IRC logs of #shogun for Thursday, 2011-06-16

--- Log opened Thu Jun 16 00:00:51 2011
-!- blackburn [~blackburn@188.168.5.186] has quit [Quit: Leaving.]00:15
CIA-32shogun: Soeren Sonnenburg master * r218bd89 / src/libshogun/features/SimpleFeatures.h : add support for SGMatrix in constructor - http://bit.ly/ilyzst00:17
CIA-32shogun: Soeren Sonnenburg master * ra6e3ec9 / examples/undocumented/java_modular/classifier_knn_modular.java : make path relative to the java_modular examples dir - http://bit.ly/montNU00:17
CIA-32shogun: Soeren Sonnenburg master * r301896d / src/modular/Regression.i : fix compilation error in python modular - http://bit.ly/m4y1d500:17
-!- f-x [~user@117.192.223.246] has quit [Remote host closed the connection]00:40
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds]02:48
-!- in3xes [~in3xes@180.149.49.227] has joined #shogun02:48
-!- in3xes_ [~in3xes@59.163.196.121] has joined #shogun04:06
-!- in3xes [~in3xes@180.149.49.227] has quit [Ping timeout: 240 seconds]04:08
-!- in3xes_ is now known as in3xes05:09
-!- sploving [~root@124.16.139.196] has joined #shogun10:00
splovinghello sonney2k, when i installed atlas, it said, /root/atlas/../ATLAS//CONFIG/src/backend/comptestC.c:1: not implement to support 64 bit10:01
-!- f-x [~user@117.192.207.37] has joined #shogun11:33
-!- blackburn [~blackburn@188.168.3.125] has joined #shogun11:45
blackburngot 4 :)11:47
splovingblackburn, congratuation12:33
splovingblackburn, do you know how to install atlas/12:34
blackburnsploving: thanks13:13
blackburnsorry fell asleep13:13
blackburnwell I just have installed it from my ubuntu repo13:13
-!- f-x [~user@117.192.207.37] has quit [Remote host closed the connection]13:42
-!- f-x [~user@117.192.207.37] has joined #shogun14:09
splovingblackburn, use apt-get?14:15
splovingmy computer is fedora, and my yum is broken. so I have to install it from source.14:24
in3xesI've never seen a working yum :P14:27
splovingwhich system you use? in3xes14:28
in3xesI don't work on fedora much, but when ever I used...it was broken14:28
in3xessploving, ubuntu.14:28
in3xesThe bloated one :P14:29
splovingI would not like to change my system as that might cost lot of time14:29
splovinghello blackburn, another issue, are you around?15:36
-!- sploving [~root@124.16.139.196] has left #shogun []15:43
-!- f-x [~user@117.192.207.37] has quit [Remote host closed the connection]20:04
-!- f-x [~user@117.192.198.36] has joined #shogun20:46
@sonney2kblackburn, congrats22:02
blackburnsonney2k: thanks22:03
blackburnsonney2k: today was asked by Chris about multi-core routines for eigenproblems22:03
blackburnyour opinion is use lapack by default, right?22:04
@sonney2kblackburn, there is some parallel lapack22:04
@sonney2kscalapack IIRC22:04
blackburnsonney2k: it is distributed-memory based lapack22:04
blackburnthere is PARPACK (P stands for parallel)22:05
blackburnit is in debian repo22:05
blackburnso not 'rare' lib22:05
blackburnsonney2k: well defaults is: we use BLAS for matrix and vector operations and LAPACK for some LA routines, right?22:07
@sonney2kblackburn, yes22:07
blackburnand if possible other libraries22:07
@sonney2kblackburn, anything compatible to lapack / blas can be used22:07
@sonney2kyou actually don't even need to recompile22:08
blackburnI see22:08
blackburnunfortunately lapack doesn't provide lanczos22:08
blackburnso from mid of next week (just after my last exam) I'm going to start integrating it somehow22:09
blackburnsonney2k: I guess I will need some help from you - don't know much about how makefile provides these things22:10
@sonney2kblackburn, as I said just give me a snipped with arpack22:10
@sonney2kit is trivial22:10
blackburnsnipped?22:11
@sonney2ksnippet22:11
blackburni.e. some code using arpack, right?22:12
@sonney2kyes22:12
blackburnokay22:12
blackburnparallel things is so scary forefront of development!22:15
blackburnI guess we should do heavy parallel computing before 1.0.022:15
blackburnshouldn't*!22:15
@sonney2kack22:16
@sonney2kblackburn, btw I stumbled across a problem with SGMatrix22:16
@sonney2kfor example the kernel has a function get_kernel_matrix()22:16
blackburnyes22:16
@sonney2kwhich obviously should return a SGMatrix22:16
blackburnexactly22:17
@sonney2kthe problem is that this matrix will always be newly allocated22:17
@sonney2kand has to be freed22:17
@sonney2kthis is in contrast to all the other get* functions we have22:17
@sonney2khappens so far in get_distance_matrix / get_kernel_matrix22:18
blackburnyes I understand22:18
blackburnsonney2k: well why not put it into destructor of SGMatrix?22:19
@sonney2kblackburn, but how does it know that it has to desctruct?22:20
blackburnah yes, interesting problem22:21
blackburnsonney2k: why not store matrix in distance instance?22:22
@sonney2kor return a customkernel22:24
@sonney2kwhich has the kernel matrix inside22:24
blackburnlooks ugly22:25
@sonney2kbut is at least consistent22:26
blackburnsonney2k: well how you will get matrix if you will just return custom kernel?22:27
@sonney2kthe function would have to go22:27
@sonney2kand one then uses k=CustomKernel(someotherkernelobj)22:27
@sonney2kk.get_kernel_matrix()22:27
blackburnwe should think about it.. it works but doesn't seems very consistent to me22:29
@sonney2kblackburn, this problem will re-occur whenever we have some function that returns sth that is on-the-fly computed22:30
blackburnwill create a ticket :)22:31
blackburnnah later22:34
blackburnhave slept less than I need :D22:34
blackburngoing sleep now, see you sonney2k22:35
-!- blackburn [~blackburn@188.168.3.125] has quit [Quit: Leaving.]22:35
CIA-32shogun: Baozeng Ding master * r38fe3d6 / (7 files): make path relative to the java_modular examples dir and output the out_labels (+9 more commits...) - http://bit.ly/iDtjTc23:43
--- Log closed Fri Jun 17 00:00:53 2011

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