Open in new window / Try shogun cloud
--- Log opened Thu Feb 16 00:00:19 2012
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]00:20
CIA-18shogun: Soeren Sonnenburg master * r3649f3b / (31 files in 9 dirs):03:25
CIA-18shogun: Merge pull request #367 from karlnapf/master03:25
CIA-18shogun: A draft for training on fixed kernel matrices/data in general - http://git.io/qv89yQ03:25
CIA-18shogun: Soeren Sonnenburg master * r3e3db13 / (33 files in 14 dirs):03:25
CIA-18shogun: add linear least squares and ridge regression03:25
CIA-18shogun: This add linear ridge regression and a convenience class03:25
CIA-18shogun: CLeastSquaresRegression calling CLinearRidgeRegression with03:25
CIA-18shogun: regularization parameter tau=0. To not cause confusion KRR is03:25
CIA-18shogun: renamed to KernelRidgeRegression throughout examples/code. - http://git.io/pQJ0OA03:25
CIA-18shogun: Soeren Sonnenburg master * r0338587 / examples/undocumented/python_modular/regression_linear_ridge_modular.py : remove unused gaussian kernel from example - http://git.io/Sf2o5w03:25
CIA-18shogun: Soeren Sonnenburg master * rc612a4c / (26 files in 7 dirs):03:28
CIA-18shogun: Merge pull request #373 from karlnapf/master03:28
CIA-18shogun: a bunch of improvements (+27 more commits...) - http://git.io/0IzzXg03:28
-!- shogun-buildbot_ [~shogun-bu@7nn.de] has quit [Quit: buildmaster reconfigured: bot disconnecting]03:35
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun03:36
-!- nageshp [~nagesh@triband-mum-120.61.0.77.mtnl.net.in] has joined #shogun04:06
-!- nageshp [~nagesh@triband-mum-120.61.0.77.mtnl.net.in] has left #shogun []04:12
shogun-buildbotbuild #316 of ruby_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/ruby_modular/builds/31605:08
-!- dfrx [~f-x@inet-hqmc05-o.oracle.com] has joined #shogun05:36
shogun-buildbotbuild #344 of octave_modular is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/344  blamelist: viktor.gal@maeth.com, heiko.strathmann@gmail.com, fernando.iglesiasg@gmail.com, slianoglou@gmail.com, blackburn91@gmail.com, sonne@debian.org06:25
-!- n4nd0 [82ed3d7d@gateway/web/freenode/ip.130.237.61.125] has joined #shogun08:30
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]08:51
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:02
CIA-18shogun: Soeren Sonnenburg master * r32bfe20 / examples/undocumented/octave_modular/regression_kernel_ridge_modular.m : fix missing rename KRR -> KernelRidgeRegression - http://git.io/rcpScw09:24
-!- n4nd0 [82ed3d7d@gateway/web/freenode/ip.130.237.61.125] has quit [Quit: Page closed]10:02
shogun-buildbotbuild #345 of octave_modular is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/34510:03
wikingdoh, why SGVector does not have an iterator class around it... ;(11:24
sonne|workwiking: SGVector used to be just a float ptr + length for storing data...11:26
wikingwell i just kind of like wish there'd be an iterator struct for it now :P11:27
sonne|workI wish we introduced proper reference counts in SGVector & co11:29
sonne|workneed that now but at some point 'every byte counted'11:29
-!- n4nd0 [~n4nd0@n145-p102.kthopen.kth.se] has joined #shogun12:10
-!- dfrx [~f-x@inet-hqmc05-o.oracle.com] has quit [Quit: Leaving.]12:14
wikingsonne|work: are there any rules of adding BSD licensed code to shogun?13:00
wikingi mean most of the code is derived from a bsd licensed file that i'd like to commit soon, that's why i'm asking13:01
sonne|workwiking: anything GPL compatible is fine13:24
wikingcool13:24
-!- stratis [9232285a@gateway/web/freenode/ip.146.50.40.90] has joined #shogun14:14
stratishi!!!14:14
stratisto all of you14:14
stratisi m a new user of shogun14:14
stratisand I am working with the matlab version14:14
stratisso, i cant seem to find the function for calculating the area and the precision-recall curve14:15
stratisor at least how to get the return distances from the svm14:15
stratisi have a trained a multi-class svm14:16
sonne|workstratis: hi - the perf measures are only available in modular interfaces14:16
stratisoh ok14:16
stratisso14:16
stratisi have a multi class svm14:16
stratiscan i at least some how get14:16
sonne|workstratis: IIRC you can do get_classifier with an index14:16
stratisthe distances for the samples14:16
sonne|worksg('get_classifier',0) or so14:16
stratisok i ll try that14:17
stratiswhen i try14:20
stratissg('get_classifier',0)14:20
stratisit says14:20
stratis[bias, weights]=sg('get_classifier', [index in case of MultiClassSVM])14:20
stratisthe same when I don't use an index14:20
stratishow can I see the variables stored in "sg" ?14:20
sonne|workwell you need to do14:24
sonne|work [bias, weights]=sg('get_classifier',0)14:24
sonne|work:)14:24
-!- stratis [9232285a@gateway/web/freenode/ip.146.50.40.90] has quit [Ping timeout: 245 seconds]14:27
-!- n4nd0 [~n4nd0@n145-p102.kthopen.kth.se] has quit [Quit: Leaving]15:29
wikingmmm why isn't it possible to define a class that has no default constructor?15:48
sonne|workwiking: for serialization we need that16:07
wikingmmm that'll require then some amount of error checking now :P16:23
wikingsonne|work:  if i define an enum within a class would that be revealed to the modular API?17:05
-!- dfrx [~f-x@inet-hqmc04-o.oracle.com] has joined #shogun18:06
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has joined #shogun19:20
-!- dfrx [~f-x@inet-hqmc04-o.oracle.com] has left #shogun []19:49
@sonney2kwiking, no nested classes /enums etc won't work20:14
wikinghow can i have constants ?20:15
wikingthat one could use from modular as well20:15
@sonney2kuse enums20:22
@sonney2kjust not nested within a class20:22
wikingmmmhm20:27
wikingand how do i call it in python20:27
wikingbecause i've tried to call for LibSVM LIBSVM_C_SVC20:28
wikingbut python was crying that it's not defined20:28
wikingdoh, what's this: TypeError: not a numpy vector of appropriate type20:38
wiking:>20:38
@sonney2kwiking, not sure what you are trying to do21:04
@sonney2kjust use the enum as usual - it is just an int also in python21:04
wikingwell just that i'm creating a new preprocessor21:04
wikingand i need to use some enums21:04
wikingand i was wondering if those enums are going to be defined in the modular part as well21:04
wikingsince now as far as i can see21:05
wikingi can just use in python only the corresponding int value of the enum21:05
wikingbut not the macro for it21:05
wikingso in case of LibSVM there's a 2 element enum21:05
wikingand i was trying in python modular part whether i could use LIBSVM_C_SVC21:05
wikingor how is it possible at all to use that enum21:06
shogun-buildbotbuild #151 of nightly_none is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/15121:08
@sonney2kwiking, you should just see the enum in python_modular21:09
@sonney2kexaclty like in C++21:09
wikingreally21:24
wikingand you would call it like LIBSVM_C_SVC21:24
wikingbecause if you try print LIBSVM_C_SVC21:24
wikingit'll say that no such thing exist... in python21:24
shogun-buildbotbuild #150 of nightly_all is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/15021:25
@sonney2kwiking, yes really - look at some of the examples, e.g. classifier_liblinear_modular.py21:33
@sonney2kL2R_L2LOSS_SVC_DUAL is an enum21:33
@sonney2kalright n821:33
shogun-buildbotbuild #137 of nightly_default is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/13721:54
-!- blackburn [~qdrgsm@188.168.5.29] has joined #shogun21:55
wikingblackburn: maaan22:11
blackburnyeah?22:11
wikingi'm about to finish the homkermap22:11
blackburnhuh22:11
blackburnyou are fast22:11
wikingjust trying to test some shit with it22:11
wikingbecause for sure i've fucked up some stuff22:11
wikingbut yeah i still need to clean the api22:11
wikingbecause now the constructor is like this: CHomogeneousKernelMap (HomogeneousKernelType kernel, HomogeneousKernelMapWindowType wType, float64_t gamma, uint64_t order, float64_t period);22:12
wikingbtw what's the consensus22:12
wikingfor the parameters of a class22:12
blackburnhmm22:12
wikingthere should be getter/setter function pairs... ? for each of them?22:13
blackburnyeah would be nice22:13
wikingaha22:13
wikingbtw22:13
wikingm_parameters->add thingy22:14
wikingis for?22:14
wikingi didn't get that yet...22:14
blackburnserialization stuff22:14
wikingwhen does serialization happen?22:14
wikingso when is a class written to file?22:14
blackburnhmm when you call it in python22:14
blackburnyes22:14
blackburnor octave or etc22:14
wikingaha so modular22:15
blackburnsure, static sucks22:15
wikingso each private variables should be serialized, right? :)22:15
blackburnsure22:16
wikingmmhm22:16
blackburnwell I don't think it is so necessary for onw22:16
blackburnabout your constructor22:16
wikingwell i'm trying to test the thing in python22:16
blackburnbecause now the constructor is like this: CHomogeneousKernelMap (HomogeneousKernelType kernel, HomogeneousKernelMapWindowType wType, float64_t gamma, uint64_t order, float64_t period);22:16
wikingso i'll have to do it22:16
blackburnis HomogenousKernelType an enum?22:16
wikingyep22:16
blackburnsame for window type?22:16
wikingyep22:17
wikingif i call m_parameters->add(&m_table, "", "");22:17
wikingwhere SGVector<float64_t> m_table;22:17
wikingwill it know how to serialize it/22:17
wiking?22:17
blackburnyes22:17
blackburnadd_vector IIRC22:17
blackburnor simply add if it works22:17
wikingok let's see22:21
wikingrecompiling ;)22:21
wikinghahaha22:21
wikinghow do you serialize an enum? :)22:21
wikingshould i cast it into int?22:21
blackburnehmm22:21
wikingSG_ADD((machine_int_t*) &liblinear_solver_type, "liblinear_solver_type",22:22
wiking                        "Type of LibLinear solver.", MS_NOT_AVAILABLE);22:22
wikingthat's in liblinear22:22
blackburnyeah22:22
blackburndo the same22:22
wikingdoh22:24
wikingadd_vector is not working as i thought it would22:24
wikingok now...22:26
wikingrecompiling :P22:26
wikingok one more question22:26
wikingso in an SGMatrix22:26
wikingthe number of rows are the number of examples? or that the number of columns?22:27
blackburnre22:35
blackburnwiking: ehmm what is the matrix you are talking about?22:36
wikingimplementing the apply_to_feature_matrix22:36
wikingso there i'm returning a SGMatrix<float64_t>22:36
wikingand now i have22:36
wikingCSimpleFeatures<float64_t>* simple_features = (CSimpleFeatures<float64_t>*)features;22:36
wikingint32_t num_vectors = simple_features->get_num_vectors ();22:36
wikingint32_t num_features = simple_features->get_num_features ();22:36
wikingSGMatrix<float64_t> result (num_vectors, num_features*(2*m_order+1));22:36
wikingand i wonder if the arguments of result should be swapped or not22:37
blackburnfeature matrix contains feature vectors row-wise22:37
blackburnso dim = rows count22:37
blackburnand number of vectors = cols count22:37
wikingahaaa22:38
wikingso it should b swapped in that case22:38
wikingor wait22:38
blackburnyes, should be22:38
wikingyey recompile ;P22:40
wikingfuckthisshiat22:40
blackburndid you simply integrate their code?22:42
blackburnbtw I've read paper recently, now understand things much better22:43
wikingyep basically it's a wrapping22:46
wikingi just want to finish it up22:47
wikingbecause then i can start adding new mappings22:47
wikingor at least to test new mappings22:47
wikingfor other kernels..22:47
blackburnyeah22:48
blackburnwiking: btw thanks for your efforts22:57
blackburnokay power off now :)22:58
blackburnsee you22:58
-!- blackburn [~qdrgsm@188.168.5.29] has quit [Quit: Leaving.]22:58
--- Log closed Fri Feb 17 00:00:19 2012