--- Log opened Tue Aug 28 00:00:17 2012 | ||
-!- romi_ [~mizobe@201-0-61-96.dsl.telesp.net.br] has joined #shogun | 03:06 | |
-!- romi_ [~mizobe@201-0-61-96.dsl.telesp.net.br] has quit [Remote host closed the connection] | 04:37 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 06:47 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 07:32 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has joined #shogun | 12:01 | |
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun | 14:29 | |
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Client Quit] | 14:32 | |
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun | 14:43 | |
yoo | hi | 14:43 |
---|---|---|
yoo | I have a question about mkl classification | 14:44 |
yoo | using p-norm mkl and the example below | 14:46 |
yoo | http://pastebin.com/CZY9NbYs | 14:46 |
yoo | with p=1 no problem | 14:46 |
yoo | with p>1 I get different results according of orders the kernels are "append" to CombinedKernel | 14:47 |
yoo | regardless to the order the kernels are added, the first kernel always get a big weight ... | 14:48 |
n4nd0 | yoo: what classifier are you using? | 14:57 |
n4nd0 | it looks like there are some bugs with MKL and SVMLight | 14:57 |
yoo | I am using MKLMulticlass | 14:58 |
yoo | Could you confirm me the "bug" I get running the code below | 14:59 |
yoo | http://pastebin.com/CZY9NbYs | 14:59 |
yoo | despite those results, I dont know if its a real bug | 14:59 |
yoo | Is there another mkl multiclass machine than CMKLMulticlass ? | 15:01 |
n4nd0 | give me a sec | 15:03 |
yoo | Maybe CMKLMulticlass support only 1-norm ? | 15:05 |
n4nd0 | I am not sure about that, I have never used this part of the project but I am checking | 15:05 |
yoo | thanks | 15:05 |
n4nd0 | about other MKL multiclass classes | 15:05 |
n4nd0 | I think there is an MKLMulticlassGradient and MKLMulticlassGLPK | 15:06 |
n4nd0 | did you see those? | 15:06 |
n4nd0 | yoo: in the doc of MKLMulticlass, it says that it is a class for L1-norm - so yes, I guess it is only for L1-norm | 15:07 |
yoo | yes but MKLMulticlassGradient and MKLMulticlassGLPK are helpers for MKLMulticlass | 15:07 |
n4nd0 | yoo: aham, I understand | 15:07 |
yoo | mmmh | 15:09 |
n4nd0 | however there are some things related about pnorm begin >= 1 | 15:09 |
n4nd0 | being, sorry | 15:09 |
yoo | in the MKLMulticlass norm | 15:10 |
yoo | void CMKLMulticlass::set_mkl_norm(float64_t norm) { pnorm=norm; if(pnorm<1 ) SG_ERROR("CMKLMulticlass::set_mkl_norm(float64_t norm) : parameter pnorm<1"); } | 15:10 |
-!- n4nd0_ [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 15:10 | |
-!- n4nd0 [53b32c87@gateway/web/freenode/ip.83.179.44.135] has quit [Quit: Page closed] | 15:10 | |
-!- n4nd0_ is now known as n4nd0 | 15:10 | |
n4nd0 | yeah ok so the norm shouldn't be less than 1 | 15:11 |
n4nd0 | but your problem was with norm greater than 1 right? | 15:11 |
yoo | yes of course | 15:11 |
n4nd0 | I think it should work fine with pnorm>=1 | 15:12 |
yoo | :) | 15:13 |
n4nd0 | mmm | 15:13 |
yoo | allright could you just run the code I sent you, set p-norm with p=1.5 for example | 15:13 |
n4nd0 | sure | 15:13 |
n4nd0 | let's see | 15:13 |
yoo | and then change the order of the kernels | 15:13 |
yoo | "the order you call cker->append_kernel()" | 15:14 |
yoo | and tell me if you get correct results : meaning same weights with different corresponding orders | 15:14 |
n4nd0 | ok | 15:15 |
yoo | (thanks :p ) | 15:15 |
n4nd0 | there was a compile error btw, line 21 missing a " | 15:15 |
yoo | certainly when passing code to pastebin | 15:15 |
yoo | yes, because you need to set your path to shogun example data | 15:16 |
n4nd0 | yeah, the weights are different | 15:16 |
n4nd0 | firs time | 15:16 |
n4nd0 | weights=[0.772121228165275508,0.355850164735586394,0.228543243936139867] | 15:17 |
n4nd0 | second | 15:17 |
n4nd0 | weights=[0.807646890960066477,0.372222999073939242,0.13038554325345969] | 15:17 |
n4nd0 | and I just switched two calls to append | 15:17 |
yoo | thats strange | 15:17 |
n4nd0 | I certainly don't know | 15:17 |
n4nd0 | I don't know much about MKL apart from what the acronym means :S | 15:17 |
yoo | because linear kernel give poor results but when you place it in the first position, it takes 80% of weights .. | 15:18 |
yoo | I am not very used to it too | 15:18 |
yoo | maybe I can come back later. Could you keep the attached code and ask to people who wrote the code ? | 15:19 |
n4nd0 | have you checked that the other parts of your example are correct? | 15:21 |
n4nd0 | for example, I am not very sure if I understand why before each call to append_kernel you call append_feature_obj? | 15:21 |
yoo | yes I think this part is correct. append_features is called for each kernel since the features could be different for each kernels. | 15:38 |
yoo | from doc "It keeps pointers to the added sub-features and is especially useful to combine kernels working on different domains (c.f. CCombinedKernel) and to combine kernels looking at independent features." | 15:40 |
n4nd0 | I see | 15:45 |
n4nd0 | but the thing is that for the kernels you are using, all of them use the same features | 15:47 |
n4nd0 | then I am not really sure why to append the same features several times | 15:48 |
yoo | because you need CCombinedFeatures object ? | 15:48 |
yoo | if not : segfault " Number of features/kernels does not match - bailing out " | 15:49 |
n4nd0 | aham ok, you are probably right | 15:49 |
yoo | but one more time, I am not an expert .. | 15:49 |
n4nd0 | me neither | 15:49 |
n4nd0 | take a look at python_modular example though | 15:50 |
n4nd0 | mkl_multiclass_modular.py | 15:50 |
yoo | ok i'll do that right now | 15:50 |
yoo | give me a minute | 15:50 |
n4nd0 | as you can see there the features are not given in the constructor to the individual kernels | 15:50 |
n4nd0 | maybe that doesn't make a difference but let's better be sure | 15:50 |
yoo | and they test p-norm with p =1.5 or 2.0 | 15:54 |
n4nd0 | true | 15:55 |
yoo | but I can see : feats_train.append_feature_obj(subkfeats_train) | 15:55 |
n4nd0 | yes | 15:55 |
yoo | 3 time fr each kernel | 15:55 |
yoo | what d you mean "he features are not given in the constructor to the individual kernels" ? | 15:55 |
yoo | oh I see | 15:56 |
yoo | ! | 15:56 |
n4nd0 | I have run the test like that | 15:56 |
n4nd0 | and it seems that the weights still vary | 15:56 |
n4nd0 | if one changes the order of them | 15:57 |
n4nd0 | of the kernels I mean | 15:57 |
yoo | yes me too | 15:59 |
yoo | did you run it in python or c++ ? | 15:59 |
n4nd0 | C++ | 16:00 |
n4nd0 | have you tried changing them in python? | 16:00 |
yoo | not yet | 16:01 |
n4nd0 | let me know when you have | 16:01 |
yoo | is there a get_kernel_weights method in python ? | 16:05 |
n4nd0 | yeah sure | 16:08 |
n4nd0 | if you are using it from libshogun interface, it should be available from python modular | 16:09 |
n4nd0 | yoo: did you find any problem trying it? | 16:09 |
yoo | here is the code if you want to test it | 16:13 |
yoo | http://pastebin.com/VGMCb2UY | 16:13 |
yoo | change the path to inputs | 16:13 |
n4nd0 | so what did you find? | 16:14 |
yoo | the same pb .. | 16:15 |
yoo | work well only with 1-norm | 16:15 |
n4nd0 | mmmm | 16:16 |
n4nd0 | I have just tested it in python | 16:16 |
n4nd0 | changing the other of the kernels introduced | 16:17 |
yoo | and ? | 16:17 |
n4nd0 | using 1.5 norm | 16:17 |
n4nd0 | the weights do not change | 16:17 |
n4nd0 | sometimes there's a small variation in the last weight | 16:17 |
n4nd0 | but it is about e-13 so I guess that is normal | 16:17 |
yoo | but they should right ? if you change the order | 16:17 |
yoo | values should stay the same in different ordrers | 16:18 |
n4nd0 | well I guess they are given in that ascending order | 16:18 |
yoo | no way | 16:18 |
yoo | they are given in the order you ve called append | 16:18 |
yoo | just try with 1-norm, you will see | 16:18 |
n4nd0 | ok | 16:19 |
n4nd0 | lol this is a funny one | 16:20 |
n4nd0 | when I change the norm from 1.5 to 1 | 16:20 |
n4nd0 | SystemError: [ERROR] glpk.h from GNU glpk not included at compile time necessary here | 16:20 |
yoo | ouch | 16:21 |
yoo | dont get that | 16:21 |
yoo | outputs are | 16:21 |
yoo | [ 0.97250202 0.02749798 0. ] | 16:21 |
yoo | [ 0.97250202 0. 0.02749798 ] | 16:21 |
yoo | works fine | 16:21 |
n4nd0 | for which example? | 16:24 |
n4nd0 | norm 1 you mean right? | 16:24 |
yoo | yes | 16:25 |
n4nd0 | aham | 16:26 |
n4nd0 | yeah what I said about the ascending order is probably stupid | 16:27 |
n4nd0 | sonney2k: ping | 16:27 |
n4nd0 | sonney2k: do you know something about a problem using MKL with norm>=1? | 16:27 |
n4nd0 | it looks like yoo gets different weights depending on the order he includes the subkernels | 16:28 |
yoo | http://pastebin.com/CZY9NbYs and http://pastebin.com/VGMCb2UY for reproducing in c++/python | 16:32 |
CIA-52 | shogun: Sergey Lisitsyn master * rb630803 / (8 files in 3 dirs): Renamed L1L2 to L12 - http://git.io/japeKA | 17:21 |
yoo | n4nd0: could you keep me inform if you have news on this issue ? (thereisnoknife@gmail.com) | 17:24 |
n4nd0 | yoo: ah so you are the HM-SVM guy too! :) | 17:25 |
n4nd0 | yeah sure | 17:25 |
n4nd0 | but probably they will say something about this here at the chat | 17:25 |
n4nd0 | it will be in the logs | 17:25 |
n4nd0 | but I'll tell you if I see any answer about it posted | 17:26 |
-!- blackburn [~blackburn@188.168.5.210] has joined #shogun | 17:41 | |
-!- heiko1 [~heiko@host86-176-252-249.range86-176.btcentralplus.com] has joined #shogun | 17:41 | |
yoo | blackburn: hi | 17:48 |
yoo | blackburn: I have not finished the compilation procedure of shogun using cmake, it turns out there is severals macro to handle | 17:49 |
blackburn | yoo: macro like? | 17:50 |
yoo | blackburn: I have encountered pbs testing mklmulticlass as you can see in the chat logs, have you already experienced smthg similar ? | 17:50 |
heiko1 | blackburn, how to use this python script you sent around for extracting code? | 17:51 |
yoo | -DUSE_* | 17:51 |
heiko1 | could you give me your call? | 17:51 |
blackburn | yoo: yes I've seen | 17:51 |
blackburn | heiko1: hmm let me remember | 17:51 |
blackburn | heiko1: python % -s [since_date] -b [end of gsoc date] | 17:53 |
n4nd0 | heiko1: where is that script? | 17:53 |
heiko1 | blackburn, which format is the date? | 17:53 |
heiko1 | and do I have to put my name somewhere? | 17:53 |
blackburn | heiko1: oops | 17:53 |
blackburn | n4nd0: heiko1: http://inspirated.com/2011/08/23/tarball-generator-for-git-commits | 17:55 |
n4nd0 | thanks | 17:57 |
blackburn | ? ?????????????? ????? | 17:59 |
blackburn | oops I was cleaning my kboard | 17:59 |
-!- blackburn1 [~blackburn@188.168.4.151] has joined #shogun | 18:01 | |
-!- blackburn [~blackburn@188.168.5.210] has quit [Ping timeout: 244 seconds] | 18:03 | |
heiko1 | damn it doesnt work | 18:04 |
heiko1 | strange error messages | 18:04 |
n4nd0 | too bad | 18:04 |
n4nd0 | shouldn't there be something built in git to extract a patch? | 18:04 |
heiko1 | n4nd0 does it work for you? | 18:05 |
n4nd0 | didn't try | 18:06 |
heiko1 | argh, always this messy stuff ;) | 18:07 |
yoo | n4nd0: yes I am testing most of the training machines and get used to those | 18:07 |
-!- blackburn [~blackburn@188.168.3.80] has joined #shogun | 18:08 | |
-!- blackburn2 [~blackburn@188.168.128.97] has joined #shogun | 18:08 | |
n4nd0 | heiko1: I have downloaded the tar.gz and it is empty... | 18:09 |
-!- yoo_ [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun | 18:09 | |
-!- blackburn1 [~blackburn@188.168.4.151] has quit [Ping timeout: 240 seconds] | 18:09 | |
heiko1 | blackburn, did the script work for you? | 18:10 |
heiko1 | I get tar error messages | 18:10 |
heiko1 | Think I am too stupid to use it | 18:10 |
n4nd0 | but isn't it the tar.gz empty? | 18:11 |
blackburn2 | heiko1: well I used it last year but no I didn't try yet | 18:11 |
heiko1 | blackburn, I see | 18:11 |
heiko1 | n4nd0 which download? | 18:11 |
heiko1 | I use the once from sergeys mail | 18:11 |
n4nd0 | aah I was checking the one from the link he sent | 18:12 |
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Ping timeout: 245 seconds] | 18:12 | |
n4nd0 | I don't have that one from the mail | 18:12 |
heiko1 | Ill send it | 18:12 |
-!- blackburn [~blackburn@188.168.3.80] has quit [Ping timeout: 244 seconds] | 18:12 | |
n4nd0 | thanks | 18:12 |
n4nd0 | it seems that I completely ignored that mail :O | 18:13 |
blackburn2 | heiko1: it must be the same | 18:13 |
heiko1 | blackburn, any updates on the release? I feel bad for having so little time currently ... :( | 18:14 |
blackburn2 | heiko1: I would say we are mostly ready | 18:14 |
heiko1 | blackburn, nice, I will add a few things today which I discussed with Arthur .... | 18:15 |
heiko1 | still the serialisation is missing right? | 18:15 |
blackburn2 | heiko1: I am missing its status to be honest | 18:15 |
blackburn2 | what is wrong with serialization? | 18:16 |
heiko1 | blackburn, memory leaking :) | 18:16 |
heiko1 | reference counts not being stored | 18:16 |
heiko1 | examples failing | 18:16 |
blackburn2 | super :D | 18:16 |
heiko1 | (leaking=failing in this case) | 18:16 |
yoo_ | have to go, see you later | 18:23 |
heiko1 | n4nd0, blackburn got it :) | 18:24 |
heiko1 | (the git extract) | 18:24 |
heiko1 | the script fails if the git log ist is empty | 18:24 |
heiko1 | this can be checked with git log --author= .... | 18:24 |
-!- yoo_ [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Quit: Page closed] | 18:25 | |
n4nd0 | heiko1: how did you do? | 18:25 |
n4nd0 | I get | 18:25 |
heiko1 | I had a typo in my name before | 18:25 |
n4nd0 | tar: *: Cannot stat: No such file or directory | 18:25 |
n4nd0 | tar: Exiting with failure status due to previous errors | 18:25 |
n4nd0 | test.tar.gz created | 18:25 |
heiko1 | yes | 18:25 |
n4nd0 | so we have to use our git name | 18:26 |
heiko1 | git log --author='Heiko Strathmann <heiko.strathmann@gmail.com>' | 18:26 |
heiko1 | if this works, plug it into the script | 18:26 |
heiko1 | oh | 18:27 |
heiko1 | and you have to delete line 35 in the script | 18:27 |
heiko1 | process.wait() | 18:27 |
heiko1 | this causes a deadlock | 18:27 |
blackburn2 | haha sorry I thought it is working | 18:27 |
heiko1 | Ill send an example to the list | 18:29 |
heiko1 | n4nd0, blackburn, see mail | 18:31 |
n4nd0 | heiko1: thanks! | 18:32 |
blackburn2 | thanks | 18:32 |
n4nd0 | nice | 18:33 |
n4nd0 | heiko1: are you going to upload that as code sample? | 18:33 |
heiko1 | n4nd0, just asking myself that :) | 18:34 |
heiko1 | there have been many emails on the list | 18:34 |
heiko1 | but I deleted all of them :) | 18:34 |
n4nd0 | haha | 18:34 |
n4nd0 | it looks like some people just submitted this | 18:35 |
n4nd0 | I think I will submit this | 18:35 |
n4nd0 | and some simpler example | 18:35 |
blackburn2 | probably nobody takes care of that | 18:35 |
n4nd0 | like the things we prepare for python_modular | 18:35 |
blackburn2 | :D | 18:35 |
heiko1 | oh n4nd0 | 18:35 |
heiko1 | sorry | 18:35 |
heiko1 | I did read your question wrong | 18:35 |
heiko1 | yes you can submit that | 18:35 |
heiko1 | We all did that last year | 18:36 |
n4nd0 | ok | 18:36 |
heiko1 | but where? | 18:36 |
heiko1 | (thought you were asking that) | 18:36 |
n4nd0 | heiko1: in melange | 18:36 |
n4nd0 | in the page of your project | 18:36 |
n4nd0 | just press edit there | 18:36 |
n4nd0 | heiko1: the script didn't create the tar.gz for me directly, I had to give it an output like /home/....../gsoc.tar.gz | 18:37 |
heiko1 | n4nd0, edit? | 18:37 |
heiko1 | ah found it | 18:38 |
-!- blackburn2 [~blackburn@188.168.128.97] has quit [Quit: Leaving.] | 18:39 | |
n4nd0 | exactly :) | 18:40 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 19:04 | |
-!- blackburn [~blackburn@188.168.5.212] has joined #shogun | 20:25 | |
-!- blackburn1 [~blackburn@83.234.54.19] has joined #shogun | 20:42 | |
-!- blackburn [~blackburn@188.168.5.212] has quit [Ping timeout: 256 seconds] | 20:44 | |
CIA-52 | shogun: Heiko Strathmann master * r3ec005c / (13 files in 5 dirs): Merge pull request #760 from karlnapf/master (+6 more commits...) - http://git.io/Bp_IEA | 20:53 |
CIA-52 | shogun: iglesias master * r68061ea / (2 files): * fix compile errors in libshogun hmsvm mosek examples - http://git.io/OoSJJw | 20:59 |
CIA-52 | shogun: Sergey Lisitsyn master * rc35b5ed / (2 files): Merge pull request #759 from iglesias/master - http://git.io/-LMimQ | 20:59 |
-!- heiko1 [~heiko@host86-176-252-249.range86-176.btcentralplus.com] has left #shogun [] | 21:22 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 21:34 | |
-!- blackburn1 [~blackburn@83.234.54.19] has quit [Quit: Leaving.] | 21:46 | |
--- Log closed Wed Aug 29 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!