Open in new window / Try shogun cloud
--- Log opened Sun May 27 00:00:41 2012
@sonney2kblackburn, I added some stuff and at least now kernel.kernel() works00:02
blackburnwhat's that?00:02
@sonney2kgives you k[i,j] :D00:03
@sonney2kI don't udnerstand the error though00:03
blackburninit one?00:03
@sonney2kthere is no init00:03
@sonney2kthe error is in get_kernel_matrix()00:04
@sonney2kand that is just calling kernel(i,j)00:04
blackburnah so what is the error?00:05
@sonney2kRuntimeError: maximum recursion depth exceeded while calling a Python object00:05
@sonney2kbut at least it still says returning kernel matrix of size 5x500:07
@sonney2kso it enters get_kernel_matrix00:07
blackburnsonney2k: looks like known bug of swig?00:07
@sonney2kwait not true00:08
@sonney2kthat was for the get-kernel_matroix of the linear kernel00:08
@sonney2kblackburn, ?00:08
blackburnI googled - some issues00:08
@sonney2ksure but why does kernel(i,j) work?00:08
blackburndifferent way of call I'd say00:08
@sonney2khmmhh, one is using typemaps...00:09
@sonney2kit returns SGMatrix00:09
@sonney2kkernel(i,j) just double00:09
@sonney2klet me try sth00:09
@sonney2kblackburn, I now created a simple get_km()00:12
@sonney2kreturning some sgmatrix00:12
@sonney2kdirectly from directorkernel00:12
@sonney2kif that fails we will see00:12
@sonney2kcompile time is much higher with directors...00:12
@sonney2kwrapper is much bigger00:12
@sonney2kblackburn, yes that works00:13
blackburnso what is the reason?00:14
@sonney2kblackburn, no idea00:19
@sonney2k0.7 sec for a 1000x1000 matrix w/ shogun00:19
@sonney2k15 sec with directors from python00:19
blackburnsonney2k: what is the reason between get_km and get_kernel_matrix?00:20
@sonney2kblackburn, again no idea00:20
@sonney2kbtw the example crashes at the end00:20
@sonney2kblackburn, well I set threads=100:20
@sonney2k(of course)00:20
@sonney2kit doesn't crash with gdb00:21
@sonney2kblackburn, also doesn't crash with valgrind00:22
blackburnyes that happens sometimes00:22
@sonney2kblackburn, do you think gsomix can find out what is wrong?00:23
blackburngive a chance :)00:23
@sonney2kI will mail him the example then00:23
CIA-113shogun: Soeren Sonnenburg master * rd8a7270 / (2 files): add some helper functions for director kernel -
@sonney2kuhh ohh00:29
@sonney2kI have to sleep00:29
-!- blackburn [~blackburn@] has quit [Ping timeout: 240 seconds]03:36
-!- gsomix [~gsomix@] has joined #shogun08:33
gsomixsonney2k, I got mail. I got mail. I can try to solve issue in evening09:15
gsomixsorry, need to go09:15
-!- ZiYuan [~ZiYuan@] has joined #shogun10:52
-!- ZiYuan [~ZiYuan@] has quit [Client Quit]10:52
-!- ZiYuan [~ZiYuan@] has joined #shogun10:53
-!- n4nd0 [] has joined #shogun10:53
-!- ZiYuan [~ZiYuan@] has quit []10:59
-!- ZiYuan [~ZiYuan@] has joined #shogun10:59
-!- ZiYuan [~ZiYuan@] has quit [Client Quit]11:00
-!- n4nd0 [] has quit [Quit: Lost terminal]11:11
-!- blackburn [~blackburn@] has joined #shogun11:27
-!- n4nd0 [] has joined #shogun11:40
-!- blackburn [~blackburn@] has quit [Ping timeout: 260 seconds]11:50
-!- gsomix [~gsomix@] has quit [Quit: Ex-Chat]14:14
-!- blackburn [~blackburn@] has joined #shogun19:14
-!- n4nd0 [] has quit [Quit: leaving]19:34
-!- blackburn [~blackburn@] has quit [Quit: Leaving.]19:36
-!- blackburn [~blackburn@] has joined #shogun19:37
-!- gsomix [~gsomix@] has joined #shogun20:16
gsomixfuf, I have returned to dorm.20:16
gsomixtime to work20:16
gsomixsonney2k, blackburn,  hi20:17
-!- n4nd0 [] has joined #shogun20:35
blackburnn4nd0: hey20:56
blackburnI tweaked it up to 97.82 :D20:56
n4nd0blackburn: hey man, how is it going?20:56
n4nd0blackburn: wow nice tweak then :)20:56
n4nd0blackburn: what value were you aiming at?20:57
blackburnno idea I just needed to overwhelm 97.00%20:57
n4nd0objective achieved then20:57
blackburnwell.. before (in paper for that russian journal) I stated I 'can' recognize with 97.43 accuracy20:57
blackburnso objective is rather 97.4320:57
blackburnbut achieved yeah20:58
n4nd0congrats ;)20:58
blackburnhowever I have not too much time now to get thesis itself done20:59
n4nd0blackburn: do you have to write it completely?21:02
blackburnyes needs to be 100% done before 7th June21:03
n4nd0blackburn: you write it in Russian, right?21:08
n4nd0:( it would be cool to read it otherwise21:08
blackburnhah nothing special21:09
blackburnI have nothing but HOG and SVM here21:09
blackburnCrammer-Singer SVM + Homogeneous Kernel Map + HOG21:09
n4nd0blackburn: I insist, probably interesting to read ;)21:11
blackburngood thing being russian is direct access to vapnik's stuff21:12
-!- n4nd0 [] has quit [Quit: leaving]21:15
gsomixsonney2k, around21:41
gsomixhmm, nevermind :)21:42
@sonney2kgsomix, any news?23:05
gsomixsonney2k, yep.23:05
gsomixget_kernel_matrix works.23:05
blackburnsonney2k: to redeclare virtuals is the key :D23:06
gsomixblackburn, tnx for fast description. :)23:08
@sonney2kno more crashes?23:10
gsomixsonney2k, hmm, I have not noticed more. but work in progress, need 15mins.23:11
blackburnsonney2k: what do you think about BSVM?23:27
@sonney2kblackburn, don't think it is worth it23:31
@sonney2kbut I haven't read the paper/mail :)23:31
@sonney2kgmnpsvm should be doing better23:31
@sonney2kwe need something massively multiclass23:32
blackburnMDM is cool yes but I do not think BSVM is slower23:32
blackburnsonney2k: ok then another thing - strprtool has 5 solver for gmnp23:33
blackburnwhy do we have only mdm?23:34
blackburndo not say ask vojtech :D23:39
gsomixsonney2k, ok, works. but this requires cleaning up23:55
gsomixbtw, feats_train.parallel.set_num_threads(2) -> segfault23:56
gsomixsonney2k, can I commit/push it tomorrow?23:59
gsomixtired, time to sleep =__=23:59
--- Log closed Mon May 28 00:00:41 2012