Open in new window / Try shogun cloud
--- Log opened Sat Jun 02 00:00:41 2012
heiko3heiko2: so did you submit?00:01
heiko2its done :)00:01
heiko2lots of work00:01
heiko3and exam?00:01
heiko2thanks :)00:01
heiko2exam was nice00:01
heiko2also by arthur btw :)00:01
-!- heiko1 [] has quit [Ping timeout: 260 seconds]00:01
heiko3hahah exam by supervisor??00:01
heiko2yes, funny isnt it? :)00:01
heiko3that should be unfair a little00:01
heiko2oh, my girl just returned home, gotta go00:02
heiko2no, he gave me no hints ;)00:02
-!- heiko3 is now known as blackburn00:02
-!- heiko2 [] has quit [Ping timeout: 260 seconds]00:06
n4nd0good night!01:11
-!- n4nd0 [] has quit [Quit: leaving]01:13
-!- blackburn [~blackburn@] has left #shogun []02:19
CIA-9shogun: Soeren Sonnenburg master * r78f9ac6 / src/configure : fix octave detection for octave 3.6.x (Closes issue #383) -
CIA-9shogun: Soeren Sonnenburg master * r8a15088 / src/configure : fix ruby detection in configure -
CIA-9shogun: Soeren Sonnenburg master * r7344caa / (2 files in 2 dirs): add empty define for ignore in class list -
-!- n4nd0 [] has joined #shogun09:03
-!- n4nd0 [] has quit [Quit: leaving]11:25
-!- blackburn [~blackburn@] has joined #shogun12:44
-!- n4nd0 [] has joined #shogun14:11
-!- blackburn [~blackburn@] has quit [Ping timeout: 244 seconds]14:12
-!- n4nd0 [] has quit [Quit: leaving]15:12
-!- blackburn [~blackburn@] has joined #shogun18:59
-!- zxtx [] has quit [Remote host closed the connection]20:17
CIA-9shogun: Soeren Sonnenburg master * rb921a7d / (2 files in 2 dirs): add proper copyrights in Director* stuff -
CIA-9shogun: Soeren Sonnenburg master * r80b8013 / (src/shogun/lib/Hash.cpp src/shogun/lib/Hash.h): fix compile warning in md5sum -
CIA-9shogun: Soeren Sonnenburg master * rf0a2a12 / (5 files in 3 dirs): add empty liblinear regressor -
-!- zxtx [] has joined #shogun20:24
-!- zxtx [] has quit [Remote host closed the connection]20:27
-!- zxtx [] has joined #shogun20:31
-!- zxtx [] has quit [Remote host closed the connection]20:37
-!- gsomix [~gsomix@] has joined #shogun21:18
gsomixmechanics is full of pain21:19
gsomixdepressive lagrangians, suffering hamiltonians and many integrals of fear21:21
gsomixthey come to me in dreams21:24
gsomixand organize an orgy of terror21:25
blackburnwell I still feel excited about Lagrange's stuff21:27
gsomixblackburn, my favorite subject is multi-dimensional oscillators21:38
-!- KMcQuisten [d8338942@gateway/web/freenode/ip.] has joined #shogun22:15
KMcQuistenAfternoon Sergey22:15
blackburn(its me)22:16
KMcQuistenI'd guessed :)22:16
KMcQuistenSo, any ideas as to what the problem is?22:16
blackburnokayy let me try to guess22:16
blackburncould you please first say what is classifier you do use?22:17
KMcQuistenI'm working with the MKLRegression22:17
blackburnlet me check the code22:17
blackburnokay and the problem appears only with combined features, right?22:18
KMcQuistenTo the best of my knowledge yes.  It worked fine when I gave a single kernel to LibSVR, but I dind't try putting a single kernel into MKLRegression22:20
blackburnlet me try to come with some reproducing example22:20
blackburnKMcQuisten: should it look similar to python_modular/
-!- gsomix [~gsomix@] has quit [Quit: Ex-Chat]22:26
-!- gsomix [~gsomix@] has joined #shogun22:26
KMcQuistenYes, essentially, but using StringWordFeatures and the CommWordStringKernel in the Comined objects.22:26
blackburnI changed to regression - works22:27
blackburnlets see if stringfeatures change the game22:28
blackburnKMcQuisten: can I ask you to modify some code meanwhile? some simple thing to debug22:31
blackburnI just want to know where it wants to stop22:31
KMcQuistenJust let me know where I need to modify things22:32
blackburnreplace SG_ERROR with SG_WARNING and try to recompile22:32
blackburnah and %d after "is" word in msg (I forgot that)22:33
KMcQuistenOk, done.  Now I'll recompile22:34
KMcQuistenOk, here's what I got.  It looks like we have an index issue:22:38 RuntimeWarning: [WARN] Requested feature vector with index 50 while total num is 50   testsvm.train(combfeats) Traceback (most recent call last):   File "", line 66, in <module>     testsvm.train(combfeats) SystemError: [ERROR] Trying to access string[50] but num_str=5022:38
blackburnonly 50, but no more, right?22:40
KMcQuistenYes.  I only put in 50 data points as a toy set.  Let me check with the whole dataset22:40
blackburnwait wait22:40
blackburnERROR is raised somewhere22:41
KMcQuistenYes.  I think it's trying to access index 50 on a vector that's 0 indexed with 50 items in ti22:41
blackburnno, that can go for bigger values (it stopped because of the SG_ERROR)22:42
KMcQuistenHow is that possible?  If you have a vector of length 50 its highest index is 49.  I get the same kind of error with the full set of 2430 sequences22:44
blackburnI mean it can go for 51, etc22:44
KMcQuistenSystemError: [ERROR] Trying to access string[2430] but num_str=243022:45
KMcQuistenRIght.  I know i'm not limited to 50 data points22:45
blackburnok let me find this ERROR22:45
blackburnI tried to combine string and real features btw22:45
blackburnKMcQuisten: could you please replace SG_ERROR with SG_WARNING at line 348 too?22:48
KMcQuistenI will also try and comine the real and the StringWord22:51
KMcQuistenand see if it works for me too22:51
blackburnwhat do you combine?22:52
blackburnstring and string?22:52
KMcQuistenStringWord and StringWord.   DNA sequences mapped into the unsigned 16bit integers.  In python, it's StringWordFeatures, but in the C++ it's StringFeatures<uint_16>22:53
KMcQuistennot StringFeatures<char>22:53
blackburnKMcQuisten: okay I managed to reproduce22:56
blackburntwo StringWord feautres22:56
KMcQuistendoes it work for you?22:59
blackburnno dies on the same error22:59
KMcQuistenThat makes me a happy cmaper22:59
KMcQuistenat least it's reproduceable23:00
blackburnyeah I'll try to figure out what is wrong23:00
KMcQuistenthanks so much!23:00
KMcQuistenI've got plenty of other work and writing to do, so i'll just camp here for a while23:01
blackburnok sure23:01
blackburnI'll let you know on any progress23:02
KMcQuistenThank you23:02
gsomixgood night guys23:11
blackburnKMcQuisten: hmm as I thought it stopped at 18323:23
-!- gsomix [~gsomix@] has quit [Ping timeout: 260 seconds]23:25
KMcQuistenWhat does that indicate?23:26
blackburnsome code adds number of feature vectors somehow I guess23:28
blackburnbut it is really strange23:28
KMcQuistenWhat's the significance of the 92 * 2?23:28
blackburnnumber of feature vectors in test example is 9223:28
blackburnand two string word features23:29
blackburnKMcQuisten: good thing I detected a bug :D23:30
KMcQuistenIndeed.  I changed that error to a warning as you asked before and recompiled, got the following:23:31 RuntimeWarning: [WARN] Requested feature vector with index 2430 while total num is 243023:32 RuntimeWarning: [WARN] Requested feature vector with index 2431 while total num is 243023:32 RuntimeWarning: [WARN] Requested feature vector with index 2432 while total num is 243023:32 RuntimeWarning: [WARN] Trying to access string[2432] but num_str=243023:32
KMcQuistenSegmentation fault (core dumped)23:33
KMcQuistenDoes the Combinedfeats object adress all the features as thogh they are one long list?23:33
blackburnsomething like that but I do not understand what is the difference between real+string and string+string23:34
CIA-9shogun: Sergey Lisitsyn master * r3e1dc47 / src/shogun/regression/svr/SVRLight.cpp : Fixed wrong free in SVRLight -
blackburnKMcQuisten: okay I'm pretty near I think23:57
blackburnsvr bug23:57
--- Log closed Sun Jun 03 00:00:41 2012