] has joined #shogun08:50
-!- f-x` [~user@] has joined #shogun08:51
-!- in3xes [~in3xes@] has quit [Ping timeout: 252 seconds]09:05
-!- in3xes [~in3xes@] has joined #shogun09:43
-!- in3xes [~in3xes@] has quit [Ping timeout: 276 seconds]09:50
-!- in3xes [~in3xes@] has joined #shogun09:55
-!- in3xes [~in3xes@] has quit [Ping timeout: 258 seconds]10:00
-!- in3xes [~in3xes@] has joined #shogun10:08
-!- in3xes [~in3xes@] has quit [Ping timeout: 264 seconds]10:15
-!- f-x` [~user@] has quit [Ping timeout: 260 seconds]10:51
-!- f-x [~user@] has quit [Ping timeout: 260 seconds]10:51
-!- f-x [~user@] has joined #shogun11:00
gsomix40C. It's hot. =___=11:10
-!- blackburn [~blackburn@] has joined #shogun11:28
blackburnhooray new keyboard11:31
blackburnnot key6oard!11:36
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]11:39
CIA-87shogun: Alesis Novik master * r8af741f / (3 files in 2 dirs): Initial SMEM implementation. With bugs, fix will follow. -
CIA-87shogun: Alesis Novik master * r104f0f0 / src/shogun/clustering/GMM.cpp : Some memory leaks and general fixes. Mostly works now. -
CIA-87shogun: Alesis Novik master * rfc4baa4 / (2 files): Added SMEM GMM examples -
CIA-87shogun: Soeren Sonnenburg master * ra6e9a4a / (5 files in 3 dirs):11:46
CIA-87shogun: Merge pull request #229 from alesis/gmm11:46
CIA-87shogun: SMEM for GMM -
-!- blackburn [~blackburn@] has joined #shogun11:46
@sonney2k18 C - freezing11:47
blackburnsonney2k: you could warm up berlin with your macbook11:48
gsomixWe are using air conditioning at home. But the outside is very bad to work.11:52
@sonney2kblackburn, did you manage to look at the java examples?11:53
blackburnsonney2k: going to test all of them today11:53
@sonney2kand gsomix despite the wheather how is wresting with C5.0 going?11:54
gsomixI'm working on some data structures now. Blackburn and I have no idea how to integrate c5.0. =___=11:57
CIA-87shogun: Soeren Sonnenburg master * rcdc1dc6 / README : mention modular interfaces at the top and static ones below -
@sonney2kgsomix, I see11:58
@sonney2kgsomix, do you perhaps have a C5.0 branch somewhere?11:59
-!- gsomix [~gsomix@] has quit [Read error: Connection reset by peer]12:02
-!- gsomix [~gsomix@] has joined #shogun12:04
gsomixsonney2k, only on my hard disk. If I understand the phrase.12:04
@sonney2kgsomix, mind pushing it to github such that I can have a look?12:05
blackburnI guess he have nothing to push yet12:05
blackburnwith C5.0 I mean12:05
gsomixOnly source code of C5.0. No more.12:07
gsomixsonney2k, send the code of C5.0 to github? Specify.12:13
@sonney2kgsomix, I meant you have a local branch w/ c5.0 that you could just push - but it seems not.12:19
@sonney2kso forget it12:19
blackburnsonney2k: ok to merge?
@sonney2kblackburn, you have the OK from the author or where does this come from?13:14
blackburnsonney2k: he rewrote it ;)13:14
blackburngsomix: did you?13:15
blackburnsonney2k: so I guess no license issues13:15
@sonney2ktrue then13:15
@sonney2klots of work ...13:15
blackburnit could speed up isomap when I use it for dijkstra13:16
blackburnsonney2k: could you merge it? I'm messed with some QR now13:16
blackburnsonney2k: ah, nevermind, I'll do13:18
gsomixsonney2k, all ok? could you merge it?13:39
-!- blackburn [~blackburn@] has quit [Ping timeout: 255 seconds]13:42
-!- in3xes [~in3xes@] has joined #shogun13:45
-!- in3xes [~in3xes@] has quit [Quit: Leaving]14:39
-!- blackburn [~blackburn@] has joined #shogun14:45
@sonney2kI did add some comments15:00
blackburnsonney2k: they all are fixed15:01
CIA-87shogun: Evgeniy Andreev master * r194468a / (2 files): Rewritten and added CFibonacciHeap class. -
CIA-87shogun: Evgeniy Andreev master * rbb39c2f / (2 files): Some fixes. -
CIA-87shogun: Evgeniy Andreev master * r1ba5573 / (2 files): Some fixes. -
CIA-87shogun: Soeren Sonnenburg master * r96ba02b / (2 files):15:06
CIA-87shogun: Merge pull request #230 from gsomix/master15:06
CIA-87shogun: Implemented Fibonacci heap. -
gsomixYahoo. My first real code in the project. :)15:08
blackburnokay let the great java test begin15:08
@sonney2kgsomix, thanks15:15
blackburnPCA differs15:47
blackburnokay LoadMatrix is definitely wrong15:48
-!- f-x [~user@] has quit [Ping timeout: 260 seconds]15:53
blackburnhmm no, println produces flat matrix15:58
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]16:40
-!- f-x [~user@] has joined #shogun17:00
-!- blackburn [~blackburn@] has joined #shogun17:23
-!- in3xes [~in3xes@] has joined #shogun18:18
-!- blackburn [~blackburn@] has quit [Ping timeout: 255 seconds]18:32
-!- f-x [~user@] has quit [Ping timeout: 260 seconds]18:52
-!- in3xes [~in3xes@] has quit [Read error: Connection reset by peer]19:41
-!- in3xes [~in3xes@] has joined #shogun19:57
-!- blackburn [~blackburn@] has joined #shogun20:08
-!- [1]warpy [] has joined #shogun20:16
blackburnsonney2k: I got rid of wi-fi lost connection. the solution is to use another kernel :D20:22
CIA-87shogun: Alesis Novik master * r81c0c6d / examples/undocumented/python_modular/graphical/ : Small example changes -
CIA-87shogun: Alesis Novik master * r220e8bf / (2 files in 2 dirs): Bug fixes for Gaussian and example -
CIA-87shogun: Alesis Novik master * r74f3665 / examples/undocumented/python_modular/graphical/ : A more illustrative example of SMEMs effect. -
CIA-87shogun: Soeren Sonnenburg master * ra75874e / (3 files in 2 dirs):20:35
CIA-87shogun: Merge pull request #231 from alesis/gmm20:35
CIA-87shogun: Important EM bug fix and nice examples -
@sonney2kblackburn, congrats20:36
@sonney2kblackburn, is java loadmatrix working or not?20:38
blackburnsonney2k: it is ok, I was wrong20:39
-!- serialhex_ is now known as serialhex20:39
blackburnwhen printing20:39
blackburnit produces strange output20:39
@sonney2kyou mean toString()20:40
blackburnprintln, yes21:08
blackburnsonney2k: difficult to compare randomized algo like landmarkisomap :D21:09
@sonney2kblackburn, why?21:09
blackburnhow can I fix landmarks if they are all random21:09
@sonney2kblackburn, there is no randomness in this world21:11
@sonney2kjust init the rnd generator with the same seed21:11
blackburnsonney2k: I just ran all of examples without > /dev/null -- looks pathetic21:12
@sonney2kblackburn, which means?21:13
blackburnmany asserts fail21:14
blackburnwill check regression now21:14
@sonney2kblackburn, so lets concentrate on simple examples first21:14
blackburnhm okay21:14
@sonney2kyou should check if you can get/set features and result is the OK21:14
* sonney2k was afraid of that21:16
blackburnsonney2k: ufff just take a look at features_simple_modular.java21:17
@sonney2kblackburn, yes?21:19
blackburnaren't you getting confused with adding a matrix asList?21:20
@sonney2kblackburn, you mean this Arrays.asList(A) ?21:21
@sonney2klooks weird indeed21:21
blackburnsonney2k: yes21:21
blackburnuh oh21:22
@sonney2kbetter use an ArrayList of DoubleMatrix21:22
blackburnI guess Arrays.asList(A) doesn't produce doublematrix21:23
blackburnokay I'll just try to set/get features21:25
blackburnsonney2k: any way to map from "get_feature_matrix" style to "getFeatureMatrix" style automagically?21:29
@sonney2kblackburn, no21:30
@sonney2kI don't thinks os21:30
blackburnsonney2k:ok, set/get RealFeatures - works21:30
@sonney2kI know this is troublesome for pure java people but at least for me having to program in python java etc etc - I will die21:31
@sonney2kok - at least21:31
@sonney2kblackburn, next level would be a kernel matrix / distance matrix21:31
blackburnI guess I have to rewrite simple features test21:32
blackburnand then distance,yes21:32
blackburnsonney2k: is it ok to use DoubleMatrix everywhere?21:39
@sonney2kblackburn, jblas simply doesn't support other data types than float/double21:42
@sonney2kso it is the only way to 'emulate' other types - so yes21:42
blackburnso, e.g. new bytefeatures(doublematrix) is ok?21:44
@sonney2kother things will only work with ujmp21:51
@sonney2ks/other things/native types/21:51
blackburnI see21:52
@sonney2kblackburn, so do kernels work?22:09
blackburnnot tested yet22:09
CIA-87shogun: Sergey Lisitsyn master * r4a479cc / (3 files in 2 dirs): Merge branch 'master' of -
CIA-87shogun: Sergey Lisitsyn master * r32df164 / (4 files): Replaced simple features java test with one -
CIA-87shogun: Sergey Lisitsyn master * rc1bbc5d / (3 files): Split isomap java example -
CIA-87shogun: Sergey Lisitsyn master * ra844225 / examples/undocumented/java_modular/ : Removed unnecessary println at landmark isomap java example -
blackburnsonney2k: looks like it is ok with gaussian kernel22:23
@sonney2khurray - at least sth22:24
@sonney2kblackburn, next level classifier svm + kernel22:25
blackburnnot, sth is wrong22:26
blackburnI replaced data with22:28
blackburn2.0 2.022:28
blackburn3.0 3.022:28
blackburn2.5 2.522:28
blackburnin python got22:28
blackburn[[ 1.          0.82505297  0.38230427]22:28
blackburn [ 0.82505297  1.          0.46336937]22:28
blackburn [ 0.38230427  0.46336937  1.        ]]22:28
blackburnin java22:28
blackburn[1.0, 0.3823042728920807, 0.3823042728920807;22:28
blackburn0.3823042728920807, 1.0, 0.6807123983233854;22:28
blackburn0.3823042728920807, 0.6807123983233854, 1.0]22:28
-!- blackburn was kicked from #shogun by bettyboo [flood]22:28
-!- blackburn [~blackburn@] has joined #shogun22:29
blackburnsonney2k: ^22:29
@sonney2kblackburn, please check if features are indeed the same after you got the kernel matrix22:31
@sonney2ksth like22:31
blackburnsonney2k: trouble is get_lhs return features..22:40
@sonney2kblackburn, ahh yes true - that is troublesome under python22:41
@sonney2kthen just check if the features you create still contain the values22:41
blackburnsonney2k: hmm it is ok22:43
blackburnI guess 'reshape' works different in java and python22:43
@sonney2kboth in java and python?22:43
@sonney2kwhat for?22:43
blackburnload uses it22:44
blackburnyes, different way of reshape22:45
blackburn*very* strange way of reshape in java22:45
blackburn2.0 2.022:45
blackburn3.0 3.022:45
blackburn2.5 2.522:45
blackburnin python:22:45
blackburn[[ 2.   2.   3. ]22:45
blackburn [ 3.   2.5  2.5]]22:45
blackburnin java22:45
blackburn[2.0, 2.5, 3.0; 3.0, 2.0, 2.5]22:46
@sonney2kwhat is the input?22:46
blackburn2.0 2.022:47
blackburn3.0 3.022:47
blackburn2.5 2.522:47
blackburnI modified fm_train_real.dat22:47
@sonney2khow can that be reshaped?22:47
@sonney2kit is a matrix already?!22:47
@sonney2kshouldn't this be transposed?22:47
blackburnno, it reads not matrix but vector I guess22:48
blackburnjust take a look how it works in python and java22:48 and or so22:49
@sonney2kblackburn, looks like the python thing is totally broken23:00
@sonney2kit should be numpy.loadtxt('../data/fm_train_real.dat').T23:00
@sonney2kinstead of whatever is done in load_numbers23:00
@sonney2kI now understand why this happened23:02
@sonney2kat some point we had data in matlab compatible format23:02
@sonney2kascii format23:02
@sonney2kthat means one example == one column23:03
@sonney2kbut this is stupid for ascii23:03
@sonney2kso we transposed this23:03
@sonney2kbut the python function was never changed obviously23:03
blackburnsonney2k: so, what should I/you do?23:07
@sonney2kblackburn, please go on23:08
blackburnsonney2k: with java?23:08
CIA-87shogun: Soeren Sonnenburg master * rd8c70f9 / examples/undocumented/python_modular/tools/ :23:09
CIA-87shogun: fix loading of matrices (we now have one example per line - no longer23:09
CIA-87shogun:  one per column) -
CIA-87shogun: Soeren Sonnenburg master * r020d909 / (7 files): Merge branch 'master' of -
blackburnthe most lame weekly report ever23:18
blackburnI should raise my efficiency, I feel sorry for such low productivity23:18
-!- gsomix [~gsomix@] has quit [Ping timeout: 260 seconds]23:18
@sonney2kblackburn, do kernels now work as they should?23:28
blackburnsonney2k: will test in a min23:28
CIA-87shogun: Sergey Lisitsyn master * r480b717 / examples/undocumented/python_modular/tools/ : Fixed -
blackburnsonney2k: ok, different matrices again23:31
@sonney2kdifferent in which respect?23:31
blackburn.dat is23:32
blackburn2.0 2.023:32
blackburn3.0 3.023:32
blackburn2.5 2.523:32
blackburn[[ 2.   3.   2.5]23:32
blackburn [ 2.   3.   2.5]]23:32
blackburn[2.0, 2.5, 3.0; 3.0, 2.0, 2.5]23:32
blackburnsonney2k: java one is wrong I guess?23:36
@sonney2kblackburn, my guess is you need to transpose not reshape23:38
* sonney2k wonders if blackburn needs some more fuel23:39
@sonney2kaka the V-word23:40
blackburnwhy I need? :(23:40
blackburnremember I work not on my project and I don't know much about it23:40
@sonney2kI think you know pretty well what a matrix is :)23:42
blackburnbut don't know much about jblas?23:43
CIA-87shogun: Sergey Lisitsyn master * rf70a405 / examples/undocumented/java_modular/ : Fixed load matrix for java modular -
@sonney2kblackburn, me neither but it is trivial...23:43
blackburnokay fixed23:43
@sonney2kat least google finds me everything I need23:43
blackburnyes yes23:43
@sonney2kdoes it now return the same thing?23:43
blackburnI'm such a stupid russian okay :D23:43
blackburnyes, kernel matrix is the same now23:44
blackburnsonney2k: toString in println! such a useless thing. everywhere23:47
@sonney2kblackburn, is this automagically called?23:47
blackburnof course it is23:48
@sonney2kI know it is in python when __str__ or __repr__ is there23:48
blackburnyes, just like in python23:48
@sonney2kanyway, please call some libsvm classifier thingy23:49
@sonney2kthat should be easy enough to see if things work at all23:49
blackburnnow try gaussian naive bayes, then libsvm23:49
blackburnhm gaussian naive bayes is broken I gues23:53
blackburnsonney2k: libsvm minimal produces ~0.08 mean error both for java and python23:57
@sonney2kblackburn, please compare label - outputs not just accuracy23:57
@sonney2klabel outputs must match exactly23:58
blackburndamn that day I say I could work on java examples23:58
--- Log closed Mon Jul 25 00:00:12 2011