Open in new window / Try shogun cloud
--- Log opened Fri Jun 03 00:00:52 2011
-!- blackburn [~qdrgsm@188.122.237.128] has quit [Quit: Leaving.]01:36
-!- sploving [~root@124.16.139.196] has left #shogun []03:02
-!- CIA-32 [cia@cia.atheme.org] has quit [Read error: Operation timed out]05:20
-!- CIA-18 [cia@cia.atheme.org] has joined #shogun05:23
-!- blackburn [~qdrgsm@188.122.237.128] has joined #shogun09:33
-!- blackburn [~qdrgsm@188.122.237.128] has quit [Ping timeout: 250 seconds]09:51
-!- sploving [~root@124.16.139.196] has joined #shogun10:05
splovingsonney2k, around?10:05
splovinganyone?10:05
splovingit still can not work. :(10:05
-!- sploving [~root@124.16.139.196] has left #shogun []10:24
-!- sploving [~root@124.16.139.196] has joined #shogun10:39
splovingthx sonney2k, all my work dir is in root dir. It is not easy for me to move it to another user10:43
@sonney2ksploving, look at my email10:44
@sonney2kit works here10:44
@sonney2kgtg10:44
splovingI saw. maybe my swig_type_map.i. I write the ujmp typemap. so it is still have the error. sonney2k, i need have a check my typemap10:45
splovinghmm, not work11:10
splovingI think, i need to dive swig java doc more deeply11:10
-!- sploving [~root@124.16.139.196] has left #shogun []11:10
-!- f-x [~shashwat@117.192.209.39] has joined #shogun15:40
-!- f-x [~shashwat@117.192.209.39] has quit [Ping timeout: 258 seconds]17:08
-!- blackburn [~qdrgsm@188.122.237.128] has joined #shogun18:05
-!- epps [~epps@84.18.157.200] has joined #shogun18:59
-!- epps [~epps@84.18.157.200] has quit [Changing host]18:59
-!- epps [~epps@unaffiliated/epps] has joined #shogun18:59
-!- epps [~epps@unaffiliated/epps] has left #shogun ["Leaving"]19:29
serialhexhey blackburn, whats going on?19:38
blackburnhey, going fine! just get classic mds to work19:38
blackburnwill paste a link in a minute19:38
blackburnwhat's up with you?19:38
serialhexnot too much, about to get to work wrtn some ruby examples19:38
blackburnhow is your work going on?19:39
blackburnsold many ones? :)19:39
blackburnwell, data for mds: http://dl.dropbox.com/u/10139213/share/mds_data.png19:40
serialhexehh, i've sold ~$400 worth of knives, i need more appts19:40
blackburnand result: http://dl.dropbox.com/u/10139213/share/mds_data_transformed.png19:40
serialhexright now i'm below average :(19:40
serialhexcool cool, it's spiraly!!!19:41
blackburnoh it is more fast than LLE19:48
blackburn~65s for 2000 3d examples19:52
serialhexcool cool20:33
serialhexso blackburn, how does the dim reduction work for cases where i might need that third dimension??20:35
blackburnjust the same20:36
blackburnthey all are hmmm20:36
blackburneh.. how to say that :)20:36
serialhexso like if the spiral didnt contract/expand, it sould simply look like a circle20:36
blackburnwell if given data X20:36
blackburnand we got x,y,z coordinates of embedding20:37
blackburnso x,y will be 2d spiral20:37
blackburnand z will be just like z in X20:37
blackburni.e. if map is 3d->3d it will stay the same20:37
serialhextrue20:37
serialhexbut if - as in your example - the 3d spiral dosnt shrink, the resulting 2d image will simply be a circle20:38
blackburni dk20:38
blackburn'll test it right now20:38
serialhexi'm not saying what you're doing is wrong, but the overall theory is kind of wonky imo... :P20:38
blackburnnot it should be a circle20:38
blackburnit preserves distances20:39
serialhexoh, ok20:39
blackburnthe key idea of MDS is to preserve distance20:39
blackburn*pairwise distances20:39
serialhexok, that makes sense... because in your current example, if looked at from the top it might look like a couple of zig-zag lines... tho a spiffy alg would change that :D20:41
blackburnhmm interesting thing with just flat spiral20:42
blackburnhttp://dl.dropbox.com/u/10139213/share/spiral.png20:43
blackburnhttp://dl.dropbox.com/u/10139213/share/spiral_transformed.png20:43
blackburnserialhex: well these pictures show that idea more clearly20:44
blackburn3d->2d is just the same spiral20:44
blackburnin 2d*20:44
serialhexhmm, ok20:45
blackburnidk exactly how it will be proceed on different data20:46
blackburnin fact it presents some suprises sometimes20:47
blackburnfor example when flat 3d data is given it have only 2 non-zero eigenvalues20:47
blackburnif it is not flat - 3 non-zero eigs20:48
blackburnlooks like it reflects initial dimension somehow20:48
serialhexhmm20:48
blackburnI speak some broken way now :)20:49
serialhexheh, well it's ok, you're trying to explain funky technical things, which is hard sometimes in ones native language :D20:50
@bettyboofunny?20:50
serialhexyes it is funny bettyboo20:50
@bettybooserialhex: for static interfaces like matlab you have to modify libshogunui/GUIKernel.cpp / SGInterface.cpp - not as nice but will work :)20:50
blackburnserialhex: just tried for 3d parabola20:52
serialhexcool20:52
blackburn.. and it creates 3d parabola20:52
blackburn:D20:52
@bettyboo;D blackburn20:52
blackburnhuh how useless it is for some data20:52
serialhexyou cant squinch to 2d??20:52
blackburni can't but it is done not that way I want20:53
blackburnfrom side or so20:53
blackburnright mapping should be just like circle20:53
serialhexhow about the top?20:53
serialhexlooking into the parabola...20:53
blackburnaha20:53
blackburnshould be this way20:53
blackburnI think the difference lies in parabola height20:54
blackburnbtw MDS for euclidean distance is equal to PCA :D20:54
blackburnserialhex: well multiplying Z by 0.005 gave right mapping to 2d20:56
blackburnhaha21:30
blackburnI had been using *deleted* array21:31
blackburnserialhex: around?21:34
blackburnah, nevermind I found :)21:36
-!- alesis-novik [~alesis@188.74.87.84] has joined #shogun21:47
@sonney2kwtf?!22:00
@sonney2know I am also getting baozengs error22:00
blackburnwith C++ you can not only make errors but also inherit it hehe22:13
@bettyboohrhr22:13
serialhexhey blackburn i'm here now, whats up??22:32
blackburnserialhex: just english issue :)22:32
serialhexoh, ok :)22:33
blackburnforgot how to say about eigenvector's eigenvalue22:33
serialhexsonney2k, baozeng wanted me to rub the java tests, how do i do that??  (i have 0 java expierence)22:33
blackburnassociated one22:34
serialhexoh, ok... eigenvalues & such still kind of confuse me, mostly cause i havnt needed/studied them at all22:34
blackburnserialhex: just wanted how to say that they are associated - now I know :)22:35
blackburnwanted to know822:35
blackburn*22:35
serialhexok, cool22:35
serialhexhmm... apparently the documentation is out of date, even when you generate it :(22:39
serialhexSTEAK!!!22:40
-!- blackburn [~qdrgsm@188.122.237.128] has quit [Remote host closed the connection]22:59
-!- blackburn [~qdrgsm@188.122.237.128] has joined #shogun23:00
-!- blackburn1 [~qdrgsm@188.122.237.128] has joined #shogun23:04
-!- blackburn [~qdrgsm@188.122.237.128] has quit [Read error: Connection reset by peer]23:04
-!- blackburn1 [~qdrgsm@188.122.237.128] has quit [Ping timeout: 246 seconds]23:11
--- Log closed Sat Jun 04 00:00:54 2011