--- Log opened Mon May 28 00:00:41 2012 | ||
gsomix | sonney2k, blackburn good night | 00:04 |
---|---|---|
blackburn | nite | 00:04 |
-!- blackburn [~blackburn@31.28.59.65] has quit [Ping timeout: 244 seconds] | 02:06 | |
-!- cheng [80fa500f@gateway/web/freenode/ip.128.250.80.15] has joined #shogun | 02:48 | |
-!- cheng [80fa500f@gateway/web/freenode/ip.128.250.80.15] has quit [Ping timeout: 245 seconds] | 05:33 | |
-!- gsomix [~gsomix@188.168.5.87] has quit [Quit: Ex-Chat] | 06:38 | |
-!- gsomix [~gsomix@188.168.5.87] has joined #shogun | 06:39 | |
@sonney2k | gsomix, nice ok | 06:41 |
gsomix | sonney2k, good morning :) | 06:42 |
-!- cheng [80fa500f@gateway/web/freenode/ip.128.250.80.15] has joined #shogun | 06:48 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 08:02 | |
-!- gsomix [~gsomix@188.168.5.87] has quit [Ping timeout: 244 seconds] | 08:06 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 240 seconds] | 08:24 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has joined #shogun | 08:38 | |
-!- uricamic [~uricamic@2001:718:2:1634:d981:6e8d:30f8:8a27] has joined #shogun | 08:48 | |
-!- cheng [80fa500f@gateway/web/freenode/ip.128.250.80.15] has quit [Ping timeout: 245 seconds] | 09:00 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 245 seconds] | 09:55 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has joined #shogun | 10:11 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 245 seconds] | 10:20 | |
-!- dfrx [~f-x@inet-hqmc04-o.oracle.com] has joined #shogun | 10:41 | |
-!- dfrx [~f-x@inet-hqmc04-o.oracle.com] has quit [Quit: Leaving.] | 10:42 | |
-!- heiko [~heiko@host86-177-178-36.range86-177.btcentralplus.com] has joined #shogun | 11:10 | |
-!- heiko1 [~heiko@host86-180-159-191.range86-180.btcentralplus.com] has joined #shogun | 11:25 | |
-!- heiko [~heiko@host86-177-178-36.range86-177.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 11:26 | |
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun | 11:29 | |
-!- blackburn [~blackburn@31.28.59.65] has quit [Client Quit] | 11:30 | |
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun | 11:32 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has joined #shogun | 11:42 | |
n4nd0 | sonne|work, hey! are you ok if we just link to our blog for our weekly reports in the mailing list? | 11:55 |
-!- blackburn [~blackburn@31.28.59.65] has quit [Ping timeout: 260 seconds] | 12:41 | |
-!- heiko [~heiko@host86-174-151-230.range86-174.btcentralplus.com] has joined #shogun | 12:45 | |
-!- heiko1 [~heiko@host86-180-159-191.range86-180.btcentralplus.com] has quit [Ping timeout: 260 seconds] | 12:48 | |
-!- heiko [~heiko@host86-174-151-230.range86-174.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 13:29 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has quit [Quit: Ex-Chat] | 14:12 | |
-!- heiko [~heiko@host86-174-151-230.range86-174.btcentralplus.com] has joined #shogun | 14:57 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has joined #shogun | 16:35 | |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 17:25 | |
n4nd0 | sonney2k, around? | 17:51 |
n4nd0 | sonne|work, ^? | 17:51 |
-!- uricamic [~uricamic@2001:718:2:1634:d981:6e8d:30f8:8a27] has quit [Quit: Leaving.] | 18:00 | |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 244 seconds] | 18:31 | |
-!- gsomix [~gsomix@83.234.54.199] has joined #shogun | 18:38 | |
gsomix | hi | 18:38 |
-!- heiko [~heiko@host86-174-151-230.range86-174.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 18:44 | |
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun | 18:48 | |
gsomix | blackburn, yo | 18:58 |
blackburn | hey | 18:59 |
-!- heiko [~heiko@host86-177-178-176.range86-177.btcentralplus.com] has joined #shogun | 19:03 | |
@sonney2k | gsomix, don't forget about the weekly report... | 19:04 |
gsomix | sonney2k, thanks, I remember :) | 19:05 |
@sonney2k | gsomix, heh | 19:05 |
@sonney2k | gsomix, so what is the directorkernel state now? | 19:05 |
@sonney2k | do you have a patch I can try? | 19:05 |
gsomix | sonney2k, compiling... | 19:06 |
@sonney2k | ok | 19:07 |
@sonney2k | n4nd0, just post an email to the ML with a pointer to your report for the week on your blog. | 19:10 |
* sonney2k checks whose report is missing | 19:14 | |
@sonney2k | looks like gsomix's n4nd0's and wikings... | 19:15 |
gsomix | sonney2k, check github, please. | 19:19 |
-!- heiko [~heiko@host86-177-178-176.range86-177.btcentralplus.com] has quit [Ping timeout: 245 seconds] | 19:27 | |
CIA-113 | shogun: Evgeniy Andreev master * r1a1a30c / (2 files in 2 dirs): added missing virtual function to DirectorKernel - http://git.io/i5huug | 19:41 |
CIA-113 | shogun: Evgeniy Andreev master * r92b1eec / src/shogun/kernel/Kernel.h : deleted spaces - http://git.io/WFTOOw | 19:41 |
CIA-113 | shogun: Soeren Sonnenburg master * r2dc985c / (3 files in 2 dirs): Merge pull request #557 from gsomix/director_kernel - http://git.io/BYe31g | 19:41 |
gsomix | sonney2k, works? =___= | 19:44 |
@sonney2k | gsomix, why is get_kernel_row() not needed? | 19:45 |
@sonney2k | I mean it is also defined virtula in ckernel? | 19:45 |
@sonney2k | gsomix, compiling... | 19:45 |
gsomix | sonney2k, oops. | 19:45 |
@sonney2k | gsomix, ? | 19:47 |
@sonney2k | not needed or needed? | 19:47 |
@sonney2k | gsomix, why are virtual functions needed? | 19:47 |
gsomix | sonney2k, needed. just a moment. | 19:47 |
gsomix | sonney2k, oh. I really dont know, but I have some hypothesis. | 19:48 |
-!- n4nd0 [~n4nd0@s83-179-44-135.cust.tele2.se] has joined #shogun | 19:48 | |
@sonney2k | gsomix, then tell me about it :D | 19:49 |
gsomix | it seems, that swig "director" (verb, hehe) virtual functions from CKernel. | 19:49 |
gsomix | but... there is table with DirectorKernel's virtual methods. | 19:50 |
gsomix | and there are not virtual methods from CKernel in this table. | 19:51 |
@sonney2k | gsomix, ok | 19:51 |
gsomix | in my "home" sample this situation is throwing exception | 19:51 |
@sonney2k | gsomix, btw the has_features check is a bit questionable | 19:52 |
@sonney2k | if one uses e.g. only numpy | 19:52 |
@sonney2k | gsomix, so why did the endless recursion happen w/ init? | 19:52 |
@sonney2k | n4nd0, (reposting what I said above: just post an email to the ML with a pointer to your report for the week on your blog.) | 19:53 |
n4nd0 | sonney2k, yeah, I just saw it in the logs, thank you :) | 19:53 |
gsomix | sonney2k, get_num_vec_rhs/lhs in methods is reason | 19:53 |
gsomix | sonney2k, e.g in get_kernel_matrix. | 19:54 |
@sonney2k | gsomix, well I couldn't train w/o doing dkernel.init(lhs,rhs) | 19:54 |
@sonney2k | I would also want to set num_lhs /rhs | 19:54 |
@sonney2k | and do without even setting features | 19:55 |
gsomix | sonney2k, mmm? | 19:55 |
gsomix | just override has_features | 19:55 |
@sonney2k | gsomix, well if I use a numpy array as in the example I send to you | 19:55 |
@sonney2k | gsomix, that is too slow | 19:55 |
@sonney2k | it will be checked all the time | 19:55 |
@sonney2k | gsomix, I think either adding a flag to constructor | 19:56 |
@sonney2k | yeah well that would be best | 19:56 |
@sonney2k | gsomix, add some bool to constructor no_features_required | 19:56 |
@sonney2k | if false - do like we do now | 19:56 |
@sonney2k | and otherwise return true on has_features | 19:57 |
gsomix | aha | 19:57 |
@sonney2k | gsomix, btw 7 seconds for director kernel here for 1000x1000 matrix | 19:57 |
@sonney2k | 0.7 for shogun's | 19:57 |
gsomix | sonney2k, aha, slow... | 19:57 |
blackburn | 10? is ok | 19:58 |
@sonney2k | (for 1 thread) | 19:58 |
@sonney2k | for 2 threads 0.4 s | 19:58 |
@sonney2k | :) | 19:58 |
@sonney2k | 10x is ok | 19:58 |
@sonney2k | I think so too | 19:58 |
@sonney2k | considering how complex the thing is | 19:58 |
blackburn | development can take 10000000x | 19:58 |
@sonney2k | gsomix, question would be if this director stuff also works from octave etc | 19:59 |
gsomix | sonney2k, ok, I'll check it. | 19:59 |
blackburn | so being able to prototype even slow kernel is soo awesome | 19:59 |
@sonney2k | blackburn, we now have to decide which things we want to be director-enabled | 19:59 |
blackburn | features, kernels, distances | 19:59 |
blackburn | multiclass strategies? | 19:59 |
blackburn | reject strategies | 19:59 |
blackburn | preprocessors | 20:00 |
blackburn | machines | 20:00 |
blackburn | svm especially | 20:00 |
blackburn | may be some splitting strategy or so | 20:01 |
blackburn | for xval | 20:01 |
blackburn | loss for SO | 20:01 |
@sonney2k | blackburn, so the way to proceed would be to have an extra class for each thing right? | 20:01 |
blackburn | yes I think so | 20:01 |
@sonney2k | DirectorKernelMachine etc? | 20:01 |
@sonney2k | blackburn, not so sure about sparse etc features... | 20:02 |
blackburn | ok so DirectorDotFeatures, DirectorKernel, DirectorDistance, DirectorMulticlassStrategy, DirectorRejectStrategy, DirectorKernelMachine, DirectorLinearMachine, DirectorStructuredLoss | 20:02 |
@sonney2k | blackburn, uhh ohh | 20:03 |
@sonney2k | for dotfeatures we have no way to support the api | 20:03 |
blackburn | hm hmhm | 20:04 |
@sonney2k | and I see lots of problems with protected etc... | 20:04 |
@sonney2k | n4nd0, ok so where is your email then? | 20:04 |
* sonney2k is missing wiking :( | 20:04 | |
blackburn | we need DirectorDotFeatures | 20:04 |
@sonney2k | does anyone here have contact to him? | 20:05 |
blackburn | pakdd! | 20:05 |
n4nd0 | sonney2k, working on it right now, making ac couple of updates to the blog | 20:05 |
@sonney2k | blackburn, still not at the meeting, not in irc, no email no nothing. something is wrong :( | 20:06 |
* sonney2k tries to compile the swig interface file with optimizations enabled | 20:09 | |
* sonney2k waits for minutes... | 20:09 | |
blackburn | swig with opts? | 20:11 |
blackburn | why? | 20:11 |
@sonney2k | blackburn, to see if it helps directorkernel | 20:11 |
blackburn | ahh | 20:11 |
@sonney2k | but no change | 20:11 |
@sonney2k | 7s | 20:11 |
@sonney2k | 0.4 w/o | 20:11 |
blackburn | sonney2k: try to profile it somehow :) | 20:11 |
@sonney2k | haha | 20:12 |
@sonney2k | blackburn, might be that even doing this from python is as slow as this | 20:12 |
blackburn | sonney2k: then measure sth like 'for i in xrange(N): for j in xrange(N): kernel(i,j)' | 20:13 |
n4nd0 | weekly report, there you go | 20:23 |
@sonney2k | blackburn, boah 10.3s! | 20:29 |
blackburn | sonney2k: fors? | 20:30 |
@sonney2k | blackburn, yes | 20:32 |
blackburn | hmm | 20:32 |
blackburn | sonney2k: and dot(X,X.T)? | 20:33 |
blackburn | or X.T, X :D | 20:33 |
@sonney2k | blackburn, 0.0 s | 20:36 |
blackburn | 0.0?? | 20:36 |
@sonney2k | yup | 20:37 |
blackburn | ehm so we have huge overhead out there :D | 20:37 |
@sonney2k | blackburn, it is not so clear if directors affect the rest of shogun... | 20:37 |
blackburn | sonney2k: what is d? | 20:38 |
@sonney2k | 1000 | 20:38 |
@sonney2k | and ndata=1000 too | 20:38 |
@sonney2k | I think calling a blas operation is the fastest one can do | 20:38 |
@sonney2k | and that is what numpy does :) | 20:39 |
blackburn | 0.97 here | 20:41 |
blackburn | however dot takes 0.6 | 20:41 |
blackburn | sonney2k: ^ how can that be? | 20:42 |
blackburn | >>> start = time.time(); x = numpy.dot(data.T,data); end = time.time(); print end-start | 20:43 |
blackburn | 0.607996940613 | 20:43 |
blackburn | >>> start = time.time(); x = k.get_kernel_matrix(); end = time.time(); print end-start | 20:43 |
blackburn | 0.961124897003 | 20:43 |
@sonney2k | blackburn, what do you mean? | 20:48 |
@sonney2k | it is just a 1000x1000 matrix | 20:48 |
@sonney2k | so 0.3s difference don't matter | 20:48 |
blackburn | sonney2k: why do you have 0.0? | 20:48 |
@sonney2k | my machine is faster :D | 20:49 |
blackburn | but why I have 0.3 difference but you have 0.6 difference | 20:49 |
@sonney2k | blackburn, did you compile atlas for your machine? | 20:50 |
blackburn | no | 20:50 |
@sonney2k | well... | 20:50 |
@sonney2k | then of course | 20:50 |
@sonney2k | speedup will be factor >10 | 20:50 |
blackburn | what do numpy use? | 20:50 |
@sonney2k | the one that is in your system | 20:50 |
@sonney2k | so when you compiled | 20:50 |
@sonney2k | atlas using the machinism provided by debian | 20:51 |
@sonney2k | then it will be a lot faster | 20:51 |
blackburn | hmm | 20:51 |
blackburn | sonney2k: is it necessary to *compile* it? | 20:51 |
@sonney2k | yes | 20:52 |
blackburn | sonney2k: IIRC you said one in debian is b0rken | 20:52 |
@sonney2k | there are no optimized atlas packages | 20:52 |
@sonney2k | in debian/ubuntu | 20:52 |
@sonney2k | blackburn, no | 20:52 |
blackburn | is there any reason? | 20:52 |
@sonney2k | blackburn, yes too many hardware versions | 20:53 |
@sonney2k | and debians is i686 specific | 20:53 |
@sonney2k | so sse2/3 whatever won't be enabled | 20:53 |
@sonney2k | and that makes a *HUGE* difference | 20:53 |
* sonney2k is off for a swim | 20:53 | |
blackburn | heh let me check | 20:53 |
@sonney2k | blackburn, just run the jblas benchmark btw | 20:54 |
@sonney2k | LD_LIBRARY_PATH=/usr/lib/jni java -jar /usr/share/java/jblas.jar | 20:55 |
@sonney2k | Running benchmark "Java matrix multiplication, double precision". | 20:56 |
@sonney2k | n = 10 : 0.854 GFLOPS (2134302 iterations in 5.0 seconds) | 20:56 |
@sonney2k | n = 100 : 1.682 GFLOPS (4206 iterations in 5.0 seconds) | 20:56 |
@sonney2k | n = 1000 : 1.114 GFLOPS (3 iterations in 5.4 seconds) | 20:56 |
@sonney2k | Running benchmark "Java matrix multiplication, single precision". | 20:56 |
@sonney2k | n = 10 : 0.747 GFLOPS (1868112 iterations in 5.0 seconds) | 20:56 |
@sonney2k | n = 100 : 1.668 GFLOPS (4170 iterations in 5.0 seconds) | 20:56 |
@sonney2k | n = 1000 : 1.334 GFLOPS (4 iterations in 6.0 seconds) | 20:57 |
@sonney2k | Running benchmark "ATLAS matrix multiplication, double precision". | 20:57 |
@sonney2k | n = 10 : 0.773 GFLOPS (1931439 iterations in 5.0 seconds) | 20:57 |
@sonney2k | n = 100 : 5.010 GFLOPS (12525 iterations in 5.0 seconds) | 20:57 |
@sonney2k | n = 1000 : 9.299 GFLOPS (24 iterations in 5.2 seconds) | 20:57 |
@sonney2k | Running benchmark "ATLAS matrix multiplication, single precision". | 20:57 |
@sonney2k | n = 10 : 0.812 GFLOPS (2030799 iterations in 5.0 seconds) | 20:57 |
@sonney2k | n = 100 : 9.997 GFLOPS (24993 iterations in 5.0 seconds) | 20:57 |
@sonney2k | n = 1000 : 15.100 GFLOPS (38 iterations in 5.0 seconds) | 20:57 |
@sonney2k | do this for optimized / not optimized atlas | 20:57 |
@sonney2k | you will notice a big difference | 20:57 |
-!- blackburn [~blackburn@31.28.59.65] has quit [Quit: Leaving.] | 21:03 | |
* gsomix have returned | 21:06 | |
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun | 21:08 | |
shogun-buildbot | build #565 of python_modular is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_modular/builds/565 blamelist: gsomix@gmail.com | 21:32 |
gsomix | fail | 21:38 |
gsomix | =____= | 21:38 |
-!- gsomix is now known as failman | 21:41 | |
n4nd0 | :D | 21:41 |
-!- failman is now known as fail_man | 21:41 | |
fail_man | >> This nickname is registered. | bebebe :( | 21:41 |
@sonney2k | blackburn, so did you try? | 21:47 |
blackburn | sonney2k: yeah but I think needs tweaking in configure | 21:48 |
blackburn | became much slower :D | 21:48 |
CIA-113 | shogun: Soeren Sonnenburg master * r80e75c8 / examples/undocumented/python_modular/kernel_director_linear_modular.py : let example exit sanely when shogun wasn't compiled with directors - http://git.io/b9ZINQ | 21:51 |
@sonney2k | blackburn, ? | 21:51 |
blackburn | sonney2k: I just did configure with no options | 21:52 |
@sonney2k | blackburn, I am talking about jblas... | 21:52 |
blackburn | ahhhhh | 21:52 |
blackburn | let me try | 21:52 |
@sonney2k | blackburn, yes directors have 100% overhead or so | 21:52 |
blackburn | sonney2k: | 21:53 |
blackburn | Running benchmark "Java matrix multiplication, double precision". | 21:53 |
blackburn | n = 10 : 0.411 GFLOPS (1027007 iterations in 5.0 seconds) | 21:53 |
blackburn | n = 100 : 0.630 GFLOPS (1574 iterations in 5.0 seconds) | 21:53 |
blackburn | n = 1000 : 0.439 GFLOPS (2 iterations in 9.1 seconds) | 21:53 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds] | 21:55 | |
@sonney2k | blackburn, that is just java | 21:57 |
@sonney2k | so where is the atlas part? | 21:58 |
blackburn | argh | 21:58 |
blackburn | Running benchmark "ATLAS matrix multiplication, double precision". | 21:59 |
blackburn | n = 10 : 0.368 GFLOPS (918914 iterations in 5.0 seconds) | 21:59 |
blackburn | n = 100 : 0.995 GFLOPS (2488 iterations in 5.0 seconds) | 21:59 |
blackburn | n = 1000 : 1.131 GFLOPS (3 iterations in 5.3 seconds) | 21:59 |
@sonney2k | n = 1000 : 9.299 GFLOPS (24 iterations in 5.2 seconds) | 22:06 |
@sonney2k | haha | 22:06 |
@sonney2k | blackburn, what machine do you have? | 22:06 |
blackburn | sonney2k: | 22:07 |
blackburn | vendor_id: GenuineIntel | 22:07 |
blackburn | cpu family: 6 | 22:07 |
blackburn | model: 23 | 22:07 |
blackburn | model name: Pentium(R) Dual-Core CPU T4300 @ 2.10GHz | 22:07 |
@sonney2k | blackburn, Intel(R) Core(TM)2 Duo CPU T9900 @ 3.06GHz | 22:07 |
@sonney2k | here | 22:07 |
blackburn | well not 9 times faster | 22:07 |
@sonney2k | certainly not | 22:07 |
blackburn | sonney2k: I just ran configure with no options at all | 22:08 |
blackburn | I do think it can cause low performance | 22:08 |
fail_man | sonney2k, ufff, I sent a little report. | 22:09 |
@sonney2k | fail_man, heh | 22:16 |
fail_man | =___= | 22:16 |
@sonney2k | fail_man, so lets quickly plan the next steps: | 22:16 |
@sonney2k | you should introduce these Director* classes for lots more similar classes | 22:17 |
@sonney2k | like blackburn said | 22:17 |
@sonney2k | kernel was a good start, maybe some linear / kernel machine should be next | 22:17 |
fail_man | ok | 22:20 |
@sonney2k | fail_man, as you know I will be away the week after next wek | 22:25 |
@sonney2k | so you should one by one do more director* classes | 22:25 |
@sonney2k | should be rather easy now that we know why things didn't wotk | 22:26 |
@sonney2k | work | 22:26 |
@sonney2k | fail_man, btw the amount of information in your update email is just what we want! | 22:32 |
@sonney2k | so good job | 22:32 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 22:36 | |
fail_man | sonney2k, thanks. | 22:40 |
fail_man | good night guys | 23:02 |
fail_man | sonney2k, I will push DirectorKernel with no_features tomorrow. | 23:02 |
n4nd0 | fail_man, good night! | 23:24 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has quit [Ping timeout: 245 seconds] | 23:33 | |
--- Log closed Tue May 29 00:00:41 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!