Open in new window / Try shogun cloud
--- Log opened Thu Sep 15 00:00:32 2011
-!- sonney2k [~shogun@7nn.de] has quit [Ping timeout: 276 seconds]00:36
-!- sonney2k [~shogun@7nn.de] has joined #shogun00:36
-!- sonne|work [~sonnenbu@194.78.35.195] has joined #shogun09:10
[1]warpysonne|work, did you see my msg ?09:19
sonne|workyes just saw it09:19
sonne|workwe don't have free manpower for that currently09:19
sonne|workwe will be getting windows / osx build bots09:20
sonne|workso when these are there I will at some point make sure shogun compiles on osx/windows09:20
[1]warpyany estimation on that, time frame wise ?09:20
-!- serialhex_ [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has joined #shogun09:26
-!- serialhex [~quassel@99-101-148-183.lightspeed.wepbfl.sbcglobal.net] has quit [Ping timeout: 258 seconds]09:27
-!- aelray [d472ce8a@gateway/web/freenode/ip.212.114.206.138] has joined #shogun09:31
aelrayhi09:33
aelraysomebody here?09:34
sonne|workyes09:40
sonne|workthough busy and at work09:40
sonne|workjust shoot09:40
[1]warpysonne|work, no work plan / time frame yet ?09:50
[1]warpyalso is anynone else working on the c# examples beside us ?09:51
aelrayi have a basic question on ML and am not shure if it is solveable with shogun09:53
aelraymy problem is i have to detect polygonal regions with only 90 degree corners in greyscale images09:54
aelrayvalues inside this regions have usually higher amplitude than outside, the outside can be considered somehow black, except for noise09:55
aelrayi need to extract the exact corner positions, can a SVM do this?09:56
sonne|workaelray: you could extract some context around the corners, like +/- 10 pixels or so and then using that 10x10 image try to detect if there is a corner in the middle10:01
sonne|workand then slide that thing around10:01
aelrayok, and this is realizable with sn SVM?10:06
-!- [1]warpy [~warpy@bzq-79-181-14-161.red.bezeqint.net] has quit [Quit: Try HydraIRC -> http://www.hydrairc.com <-]10:38
sonne|workaelray: you have to extract these windows manually - but then classifying e.g. with gaussian kernel could already give you reasonable results10:38
aelrayok, so i have to extract a lot of small like 10x10 pictures from my datasets with and without corners? how much do i need ca.?10:42
sonne|workthe more the better / I guess limited only by computation time / storage :)10:43
sonne|workthousands10:43
aelraysounds like fun10:46
aelrayis this working even when the corner is not visible to human eyes but one knows, because of geometry, that there must be one?10:47
sonne|worktrial and error...10:57
aelrayok, thanks so far10:57
-!- sonne|work [~sonnenbu@194.78.35.195] has left #shogun []11:41
-!- sonne|work [~sonnenbu@194.78.35.195] has joined #shogun13:38
-!- aelray [d472ce8a@gateway/web/freenode/ip.212.114.206.138] has quit [Ping timeout: 252 seconds]15:08
-!- blackburn [~blackburn@188.168.2.36] has joined #shogun15:28
-!- cwidmer [~quassel@connect.tuebingen.mpg.de] has joined #shogun16:48
-!- cwidmer [~quassel@connect.tuebingen.mpg.de] has quit [Remote host closed the connection]16:58
blackburnshit missed chris again18:17
-!- mode/#shogun [+o sonney2k] by ChanServ19:16
@sonney2kAnyone here, blackburn?19:50
blackburnsonney2k: here20:29
blackburnhellloooo20:38
-!- alesis-novik [~alesis@cpat001.wlan.net.ed.ac.uk] has joined #shogun20:47
@sonney2kblackburn, still there?21:04
blackburnsonney2k: yeap21:04
@sonney2kblackburn, do you remember how big the modshogun_wrap.cxx is for python_modular?21:17
blackburnsonney2k: it is 16.7 mb on my git21:18
@sonney2kblackburn, I meant number of lines21:18
blackburnah21:18
blackburnsonney2k: 508k21:20
@sonney2kexact number?21:20
blackburnsonney2k: 50888821:22
@sonney2k$ wc -l interfaces/python_modular/modshogun_wrap.cxx21:33
@sonney2k516531 interfaces/python_modular/modshogun_wrap.cxx21:33
@sonney2k?!21:34
blackburnblackburn@blackburn-laptop:~/shogun/shogun/src$ wc -l interfaces/python_modular/modshogun_wrap.cxx21:34
blackburn508889 interfaces/python_modular/modshogun_wrap.cxx21:34
@sonney2kblackburn, how did yo configure?21:36
@sonney2k./configure --prefix=/usr/local/stow/shogun-1.0.0 --disable-optimization --enable-debug ?21:36
@sonney2kor sth21:36
blackburnsonney2k: I don't remember21:36
blackburnshall I reconfig it?21:36
@sonney2kyes please21:37
blackburncompiling21:40
@sonney2kblackburn, you don't have to compile it to the end21:44
@sonney2kjust when the *.cxx file is there is sufficient21:44
@sonney2kblackburn, so how big is the *.cxx?21:55
blackburnblackburn@blackburn-laptop:~/shogun/shogun/src$ wc -l interfaces/python_modular/modshogun_wrap.cxx22:04
blackburn508889 interfaces/python_modular/modshogun_wrap.cxx22:04
blackburnsonney2k: the same22:04
@sonney2khmmmhh let me treat the remaining classes in features/* and then git push22:05
@sonney2kblackburn, then please check again22:05
@sonney2kI removed a lot of code from the headers22:05
blackburnhmm I forgot to commit some changes22:05
@sonney2kso it cannot possibly grow but what shall I say...22:05
@sonney2kblackburn, today I tried your GNB22:06
blackburnsonney2k: and how it was? :)22:06
@sonney2kfor two-class I would love to have floating point values22:07
@sonney2knot 0 and + 122:07
blackburnpretty hard to get comprehensible map from probabilities to resulting labels22:07
@sonney2k?22:08
blackburnbtw22:08
@sonney2kjust  prob(x|theta_+) - prob(x|theta_-)22:08
blackburnfor twoclass it is better to use logistic regression22:08
blackburnit is almost equal but faster22:08
@sonney2kwhere theta+/- are the model params22:09
blackburnsonney2k: I'm going to sleep now, am I needed now?22:11
blackburnwith .cxx issues22:11
@sonney2kif you can wait for a few more minutes22:12
@sonney2kI would be very interested in knowing whether the situation improved22:12
blackburnyeah sure22:12
CIA-3shogun: Soeren Sonnenburg master * r9c69c84 / src/shogun/preprocessor/Preprocessor.h : fix comment type to get it integrated in doxygen - http://git.io/N-RhJw22:14
CIA-3shogun: Soeren Sonnenburg master * r5fee0e6 / src/README.developer : mention that one should avoid inline functions - http://git.io/yLgAfw22:14
CIA-3shogun: Soeren Sonnenburg master * r8b4ce6c / (33 files):22:14
CIA-3shogun: remove most inline functions in features22:14
CIA-3shogun: - exceptions are some single line functions22:14
CIA-3shogun: - templated classes - http://git.io/vCcP0w22:14
@sonney2kblackburn, please compile again22:14
@sonney2kwith the update ^22:14
blackburnsonney2k: yeah but first will commit some changes I did before22:15
@sonney2kheh22:15
@sonney2kblackburn, I think the rule should be one of us two doing one (small) git push every day22:15
blackburneh?22:15
@sonney2kthe last week the project looked pretty dead22:16
blackburnah22:16
blackburnyeah22:16
@sonney2kso 1 commit / day improving minor things is helpful22:17
blackburnokay got it22:20
blackburnI won't commit, I didn't manage to finish it yet22:20
@sonney2kblackburn, then create a local branch and configure compile again22:21
@sonney2kI would be interested if what I did changed anything22:22
blackburnsonney2k: I just stashed it22:23
@sonney2kblackburn, I think I will do the header changes removing all unnecessary includes / putting things into .cpp files and thus libshogun.so22:26
@sonney2kso that's my task for the next weeks22:26
blackburnI see22:26
@sonney2kand bug fixes here and there as usual22:27
@sonney2k(btw I am still messing with the debian package - tough...)22:27
@sonney2kblackburn, does it compile?22:28
@sonney2kthe new stuff?22:28
blackburnsonney2k: yeah was troubles with some .cpp I forgot to remove22:28
@sonney2kI mean is the *.cxx already generated?22:28
@sonney2kok22:28
blackburngenerating22:28
blackburnsonney2k: my tasks are still get preprocessor faster, etc22:29
@sonney2kGNB22:29
@sonney2k:D22:29
blackburnI guess you can fix/improve gnb if you will need that :)22:30
@sonney2kbtw the nice graphical gnb example you had could be in shogun examples22:30
blackburnI'm in rush with potential paper22:30
blackburnso focused on algorithm things22:30
blackburnin fact I was reading about superlu on my job for 2 hours :D22:31
blackburnsonney2k:22:31
blackburnblackburn@blackburn-laptop:~/shogun/shogun/src$ wc -l interfaces/python_modular/modshogun_wrap.cxx22:31
blackburn508785 interfaces/python_modular/modshogun_wrap.cxx22:31
blackburnsonney2k: but looks faster22:32
@sonney2kso 100lines less22:32
@sonney2kok22:32
@sonney2kthere is more now in libshogun.so22:32
@sonney2kthe headers are just mere definitions22:32
blackburnI see22:33
@sonney2kthis will help removing includes too btw22:33
@sonney2kand so speed up compilation hopefully...22:33
blackburnokay see you tomorrow22:33
@sonney2kand will solve the problems with lapack being required or so22:33
@sonney2kyeah22:33
@sonney2khave a good sleep22:33
blackburnthanks22:33
@sonney2kthanks for the report!22:33
blackburnyou too22:33
blackburnreport about?22:33
-!- blackburn [~blackburn@188.168.2.36] has quit [Quit: Leaving.]22:34
CIA-3shogun: Soeren Sonnenburg master * rc834d2b / (4 files): remove remaining inline functions in features/ - http://git.io/tYKBiw23:33
CIA-3shogun: Soeren Sonnenburg master * r4f9566f / src/interfaces/python_modular/swig_typemaps.i : start to simplify typemaps - extracting all code into functions - http://git.io/cHdaLg23:33
CIA-3shogun: Soeren Sonnenburg master * ra49b0f6 / (19 files in 3 dirs): remove unnecessary includes from .h files - http://git.io/xh2KxQ23:33
CIA-3shogun: Soeren Sonnenburg master * r4a8d99d / (2 files): (log message trimmed)23:34
CIA-3shogun: Put implementations of templated functions into .cpp files23:34
CIA-3shogun: This is possible when strictly following23:34
CIA-3shogun: http://www.parashift.com/c++-faq-lite/templates.html#faq-35.1523:34
CIA-3shogun: i.e. we must know beforehand which types we want to support and specify23:34
CIA-3shogun: them like23:34
CIA-3shogun: template class Foo<int32_t>;23:34
--- Log closed Fri Sep 16 00:00:37 2011