IRC logs of #shogun for Monday, 2011-09-19

« Sunday, 2011-09-18 Index Tuesday, 2011-09-20 »
--- Log opened Mon Sep 19 00:00:48 2011
-!- blackburn [~blackburn@188.168.5.156] has quit [Quit: Leaving.]00:48
-!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has joined #shogun03:16
-!- blackburn [~blackburn@188.168.4.53] has joined #shogun08:41
-!- sonne|work [~sonnenbu@194.78.35.195] has joined #shogun09:59
-!- blackburn [~blackburn@188.168.4.53] has quit [Quit: Leaving.]10:13
-!- blackburn [~blackburn@188.168.4.132] has joined #shogun18:50
blackburnsonney2k: I have found worse doc than shogun have: netcracker :D19:14
@sonney2kblackburn, heh - I guess the project is bigger so it is more difficult to have doc coverage19:32
@sonney2kblackburn, btw how is your 1 patch per day going :D19:33
CIA-3shogun: Evan Shelhamer master * r486a452 / (doc/pages/Installation.mainpage src/INSTALL): Updated install docs w/ current interfaces names, homebrew suggestion for OSX - http://git.io/YSZEtA20:03
CIA-3shogun: Soeren Sonnenburg master * r8adcc9d / (doc/pages/Installation.mainpage src/INSTALL):20:03
CIA-3shogun: Merge pull request #322 from shelhamer/install_docs20:03
CIA-3shogun: Updated install docs w/ current interfaces names, homebrew suggestion - http://git.io/xwyiHQ20:03
blackburnsonney2k: in progress20:50
blackburnshiiit I've got vertical stripe on my lcd screen20:54
@sonney2kI guess you need to reboot21:13
blackburnI need to replace my screen..21:14
blackburnsonney2k: superlu success21:15
blackburn33s for 6K21:15
blackburnsklearn took 90s21:15
@sonney2ksounds good21:17
@sonney2k:)21:17
blackburnbut I have to investigate why the fuck the results *are* different21:17
blackburngot it21:28
blackburnsonney2k:21:29
blackburnSHOGUN Took 58.796179s21:29
blackburnSKLEARN Took 236.027341s21:29
@sonney2k4 times faster - neat21:30
blackburnwho's your daddy21:32
blackburn:D21:32
blackburnhatehatehate21:57
@sonney2khate?22:21
CIA-3shogun: Soeren Sonnenburg master * rf694cb6 / (152 files in 10 dirs):22:21
CIA-3shogun: Replace void foo(void) -> void foo() and finish template .h -> .cpp22:21
CIA-3shogun: transition in shogun/features/* - http://git.io/5Mg7XQ22:21
@sonney2kwho is to be hated?22:21
blackburnsonney2k: I've been hating myself for a while22:32
blackburnI broke arpack code and it took a 0.5h to fix that22:32
blackburnstill unstable ^(22:32
blackburn:(22:32
@sonney2khmm ok22:32
blackburnsonney2k: I had a conversation with one professor here, may be I will do some road sign recognition as my bachelor work22:33
blackburnhe is the only man knowing what SVM is :D22:33
@sonney2kfun :)22:34
@sonney2kI wish I was young again to spend some more time :)22:35
@sonney2kon such things22:35
blackburnV. A. Fursov, I. A. Kulagina, and N. E. Kozin. Building of Classifiers Based on Conjugation Indices. Optical Memory and Neural Networks (Information Optics) Vol. 16 No. 3 200722:35
blackburnVladimir Fursov, IrinaKulagina, Nikita Kozin.. Building of Classifier Based on Conjugation Indexes. Proceedings of The 5-th International Conference on Machine Lerning and Data Mining. Leipzig, Germany, 18 - 20 July, 2007, p. 231-235.22:35
blackburnthat is two papers describing some idea he have22:36
blackburnI'm sure it will be in shogun if idea worths something :)22:37
blackburnsonney2k: I wish to commit superlu stuff but it is not stable yet, damn!22:38
@sonney2kthen get it stable!22:40
blackburnsonney2k: no idea what is wrong22:41
blackburnthere are near-singular matrix being LU-factorized and etc22:41
blackburna lot of numerical issues22:41
blackburnsonney2k: got it!23:00
@sonney2kok23:00
blackburn+1e-9 save me23:00
blackburn:D23:00
@sonney2kI would have suggested that :D23:01
blackburnComputing embedding of 6000 vectors using LTSA23:04
blackburnSHOGUN Took 7.369902s23:04
blackburnSKLEARN Took 85.428765s23:04
blackburnsonney2k: okay who am I? :D23:04
@sonney2kdaaaaady23:04
blackburnhaha23:05
blackburnsurprise to get such a small time for ltsa23:05
@sonney2kdue to the +1e-9 trick?23:06
blackburnnot really, it just stabilized23:06
blackburnbefore it was measures for LLE23:06
blackburnLTSA is computationally easier23:06
blackburnI have no idea why it is so slow in sklearn23:06
blackburnsonney2k: some more improvements and it is worth to write a paper I hope23:08
@sonney2kblackburn, btw I got rid of all templated code from .h file in shogun/features23:08
blackburnyeah I have seen that23:08
@sonney2kthe code is now much better to digest now23:08
blackburnsonney2k:23:09
blackburnComputing embedding of 8000 vectors using LTSA23:09
blackburnSHOGUN Took 12.350444s23:09
blackburnSKLEARN Took 189.827115s23:09
@sonney2kbut result is the same?23:10
blackburnexactly23:10
@sonney2kblackburn, let them know23:10
@sonney2k... if you want competition :D23:10
blackburnsonney2k: I'm doing it in attempt to write a paper with chris23:11
blackburnso they would see how slow their ltsa is in a paper23:11
blackburndrtoolbox (that thing with INFINITY methods) have pretty fast code btw23:12
blackburnI will measure it too23:12
@sonney2kok23:13
blackburnsonney2k: what do you think about 'options' structures?23:14
blackburnhttp://pastebin.com/xMm9tpRZ23:16
blackburnsomething like that23:16
@sonney2kI don't understand - you want to set options for arpack or what?23:21
blackburnsonney2k: yeah I would love to use struct instead of a lot of args23:22
blackburneasier to extend, etc23:22
@sonney2kit makes sense when you simultaneously have to set many options23:31
blackburnsonney2k: already ~10 params23:32
--- Log closed Tue Sep 20 00:00:51 2011
« Sunday, 2011-09-18 Index Tuesday, 2011-09-20 »

Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!