--- Log opened Mon Dec 17 00:00:38 2012 | ||
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 01:21 | |
-!- zxtx [~zv@75.83.151.252] has joined #shogun | 02:35 | |
shogun-buildbot | build #182 of nightly_none is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/182 | 03:00 |
---|---|---|
shogun-buildbot | build #220 of nightly_default is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/220 | 03:06 |
-!- zxtx [~zv@75.83.151.252] has quit [Ping timeout: 272 seconds] | 05:12 | |
-!- blackburn [~blackburn@188.168.5.28] has quit [Quit: Leaving.] | 07:48 | |
-!- sonne|work [~sonnenbu@194.78.35.195] has joined #shogun | 08:55 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 09:14 | |
sonne|work | wiking: whats up with multiclass? | 10:47 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 11:11 | |
wiking | sonne|work: libsvm multiclass doesn't work with 3 classes | 11:23 |
sonne|work | what? | 11:23 |
sonne|work | so basically all OvO MC stuff is b0rken right now? | 11:23 |
sonne|work | or just libsvm? | 11:24 |
wiking | libsvm | 11:24 |
wiking | sonne|work: indexing problem | 11:36 |
sonne|work | wiking: you have your verbose day today? | 11:36 |
wiking | ? | 11:37 |
wiking | verbose day?:) | 11:37 |
sonne|work | yeah - 3 words withing 12 minutes! | 11:38 |
wiking | lol | 11:38 |
wiking | yeah working in the meanwhile ;) | 11:38 |
sonne|work | I almost have an idea what you are talking about | 11:38 |
wiking | so yeah the problem is with multiclass libsvm, having only 3 classes | 11:38 |
wiking | it complains about indexing | 11:38 |
sonne|work | and w/ 4 classes or 2 it works correctly? | 11:39 |
wiking | REQUIRE(set_svm(idx, svm),"SVM set failed"); | 11:39 |
wiking | this line fails | 11:39 |
wiking | MulticlassLIBSVM.cpp:194 | 11:39 |
sonne|work | I am a bit concerned that everything is broken ... | 11:39 |
wiking | it works with 4 classes, i haven't tested 2 classes | 11:39 |
sonne|work | wiking: correctly even? | 11:40 |
sonne|work | or just 'works' as in doesn't crash | 11:40 |
wiking | basically i know that n > 3 classes works correctly | 11:40 |
sonne|work | ok | 11:40 |
wiking | well it gives me good result | 11:40 |
wiking | but maybe it doesn't give me as good as i would like it ;) | 11:40 |
wiking | haven't tested using libsvm purely if that's what you r asking | 11:40 |
wiking | i don't know if there's a difference between our libsvm and the original implementation's ouptut | 11:41 |
wiking | *output | 11:41 |
-!- ptizoom [~christian@85.210.82.84] has joined #shogun | 15:11 | |
-!- Hausas [~admin@unaffiliated/hausas] has quit [Ping timeout: 248 seconds] | 15:12 | |
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun | 15:19 | |
wiking | sonne|work: hey here ? | 15:41 |
wiking | sonne|work: i was just wondering that for how long we should archive the static analyzer output? | 15:42 |
wiking | sonne|work: as it now starts to eat up significant amount of space ;) | 15:42 |
sonne|work | wiking: do as you wish ... some kind of binary granularity? | 16:30 |
sonne|work | like every day for 1 week | 16:30 |
sonne|work | then say 8 more weeks but just weekly? | 16:31 |
wiking | sonne|work: now we have december accessible right away and the rest in tar.bz2 | 16:33 |
wiking | yeah i'll do some kind of a weekly thing after 1 month or so | 16:33 |
wiking | since keeping all that tar.bz2 is kind of waste | 16:33 |
wiking | i reckon nobody will dig that out :P | 16:33 |
sonne|work | wiking: just remove it then | 16:36 |
sonne|work | it doesn't really help us to have a long history here | 16:36 |
-!- Hausas [~admin@185.6.154.4] has joined #shogun | 17:41 | |
-!- Hausas [~admin@185.6.154.4] has quit [Changing host] | 17:41 | |
-!- Hausas [~admin@unaffiliated/hausas] has joined #shogun | 17:41 | |
-!- blackburn [~blackburn@83.234.54.175] has joined #shogun | 18:09 | |
-!- hoijui [~hoijui@dslb-092-078-189-197.pools.arcor-ip.net] has joined #shogun | 19:24 | |
-!- hoijui [~hoijui@dslb-092-078-189-197.pools.arcor-ip.net] has quit [Quit: Leaving] | 19:35 | |
-!- FSCV [~FSCV@187.210.54.162] has joined #shogun | 19:36 | |
-!- nona [~nona@cpc5-haye17-2-0-cust251.haye.cable.virginmedia.com] has joined #shogun | 20:26 | |
-!- nona is now known as Guest80301 | 20:26 | |
Guest80301 | Hi | 20:26 |
blackburn | hi | 20:27 |
Guest80301 | I would like to work with spectrum RBF kernel, anybody work with this before? | 20:27 |
Guest80301 | I am a bit puzzled it is not working from R and based on my information, if something implemeted in shogun it has to work from all interfaces, Am I right or not? | 20:29 |
Guest80301 | anybody ineterested to talk about this? | 20:30 |
blackburn | Guest80301: I answered you in the mailing list | 20:30 |
Guest80301 | where? | 20:31 |
Guest80301 | ok I just go and check | 20:31 |
blackburn | ah | 20:31 |
blackburn | you answered already okay | 20:31 |
blackburn | Guest80301: what if you try WEIGHTEDDEGREERBF? | 20:33 |
blackburn | I see no reason not to work with R but I do not remember if case matters | 20:34 |
Guest80301 | it seems to work I don't get the same error | 20:35 |
Guest80301 | ump <- sg('set_kernel', 'WEIGHTEDDEGREERBF') Error in sg("set_kernel", "WEIGHTEDDEGREERBF") : [ERROR] Usage: set_kernel sg('set_kernel', type, size[, kernel-specific parameters]) | 20:35 |
blackburn | there are a few parameters here | 20:36 |
Guest80301 | let's see if I set up the parameters it works | 20:37 |
blackburn | okay | 20:37 |
blackburn | bad you can't use modular interface | 20:37 |
blackburn | it is much comprehensive | 20:37 |
Guest80301 | I am using static interface | 20:38 |
blackburn | I know | 20:39 |
Guest80301 | MY data set the the dna data set which is in the 'data' folder of shogun | 20:39 |
Guest80301 | dump <- sg('set_kernel', 'WEIGHTEDDEGREERBF', 'CHAR', size_cache, degree) > km <- sg('get_kernel_matrix', 'TRAIN') Error in sg("get_kernel_matrix", "TRAIN") : [ERROR] Kernel can not process this train feature type: 30 10. | 20:39 |
Guest80301 | It is hard to figure it out without any example :( | 20:41 |
blackburn | I am not sure weight degree rbf is the choice for string data | 20:41 |
blackburn | I can even say it won't work with strings | 20:41 |
blackburn | Guest80301: there are other string kernels so please rather try them | 20:43 |
Guest80301 | it seems to be a good choice for me since I don't know the structure of my string and this weighted degree kernels let me to have shifts and mis matches | 20:44 |
Guest80301 | I am not sure about rbf part | 20:46 |
blackburn | drop rbf and it will serve you | 20:46 |
Guest80301 | If you have a check on the mailing list I am also trying it without rbf but still not working | 20:47 |
blackburn | Guest80301: could that example help you? https://github.com/shogun-toolbox/shogun/blob/master/examples/undocumented/r_static/kernel_weighteddegreestring.R | 20:48 |
Guest80301 | I have been thorough this example but as you can see the only defined argument for kernel is degree | 20:50 |
Guest80301 | therefor there is nothing about mis-match and shifts | 20:51 |
Guest80301 | these axamples need some work to make them more helpful | 20:51 |
blackburn | I am afraid nobody will touch that ever :) | 20:52 |
Guest80301 | :(( | 20:53 |
blackburn | we have neither power nor will to support static interfaces | 20:53 |
Guest80301 | what is a systematic way to find out the parameters of a kernel function written in shogun | 20:54 |
blackburn | it works like a charm with modular I think ;) | 20:54 |
blackburn | well | 20:54 |
Guest80301 | ok that means it is a self discovery | 20:54 |
blackburn | probably I am not sure | 20:55 |
blackburn | it is pretty clear with modular as it is just as is | 20:55 |
Guest80301 | I have never used modular ones | 20:56 |
Guest80301 | another question does shogun work with matlab under windows? | 20:57 |
blackburn | https://github.com/shogun-toolbox/shogun/blob/master/examples/undocumented/python_modular/kernel_weighted_degree_string_modular.py | 20:57 |
blackburn | IIRC it should but no guarantees | 20:57 |
Guest80301 | this is an example in python I saw similar example for matlab | 20:59 |
Guest80301 | thanx I am going to work on it | 21:42 |
Guest80301 | bye now | 21:42 |
-!- hoijui [~hoijui@dslb-092-078-189-197.pools.arcor-ip.net] has joined #shogun | 22:02 | |
-!- hoijui [~hoijui@dslb-092-078-189-197.pools.arcor-ip.net] has quit [Quit: Leaving] | 22:34 | |
--- Log closed Tue Dec 18 00:00:38 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!