--- Log opened Fri May 10 00:00:38 2013 | ||
@wiking | sonney2k: but buildbot failed | 00:15 |
---|---|---|
@wiking | :) | 00:15 |
@wiking | lol | 00:15 |
@wiking | preprocessor_kpca.py setting 1/2 | 00:15 |
@wiking | AFTER AAAALL that PRs | 00:16 |
@lisitsyn | haha | 00:16 |
@wiking | it's crazy | 00:17 |
@wiking | ok let's see what does git flow feature finish does | 00:17 |
@wiking | lisitsyn: go? | 00:18 |
@lisitsyn | wiking: it merges | 00:18 |
@lisitsyn | and deletes | 00:18 |
@lisitsyn | your branch | 00:18 |
@wiking | yep | 00:18 |
@wiking | i guessed | 00:18 |
@wiking | shall i hit enter? | 00:19 |
@lisitsyn | wiking: what are you merging? | 00:19 |
@sonney2k | wiking, do it | 00:19 |
@wiking | feature/PRNG | 00:19 |
@lisitsyn | ahh yeah merge | 00:19 |
@wiking | No branch matches prefix 'feature/PRNG' | 00:20 |
@wiking | lol | 00:20 |
@wiking | ah w/o feature/ | 00:20 |
@lisitsyn | yes of course | 00:20 |
@wiking | done | 00:20 |
@wiking | pushing | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 35b7254 / src/shogun/lib/external/SFMT/ (16 files): https://github.com/shogun-toolbox/shogun/commit/35b7254d4e861b78a24ae35323450a9071b55d99 | 00:20 |
shogun-notifier- | shogun: Import SFMT PRNG | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 31da88a / src/shogun/mathematics/Random.cpp,src/shogun/mathematics/Random.h: https://github.com/shogun-toolbox/shogun/commit/31da88aac065398f6aaf0ad0a0ff897fbb72c151 | 00:20 |
shogun-notifier- | shogun: Create class for PRNG | 00:20 |
shogun-notifier- | shogun: wrapper class for SFMT | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 81599e1 / src/configure: https://github.com/shogun-toolbox/shogun/commit/81599e1c93ab83d54c1599697990dfd9c713d5fb | 00:20 |
shogun-notifier- | shogun: Add --mexp and HAVE_SSE2 defines in ./configure | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 4736671 / src/shogun/mathematics/Random.cpp,src/shogun/mathematics/Random.h: https://github.com/shogun-toolbox/shogun/commit/473667152bc96e8bbbe2f1438b997ac918d7897f | 00:20 |
shogun-notifier- | shogun: Add parameters of Random and add HAVE_PTHREAD macro check | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 7dd6b47 / src/shogun/mathematics/Random.cpp,src/shogun/mathematics/Random.h: https://github.com/shogun-toolbox/shogun/commit/7dd6b47b92a73cf282135a6a936a060f622e0b12 | 00:20 |
shogun-notifier- | shogun: Make Random unserializable | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * c204e89 / src/shogun/base/init.cpp,src/shogun/base/init.h: https://github.com/shogun-toolbox/shogun/commit/c204e89ad4828227c7c648bd0d3d5fc4903ebc1e | 00:20 |
shogun-notifier- | shogun: Create a global Random object in init | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 972c7ea / src/shogun/lib/external/SFMT/ (4 files): https://github.com/shogun-toolbox/shogun/commit/972c7eaaf01491cbb768b9be246caa58a8927b72 | 00:20 |
shogun-notifier- | shogun: SFMT: add corrent shogun path for includes | 00:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 9b9866d / src/shogun/mathematics/Random.cpp,src/shogun/mathematics/Random.h: https://github.com/shogun-toolbox/shogun/commit/9b9866d33d17fd74e88473084356231f8793b02c | 00:20 |
shogun-notifier- | shogun: Several changes in Random PRNG API | 00:20 |
shogun-notifier- | shogun: move state check in the lock part to avoid invalid state checks | 00:20 |
@wiking | lol | 00:21 |
@wiking | :) | 00:21 |
shogun-notifier- | shogun: Change include paths in dSFMT for shogun | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * e659378 / src/shogun/lib/external/dSFMT/dSFMT.h: https://github.com/shogun-toolbox/shogun/commit/e659378eada060dcaaa1870c6d4738367106c02d | 00:21 |
shogun-notifier- | shogun: Ignore dSFMT in doxygen generation | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * c5f7097 / src/shogun/lib/external/dSFMT/ (3 files): https://github.com/shogun-toolbox/shogun/commit/c5f7097cdad5d1219acc9c6928a71e90cf999f69 | 00:21 |
shogun-notifier- | shogun: Fix redefinition of W128_T in dSFMT | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 41f4a1d / src/shogun/mathematics/Random.cpp,src/shogun/mathematics/Random.h: https://github.com/shogun-toolbox/shogun/commit/41f4a1d8303f50f2c767797c553078a8c128ca38 | 00:21 |
shogun-notifier- | shogun: Add dSFMT wrapper API to Random | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 3232bae / src/configure: https://github.com/shogun-toolbox/shogun/commit/3232bae41ec20e09e69458398a89be1ac0fdba5a | 00:21 |
shogun-notifier- | shogun: Define DSFMT_MEXP same as SFMT_MEXP | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 6a1ca34 / src/shogun/mathematics/Random.cpp: https://github.com/shogun-toolbox/shogun/commit/6a1ca34cf39533a803c2ded3422a5d952c49cf7f | 00:21 |
shogun-notifier- | shogun: Random: Initialize class variables to NULL | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 012c4fe / tests/unit/mathematics/Random_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/012c4fe39f73d79e6c00bf8e8fefc7d9d6c9a958 | 00:21 |
shogun-notifier- | shogun: Add unit test for Random | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * e270af7 / examples/undocumented/libshogun/Makefile: https://github.com/shogun-toolbox/shogun/commit/e270af7eff0a9a9c6333e4861001eee375fda949 | 00:21 |
shogun-notifier- | shogun: Remove statistics_hsic example run | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 43af113 / src/shogun/mathematics/Math.cpp,src/shogun/mathematics/Math.h: https://github.com/shogun-toolbox/shogun/commit/43af113771b39b83fd4409f40ef28f9bf67bab27 | 00:21 |
shogun-notifier- | shogun: Change CMath::init_random to use CRandom's set_seed | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 9ea5c08 / tests/unit/statistics/QuadraticTimeMMD_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/9ea5c08da92a1a9d6f0492676c9e3ad77aa2e27f | 00:21 |
@wiking | FLOOD TIME! | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 801bb93 / src/configure,src/shogun/mathematics/Random.cpp: https://github.com/shogun-toolbox/shogun/commit/801bb935307de36df78abd37d881e8429fdb573c | 00:21 |
shogun-notifier- | shogun: Random use d/SFMT fill array functions only when memory is aligned | 00:21 |
shogun-notifier- | shogun: Viktor Gal :develop * 0ad6906 / / (40 files): https://github.com/shogun-toolbox/shogun/commit/0ad6906c052fd79d3c8619ba268bfe900b21fff3 | 00:21 |
shogun-notifier- | shogun: Merge branch 'feature/PRNG' into develop | 00:21 |
@sonney2k | wiking goes berserk | 00:21 |
@wiking | remotes/upstream/c50 | 00:21 |
@sonney2k | wiking, is the remote branch removed now too? | 00:21 |
@sonney2k | wiking, keep that | 00:22 |
@lisitsyn | sonney2k: why did you merge that :D | 00:22 |
@sonney2k | lisitsyn, :P | 00:22 |
@wiking | sonney2k: remote branch is not deleted | 00:22 |
@wiking | shall i? | 00:22 |
@lisitsyn | wiking: yeah just KILL IT | 00:22 |
@lisitsyn | :D | 00:22 |
@sonney2k | wiking, sure | 00:22 |
@lisitsyn | sonney2k: lets drop that example | 00:22 |
@wiking | deleted | 00:23 |
@sonney2k | lisitsyn, yeah drop it and the data I just created - hmmhh maybe add a bit to preprocessor_kernelpca.py | 00:23 |
@lisitsyn | sonney2k: why? | 00:23 |
@sonney2k | lisitsyn, look at the examples! | 00:24 |
@lisitsyn | I am sorry I lost the track on that | 00:24 |
@lisitsyn | but I have no idea what is it for | 00:24 |
@sonney2k | lisitsyn, yeah me neither o_O | 00:24 |
@lisitsyn | sonney2k: we now have two similar examples | 00:24 |
@lisitsyn | preprocessor_kpca_modular.py | 00:24 |
@sonney2k | lisitsyn, no | 00:24 |
@lisitsyn | and one more | 00:24 |
@lisitsyn | in graphical | 00:24 |
@lisitsyn | they are just about the same (circles) | 00:24 |
@sonney2k | in graphical? | 00:25 |
@lisitsyn | yes | 00:25 |
@sonney2k | hmmhh | 00:25 |
@sonney2k | we shuold probably do a kpca example on digits | 00:25 |
@lisitsyn | sonney2k: it has the same code but just plots | 00:25 |
@sonney2k | lisitsyn, ahh ok | 00:25 |
@sonney2k | then drop the kpca one | 00:25 |
@sonney2k | including data! | 00:25 |
@sonney2k | but still move some stuff over | 00:26 |
@sonney2k | to kernelpca | 00:26 |
@sonney2k | but comment it | 00:26 |
@lisitsyn | ?? | 00:26 |
@lisitsyn | what stuff? | 00:26 |
@sonney2k | #preprocessor.set_target_dim(2) | 00:26 |
@lisitsyn | it is default already | 00:26 |
@sonney2k | yeah still it is an invisible option | 00:27 |
@sonney2k | just put it in ther eand then I am fine | 00:27 |
shogun-buildbot | build #780 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/780 blamelist: Viktor Gal <viktor.gal@maeth.com> | 00:27 |
@lisitsyn | sonney2k: anytthing else? | 00:27 |
@sonney2k | lisitsyn, no | 00:27 |
@sonney2k | lisitsyn, btw how expert are you with GPs? | 00:29 |
@sonney2k | I keep wondering why I don't get the variance prediction below 1 | 00:29 |
@lisitsyn | sonney2k: like giraffe in SVMs | 00:29 |
@lisitsyn | where? | 00:30 |
@sonney2k | what? | 00:30 |
@sonney2k | in the example failing on travis | 00:30 |
@lisitsyn | sonney2k: I know as much as giraffe about SVMs | 00:30 |
@sonney2k | yeah well | 00:30 |
@sonney2k | GPs | 00:30 |
@lisitsyn | sonney2k: they have old eigen there | 00:31 |
@lisitsyn | just guess | 00:31 |
@sonney2k | lisitsyn, ?? | 00:31 |
@sonney2k | no I mean not why it fails | 00:32 |
shogun-notifier- | shogun: Sergey Lisitsyn :develop * 51cf598 / data,examples/undocumented/python_modular/preprocessor_kpca.py: https://github.com/shogun-toolbox/shogun/commit/51cf598e29237beeee31ab03084a9b630f857e1b | 00:32 |
shogun-buildbot | build #781 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/781 blamelist: Viktor Gal <viktor.gal@maeth.com> | 00:32 |
shogun-notifier- | shogun: Removed kpca example | 00:32 |
@sonney2k | but I am wondering why I don't get variance < 1 | 00:32 |
@sonney2k | in this example | 00:32 |
shogun-notifier- | shogun: Sergey Lisitsyn :develop * 024eca5 / examples/undocumented/python_modular/preprocessor_kernelpca_modular.py: https://github.com/shogun-toolbox/shogun/commit/024eca5d3f65cc8c1456391986307fd6a43255f6 | 00:32 |
@sonney2k | if I put > 1000 points in there | 00:32 |
shogun-notifier- | shogun: Explicit target dimensionality of kernel pca example set to 2 | 00:32 |
@lisitsyn | ahh | 00:33 |
@sonney2k | I would expect the GP to be pretty sure of its estimate | 00:33 |
@sonney2k | I mean it is 2d | 00:33 |
@lisitsyn | I get what you mean | 00:33 |
shogun-buildbot | build #782 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/782 blamelist: Viktor Gal <viktor.gal@maeth.com> | 00:33 |
@sonney2k | and uncertainty also doesn't grow above 2 | 00:33 |
@sonney2k | which is kind of weird | 00:34 |
@sonney2k | when you go far away and you never have seen anything in training there | 00:34 |
shogun-buildbot | build #783 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/783 blamelist: Viktor Gal <viktor.gal@maeth.com> | 00:34 |
@sonney2k | HeikoS, is the expert but I guess sleeping | 00:34 |
@lisitsyn | yeah | 00:35 |
@sonney2k | lisitsyn, what was the nickname of the student working on it? | 00:36 |
@sonney2k | he might know too... | 00:36 |
@lisitsyn | sonney2k: votjakovr | 00:36 |
@sonney2k | ok I will ask him tomorrow | 00:36 |
@lisitsyn | sonney2k: he is in syberia so sleeping too for sure | 00:36 |
shogun-notifier- | shogun: Sergey Lisitsyn :develop * 5fa05de / src/shogun/lib/tapkee/ (4 files): https://github.com/shogun-toolbox/shogun/commit/5fa05def44b3f344d9739d6a479b1edb0450c7ca | 00:37 |
shogun-notifier- | shogun: Update for tapkee library | 00:37 |
shogun-notifier- | shogun: | 00:37 |
shogun-notifier- | shogun: - Fix for compilation issue with clang - | 00:37 |
shogun-notifier- | shogun: struct template specialization instead of | 00:37 |
shogun-notifier- | shogun: boolean value specialization | 00:37 |
@sonney2k | wiking, ok so standard stuff works | 00:37 |
@sonney2k | ohh I have to sleep | 00:37 |
@sonney2k | cu | 00:37 |
@lisitsyn | I received a mail today | 00:37 |
@lisitsyn | guy said he had a lot of fun reading tapkee code | 00:38 |
@lisitsyn | :D | 00:38 |
@lisitsyn | not sure if it is bad or good | 00:38 |
@lisitsyn | okay I fixed an issue which could happen on clang | 00:38 |
@lisitsyn | now works better | 00:38 |
-!- travis-ci [~travis-ci@ec2-107-20-28-143.compute-1.amazonaws.com] has joined #shogun | 00:41 | |
travis-ci | [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7036124 | 00:41 |
-!- travis-ci [~travis-ci@ec2-107-20-28-143.compute-1.amazonaws.com] has left #shogun [] | 00:41 | |
shogun-buildbot | build #939 of deb2 - static_interfaces is complete: Failure [failed test python_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/939 blamelist: Viktor Gal <viktor.gal@maeth.com> | 00:53 |
-!- travis-ci [~travis-ci@ec2-107-20-28-143.compute-1.amazonaws.com] has joined #shogun | 00:58 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7036398 | 00:58 |
-!- travis-ci [~travis-ci@ec2-107-20-28-143.compute-1.amazonaws.com] has left #shogun [] | 00:58 | |
shogun-buildbot | build #1078 of deb3 - modular_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1078 | 01:17 |
shogun-buildbot | build #784 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/784 | 01:19 |
shogun-buildbot | build #940 of deb2 - static_interfaces is complete: Failure [failed test python_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/940 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 01:25 |
-!- travis-ci [~travis-ci@ec2-23-22-160-141.compute-1.amazonaws.com] has joined #shogun | 01:32 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7036517 | 01:32 |
-!- travis-ci [~travis-ci@ec2-23-22-160-141.compute-1.amazonaws.com] has left #shogun [] | 01:32 | |
shogun-buildbot | build #1079 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1079 blamelist: Viktor Gal <viktor.gal@maeth.com> | 01:42 |
shogun-buildbot | build #1080 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1080 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 02:05 |
-!- nube [~rho@49.244.137.73] has joined #shogun | 02:30 | |
shogun-buildbot | build #335 of nightly_none is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/335 | 03:20 |
shogun-buildbot | build #327 of nightly_all is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/327 | 03:31 |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 03:37 | |
shogun-buildbot | build #390 of nightly_default is complete: Failure [failed test] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/390 | 04:13 |
-!- FSCV [~FSCV@204.45.132.11] has quit [Quit: Leaving] | 04:16 | |
-!- foulwall_ [~foulwall@2001:da8:215:503:f1e8:e3c3:e141:20f0] has joined #shogun | 05:18 | |
-!- nube [~rho@49.244.137.73] has quit [Ping timeout: 248 seconds] | 05:47 | |
-!- gsomix [~Miranda@188.168.3.6] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 05:53 | |
-!- nube [~rho@49.244.9.98] has joined #shogun | 06:02 | |
-!- foulwall_ [~foulwall@2001:da8:215:503:f1e8:e3c3:e141:20f0] has quit [Remote host closed the connection] | 06:17 | |
-!- van51 [~van51@195.251.235.94] has joined #shogun | 09:34 | |
@sonney2k | HeikoS, lisitsyn please work on the spreadsheet... | 09:35 |
van51 | sonney2k: hey, about the rewrite of the CCombinedKernel class, I had only retained a get_kernel(index) method since it is backed by an array now | 09:36 |
@sonney2k | van51, makes totally sense | 09:38 |
van51 | yeah, but other classes that used it in a way of get_first_kernel(), get_next_kernel() break now | 09:40 |
van51 | should I rewrite the loops in those or maintain those methods? | 09:40 |
@sonney2k | van51, *sigh* yes these methods need a rewrite - which ones are this? | 09:43 |
@sonney2k | I guess sth. in svmlight? | 09:43 |
@sonney2k | and mkl | 09:46 |
van51 | for starters there are somein SVRLight | 09:47 |
van51 | ok I can rewrite them | 09:47 |
van51 | it isn't something hard | 09:47 |
van51 | just a small change in the beginning of some loops | 09:47 |
@sonney2k | lisitsyn, you had 43 patches submitted when we accepted you for gsoc'11 | 09:48 |
@sonney2k | crazy! | 09:48 |
@lisitsyn | sonney2k: haha | 09:48 |
@sonney2k | others had 22 / 17 / 2 /1 | 09:49 |
@sonney2k | gue3ss who had 22? | 09:49 |
@lisitsyn | sonney2k: heiko? | 09:49 |
@sonney2k | yes HeikoS | 09:49 |
@lisitsyn | but who had 17? | 09:49 |
@sonney2k | alesis | 09:50 |
@lisitsyn | ahh | 09:50 |
@sonney2k | lisitsyn, back then my estimate was that you will be a long time contributor | 09:56 |
@sonney2k | for the others I was like 50% | 09:56 |
@lisitsyn | heh | 09:56 |
van51 | I have another question, I see that the method CCombinedKernel::get_num_subkernels() may return a number greater than the number of kernels contained in its array(or list) | 10:30 |
van51 | but I have seen it used in a for loop, as the max index, in order to traverse the list of a CombinedKernel | 10:32 |
van51 | so I have added another method called get_num_kernels() to return only the number of elements in the array | 10:32 |
van51 | but, am I missing something? | 10:32 |
-!- nube1 [~rho@49.244.9.98] has joined #shogun | 10:42 | |
-!- Netsplit *.net <-> *.split quits: nube | 10:46 | |
-!- Netsplit *.net <-> *.split quits: nube1 | 10:59 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 10:59 | |
-!- Netsplit over, joins: nube1 | 11:01 | |
-!- sijin [~smuxi@144.214.222.109] has quit [Excess Flood] | 11:01 | |
van51 | anyway I got to go now, if you reply I'll see it at the logs | 11:02 |
van51 | otherwise see you tonight | 11:03 |
-!- van51 [~van51@195.251.235.94] has left #shogun ["QUIT :Leaving."] | 11:03 | |
-!- sijin [~smuxi@144.214.222.109] has joined #shogun | 11:03 | |
@wiking | mmm | 11:39 |
@wiking | we need to regenerate the data | 11:39 |
@lisitsyn | HeikoS: u can't touch this! | 11:39 |
@lisitsyn | wiking: all the data? | 11:39 |
@HeikoS | lisitsyn: man, are you still raving on this? :) | 11:39 |
@lisitsyn | HeikoS: whole.night. | 11:39 |
@lisitsyn | :D | 11:39 |
@lisitsyn | hah | 11:39 |
@wiking | lisitsyn: afaik not... first the one for kmeans | 11:39 |
@lisitsyn | wiking: why? | 11:39 |
@wiking | lisitsyn: well kmeans in integration uses random | 11:40 |
@lisitsyn | ahhh | 11:40 |
@wiking | and now since it's a different PRNG | 11:40 |
@wiking | we need to regenerate a new data for it | 11:40 |
@lisitsyn | wiking: da | 11:40 |
@wiking | lisitsyn: tavaris? | 11:40 |
@lisitsyn | wiking: tovarisch! | 11:40 |
@wiking | ookok dont go into that | 11:40 |
@lisitsyn | wiking: hahah | 11:40 |
@wiking | you write in cyrilic | 11:40 |
@wiking | so there's no 'good way to write in latin' | 11:40 |
@lisitsyn | wiking: not yet | 11:40 |
@wiking | :) | 11:40 |
@wiking | i mean in russian | 11:41 |
@wiking | the official alphabet is cyrilic | 11:41 |
@wiking | ;P | 11:41 |
@lisitsyn | wiking: no, why? | 11:41 |
@wiking | what no? | 11:41 |
@HeikoS | lisitsyn: whats the story with the merged/removed kpkca test? | 11:41 |
@wiking | yes | 11:41 |
@lisitsyn | we can easily translate | 11:41 |
@lisitsyn | transliterate | 11:41 |
@wiking | lisitsyn: ye tthere's almost a 1-1 mapping | 11:41 |
@wiking | but for example english alphabet has some missing letters | 11:42 |
@wiking | :P | 11:42 |
@wiking | lisitsyn: dontfuckwithme i know what i'm saying | 11:42 |
@wiking | :DDD | 11:42 |
@lisitsyn | wiking: yeah but they can be emulated | 11:42 |
@lisitsyn | HeikoS: well | 11:42 |
@lisitsyn | HeikoS: that 13 attempt PR | 11:42 |
@lisitsyn | introduced an example which was here already | 11:43 |
@wiking | HeikoS: 13 attempts aaaaand STILL FAILED! | 11:43 |
@HeikoS | why was it merged then? | 11:43 |
@HeikoS | does the other example have this nice circle data? | 11:43 |
@HeikoS | becazuse I liked that | 11:43 |
@lisitsyn | HeikoS: sonney2k was undre denial of service attack | 11:43 |
@HeikoS | reproduced the results from the scikit learn doc | 11:43 |
@HeikoS | lisitsyn: haha :) | 11:43 |
@lisitsyn | 13 pr is not a joke | 11:43 |
@lisitsyn | :D | 11:43 |
@HeikoS | lisitsyn: I know | 11:43 |
@wiking | yeah DoSed | 11:43 |
@lisitsyn | one can lose his mind | 11:44 |
@HeikoS | the guy also did the 1000th pull request | 11:44 |
@HeikoS | which is sad | 11:44 |
@lisitsyn | hahaha | 11:44 |
@wiking | HeikoS: yep | 11:44 |
@HeikoS | since I wanted to celebrate it | 11:44 |
@lisitsyn | yeah wish it was | 11:44 |
@HeikoS | but kept on opening/closing | 11:44 |
@lisitsyn | ehm better | 11:44 |
@lisitsyn | HeikoS: graphical circle example is here still | 11:44 |
@HeikoS | lisitsyn: I see | 11:44 |
@lisitsyn | but non-graphical (the same code mainly) | 11:44 |
@lisitsyn | is not | 11:44 |
@HeikoS | I liked that since we now know that KPCA does reasonable things | 11:45 |
@HeikoS | anyway | 11:45 |
@HeikoS | van51, sorry just saw your message | 11:46 |
@lisitsyn | vgorbati: hey | 11:46 |
@lisitsyn | how is it going? | 11:47 |
@HeikoS | sonney2k, lisitsyn I updated the spreadsheet a bit | 11:47 |
@lisitsyn | HeikoS: yeah I am watching you | 11:47 |
vgorbati | lisitsyn: manifold? | 11:47 |
@lisitsyn | HeikoS: u can't touch this | 11:47 |
@lisitsyn | vgorbati: anything you like to tell us about ;) | 11:47 |
@HeikoS | Mannigfaltigkeit! | 11:47 |
@lisitsyn | HeikoS: is that manifold in de? | 11:48 |
@HeikoS | yep, one of my favourite words! :) | 11:48 |
@lisitsyn | come on couldn't you had a *SHORTER* word? | 11:48 |
@HeikoS | NEIN | 11:49 |
vgorbati | lisitsyn: :) If everything goes ok I will end the implementation by the end of this week. What if I submit a PR then, leaving it for you to review, while I will continue working on tests for it? | 11:49 |
@lisitsyn | HeikoS: I used to sehenswuerdigskeit as it is of the same length in russian - dostoprimechatelnost | 11:50 |
@lisitsyn | but mannigfaltigkeit.. | 11:50 |
@HeikoS | what is it in Russian? | 11:50 |
@lisitsyn | HeikoS: sightseeing | 11:50 |
@HeikoS | Mannig-falitg-keit | 11:50 |
@lisitsyn | HeikoS: yeah quite easy to pronounce but impossible to keep that in mind | 11:51 |
@lisitsyn | :D | 11:51 |
@HeikoS | what is it in Russian? | 11:51 |
@lisitsyn | HeikoS: ahhh | 11:51 |
@lisitsyn | HeikoS: mnogoobrazie | 11:51 |
@HeikoS | haha, cool stuff :) | 11:51 |
@lisitsyn | HeikoS: mnogo = many | 11:51 |
@lisitsyn | HeikoS: obraz = image | 11:51 |
@HeikoS | Mannigfaltig = many faces | 11:52 |
@lisitsyn | hah | 11:52 |
@lisitsyn | got it | 11:52 |
@HeikoS | keit = make a noun out of it | 11:52 |
@lisitsyn | vgorbati: perfectly ok | 11:52 |
@HeikoS | mannigfaltig = folded many times | 11:52 |
@lisitsyn | HeikoS: well I guess many fold | 11:52 |
@HeikoS | there are many meanings | 11:52 |
@HeikoS | yes | 11:52 |
@HeikoS | we use the word also for other things | 11:53 |
@HeikoS | though thats old language style | 11:53 |
@lisitsyn | HeikoS: actually obraz in mnogoobrazie stands for something else | 11:53 |
@lisitsyn | HeikoS: as obrazovan is sth like 'formed' | 11:53 |
@lisitsyn | HeikoS: so obrazovan (formed) from mnogo (many) | 11:53 |
@lisitsyn | HeikoS: I think it should be easier for us both to get into the logic under our languages | 11:54 |
@lisitsyn | I mean I see something similar with words construction | 11:54 |
@lisitsyn | HeikoS: do you have some rather small task in mind? | 11:56 |
@lisitsyn | vgorbati: in order to let HeikoS know about you more, could you do something what he could propose? | 11:56 |
vgorbati | lisitsyn: right now, or after the manifold sculpting task? | 11:57 |
@lisitsyn | vgorbati: after I think | 11:57 |
@lisitsyn | vgorbati: I mean I rather know your skills than not | 11:57 |
@lisitsyn | but HeikoS and sonney2k have difficulties with ranking you :) | 11:57 |
@HeikoS | lisitsyn: small task? | 11:58 |
@lisitsyn | HeikoS: yeah something to know about vgorbati a little more | 11:58 |
@lisitsyn | I see you can't rate him ;) | 11:58 |
@HeikoS | lisitsyn: some stuff related to dimension reduction? :) | 12:00 |
@HeikoS | more unit tests for tapkee | 12:00 |
@HeikoS | illustrative examples with graphics | 12:00 |
@HeikoS | there are lots of things | 12:00 |
@HeikoS | there is an integration test still blacklisted? | 12:01 |
vgorbati | HeikoS: yes,the one I added is still blacklisted - it needs refactoring | 12:01 |
@HeikoS | lisitsyn: dont you have more intro tasks for your project? | 12:02 |
@lisitsyn | HeikoS: nono | 12:03 |
@lisitsyn | I have a lot | 12:03 |
@lisitsyn | you didn't get my intention I guess :) | 12:03 |
@HeikoS | lisitsyn: I have little knowledge about these things | 12:04 |
@lisitsyn | HeikoS: nahh | 12:04 |
@lisitsyn | we don't need DR from you | 12:04 |
@lisitsyn | :D | 12:04 |
@HeikoS | I always like: tests, graphical examples which illustrate how thins work | 12:04 |
@HeikoS | documentation | 12:04 |
@lisitsyn | HeikoS: I want to let vgorbati help you with something | 12:04 |
@HeikoS | ok I get it :) | 12:04 |
@HeikoS | well then, infrastructure, since I can judge that | 12:05 |
@lisitsyn | wiking: have you ever tried multiclass logistic regression? | 12:07 |
@lisitsyn | HeikoS: need you help out there | 12:08 |
@lisitsyn | it seems I wrote multiclass LR as just OvR linear classifier | 12:08 |
@HeikoS | what? | 12:08 |
@lisitsyn | HeikoS: is that ok? | 12:08 |
@HeikoS | ah I see | 12:08 |
@lisitsyn | or may be we could make something better here? | 12:09 |
@HeikoS | ehm, well that is fine, but this can be solved more elegantly | 12:09 |
@HeikoS | you can use the softmax logit function for multiple classes | 12:09 |
@HeikoS | at least this is how GPs work for multiclass | 12:09 |
@lisitsyn | HeikoS: sounds like a task right? | 12:09 |
@lisitsyn | :D | 12:09 |
@HeikoS | lisitsyn: well, thats not easy to do | 12:09 |
@lisitsyn | HeikoS: why | 12:10 |
@lisitsyn | ? | 12:10 |
@HeikoS | complicated? | 12:10 |
@lisitsyn | HeikoS: I mean what is difficult here in particular? | 12:10 |
@lisitsyn | HeikoS: is it still difficult given already computed probabilities for each class | 12:11 |
@lisitsyn | ? | 12:11 |
@lisitsyn | solver is here already | 12:11 |
@HeikoS | lisitsyn: if you want to go with the OvR, it should be fine | 12:11 |
@HeikoS | but can't one also formulate it as a multiclass problem | 12:12 |
@HeikoS | so without having binary sub-classifier? | 12:12 |
@lisitsyn | no, we have OvR logistic regression | 12:12 |
@lisitsyn | not some pure multiclass thing (is there any?) | 12:13 |
@lisitsyn | HeikoS: I am talking about shogun/multiclass/MulticlassLogisticRegression.cpp | 12:13 |
@HeikoS | lisitsyn: one can do this without this ugly OvR :) | 12:14 |
@HeikoS | anyway | 12:14 |
@HeikoS | what what the point again ? :) | 12:15 |
@lisitsyn | HeikoS: the point is | 12:15 |
@lisitsyn | we have OvR logistic regression | 12:15 |
@lisitsyn | I thought we can use probabilities here (better than just linear classifier) | 12:15 |
@lisitsyn | you probably know how :) | 12:15 |
@HeikoS | lisitsyn: this can probably be done similar to the SVM case implemented by Shell | 12:16 |
@HeikoS | however, I don't know for sure | 12:17 |
@HeikoS | and | 12:17 |
@lisitsyn | HeikoS: but hey - what is the better way than ovr here? | 12:17 |
@HeikoS | logistic regression provides a framework so that one doesnt have to do this | 12:17 |
@HeikoS | for example | 12:18 |
@HeikoS | https://www.google.co.uk/url?sa=t&rct=j&q=&esrc=s&source=web&cd=10&sqi=2&ved=0CIsBEBYwCQ&url=http%3A%2F%2Fbooks.nips.cc%2Fpapers%2Ffiles%2Fnips19%2FNIPS2006_0216.pdf&ei=QcmMUfT3HdCZ0QWLo4GYCg&usg=AFQjCNE-X9ZetpVFFrmS-23NgVwOwSN1eg&sig2=903IZIspiVEX2R0WhHmk-A | 12:18 |
@HeikoS | lisitsyn: | 12:21 |
@HeikoS | http://web4.cs.ucl.ac.uk/staff/D.Barber/pmwiki/pmwiki.php?n=Brml.Online | 12:21 |
@HeikoS | check this book | 12:21 |
@HeikoS | and search for logistic regression | 12:21 |
@HeikoS | and then the multiple classes part | 12:22 |
@lisitsyn | HeikoS: but it is still estimating the same thing right? | 12:22 |
@HeikoS | yes, but correctly, not with a heuristic as on OvR | 12:22 |
@lisitsyn | I mean w's | 12:22 |
@HeikoS | the w vectos is different | 12:22 |
@HeikoS | larger | 12:22 |
@HeikoS | check the 318 reference in the book | 12:22 |
@HeikoS | there, its explained for GPs (which are logistic regression= | 12:23 |
@lisitsyn | HeikoS: 17.4.4 | 12:25 |
@lisitsyn | is that better than just max? | 12:25 |
@HeikoS | if you want probabilities, yes | 12:26 |
@HeikoS | since you get them directly | 12:26 |
@HeikoS | without any heuristics as in OvR | 12:26 |
@lisitsyn | let me check solver again | 12:27 |
@lisitsyn | HeikoS: solver is not just ovr | 12:27 |
@lisitsyn | it is already multiclass I mean | 12:27 |
@lisitsyn | the objective itself | 12:27 |
@HeikoS | lisitsyn: oh, well then :) | 12:27 |
@HeikoS | you should get probabilities directly right? | 12:27 |
@lisitsyn | HeikoS: oh it seems objective just sums | 12:29 |
@lisitsyn | exp (wx+b) | 12:29 |
@lisitsyn | that's not that cool | 12:29 |
@lisitsyn | okay so just log-likelihood I guess | 12:32 |
@lisitsyn | not sure will get back to it later | 12:32 |
@HeikoS | lisitsyn: okay :) | 12:33 |
@HeikoS | we will soon have gps | 12:34 |
@lisitsyn | HeikoS: I guess softmax solver should minimize softmax already, right? | 12:34 |
@HeikoS | which are better than logistic regression in every way :) | 12:34 |
@HeikoS | lisitsyn: I think so yes, but haven't look into the implementation | 12:34 |
@HeikoS | also never implemented multinomial LR myself | 12:34 |
@lisitsyn | HeikoS: yeah but in general minimizing something else sounds like a bad idea for me | 12:34 |
@lisitsyn | HeikoS: why GPs are better? | 12:34 |
@HeikoS | same approach, however, they are Bayesian and therefore not overconfident, and they are non-linear like a kernel machine | 12:35 |
@HeikoS | non-parametric non-lineat | 12:35 |
@HeikoS | r | 12:35 |
@lisitsyn | HeikoS: but O(N^3) right? | 12:36 |
@HeikoS | yes | 12:36 |
@lisitsyn | HeikoS: what about applying? | 12:36 |
@lisitsyn | what is complexitiy? | 12:36 |
@HeikoS | it all depends on what you are using | 12:37 |
@HeikoS | there are all sorts of approximations /sparse methods to make them faster on larger datasets | 12:37 |
@HeikoS | relevance vector machine | 12:37 |
@lisitsyn | HeikoS: what is worst? | 12:37 |
@HeikoS | FITC for regression | 12:37 |
@HeikoS | dont konw | 12:38 |
@HeikoS | know | 12:38 |
@HeikoS | depends | 12:38 |
@lisitsyn | like linear or more? | 12:38 |
@HeikoS | more, its a kernel machine :) | 12:38 |
@lisitsyn | linear in size of training set I mean | 12:38 |
@lisitsyn | HeikoS: so just like kernel svm O(# of train vectors)? | 12:38 |
@sonney2k | HeikoS, vgorbati - you are the GP experts here right? | 12:38 |
@lisitsyn | sonney2k: no not that guy | 12:39 |
@lisitsyn | votjakovr | 12:39 |
@sonney2k | lisitsyn, but? | 12:39 |
@sonney2k | ahh | 12:39 |
@lisitsyn | haha | 12:39 |
@lisitsyn | first letter is ok | 12:39 |
@sonney2k | too little edit distance | 12:39 |
@lisitsyn | yes | 12:39 |
@HeikoS | hehe :) | 12:39 |
@sonney2k | HeikoS, https://dl.dropboxusercontent.com/u/31480471/regression_gaussian_process_modular.py | 12:40 |
@sonney2k | please run that example | 12:40 |
@lisitsyn | HeikoS: what makes me happy is that sonney2k doesn't know about GPs too :D | 12:40 |
@lisitsyn | so I am not the only GP dummy here | 12:41 |
@HeikoS | sonney2k: I ran it | 12:41 |
@sonney2k | HeikoS, I don't get why we have a variance of 1 at minimum | 12:41 |
@sonney2k | so the violet curves are mean - cov estimation | 12:41 |
@sonney2k | and the red the mean prediction | 12:42 |
@sonney2k | blue the training points | 12:42 |
@HeikoS | which quantiale? | 12:42 |
@sonney2k | I would expect the tube around the region where training data exists to be much smaller | 12:42 |
@sonney2k | HeikoS, sorry what? | 12:42 |
@sonney2k | the example is very small | 12:42 |
@HeikoS | so you just substract the variance and add it? | 12:42 |
@sonney2k | yes | 12:42 |
@sonney2k | or is the variance in logspace? | 12:43 |
@wiking | mmmm | 12:43 |
@HeikoS | no should not | 12:43 |
@HeikoS | sonney2k: this also depends on the kernel parameters | 12:44 |
@wiking | btw how should i call the following preprocessor: scale each features by: (feature[i] - min(feature[i])/(max(feature[i])-min(feature[i])) | 12:44 |
@HeikoS | ehm, covariance parameters | 12:44 |
@HeikoS | so there is a noise parameter somewhere | 12:44 |
@HeikoS | but in shogun its not in the kernel but somewhere else | 12:44 |
@lisitsyn | wiking: good question | 12:45 |
@wiking | lisitsyn: svm-scale does this afaik | 12:46 |
@HeikoS | sonney2k: http://pastebin.com/ZeaCWtnU | 12:46 |
@HeikoS | try this | 12:46 |
@HeikoS | I changed two things: | 12:46 |
@HeikoS | plot 95% confidence interval of prediction variance | 12:46 |
@HeikoS | set noise parameter of likelihood to a smaller number | 12:46 |
@lisitsyn | wiking: that's a nice thing indeed | 12:47 |
@wiking | lisitsyn: http://en.wikipedia.org/wiki/Feature_scaling | 12:47 |
@wiking | lisitsyn: rescaling | 12:47 |
@lisitsyn | RescaleFeatures | 12:47 |
@lisitsyn | wiking: may be that& | 12:47 |
@lisitsyn | ? | 12:47 |
@wiking | lisitsyn: yep | 12:47 |
@wiking | lisitsyn: i wonder if i should write it as a template | 12:47 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds] | 12:47 | |
@HeikoS | sonney2k: and if you then also decrease the number of training samples to 5, you see correct behaviour | 12:48 |
@wiking | lisitsyn: as i really dont want to hardcode it for DenseFeatures<float64_t> | 12:48 |
@lisitsyn | wiking: I will think about that | 12:49 |
@lisitsyn | it is totally improper to have templated preprocessors here | 12:49 |
@lisitsyn | but no other way still | 12:49 |
@wiking | lisitsyn: yes but somehow we should fix this problem | 12:49 |
@wiking | to have 1 preprocessor | 12:49 |
@wiking | for all the features | 12:49 |
@wiking | i mean at least DenseFeatures<*> | 12:50 |
@wiking | to write a preproc that can handle both Dense and Sparse woudl be great too | 12:50 |
@wiking | :PPP | 12:50 |
@lisitsyn | ok be back later | 12:50 |
@sonney2k | HeikoS, what is that 1.96? | 12:53 |
@sonney2k | ohh sqrt | 12:53 |
@HeikoS | sonney2k: the GP returns the variance of a 1D Gaussian at the prediction point | 12:53 |
@HeikoS | in order to plot the 95% confidence interval, you need to do 1.96*std_deviation | 12:54 |
@sonney2k | I thought 2*std_deviation?! | 12:55 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 12:55 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Client Quit] | 12:55 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 12:56 | |
@HeikoS | sonney2k: if you do lik.set_sigma(0.000001), then the prediction at the training data will have zero variance | 12:57 |
@sonney2k | HeikoS, so the key is reducing the noise level of the gaussianlikelihood | 12:57 |
@HeikoS | sonney2k: yes | 12:57 |
@HeikoS | the 1.96 is just to have a meaningful plot | 12:57 |
@HeikoS | plotting the variance itself might be misleading | 12:57 |
@sonney2k | so for the ideal case we should set it to the same value noise_var above has right? | 12:57 |
@HeikoS | sonney2k: exactly | 12:58 |
@HeikoS | the GP can learn that | 12:58 |
@sonney2k | but it doesn't? | 12:58 |
@HeikoS | no, in this example it doesnt | 12:58 |
@HeikoS | thats why I set up all these intro tasks :) | 12:58 |
@sonney2k | ok do we have an example where it does / what does one need to change? | 12:58 |
@HeikoS | one can do gradient descent on the marginal likelihood to learn it for example | 12:58 |
@HeikoS | sonney2k: I currently dont know, there is one modelselection example for the gps | 12:59 |
@HeikoS | but dont know whether the noise is learned | 12:59 |
@HeikoS | for regression there are only two parameters to learn | 12:59 |
@HeikoS | we should have an example on this | 12:59 |
@HeikoS | sonney2k: there are a few c++ examples, but they are not good | 13:00 |
@sonney2k | HeikoS, yes indeed | 13:01 |
@sonney2k | it kind of defeats the purpose when one has to pre-specify the noise level | 13:01 |
@HeikoS | sonney2k: sometimes you know the observation noise | 13:02 |
@HeikoS | sometimes its zero | 13:02 |
@HeikoS | learning it is the best obviously | 13:02 |
@HeikoS | I will talk to roman about this | 13:03 |
@HeikoS | sonney2k: also note that the mean prediction is not changed | 13:03 |
@HeikoS | so if you only want to predict, it doesnt even matter, like with svr or svm | 13:03 |
@sonney2k | HeikoS, yes sure but then I don't need a GP | 13:05 |
@sonney2k | I can use *fast* SVR | 13:05 |
@HeikoS | and how do you select you kernel? | 13:05 |
@HeikoS | and C? | 13:05 |
@sonney2k | HeikoS, and epsilon | 13:05 |
@sonney2k | same with GPs - model selection | 13:06 |
@HeikoS | yes :) | 13:06 |
@sonney2k | or how do I select the width of the gaussian kernel | 13:06 |
@HeikoS | there is no free lunch | 13:06 |
@HeikoS | sonney2k: take the median :D | 13:06 |
@HeikoS | what I like about GPs is that these questions are answered | 13:06 |
@sonney2k | yeah sure | 13:06 |
@HeikoS | at least for regression | 13:06 |
@sonney2k | but in no shogun example | 13:06 |
@HeikoS | sonney2k: yep! | 13:06 |
@HeikoS | good point, will work on this | 13:07 |
@sonney2k | if width of the kernel and noise level were automagically chosen I would be fine | 13:07 |
@sonney2k | using them for regression and tiny data sets | 13:07 |
@HeikoS | we have to check whether the stuff works before however | 13:07 |
@HeikoS | yes, all this should be possible | 13:07 |
@HeikoS | it will also work for large regression data sets btw since we have this sparse approximation stuff | 13:08 |
@HeikoS | anyway back to work | 13:09 |
@sonney2k | HeikoS, ok then we need an example for that too | 13:09 |
@HeikoS | sonney2k: https://github.com/shogun-toolbox/shogun/issues/1080 | 13:19 |
@HeikoS | will add the other one soon | 13:19 |
@sonney2k | van51: what?? that shouldn't happen where?? | 13:24 |
@sonney2k | wiking any progress on gettting build back to green? | 13:24 |
@sonney2k | HeikoS, thanks | 13:24 |
@wiking | sonney2k: yeah i've just finished the new preproc | 13:28 |
@wiking | and now i'll generate the new data | 13:28 |
@sonney2k | wiking, ../data/clustering/KMeans_EuclidianDistance_3.mOK | 13:28 |
@wiking | sonney2k: do i have write access to data repo? | 13:28 |
@sonney2k | it doesn't fail on my machine?! | 13:28 |
@wiking | sonney2k: that is most interesting | 13:28 |
@sonney2k | wiking, could you please do ./test_one.py ../data/clustering/KMeans_EuclidianDistance_3.m | 13:28 |
@sonney2k | in | 13:28 |
@sonney2k | shogun/tests/integration/python_static | 13:28 |
@wiking | sonney2k: just a sec... building shogun | 13:28 |
@sonney2k | or wait | 13:29 |
@sonney2k | maybe I didn't make install | 13:29 |
@sonney2k | sec | 13:29 |
@sonney2k | alright | 13:29 |
@sonney2k | fails fine now :D | 13:29 |
@wiking | hehehe you almost gave me a heartattack :) | 13:30 |
@wiking | :P | 13:30 |
@sonney2k | wiking, it is weird though that only kmeans fails | 13:31 |
@sonney2k | all HMM stuff is random too | 13:31 |
@wiking | sonney2k: maybe it has initrandom call in it? | 13:31 |
@sonney2k | better, we are skipping HMM tests :D | 13:32 |
@wiking | :>>> | 13:32 |
@wiking | ah shit i always forget to add unit test | 13:32 |
@wiking | this time i remembered! :) | 13:32 |
@sonney2k | wiking, man not easy to fix | 13:44 |
@sonney2k | generators are for the old tests are well b0rken | 13:44 |
@wiking | ? | 13:49 |
@sonney2k | wiking, lots of issues. I fixed a couple but there are more | 13:52 |
@sonney2k | I will hopefully get it to work tonight | 13:52 |
@wiking | lisitsyn: ping? | 14:16 |
@sonney2k | HeikoS, about k-means - we currently don't always call store_model_features for k-means | 14:26 |
@sonney2k | HeikoS, this leads to an error when you do get_cluster_centers() | 14:26 |
@sonney2k | because then you just get all training data back | 14:26 |
@wiking | sonney2k: we dont have a fast way to access a row vector of a matrix right? | 14:27 |
@sonney2k | wiking, SGMatrix? | 14:27 |
@wiking | i.e. a given feature for all the vectors | 14:27 |
@wiking | sonney2k: it only has get_column_vector() | 14:27 |
@wiking | mmm do we have transpose? :) | 14:27 |
@wiking | yes! | 14:27 |
@wiking | goooood | 14:27 |
@sonney2k | what? slow | 14:27 |
@wiking | no | 14:28 |
@wiking | it's just that i need the row vector instead of a column vector | 14:28 |
@wiking | and i didn't know if we have an operator for that | 14:28 |
@wiking | instead of writing a for loop | 14:28 |
-!- abinash [75c8e7db@gateway/web/freenode/ip.117.200.231.219] has joined #shogun | 14:28 | |
@wiking | but then if i transpose and get_column_vectors | 14:28 |
@wiking | it's the same | 14:28 |
@wiking | ;) | 14:28 |
@sonney2k | > twice the memory though | 14:29 |
@wiking | mmm yeah shit | 14:29 |
@wiking | maybe for loop then | 14:30 |
@sonney2k | rather copy the rowvector | 14:30 |
@sonney2k | I mean introduce a get_row_vector | 14:30 |
@sonney2k | which just copies the values | 14:30 |
@wiking | T* get_column_vector(index_t col) const | 14:31 |
@sonney2k | SGVector<T> you mean yes | 14:32 |
@wiking | yep i wanted to ask | 14:33 |
@wiking | whether return SGVector or T* | 14:33 |
-!- nube1 [~rho@49.244.9.98] has quit [Quit: Leaving.] | 14:34 | |
@sonney2k | wiking, ohh but that is just a view on the data | 14:35 |
@sonney2k | so you need to be careful and disable refcounting for the returned SGVector | 14:35 |
-!- nube [~rho@49.244.9.98] has joined #shogun | 14:36 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 14:47 | |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * f68d3cb / data,tests/integration/generator/clustering.py: https://github.com/shogun-toolbox/shogun/commit/f68d3cb1cf72fb83e1c0db5e2d200d25cbf1ce85 | 14:47 |
shogun-notifier- | shogun: fix euclidean k-means test | 14:47 |
abinash | sonney2k: Hello, as per my gsoc proposal (interfacing libqp via modular interfaces). Can I submit a patch implementing that one right now? | 14:59 |
shogun-buildbot | build #785 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/785 blamelist: Soeren Sonnenburg <sonne@debian.org> | 15:02 |
-!- travis-ci [~travis-ci@ec2-54-224-131-101.compute-1.amazonaws.com] has joined #shogun | 15:08 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7051045 | 15:08 |
-!- travis-ci [~travis-ci@ec2-54-224-131-101.compute-1.amazonaws.com] has left #shogun [] | 15:08 | |
shogun-buildbot | build #941 of deb2 - static_interfaces is complete: Failure [failed test octave_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/941 blamelist: Soeren Sonnenburg <sonne@debian.org> | 15:13 |
abinash | Hello, anybody please answer my query? | 15:15 |
@sonney2k | abinash, sure | 15:15 |
abinash | sonney2k: ok, should I include that one in the Library module for python interfaces | 15:16 |
@wiking | sonney2k: did you mean something like this | 15:16 |
@wiking | template<class T> | 15:17 |
@wiking | SGVector<T> SGMatrix<T>::get_row_vector(index_t row) const | 15:17 |
@wiking | { | 15:17 |
@wiking | SGVector<T> rowv(num_cols, false); | 15:17 |
@wiking | for (index_t i = 0; i < num_cols; i++) | 15:17 |
@wiking | { | 15:17 |
@wiking | rowv[i] = matrix[i*num_rows+row]; | 15:17 |
@wiking | } | 15:17 |
@wiking | return rowv; | 15:17 |
@wiking | } | 15:17 |
@wiking | sonney2k: or u even just want to copy the pointers? | 15:17 |
@wiking | so it's actually is really only a view | 15:17 |
@wiking | sonney2k: as this way it's actually copying the matrix elements into an array | 15:18 |
@wiking | *vector | 15:18 |
-!- trailofdan [4e5b0e75@gateway/web/freenode/ip.78.91.14.117] has joined #shogun | 15:20 | |
-!- ErlendA [~ErlendA@cm-84.215.138.251.getinternet.no] has joined #shogun | 15:20 | |
@HeikoS | hi | 15:20 |
@sonney2k | HeikoS, did you get the email on the list? | 15:21 |
@sonney2k | wiking, you too? | 15:21 |
@wiking | sonney2k: yes | 15:22 |
@wiking | sonney2k: so a copy of row or really just a copy of 'pointers' | 15:22 |
@sonney2k | wiking, yeah I meant that | 15:22 |
@sonney2k | abinash, not so clear what you did... | 15:24 |
@sonney2k | typemaps? | 15:24 |
abinash | sonney2k: I just wanted to clarify in which module should the libqp interface be placed? | 15:27 |
@sonney2k | abinash, I don't understand? | 15:27 |
@sonney2k | you wrote a C++ class? | 15:28 |
@sonney2k | if so then shogun/optimization/<classname> | 15:28 |
shogun-buildbot | build #1081 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1081 blamelist: Soeren Sonnenburg <sonne@debian.org> | 15:29 |
abinash | sonney2k: No, libqp is implemented in shogun in /src/shogun/lib/external | 15:29 |
abinash | I just want to interface it through modular interfaces. | 15:30 |
@sonney2k | abinash, you should write C++ classes then in shogun/optimization | 15:38 |
@sonney2k | using standard shogun types (prefix's SG* e.g. SGVector) | 15:38 |
abinash | sonney2k: So, what I get is that, first I should write C++ classes implementing the functionality of libqp | 15:41 |
abinash | and after that I should interface it | 15:41 |
-!- HeikoS [~heiko@nat-179-224.internal.eduroam.ucl.ac.uk] has quit [Read error: Connection reset by peer] | 15:44 | |
-!- lambday [67157e4d@gateway/web/freenode/ip.103.21.126.77] has joined #shogun | 15:56 | |
-!- trailofdan [4e5b0e75@gateway/web/freenode/ip.78.91.14.117] has quit [Ping timeout: 245 seconds] | 16:02 | |
-!- ErlendA [~ErlendA@cm-84.215.138.251.getinternet.no] has quit [Quit: Leaving] | 16:06 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds] | 16:10 | |
-!- abinash [75c8e7db@gateway/web/freenode/ip.117.200.231.219] has quit [Ping timeout: 245 seconds] | 16:14 | |
@sonney2k | abinash - the C++ classes will be visible from any language | 16:17 |
@sonney2k | lambday, welcome back... | 16:17 |
lambday | sonney2k: thanks :) | 16:22 |
lambday | sonney2k: I was a bit occupied with my semester exams :( | 16:22 |
@wiking | sonney2k: back | 16:34 |
@wiking | sonney2k: ok so it's actually copying the row elements then... | 16:35 |
@wiking | sonney2k: although it still sucks for me.... because i would need an iterator on row elements :S | 16:36 |
@wiking | sonney2k: here? | 16:43 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 16:48 | |
@wiking | tada | 16:51 |
@wiking | new preprocessor | 16:51 |
shogun-notifier- | shogun: Viktor Gal :develop * 173b59d / / (8 files): https://github.com/shogun-toolbox/shogun/commit/173b59d503ee035a712d83e09f56272696e68a25 | 16:51 |
shogun-notifier- | shogun: Implement RescaleFeatures preprocessor | 16:51 |
shogun-notifier- | shogun: See the doxygen class description. | 16:51 |
shogun-notifier- | shogun: Add new get_row_vector(index) method for SGMatrix | 16:51 |
-!- gsomix [~Miranda@178.45.40.206] has joined #shogun | 16:59 | |
gsomix | hi | 16:59 |
gsomix | I'm happy today: 100+ kms and only 1.5 hours in bus | 17:00 |
-!- nube [~rho@49.244.9.98] has quit [Quit: Leaving.] | 17:01 | |
@wiking | sonney2k: the kmeans is fixed, thnx! | 17:02 |
shogun-buildbot | build #786 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/786 blamelist: Viktor Gal <viktor.gal@maeth.com> | 17:02 |
shogun-buildbot | build #942 of deb2 - static_interfaces is complete: Failure [failed test octave_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/942 blamelist: Viktor Gal <viktor.gal@maeth.com> | 17:10 |
shogun-buildbot | build #1082 of deb3 - modular_interfaces is complete: Failure [failed compile python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1082 blamelist: Viktor Gal <viktor.gal@maeth.com> | 17:11 |
-!- travis-ci [~travis-ci@ec2-23-22-160-141.compute-1.amazonaws.com] has joined #shogun | 17:13 | |
travis-ci | [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7054670 | 17:13 |
-!- travis-ci [~travis-ci@ec2-23-22-160-141.compute-1.amazonaws.com] has left #shogun [] | 17:13 | |
-!- trailofdan [4e5b0e75@gateway/web/freenode/ip.78.91.14.117] has joined #shogun | 17:22 | |
@lisitsyn | BACK | 17:43 |
@lisitsyn | wiking: pong | 17:44 |
-!- gsomix [~Miranda@178.45.40.206] has quit [Quit: Miranda IM! Smaller, Faster, Easier. http://miranda-im.org] | 17:48 | |
@lisitsyn | sonney2k: these domain hunting idiots keep being idiots | 17:52 |
-!- lambday [67157e4d@gateway/web/freenode/ip.103.21.126.77] has quit [Ping timeout: 245 seconds] | 17:54 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds] | 17:57 | |
-!- gsomix [~gsomix@178.45.40.206] has joined #shogun | 18:47 | |
@wiking | lisitsyn: can i define somewhere which headers should b ignored by doxygen? | 19:00 |
@lisitsyn | wiking: #ifndef DOXYGEN_SHOULD_SKIP_THIS | 19:00 |
@lisitsyn | your stuff | 19:00 |
@lisitsyn | #endif | 19:00 |
-!- trailofdan [4e5b0e75@gateway/web/freenode/ip.78.91.14.117] has quit [Quit: Page closed] | 19:01 | |
@lisitsyn | that's the usual way | 19:01 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 19:01 | |
@wiking | lisitsyn: kkk | 19:14 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds] | 19:20 | |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 19:24 | |
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] | 19:51 | |
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has joined #shogun | 19:52 | |
-!- HeikoS [~heiko@nat-179-224.internal.eduroam.ucl.ac.uk] has joined #shogun | 19:53 | |
-!- mode/#shogun [+o HeikoS] by ChanServ | 19:53 | |
@lisitsyn | HeikoS: wiking: what do you guys think about C* prefix of classes? | 19:58 |
@lisitsyn | I mean we first call them CSomething then rename again | 19:59 |
@HeikoS | lisitsyn: why not? | 19:59 |
@lisitsyn | I see no reason actually | 19:59 |
@HeikoS | its good for distinguishing between SGObject and others | 19:59 |
@lisitsyn | HeikoS: others? like? | 19:59 |
@HeikoS | we have some other classes | 19:59 |
@HeikoS | but not many | 20:00 |
@HeikoS | structs etc | 20:00 |
@lisitsyn | HeikoS: it is confusing | 20:00 |
@lisitsyn | I think | 20:00 |
@HeikoS | lisitsyn: maybe | 20:00 |
@lisitsyn | I mean in C++ and in python naming differs | 20:00 |
@HeikoS | yes | 20:00 |
@HeikoS | but it does anyways due to types | 20:00 |
@HeikoS | it would be ok with dropping them | 20:00 |
@HeikoS | but also ok with keeping them | 20:00 |
@HeikoS | what was the original reason? | 20:00 |
@lisitsyn | who knows, I was 8 yrs old when sonney2k started to write first classes :D | 20:01 |
@lisitsyn | next | 20:02 |
@lisitsyn | HeikoS: http://scikit-learn.org/stable/modules/pipeline.html | 20:02 |
@HeikoS | haha | 20:02 |
@HeikoS | whats that? | 20:02 |
@lisitsyn | HeikoS: I actually have strong preference on removing all SG and C things | 20:03 |
@HeikoS | what does sonney2k say to this? | 20:03 |
@lisitsyn | HeikoS: I think we need something like pipeline | 20:03 |
@lisitsyn | HeikoS: SGVector -> shogun::Vector | 20:03 |
@lisitsyn | there are namespaces.. | 20:03 |
@HeikoS | yes true | 20:04 |
@HeikoS | namespaces are much cleaner | 20:04 |
@HeikoS | why dont do pipelines by hand? | 20:04 |
@lisitsyn | I feel pretty bad with noise | 20:04 |
@lisitsyn | HeikoS: well everything could be done by hand | 20:04 |
@lisitsyn | just a matter of usability ;) | 20:04 |
@lisitsyn | wiking: ping pnig join us | 20:05 |
@HeikoS | lisitsyn: well yes but can one really handle all these cases | 20:05 |
@HeikoS | I wound rather try to simplify things | 20:05 |
@HeikoS | like modelselection api | 20:05 |
@HeikoS | etc | 20:05 |
@lisitsyn | HeikoS: yes of course | 20:06 |
@HeikoS | I mean its cool | 20:06 |
@HeikoS | but does it really help that much? | 20:06 |
@sonney2k | so HeikoS & lisitsyn & wiking you all have seen the domainhunting post to the mailinglist? | 20:07 |
@HeikoS | yes | 20:07 |
@lisitsyn | yeah | 20:07 |
@lisitsyn | sonney2k: makes no sense to buy all domains | 20:07 |
@sonney2k | for them or for us? | 20:08 |
@lisitsyn | so what we can do is just keep saying they didn't ask us | 20:08 |
@lisitsyn | sonney2k: for us | 20:08 |
@HeikoS | sonney2k: so what can we do? say that they are not authorised? | 20:08 |
@lisitsyn | HeikoS: it is the second time already.. | 20:08 |
@lisitsyn | and different company | 20:08 |
@sonney2k | I think anyone can buy a domain as long as there is no trade mark violated | 20:08 |
@HeikoS | lisitsyn: so is this a problem for us? | 20:08 |
@lisitsyn | HeikoS: I don't personally care about shogun-toolbox.cn | 20:09 |
@HeikoS | me neither | 20:09 |
@lisitsyn | or anything like that | 20:09 |
@lisitsyn | .com is free too | 20:09 |
@HeikoS | we might get hacked at some point :D | 20:09 |
@HeikoS | actually, that might happen for real | 20:09 |
gsomix | lisitsyn, шогун-тулбокс.рф | 20:09 |
@lisitsyn | hah | 20:09 |
@lisitsyn | sonney2k: I am very tempted to remove all C* E* and SG* noise in names | 20:11 |
@lisitsyn | tell us what do you think | 20:11 |
@sonney2k | lisitsyn, HeikoS what did you discuss above? | 20:11 |
@sonney2k | lisitsyn, we cannot easily | 20:11 |
@lisitsyn | sonney2k: there is no reason to call vector SGVector | 20:11 |
@lisitsyn | shogun::Vector is very enough | 20:11 |
@sonney2k | lisitsyn, not if you interface with something that does not use namespaces | 20:12 |
@lisitsyn | sonney2k: using namespace shogun *is wrong* | 20:12 |
@sonney2k | lisitsyn, so you have to be extra careful with standard names | 20:12 |
@sonney2k | the other issue is that we mark classes serializable by adding a C prefix | 20:12 |
@HeikoS | sonney2k: I do not really mind changing/not changing things | 20:13 |
@sonney2k | SG* is just for data types | 20:13 |
@HeikoS | but I would not mark that as top priority | 20:13 |
@lisitsyn | HeikoS: yeah it is my personal desire | 20:13 |
@sonney2k | about the pipeline in scikits - oh well the syntax is so cryptic | 20:14 |
@sonney2k | I would rather write a small python function than that | 20:14 |
@lisitsyn | sonney2k: ok forget pipeline, naming again | 20:14 |
@sonney2k | guys we have broken tests | 20:14 |
@sonney2k | I have to focus on that | 20:15 |
@lisitsyn | haha | 20:15 |
@HeikoS | lisitsyn, this equals method is quite important :) | 20:16 |
@HeikoS | in order to make integration tests work | 20:16 |
@HeikoS | and migration | 20:16 |
@lisitsyn | where? | 20:17 |
@lisitsyn | ehmm we have warning now | 20:17 |
@lisitsyn | about dsfmt | 20:17 |
@lisitsyn | alright reconfigure | 20:18 |
@lisitsyn | no more warning | 20:18 |
@sonney2k | HeikoS, we currently have real errors | 20:18 |
@HeikoS | uh really | 20:18 |
@sonney2k | well we changed the RNG | 20:19 |
@sonney2k | so now all examples must fail | 20:19 |
@sonney2k | that use some RNG | 20:19 |
@HeikoS | scii' codec can't encode character u'\u2212' in position 798691: ordinal not in range(128) | 20:19 |
@sonney2k | HeikoS, ohh great you are the expert for creating that thing | 20:20 |
@sonney2k | IIRC that is a zero width space | 20:20 |
@HeikoS | sonney2k: argh yes | 20:21 |
@HeikoS | but this wasnt mine | 20:21 |
@sonney2k | no minus sign this time | 20:21 |
@HeikoS | these happen when one copy pastes documentation | 20:21 |
@wiking | sonney2k: we need to define MEXPs in config.h :( | 20:34 |
@sonney2k | MEXPs? | 20:35 |
van51 | HeikoS: about the LinearTimeMMD class | 20:39 |
van51 | give me a moment to find it | 20:39 |
van51 | :p | 20:39 |
@HeikoS | van51: is it urgent? | 20:40 |
@HeikoS | since I have to run off soon | 20:40 |
van51 | not really | 20:40 |
@HeikoS | could you send en email instead? :) | 20:40 |
@HeikoS | sorry for this, but just realised that I am late to meet someone | 20:40 |
van51 | HeikoS: yea sure | 20:40 |
van51 | no problem | 20:40 |
-!- HeikoS [~heiko@nat-179-224.internal.eduroam.ucl.ac.uk] has left #shogun [] | 20:42 | |
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun | 20:48 | |
shogun-notifier- | shogun: Sergey Lisitsyn :develop * 25426f6 / src/Makefile: https://github.com/shogun-toolbox/shogun/commit/25426f637e9d99f05b70d24c4bb82c79bbf7d538 | 20:48 |
shogun-notifier- | shogun: Wrapped output of compilation options | 20:48 |
@lisitsyn | sonney2k: do you mind my replacing that long long g++ output with something smaller? | 20:57 |
@lisitsyn | I am tired to see these lines | 20:57 |
@lisitsyn | :D | 20:57 |
-!- gsomix_ [~gsomix@109.169.146.48] has joined #shogun | 20:58 | |
shogun-buildbot | build #787 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/787 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 20:58 |
-!- gsomix [~gsomix@178.45.40.206] has quit [Disconnected by services] | 21:01 | |
-!- gsomix_ is now known as gsomix | 21:01 | |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 43682de / data,tests/integration/generator/fileop.py: https://github.com/shogun-toolbox/shogun/commit/43682de4ff7b5b6a3afae1027614221009421b57 | 21:01 |
shogun-notifier- | shogun: consider int32 & friends to be a numeric value in generator - not a string | 21:01 |
@sonney2k | lisitsyn, propose something | 21:02 |
@sonney2k | I have to fix bugs in the meantime | 21:02 |
@lisitsyn | sonney2k: I'll commit if you don't mind | 21:02 |
@lisitsyn | sonney2k: Compling $filename instead of g++ .................... | 21:02 |
shogun-buildbot | build #943 of deb2 - static_interfaces is complete: Failure [failed test octave_static] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/943 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 21:03 |
@sonney2k | lisitsyn, but make sure we see warnings & errors | 21:04 |
@lisitsyn | yeah will check | 21:05 |
shogun-buildbot | build #1083 of deb3 - modular_interfaces is complete: Failure [failed compile python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1083 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 21:07 |
shogun-notifier- | shogun: Sergey Lisitsyn :develop * 9799f19 / src/Makefile.template: https://github.com/shogun-toolbox/shogun/commit/9799f19274cb778d4282f6a704b9220ea50c6929 | 21:08 |
shogun-notifier- | shogun: Show just the name of compiled object file instead of long real command that is called | 21:08 |
shogun-buildbot | build #944 of deb2 - static_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/944 | 21:10 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds] | 21:10 | |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has joined #shogun | 21:11 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7062528 | 21:11 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has left #shogun [] | 21:11 | |
@lisitsyn | mkay | 21:11 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun | 21:12 | |
shogun-buildbot | build #1084 of deb3 - modular_interfaces is complete: Failure [failed compile python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1084 blamelist: Soeren Sonnenburg <sonne@debian.org> | 21:17 |
shogun-buildbot | build #788 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/788 | 21:19 |
@sonney2k | wiking, you!!!!! | 21:25 |
@sonney2k | go to line 20 in your RescaleFeatures.h | 21:25 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 2ffcef3 / src/shogun/preprocessor/RescaleFeatures.h: https://github.com/shogun-toolbox/shogun/commit/2ffcef3d144851790c0197bdb58407212033f648 | 21:27 |
shogun-notifier- | shogun: fix doxygen error | 21:27 |
shogun-buildbot | build #1085 of deb3 - modular_interfaces is complete: Failure [failed compile python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1085 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 21:28 |
-!- travis-ci [~travis-ci@ec2-54-242-186-200.compute-1.amazonaws.com] has joined #shogun | 21:29 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7062928 | 21:29 |
-!- travis-ci [~travis-ci@ec2-54-242-186-200.compute-1.amazonaws.com] has left #shogun [] | 21:29 | |
* wiking suspcts that there's a bug in serialization | 21:31 | |
@wiking | sonney2k: checking | 21:31 |
@wiking | sonney2k: yes? | 21:31 |
@sonney2k | wiking, you bastard! | 21:32 |
@sonney2k | look at the minus sign | 21:32 |
@wiking | hahahaha NOOO WAY it's not utf8 | 21:32 |
@wiking | yes i got ya | 21:32 |
@sonney2k | it is utf8 | 21:32 |
@wiking | copy pasting is hurting | 21:32 |
@wiking | sonney2k: btw: have u tested serialization on a trained machine | 21:32 |
@wiking | ? | 21:33 |
@sonney2k | wiking, what do you mean? | 21:33 |
@wiking | sonney2k: i'm getting very weird stuff | 21:33 |
@wiking | sonney2k: but i'll now write a test for it | 21:33 |
@wiking | my trained machine if i serialize and then deserialize | 21:33 |
@wiking | it's always giving 1 as a predicted class | 21:34 |
@wiking | (liblinear) | 21:34 |
@wiking | but i'll make a unittest :P | 21:34 |
shogun-notifier- | shogun: Viktor Gal :develop * db8e225 / src/configure: https://github.com/shogun-toolbox/shogun/commit/db8e225533234c7f726cc0280ceedc4cbd306769 | 21:42 |
shogun-notifier- | shogun: Add SFMT_MEXP and DSFMT_MEXP macros to config.h | 21:42 |
@wiking | mmm GaussianBlobsDataGenerator | 21:49 |
@wiking | where's a doc about this :) | 21:50 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has joined #shogun | 21:50 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7063089 | 21:50 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has left #shogun [] | 21:50 | |
@sonney2k | wiking, the doc is called heiko | 21:50 |
@wiking | sonney2k: hehehe i supposed | 21:50 |
@wiking | it'd be cool if i'd know how i can fast generate with it samples from 2 gaussians | 21:51 |
@sonney2k | lisitsyn, http://astro.temple.edu/~tuc17157/budgetedSVM/ | 22:02 |
@sonney2k | hmm is that worth anything? | 22:02 |
@lisitsyn | sonney2k: budgeted means it is like mini-batch? | 22:03 |
@lisitsyn | I am afraid of putting one more thing to shogun | 22:04 |
@lisitsyn | too many unsupportable stuff already | 22:05 |
shogun-buildbot | build #791 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/791 blamelist: Viktor Gal <viktor.gal@maeth.com> | 22:05 |
@sonney2k | from their table I am not impressed at all | 22:05 |
-!- travis-ci [~travis-ci@ec2-54-242-186-200.compute-1.amazonaws.com] has joined #shogun | 22:06 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7063626 | 22:06 |
-!- travis-ci [~travis-ci@ec2-54-242-186-200.compute-1.amazonaws.com] has left #shogun [] | 22:06 | |
@wiking | how do i clear the features and the labels in a machine | 22:10 |
@lisitsyn | wiking: set with something else! | 22:10 |
@wiking | lisitsyn: .... | 22:11 |
@wiking | lisitsyn: need an api :P | 22:11 |
@lisitsyn | wiking: why? | 22:11 |
@lisitsyn | wiking: is that common? | 22:11 |
@wiking | lisitsyn: maybe we dont want to serialize a huge feature matrix + labels | 22:11 |
@wiking | when u want to save the machine | 22:11 |
@lisitsyn | wiking: so what do I do then? | 22:14 |
@wiking | lisitsyn: just save the machine with it's paramters | 22:15 |
@lisitsyn | clear_features is redundant | 22:15 |
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 245 seconds] | 22:17 | |
shogun-notifier- | shogun: Viktor Gal :develop * b1c6bb0 / tests/unit/base/Serialization_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/b1c6bb0b11bc0ca0e0bd2928efd353b192e6f205 | 22:17 |
shogun-notifier- | shogun: Add serialization test for classification machine | 22:17 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has joined #shogun | 22:20 | |
travis-ci | [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7064097 | 22:20 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has left #shogun [] | 22:20 | |
shogun-buildbot | build #1086 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1086 blamelist: Soeren Sonnenburg <sonne@debian.org> | 22:23 |
@sonney2k | wiking, what should that api call do? | 22:23 |
shogun-buildbot | build #792 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/792 blamelist: Viktor Gal <viktor.gal@maeth.com> | 22:25 |
@wiking | sonney2k: clear the model, i.e. drop features and labels | 22:27 |
@wiking | sonney2k: at least the reference | 22:28 |
@sonney2k | wiking, no I mean what you really want | 22:31 |
@sonney2k | (this was some solution to some problem :) | 22:32 |
@wiking | sonney2k: to serialize a machine and that it's not like 600 megabytes | 22:32 |
@wiking | because of my huge feature matrix | 22:32 |
@wiking | sonney2k: btw why liblinear is not thread-safe? | 22:32 |
@wiking | :D | 22:33 |
@sonney2k | heh | 22:33 |
@sonney2k | ok so what we would need is sth that drops everything except the trained model right? | 22:34 |
@wiking | yeps | 22:34 |
@sonney2k | so for a linear svm you just keep w,b | 22:34 |
@wiking | sonney2k: and even funnier | 22:34 |
@wiking | when u serialize a kernel machine | 22:34 |
@wiking | that's really crazy atm | 22:34 |
@sonney2k | for a kernel machine you need kernel, training examples that are SVs, b | 22:35 |
@sonney2k | but kaboom with multiclass again | 22:35 |
shogun-notifier- | shogun: Sergey Lisitsyn :develop * d26ca8e / src/Makefile.template: https://github.com/shogun-toolbox/shogun/commit/d26ca8e3576dbb6125be4072155d0fdcfb0238b7 | 22:36 |
shogun-notifier- | shogun: Hide deps generation and version string creation commands | 22:36 |
@sonney2k | ohh man blackburn hides all the cool stuff | 22:38 |
@lisitsyn | haha | 22:38 |
@sonney2k | wiking, lisitsyn we have to go through the list of failed tests http://shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1086/steps/test%20python_modular/logs/stdio | 22:38 |
@sonney2k | there are like 30 or so | 22:39 |
@lisitsyn | sounds like a great weekend deal | 22:39 |
@sonney2k | and if a test does fail - should we check that it uses a rng? | 22:39 |
@lisitsyn | I guess so | 22:39 |
@sonney2k | and then just regenerate data? | 22:39 |
@wiking | lisitsyn: don't u wanna colour them as gtest does it? :D | 22:39 |
@sonney2k | or should we just for the failed tests regenerate data? | 22:39 |
@sonney2k | I want ANSI ART! | 22:40 |
@lisitsyn | wiking: we emulate cmake :D | 22:40 |
@wiking | hehehehe | 22:40 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has joined #shogun | 22:40 | |
travis-ci | [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7065037 | 22:40 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has left #shogun [] | 22:40 | |
@sonney2k | wiking, lisitsyn so how do we do it? | 22:41 |
@lisitsyn | sonney2k: lets just regenerate | 22:41 |
@wiking | sonney2k: try regen | 22:41 |
@lisitsyn | and then go through that are left | 22:41 |
@wiking | second lisitsyn's idea | 22:41 |
@sonney2k | actually I just went through the list | 22:42 |
@sonney2k | it is obvious | 22:42 |
@wiking | btw i don't see now why actually multithread would fail with classification :S | 22:42 |
@sonney2k | all rng related | 22:42 |
@sonney2k | I mean statistics_mmd_* fails | 22:42 |
@sonney2k | random fourier features... | 22:42 |
@sonney2k | *clustering* | 22:42 |
@sonney2k | 3 files maybe not? | 22:43 |
@sonney2k | converter_factoranalysis_modular.py | 22:43 |
@sonney2k | lisitsyn, ^ ? | 22:43 |
@lisitsyn | sonney2k: ehm let me check | 22:43 |
@sonney2k | evaluation_cross_validation_multiclass_storage.py | 22:43 |
@sonney2k | evaluation_multiclassovrevaluation_modular.py | 22:43 |
@lisitsyn | sonney2k: FA is not using random data | 22:44 |
@sonney2k | lisitsyn, but? | 22:45 |
@lisitsyn | sonney2k: uses random initialization though | 22:45 |
@wiking | mmmm where should i put multi-thread tests in unittest? :) | 22:45 |
@sonney2k | you did commit sth recently | 22:45 |
-!- goldbug [~quassel@global-2-1.nat.csx.cam.ac.uk] has joined #shogun | 22:45 | |
@sonney2k | ohh a golden bug! | 22:46 |
* sonney2k goes on a bug hunt | 22:46 | |
@sonney2k | lisitsyn, but you changed sth in tapkee right after wiking did his commit | 22:48 |
@lisitsyn | sonney2k: no, not FA related for sure | 22:48 |
@sonney2k | lisitsyn, FA gives totally different results | 22:49 |
@lisitsyn | sonney2k: cool | 22:49 |
pickle27 | can someone say way sth stands for slash means | 22:49 |
@lisitsyn | pickle27: what what? | 22:50 |
pickle27 | lisitsyn: sth like sonney2k just said what does he mean | 22:50 |
@lisitsyn | pickle27: still don't get it sorry! | 22:51 |
@sonney2k | pickle27, I am also lost... | 22:51 |
pickle27 | lisitsyn: you said "but you changed sth in tapkee right after wiking did his commit" what does sth mean in that sentence? | 22:51 |
@lisitsyn | pickle27: something! | 22:51 |
pickle27 | oh! | 22:51 |
@sonney2k | hah | 22:52 |
@sonney2k | lisitsyn, please check FA! | 22:52 |
@lisitsyn | sonney2k: ok in a few minutes | 22:52 |
@lisitsyn | killing t-sne bug right now | 22:53 |
@sonney2k | lisitsyn, I am pretty sure that this bug is on you | 22:53 |
@lisitsyn | sonney2k: on fernando ;) | 22:54 |
@sonney2k | the rest really just needs to be re-generated | 22:54 |
@lisitsyn | sonney2k: I think I know the issue | 22:54 |
@sonney2k | lisitsyn, I know I know you always have some excuse :P | 22:54 |
@lisitsyn | sonney2k: haha sure - I just mean FA was written by fernando | 22:54 |
@sonney2k | lisitsyn, you know that we have channel logs ;) | 22:54 |
@lisitsyn | anything bad I said? :) | 22:55 |
@lisitsyn | sonney2k: I think I modified a few things with parameters | 22:55 |
@lisitsyn | and max iter is wrong now | 22:55 |
@lisitsyn | so it doesn't converge | 22:56 |
@sonney2k | hurray | 22:56 |
@sonney2k | shogun tests find bug in tapkee | 22:56 |
gsomix | bugs, bugs... your bugs do not know how to dance! http://www.youtube.com/watch?v=E5KC1E5NyR0 | 22:56 |
@lisitsyn | bad nobody uses tapkee :D | 22:57 |
shogun-notifier- | shogun: Soeren Sonnenburg :develop * 56ee307 / data: https://github.com/shogun-toolbox/shogun/commit/56ee307be7a17ff8124cdff453bc32a3e7c7b770 | 22:57 |
shogun-notifier- | shogun: require new data version | 22:57 |
shogun-buildbot | build #1087 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1087 blamelist: Viktor Gal <viktor.gal@maeth.com> | 22:57 |
gsomix | *does not | 22:57 |
@sonney2k | wiking, ok build should be restored once lisitsyn fixes FA | 22:57 |
@wiking | coool | 22:58 |
shogun-buildbot | build #1088 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1088 blamelist: Viktor Gal <viktor.gal@maeth.com> | 22:58 |
@sonney2k | gsomix, luckily not - seems like this dance attracts more bugs | 22:58 |
pickle27 | hey guys, I'm going to start on my Eigen3 version of MCLDA, whats the best to keep both implementations intact and choose one based on which libs the user has installed? | 22:59 |
@lisitsyn | sonney2k: oh totally random indeed | 22:59 |
@sonney2k | lisitsyn, ? | 22:59 |
@wiking | sonney2k: can i rename 'Posix thread spinlock support' to simply 'spinlock support' in configure? | 22:59 |
@wiking | sonney2k: as there's spinlock in osx for example but it's not posix | 22:59 |
@sonney2k | wiking, and you intend to support osx's spinlocks? | 23:00 |
@wiking | sonney2k: yep | 23:00 |
@wiking | sonney2k: http://developer.apple.com/library/mac/#documentation/Darwin/Reference/ManPages/man3/spinlock.3.html | 23:00 |
@wiking | it's pretty much the same as posix | 23:00 |
@sonney2k | jsut do it | 23:00 |
@lisitsyn | isn't worth | 23:00 |
@wiking | lisitsyn: :D | 23:01 |
@sonney2k | we are not worthy | 23:01 |
@lisitsyn | yeah that's for sure | 23:01 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has joined #shogun | 23:02 | |
travis-ci | [travis-ci] it's Sergey Lisitsyn's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7065521 | 23:02 |
-!- travis-ci [~travis-ci@ec2-54-242-220-92.compute-1.amazonaws.com] has left #shogun [] | 23:02 | |
@wiking | does it hurt too much if there's a ';' line in .h? | 23:02 |
@wiking | no ok it's in an inline function | 23:02 |
@wiking | heheh ok there's then an empty instruction | 23:02 |
@wiking | :) | 23:02 |
@lisitsyn | sonney2k: totally random | 23:03 |
@wiking | *&lock is like lock right?:) | 23:08 |
@wiking | "We will add general .csv readers anyways. So don't waste your time on that." | 23:08 |
@wiking | sonney2k: ^ ? | 23:08 |
@wiking | sonney2k: numpy is good enough no? | 23:08 |
@lisitsyn | sonney2k: ohh reference implementation is wrong! | 23:09 |
@lisitsyn | :D | 23:09 |
@wiking | sonney2k: ? :D | 23:09 |
@wiking | lisitsyn: ???? :) | 23:09 |
@lisitsyn | sonney2k: laurens van der maaten is who to blame! | 23:09 |
@wiking | :D | 23:09 |
@lisitsyn | he implicitly introduced a bug | 23:09 |
@wiking | what should rescalefeatures do | 23:11 |
@wiking | if there's only one feature vector? | 23:11 |
@wiking | :S | 23:11 |
@lisitsyn | wiking: sepukku! | 23:11 |
@sonney2k | wiking, die | 23:11 |
@wiking | hahahahah | 23:11 |
@wiking | you guys are great :) | 23:11 |
@wiking | can't wait hacking at the workshop | 23:12 |
@wiking | :) | 23:12 |
@sonney2k | wiking,nothing of course | 23:12 |
@lisitsyn | wiking: subtract something random | 23:12 |
@lisitsyn | and normalize using a t-student distributed | 23:12 |
@lisitsyn | using median sampled MCMC hammer quasi-newton method! | 23:12 |
@wiking | sonney2k: should we notify the user via MSG_DEBUG that we didn't do anything...? | 23:12 |
@lisitsyn | wiking: MSG_DEBUG("can't touch this"); | 23:13 |
@wiking | :> | 23:13 |
@sonney2k | wiking, yes it is but it is sweet when you can from libshogun read .csv's - so no memory duplication | 23:13 |
@sonney2k | wiking, no its all good | 23:13 |
@sonney2k | just do nothing | 23:13 |
@wiking | mmm i should gc my git | 23:14 |
shogun-notifier- | shogun: Viktor Gal :develop * 5a10d5e / src/configure,src/shogun/base/Parallel.h: https://github.com/shogun-toolbox/shogun/commit/5a10d5e1a1d08f03cc6fca9f977b9507c0ce7d26 | 23:14 |
shogun-notifier- | shogun: Add spinlock support for OSX | 23:14 |
@lisitsyn | sonney2k: I have no idea what to fix here | 23:14 |
@wiking | woooohooo this was maybe a too early push :S | 23:16 |
@wiking | segfaaaaullt | 23:17 |
@wiking | to the rescue :) | 23:17 |
@sonney2k | wiking, this and a whitespace error in line 23 of Parallel.h | 23:18 |
shogun-buildbot | build #793 of cyg1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/793 | 23:18 |
@sonney2k | man luckily we don't run the bb on an EC2 | 23:19 |
@wiking | hehehehe | 23:19 |
@sonney2k | but dedicated machine | 23:19 |
@wiking | bill would be huge | 23:19 |
@sonney2k | around the clock building is expensive | 23:19 |
@wiking | ok i think a rebuild fixed it | 23:19 |
shogun-buildbot | build #794 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/794 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 23:19 |
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has quit [Ping timeout: 245 seconds] | 23:20 | |
@wiking | fixed | 23:20 |
shogun-notifier- | shogun: Viktor Gal :develop * 710a210 / src/shogun/base/Parallel.h: https://github.com/shogun-toolbox/shogun/commit/710a21087a7609d4920655643351656adca0b462 | 23:20 |
shogun-notifier- | shogun: Fix whitespace error introduced in Parallel.h | 23:20 |
@wiking | :P | 23:20 |
@sonney2k | wiking, two whitespace errors and a crash | 23:22 |
@sonney2k | could be the name of a good movie ;P | 23:22 |
@wiking | sonney2k: heheh crash was not true | 23:22 |
@wiking | sonney2k: i just needed to recompile the unittests | 23:22 |
@sonney2k | lets hope it stays like this | 23:25 |
@wiking | anybody gets this warning: | 23:26 |
@wiking | ../shogun/distributions/LinearHMM.h:95:13: warning: 'shogun::CLinearHMM::get_likelihood_example' hides overloaded virtual function [-Woverloaded-virtual] float64_t get_likelihood_example(uint16_t* vector, int32_t len); | 23:26 |
@wiking | woah | 23:27 |
@wiking | suddenly multiclass ocas started to fail | 23:27 |
gsomix | cool, travis is very useful. I don't need to compile all interfaces at my notebook for checking tests now. :3 | 23:28 |
shogun-notifier- | shogun: Viktor Gal :develop * 3e8ef0a / src/shogun/preprocessor/ (2 files): https://github.com/shogun-toolbox/shogun/commit/3e8ef0a8723d8b4e551eda1d75ecdebeac26d1db | 23:29 |
shogun-notifier- | shogun: Fix RescaleFeatures when there's only 1 feature vector | 23:29 |
-!- travis-ci [~travis-ci@ec2-23-20-77-42.compute-1.amazonaws.com] has joined #shogun | 23:33 | |
travis-ci | [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/7066260 | 23:33 |
-!- travis-ci [~travis-ci@ec2-23-20-77-42.compute-1.amazonaws.com] has left #shogun [] | 23:33 | |
shogun-notifier- | shogun: Viktor Gal :develop * 505fa27 / tests/unit/preprocessor/RescaleFeatures_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/505fa2796f5a7d286d9a1c8a26ed523c57fcadf5 | 23:35 |
shogun-notifier- | shogun: Remove RescaleFeatures apply_to_feature_vector test | 23:35 |
@lisitsyn | sonney2k: gsomix don't you think Factory is misleading word here? | 23:35 |
@wiking | actually... | 23:35 |
@wiking | shit i've realised | 23:35 |
@wiking | with init() actually i can make the RescaleFeatures work | 23:35 |
@wiking | only on one feature vector | 23:36 |
@wiking | init on a big feature matrix and then save the parameters for each row | 23:36 |
gsomix | lisitsyn, hm, is something like "LabelsConverter" ok? | 23:36 |
-!- dvalcarce [5b757556@gateway/web/freenode/ip.91.117.117.86] has joined #shogun | 23:37 | |
gsomix | nope .___. | 23:37 |
@wiking | gsomix: *Factory | 23:37 |
@lisitsyn | why not to put them directly to labels? | 23:37 |
@sonney2k | lisitsyn, so any idea how to fix FA stuff? or what shall we do? | 23:37 |
@lisitsyn | factory is something that creates new objects | 23:37 |
gsomix | lisitsyn, into CLabels? | 23:37 |
@lisitsyn | sonney2k: I am trying to find a bug in algorithm itself | 23:37 |
@sonney2k | lisitsyn, ok | 23:38 |
@sonney2k | lisitsyn, btw have you seen - I've created the other mailinglists now | 23:38 |
@lisitsyn | sonney2k: yes of course | 23:38 |
@sonney2k | not of course | 23:38 |
@sonney2k | it is still not 100% that everything works as expected | 23:38 |
gsomix | sonney2k, lisitsyn sooo, what should I do now? | 23:40 |
@sonney2k | dvalcarce, hey there! welcome back! | 23:41 |
dvalcarce | hi all! | 23:41 |
@sonney2k | gsomix, well write getline | 23:41 |
@lisitsyn | sonney2k: oh okay it is not unique | 23:42 |
@sonney2k | like the gnu function | 23:42 |
dvalcarce | sonney2k, sorry for disappering :/ | 23:42 |
@sonney2k | lisitsyn, but it shoudl be deterministic? | 23:42 |
@lisitsyn | sonney2k: uses random initialization | 23:42 |
@lisitsyn | through eigen | 23:42 |
@sonney2k | dvalcarce, yeah kind of the wrong signal | 23:42 |
@sonney2k | lisitsyn, oh | 23:43 |
gsomix | sonney2k, hm, I'm writing factory for labels now. should I stop it? | 23:43 |
dvalcarce | sonne2k, I've just made a PR | 23:43 |
@sonney2k | gsomix, it is not much work to go I guess or? | 23:44 |
@sonney2k | dvalcarce, does it have contour lines now? | 23:44 |
shogun-buildbot | build #1090 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1090 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 23:44 |
gsomix | sonney2k, I mean, that lisitsyn have suggestions or so... | 23:45 |
dvalcarce | sonney2k, kind of, a central contour line and a colorbar | 23:45 |
gsomix | sonney2k, btw, getline for AsciiFile class, right? | 23:46 |
@sonney2k | gsomix, yes | 23:46 |
@sonney2k | we have some ugly code everywhere reproducing this | 23:47 |
shogun-buildbot | build #1089 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1089 blamelist: Soeren Sonnenburg <sonne@debian.org> | 23:47 |
@sonney2k | just stupid | 23:47 |
@sonney2k | lisitsyn, so how can you fix that then? | 23:47 |
gsomix | sonney2k, ok, got it. | 23:48 |
@lisitsyn | sonney2k: yeah working on that | 23:48 |
gsomix | sonney2k, lisitsyn, and about LabelsFactory - just write your suggestions to PR, it's better place for conversation | 23:49 |
shogun-buildbot | build #795 of cyg1 - libshogun is complete: Failure [failed configure] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/795 blamelist: Viktor Gal <viktor.gal@maeth.com> | 23:51 |
@lisitsyn | sonney2k: simplest we can do now is just to check it didn't fail | 23:55 |
@sonney2k | lisitsyn, ok just do what you think is most reasonable | 23:58 |
@sonney2k | dvalcarce, hmmh cannot get it to work? | 23:58 |
@sonney2k | dvalcarce, do you have a screenshot? | 23:58 |
--- Log closed Sat May 11 00:00:36 2013 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!