Open in new window / Try shogun cloud
--- Log opened Wed Jul 24 00:00:44 2013
-!- nube1 [~rho@36.252.126.80] has quit [Ping timeout: 246 seconds]00:22
* iglesiasg ZZZzzzZZ00:36
@iglesiasggood night!00:36
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]00:37
-!- nube [~rho@36.253.40.168] has joined #shogun00:37
-!- nube [~rho@36.253.40.168] has quit [Ping timeout: 256 seconds]00:42
-!- FSCV [~FSCV@50.7.50.60] has quit [Quit: Leaving]01:04
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has left #shogun ["QUIT :Leaving."]01:30
-!- nube [~rho@36.253.146.74] has joined #shogun01:48
-!- nube [~rho@36.253.146.74] has quit [Ping timeout: 264 seconds]02:27
-!- nube [~rho@36.253.129.33] has joined #shogun02:31
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has quit [Ping timeout: 268 seconds]02:52
shogun-buildbotbuild #402 of nightly_all is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/40203:08
-!- nube1 [~rho@36.253.255.182] has joined #shogun03:47
-!- nube [~rho@36.253.129.33] has quit [Ping timeout: 264 seconds]03:49
-!- nube1 [~rho@36.253.255.182] has quit [Ping timeout: 246 seconds]03:55
-!- nube [~rho@49.244.110.25] has joined #shogun03:58
-!- Yanglittle [b74040fc@gateway/web/freenode/ip.183.64.64.252] has joined #shogun04:04
Yanglittleexcuse me, how to use the kernel_normalizer in the python_modular?04:04
Yanglittleexcuse me, how to use the kernel_normalizer in the python_modular?04:06
shogun-buildbotbuild #467 of nightly_default is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/46704:26
-!- nube [~rho@49.244.110.25] has quit [Quit: Leaving.]04:35
-!- foulwall [~user@2001:da8:215:503:5dac:765e:def8:c0d3] has joined #shogun05:18
-!- nube [~rho@116.90.239.13] has joined #shogun05:43
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.]06:02
-!- nube [~rho@116.90.239.13] has joined #shogun06:06
-!- foulwall [~user@2001:da8:215:503:5dac:765e:def8:c0d3] has quit [Remote host closed the connection]06:11
-!- nube [~rho@116.90.239.13] has quit [Ping timeout: 246 seconds]06:19
-!- nube [~rho@116.90.239.13] has joined #shogun06:59
-!- nube [~rho@116.90.239.13] has quit [Read error: Connection reset by peer]07:24
-!- nube [~rho@116.90.239.13] has joined #shogun07:26
gsomixgood morning07:42
-!- hushell [~hushell@c-24-21-169-136.hsd1.or.comcast.net] has joined #shogun07:48
-!- gsomix_ [~gsomix@109.169.225.10] has joined #shogun08:44
-!- gsomix [~gsomix@80.234.25.58] has quit [Ping timeout: 246 seconds]08:47
-!- foulwall [~user@2001:da8:215:503:f4eb:f5fe:de2f:75f7] has joined #shogun09:19
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.]09:28
-!- nube1 [~rho@116.90.239.3] has joined #shogun09:28
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has joined #shogun09:34
-!- nube1 [~rho@116.90.239.3] has quit [Ping timeout: 248 seconds]10:18
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun10:28
-!- mode/#shogun [+o iglesiasg] by ChanServ10:28
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has quit [Ping timeout: 250 seconds]10:30
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has joined #shogun10:35
-!- nube [~rho@116.90.239.13] has joined #shogun10:36
thoralfHey GUIs10:43
thoralferr guys ;)10:43
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.]10:46
@iglesiasghehe hi thoralf!10:47
thoralfiglesiasg:)10:47
-!- nube [~rho@116.90.239.3] has joined #shogun10:48
-!- foulwall [~user@2001:da8:215:503:f4eb:f5fe:de2f:75f7] has quit [Remote host closed the connection]11:01
-!- foulwall [~user@2001:da8:215:503:f4eb:f5fe:de2f:75f7] has joined #shogun11:01
thoralfsonney2k: Didn't get java_modular running and found out, jblas is missing.  Could configure check the jblas dependency?11:21
thoralfsonney2k: I missed, that it already checks the dependecy.11:36
@sonney2kthoralf, I totally feel like a UI now :D11:55
thoralf:)11:55
@sonney2kvan51, any updates on quadratic?11:55
hushelliglesiasg: hi11:56
@iglesiasghey hushell11:56
van51sonney2k: not yet, I'm working on it11:56
van51sonney2k: I'm trying to minimize the number of calls to murmurhash11:56
@sonney2kgsomix_, are you reworking the line reader?11:56
hushelliglesiasg: did you run the recent unit test of SGObject.clone_equals_Mosek?11:56
thoralfsonney2k: I fixed check-examples for the other interfaces as well: https://github.com/shogun-toolbox/shogun/pull/124311:56
hushelliglesiasg: I got errors like this: [ RUN      ] SGObject.clone_equals_Mosek11:57
hushellProgram received signal SIGSEGV, Segmentation fault.11:57
hushell0x00007ffff3fc794b in MSK_deleteenv () from /usr/local/lib/libmosek64.so.6.011:57
thoralfsonney2k: I think it's ready to merge, as soon as travis is done.11:57
van51sonney2k: I've sent a PR btw, check it when you have time pls11:57
@iglesiasghushell: oh. No, I didn't run it11:57
@iglesiasghushell: in what line does it crash?11:57
hushelliglesiasg: this test case is generated automatically11:58
thoralfsonney2k: I made several changes to isolate trivial changes in the diff.  Shouldn't be a big deal.11:58
@iglesiasghushell: yes, I think it is what heiko and/or wiking were working on recently11:58
hushellyou can check unit/base/clone_unittest.cc11:59
@iglesiasghushell: I guess they didn't try with Mosek install and maybe that's why11:59
@iglesiasghushell: I don't have mosek install in this machine either so I cannot try right now11:59
hushellthey created 753 tests...11:59
hushelliglesiasg: okay, I'll skip it right now12:00
@iglesiasghushell: it would be nice if it could be fixed actually12:01
@iglesiasgI can try to have a look at it some time this week12:01
@iglesiasgI will write a github issue so we don't forget12:01
@sonney2kvan51, I guess you can just store the hash in an array (which you have to allocate and free each time)12:02
@sonney2kthoralf, yeah sounds good12:02
van51sonney2k: yes that's the only thing I can think of as well12:02
@sonney2khushell, that is a bug iglesiasg should fix :)12:02
@sonney2kvan51, lets hope malloc overhead is not so bad :)12:03
hushellsonney2k: we can wait Heiko install Mosek :)12:03
thoralfsonney2k: Disclaimer: Couldn't get r_modular and perl_modular running -- they were already broken before my intervention12:03
thoralfsonney2k: Didn't touch ruby_modular and csharp_modular, since I don't have (and want to) install all dependencies.12:04
gsomix_sonney2k, yep12:07
gsomix_sonney2k, I'll update PR at evening12:07
@sonney2kthoralf, bah so I have to do it12:08
@sonney2khushell, I guess Heiko doesn't like to be a hero too often :D12:09
@sonney2khushell, besides it is iglesiasg's code so his duty :)12:09
@iglesiasgsonney2k: all right, I got it :) I said I would take a look at it this week12:10
@iglesiasgsonney2k: I don't even have mosek installed here right now12:10
@sonney2kiglesiasg, I don't know anyone who has :P12:10
@iglesiasgsonney2k:  I created the issue some minutes ago and assigned it to me12:10
@iglesiasgsonney2k: well hushell should have it if he has gotten the error12:11
@sonney2k...except hushell :)12:11
hushellsonney2k: yeah I have it right now12:11
@iglesiasgsonney2k: I talked about it with Nico during the workshop, I would really like to get rid of the Mosek thing12:11
@iglesiasgsonney2k: keep the solver but use something different instead of Mosek for the QP12:11
hushelliglesiasg: we can try LBFGS for doing this12:12
@sonney2kiglesiasg, yeah would be great. Otherwise it is just dead code *not* covered by any tests12:12
@iglesiasghushell: no idea what that is, let me check12:12
@iglesiasghushell: nonlinear optimization?12:13
@iglesiasghushell: why non linear?12:13
@iglesiasgah wait that is BFGS12:13
@iglesiasgI guess the L stands for linear12:13
@iglesiasgLimited-memory BFGS, maybe not then12:14
hushelliglesiasg: should be L-BFGS my bad12:14
hushelliglesiasg: I remember I saw it somewhere in shogun, someone ported liblbfgs12:15
@iglesiasghushell: aham I see!12:15
hushellok in the optimization12:16
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun12:16
lisitsynha I broke through the shitty connection12:16
@iglesiasghushell: what I am wondering is that what I am solving with Mosek is a quadratic program. It reads that BFGS is for nonlinear optimization, so I am not sure whether it can be applied for QP12:17
@sonney2kiglesiasg, what exactly do you  need? a general qp or sth simpler like in libqp?12:18
@sonney2khttp://cmp.felk.cvut.cz/~xfrancv/libqp/html/12:19
@iglesiasgsonney2k: we already faced this issue this year, libqp was not enough12:19
@sonney2k^ iglesiasg is that sufficient?12:19
@sonney2kiglesiasg, I don't remember what you need in addition12:19
@iglesiasgsonney2k: let me have a look an refresh it12:19
@iglesiasgsonney2k:  IIRC it was something related to block constraints12:19
hushelliglesiasg: I know some people using L-BFGS for SSVM, this package is the most common one http://www.di.ens.fr/~mschmidt/Software/minFunc.html12:19
@iglesiasgsonney2k: ok, I remember now12:20
@iglesiasgsonney2k: the problem is pretty similar to "QP task with box constraints and a single linear equality constraint"12:20
@iglesiasgsonney2k: but we have a x^T a <= b12:20
-!- nube [~rho@116.90.239.3] has quit [Ping timeout: 240 seconds]12:20
hushelliglesiasg: this is also related to what I am going to do next, using SGD for SSVM12:20
@sonney2kthat is what libqp does12:20
@iglesiasgsonney2k: it does x^T a = b, we need <=12:21
@iglesiasghushell: yep12:21
@sonney2kiglesiasg, look at the homepage it seems to also do <=12:21
@iglesiasgsonney2k, hushell : Nico suggested that one could actually used SGD to solve this QP too12:21
@iglesiasgsonney2k: do you mean the <= in the first part: "QP task with simplex constraints"?12:22
@sonney2kyeah12:22
@iglesiasgsonney2k: well, but the constraints are different in there12:22
@iglesiasgsonney2k: simplex constraints, no box constraints12:22
@sonney2kahh misread you need dot with a12:23
@iglesiasgsonney2k: I am no optimization expert, but I think with the first ones you cannot couple constraints among the variables of x12:23
@iglesiasgall right, then12:23
@sonney2kiglesiasg, an sgd style solution should be very easy to do12:23
hushelliglesiasg: you mean in dual form this constraint is no longer exist12:24
@sonney2kiglesiasg, it was x^T 1 <=b anyway12:24
@iglesiasgsonney2k: it should be very easy to implement but the tuning can get hard :)12:24
@iglesiasgsonney2k: yeah sure but our a is not 1 always :)12:24
@iglesiasgmmm what was the a...12:25
@iglesiasgIIRC we get one constraint for each margin violator12:25
@iglesiasgand our x looks like: x = [w slacks]12:26
@iglesiasgso I think the a was basically ones, zeroes and one -112:26
@iglesiasgones for w, a -1 for the example that gives the constraint, zeros in the rest12:27
@iglesiasghushell: to solve it with SGD I think we don't need to go to the dual, right?12:28
@iglesiasghushell: we need to write the constrained as unconstrained12:28
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun12:29
shogun-notifier-shogun: lambday :develop * 665709d / tests/unit/base/clone_unittest.cc.py: https://github.com/shogun-toolbox/shogun/commit/665709daf9cab6de012025d952a35a96366ba6ee12:29
shogun-notifier-shogun: failing SNPStringKernel added to clone ignore list to pass all unit-tests12:29
shogun-notifier-shogun: lambday :develop * 5b20d7f / tests/unit/mathematics/logdet/LogDetEstimator_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/5b20d7f18816513616f7275166cf9d19ab6a891f12:29
shogun-notifier-shogun: reduced time and accuracy for log-det estimator unit-test12:29
shogun-notifier-shogun: lambday :develop * 0197f6a / src/shogun/mathematics/logdet/ (2 files): https://github.com/shogun-toolbox/shogun/commit/0197f6acc1c9bc299dc29e7fb1aa82940d5217e612:29
shogun-notifier-shogun: removed unnecessary template instantiation12:29
shogun-notifier-shogun: lambday :develop * 8ecc2dc / tests/unit/base/clone_unittest.cc.py: https://github.com/shogun-toolbox/shogun/commit/8ecc2dc6f2a1bb6bdc0715a24a03f514c764cd0412:29
shogun-notifier-shogun: added failing weighted degree position string kernel to ignore list12:29
shogun-notifier-shogun: Soeren Sonnenburg :develop * c76a576 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/c76a5766667d8c2ba0f82540aef3166dc714914912:29
shogun-notifier-shogun: Merge pull request #1284 from lambday/feature/log_determinant12:29
shogun-notifier-shogun:12:29
shogun-notifier-shogun: SNPStringKernel added to clone-test ignore list, few changes in log-det12:29
@iglesiasgsonney2k: Nico also suggested talking to vojtech and see if libqp could be extended to cover this problem12:30
hushelliglesiasg: yes, but usually people work with dual form and optimize with SGD12:30
@iglesiasghushell: aham!12:30
hushelliglesiasg: we can try both actually12:30
@iglesiasghushell: do you know why?12:30
hushelliglesiasg: not very clear, I need to check the papers, but maybe the dual form has simpler constraints12:34
hushelliglesiasg: we can have deep discussion when comes to the point12:34
shogun-buildbotbuild #1518 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1518  blamelist: lambday <heavensdevil6909@gmail.com>12:35
hushellI have to go to sleep now, Sorry for my absense of the mid-term meeting, I was not able to get connected12:35
@iglesiasghushell: sure. I have the intuition that structured learning would be very inefficient in the dual though12:36
hushellSorry guys, I will summarize my work in my next weekly report if you are interested in12:36
@iglesiasghushell: well not in general, bundle methods in the dual are fast. But the algorithm implemented in the PrimalMosekSOSVM I think it would be worse in the dual12:36
@iglesiasgworse as slower12:36
shogun-buildbotbuild #1519 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1519  blamelist: Soeren Sonnenburg <sonne@debian.org>12:37
@sonney2kvan51, I looked at the code it looks correct now! One thing - maybe it might make sense to do the hashing trick in some function I mean the12:38
@sonney2kuint32_t h_idx = CHash::MurmurHash3((uint8_t* ) &vec.features[i].feat_index, sizeof (index_t),12:38
@sonney2k   vec.features[i].feat_index);12:38
@sonney2kh_idx = h_idx % dim;12:38
@sonney2kthis thing somewhere in a static function that you use across all hashingdotfeatures12:38
van51sonney2k: true.12:38
van51sonney2k: will do!12:39
@sonney2kvan51, if you see other common patterns - reduce code duplication where possible12:39
@sonney2kvan51, I am merging now though do it later12:39
van51sonney2k: in the same PR or in a new one?12:39
van51sonney2k: cool :D12:39
hushelliglesiasg: what do you mean worse, I think they are equivalent in the sense KKT conditions are satisfied12:39
@iglesiasghushell: worse as that it would be much slower12:40
shogun-notifier-shogun: van51 :develop * 50fe540 / / (13 files): https://github.com/shogun-toolbox/shogun/commit/50fe540e8c961f8d7d57a7e5b30f40d8f994ac3512:40
shogun-notifier-shogun: Changed behavior in Hashed numerical classes12:40
shogun-notifier-shogun: Soeren Sonnenburg :develop * fa5d7c4 / / (13 files): https://github.com/shogun-toolbox/shogun/commit/fa5d7c46ac30933af1f74915cc7616d64f8be71a12:40
shogun-notifier-shogun: Merge pull request #1282 from van51/develop12:40
shogun-notifier-shogun:12:40
shogun-notifier-shogun: Corrected behavior of Hashed numerical classes12:40
@iglesiasghushell: let me try to elaborate12:40
@sonney2kgsomix_, to avoid misunderstandings how are you doing it now?12:41
shogun-buildbotbuild #1520 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1520  blamelist: van51 <vangelis_51@hotmail.com>12:43
@sonney2kthoralf, currently the buildbot installs stuff somewhere and then runs the make check-examples etc12:43
@iglesiasghushell: it is about the evaluation of the loss, in the dual you need to go over all the non-zero dual variables to compute the loss for one example. The number of non-zero dual variables is the order of n12:44
@sonney2kso it would be rather useful to have that at least for the test still12:44
@iglesiasghushell: and that is in general, not even for structured learning. In structured learning there are several dual variables for each data points, so it takes even more computations12:45
hushelliglesiasg: yes, this is the number of violated constraints basically12:45
@iglesiasghushell: yes12:45
shogun-buildbotbuild #1521 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1521  blamelist: Soeren Sonnenburg <sonne@debian.org>12:45
shogun-notifier-shogun: Soeren Sonnenburg :develop * 4964cbb / tests/unit/mathematics/Complex_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/4964cbbf69e48e398c79a0fc295c137e575f555012:46
shogun-notifier-shogun: make complex a bit more tolerant12:46
@sonney2kfoulwall, any updates on the readme / the ocr demo?12:46
hushelliglesiasg: let's talk based on the BMRM paper : http://users.cecs.anu.edu.au/~chteo/pub/TeoLeSmoVis07.pdf12:46
@iglesiasghushell: I don't have full understanding of this behaviour though. This is something Patrick explained some months ago12:46
hushellIIRC, primalMosek is solving eq.10, and BMRM solve eq.12 using libqp12:47
shogun-buildbotbuild #1517 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1517  blamelist: lambday <heavensdevil6909@gmail.com>12:47
hushelliglesiasg: okay, I think also too late for me to discuss this in detail12:48
hushelliglesiasg: we can find another time to figure out how to solve the particular QP efficiently12:49
@iglesiasghushell: hehe sure. Let's talk again about it when the moment to get Mosek to a better life arrives :)12:49
-!- nube [~rho@49.244.94.101] has joined #shogun12:49
@iglesiasghushell: good night!12:49
hushelliglesiasg: night, btw I updated the PR12:49
@iglesiasghushell: cool, I check right away12:49
hushelliglesiasg: okay, CU12:50
-!- hushell [~hushell@c-24-21-169-136.hsd1.or.comcast.net] has quit [Quit: WeeChat 0.3.7]12:50
@sonney2kshogun-buildbot, force build 'deb1 - libshogun'12:52
shogun-buildbotbuild forced [ETA 15m38s]12:52
shogun-buildbotI'll give a shout when the build finishes12:52
shogun-buildbotbuild #1522 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/152212:52
@sonney2kvan51, btw another good test for the hashing dense / sparse would be to compare it directly with the dense features. for sufficiently large bitsize all the ops should be the same.12:53
@sonney2kshogun-buildbot, dance12:53
shogun-buildbot<(^.^<)12:53
shogun-buildbot<(^.^)>12:53
shogun-buildbot(>^.^)>12:53
shogun-buildbot(7^.^)712:53
shogun-buildbot(>^.^<)12:53
@sonney2kman a little bit of green again!12:53
@iglesiasgso sweet :)12:54
* sonney2k goes to see some blue12:54
@sonney2kcya12:54
@iglesiasghave fun!12:54
shogun-buildbotbuild #745 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/745  blamelist: Soeren Sonnenburg <sonne@debian.org>12:54
thoralfsonney2k: "thoralf, bah so I have to do it" <-- I didn't break them.  I just couldn't change them.12:57
thoralfsonney2k: "thoralf, currently the buildbot installs stuff somewhere and then runs the make check-examples etc" <-- I don't understand.  With my changes, check-examples relies on the compiled libraries from the source directory.  Not the target location.12:58
@sonney2klamday, HeikoS ... http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/745/steps/compile/logs/stdio12:58
thoralfsonney2k: So, calling "make install" does not hurt, but has no effect on check-examples either.12:59
shogun-buildbotbuild #1058 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1058  blamelist: Soeren Sonnenburg <sonne@debian.org>12:59
thoralfsonney2k: Did I miss something?13:01
shogun-buildbotHey! build deb1 - libshogun #1523 is complete: Success [build successful]13:01
shogun-buildbotBuild details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/152313:01
shogun-buildbotbuild #1262 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1262  blamelist: Soeren Sonnenburg <sonne@debian.org>13:03
-!- HeikoS [~heiko@nat-185-60.internal.eduroam.ucl.ac.uk] has joined #shogun13:10
-!- mode/#shogun [+o HeikoS] by ChanServ13:10
@iglesiasghi HeikoS13:16
@HeikoSiglesiasg:  hi!13:16
@iglesiasgHeikoS: I am having a look at CMosek, and yeah the attributes does not seem to be initialized13:16
shogun-buildbotbuild #1381 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/1381  blamelist: Soeren Sonnenburg <sonne@debian.org>13:16
@iglesiasgHeikoS: but I am actually thinking that this class does not really need to be a CSGObject13:17
@HeikoSiglesiasg: yes that usually the reason13:17
@HeikoSiglesiasg: doesnt matter, it is now :)13:17
@HeikoSiglesiasg: if you initialise all members in the std constructor, the bug will probably go13:17
@HeikoSiglesiasg: is mosek hard to install?13:17
lisitsynI'd say easy13:17
@iglesiasgHeikoS: in the header Mosek.h I need to include some Mosek header13:17
@HeikoSlisitsyn: have you seen the two bug reports?13:18
@HeikoSits not in ubuntu repo13:18
lisitsynHeikoS: not yet13:18
lisitsynHeikoS: of course it is not13:18
@HeikoSiglesiasg: so could you fix the init bug?13:18
lisitsynHeikoS: it is proprietary $hit13:18
@iglesiasgHeikoS: yeah, sure13:18
lisitsynHeikoS: https://github.com/shogun-toolbox/shogun/issues/1265 ha!13:19
@iglesiasgHeikoS: but I think it would be a better idea not making it a CSGObject13:19
shogun-notifier-shogun: Sergey Lisitsyn :develop * b94061e / src/shogun/classifier/FeatureBlockLogisticRegression.cpp: https://github.com/shogun-toolbox/shogun/commit/b94061edad77a196439b7ac3299c5748456c618813:22
shogun-notifier-shogun: Fixed group lasso indices check. Fixes #128613:22
@HeikoSiglesiasg: maybe, but lets dont change things, too much work ;)13:27
@HeikoSlisitsyn: do you like that idea?13:27
@HeikoSlisitsyn: unit test the group lasso!13:27
@HeikoSat least this bit you fixed13:28
lisitsynHeikoS: ok will try rather soon13:28
@HeikoSlisitsyn: because these bugs are evil, scare away people13:28
@HeikoSlisitsyn: and the guy is a potential user :)13:28
@HeikoSgood oportunity to make things better for lasso13:29
lisitsynHeikoS: about setters/getters13:29
lisitsynHeikoS: I think in python we can do some magic to cast objects13:29
@HeikoSiglesiasg: btw, valgrind this program to find uninitialised memory for Mosek#13:29
lisitsynbut in e.g. java we are going to use something like13:29
lisitsynsvm.parameter("C").as_float()13:29
@HeikoSiglesiasg:  https://gist.github.com/anonymous/606977213:30
lisitsynHeikoS: I don't know if people would like that13:30
@HeikoSlisitsyn: I dont know about the whole model selectoin framework anymore13:30
@iglesiasgHeikoS: all right, thank you13:30
@HeikoSlisitsyn: its so complicated to use and maintain13:30
@HeikoSlisitsyn: and soooo easy to emulate: a double loop for SVMs with Gaussian kernel13:30
@iglesiasgHeikoS: I am going to prepare a Mosek installation in Shogun13:30
@HeikoSso I am currently more for dropping it13:30
@HeikoSlisitsyn: have you seen my issue on this?13:30
lisitsynHeikoS: drop?13:31
@HeikoSlisitsyn: yes, just keep x-validation13:31
-!- travis-ci [~travis-ci@ec2-50-16-34-49.compute-1.amazonaws.com] has joined #shogun13:31
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/942795013:31
-!- travis-ci [~travis-ci@ec2-50-16-34-49.compute-1.amazonaws.com] has left #shogun []13:31
lisitsynHeikoS: that's a great issue13:32
lisitsynHeikoS: this gets me back to thoughts that we've got to seriously change something13:32
@HeikoSlisitsyn: this one is easy13:32
@HeikoSjust drop it13:33
@HeikoSlisitsyn: not many things depend on it13:33
@HeikoSonly GPs13:33
lisitsynHeikoS: no I mean in general13:33
@HeikoSlisitsyn:  I would rather write modelselection stuff for the machines individually13:33
lisitsynI'd drop most of code yes13:33
@HeikoSgrid-search is easy so we dont need that13:33
@HeikoSlisitsyn: yes, agree, so this is a good place to start maybe13:33
@HeikoSlisitsyn: just thinking whether this block things13:33
@HeikoSlisitsyn: for example, more intelligent search algorithms for model-selection13:33
@HeikoSlisitsyn: cannot be done anymore then13:34
@HeikoSlisitsyn: but I think its infeasible to implement those anyway13:34
lisitsynHeikoS: I have a feeling this should not be done in C++13:34
@HeikoSlisitsyn: yeah exactly13:34
@HeikoSmaybe we can do this via swig in python at some point or so13:34
lisitsyneither we raise the level of abstraction13:34
lisitsynin C++ code13:34
@HeikoSsince its not even time dependent13:34
@HeikoSlisitsyn: yep agreed,13:34
lisitsynor we use something more high-level13:34
@HeikoSIll talk with Roman on how to modify the GP stuff that is based on model-selection13:35
@HeikoSlisitsyn: I totally agree13:35
lisitsynHeikoS: the problem is that we don't use high-level C++13:35
@HeikoSlisitsyn: maybe write some thoughts in the issue, to document them13:35
@HeikoSwe often forget what we talked about :)13:35
lisitsynHeikoS: yeah will try to formulate13:35
@HeikoSlisitsyn: cool!13:35
@HeikoSI have to do some work now, see you all later today13:36
lisitsynHeikoS: well we are in search of some common point13:36
@HeikoSlisitsyn: yep, lets do a documented discussion in the issue13:36
@HeikoSthen other people also can read about it and tell what they think13:36
lisitsynHeikoS: we have more resources than we use13:36
@HeikoSlisitsyn: ?13:36
lisitsynHeikoS: well if it was more high-level13:37
lisitsynwe could spend less time etc13:37
lisitsynI mean 8 gsocers + some of us13:37
lisitsynthat's a lot13:37
@HeikoSlisitsyn: yeah, its a big thing though13:37
lisitsynthis could change the world13:37
thoralf;)13:37
lisitsynbut we do something low-level13:37
@HeikoSlisitsyn: could you formalise this in an issue?13:38
@HeikoSlisitsyn:  Id really like to document all these ideas to have proper discussions13:38
lisitsynHeikoS: yeah a bit later as I have to arbeit too13:38
@HeikoSwas very useful at c-base meeting13:38
@HeikoSlisitsyn: arbeit arbeit brot brot banane :)13:38
@HeikoSlisitsyn: ok see you later then13:38
lisitsynarbeit macht frei haha13:38
@HeikoSlisitsyn: ouch!13:38
lisitsynHeikoS: hah yes not very correct I guess but that's just the language13:39
thoralfHeikoS: lisitsyn is not required to be german-PC ;)13:39
@HeikoSthoralf: well .... whatever13:39
@HeikoSsee you! :)13:39
shogun-buildbotbuild #1524 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1524  blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com>13:42
lisitsynthoralf: that's a curious thing one can't say some things just because of its usage earlier13:42
thoralflisitsyn: it's called self-censorship13:43
thoralflisitsyn: I know I wasn't responsible for this - but we learned in school to avoid this.  Overwise we could be offended by others, who practice more PC-aware speaking.13:45
lisitsynthoralf: yes sure - I just mean it has nothing to do with the meaning13:45
thoralflisitsyn: Yes.13:46
lisitsynlike hidden agenda13:46
lisitsynthoralf: I mean it is not 'kill all X' or whatever so I find it weird13:47
-!- Yanglittle [b74040fc@gateway/web/freenode/ip.183.64.64.252] has quit [Quit: Page closed]13:47
lisitsynthoralf: although I agree this should be avoided and now feel cumbersome :D13:49
lisitsynthoralf: but again I feel this agenda should be gone13:50
thoralflisitsyn: I think every "nation" has similar "overloaded" words.13:50
lisitsynthoralf: russians exalt stalin still13:50
thoralfUh.13:51
thoralfYeah, only one example.  Religion is another source for good examples. :)13:51
lisitsynthoralf: oh that's the thing I am totally PC-uncompliant13:51
lisitsyn:D13:51
thoralfHmm, having a std::istringstream - do you know how to check the status?  i.e. if I'm trying to cast an "token" to "int32" and it fails?13:54
-!- travis-ci [~travis-ci@ec2-50-16-34-49.compute-1.amazonaws.com] has joined #shogun13:54
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/942826213:54
-!- travis-ci [~travis-ci@ec2-50-16-34-49.compute-1.amazonaws.com] has left #shogun []13:54
lisitsynthoralf: there should be some methods to check if it is ok13:54
-!- FSCV [~FSCV@187.210.54.166] has joined #shogun13:54
thoralfstd::istringstream iss (line); int32_t number; iss >> number; <-- This fails, but I can't detect. :(13:55
lisitsynthoralf: well just catch it then13:55
lisitsynit is an exceptional case right?13:55
thoralflisitsyn: It fails as-in the integer is not correct in case of an overflow.  But it does not raise anything.13:56
thoralflisitsyn: Just return status false.13:56
lisitsynthoralf: no way! should throw something13:56
lisitsynor return13:56
thoralflisitsyn: But eof also gives false.13:56
lisitsynthoralf:13:57
lisitsynhttp://www.cplusplus.com/reference/istream/istream/operator%3E%3E/13:57
lisitsynthoralf: check failbit13:57
thoralflisitsyn: Thanks.13:57
thoralflisitsyn: I already read lot of stackoverflow stuff, but you're way smarter than stackoverflow. :D13:57
lisitsynthoralf: I am overflown already!13:58
lisitsyn:D13:58
thoralflisitsyn: iss.eof(), iss.good(), iss.bad(), iss.fail()14:02
thoralflisitsyn: failbit tells which bit is actually the fail bit. ;)14:02
thoralflisitsyn: But not if it failed.14:02
lisitsynthoralf: but if it is not set it is not failed right?14:05
thoralflisitsyn: Its always set.14:06
lisitsynwhat's the default then?14:06
thoralflisitsyn: Just kind of a constant.14:06
lisitsynthoralf: anyway14:06
lisitsynthoralf: (ss >> n).fail()14:06
lisitsynthoralf: I guess that14:06
lisitsynI'd go further and just14:07
thoralfiss.eofbit 2, iss.failbit 4, iss.badbit 114:07
lisitsynif (!(iss >> n).good()) { panic }14:07
lisitsynthoralf: ^ something like that I think?14:07
shogun-notifier-shogun: Fernando Iglesias :develop * 3c84872 / src/shogun/labels/MulticlassLabels.h: https://github.com/shogun-toolbox/shogun/commit/3c848725b8cf42a8ec2774e651a564c1f9e6d26b14:08
shogun-notifier-shogun: Fix MulticlassLabels doc, returned SGVector needs not be cleaned up14:08
shogun-notifier-shogun: Fernando Iglesias :develop * de7b308 / src/shogun/labels/MulticlassLabels.h: https://github.com/shogun-toolbox/shogun/commit/de7b308e9f6e927d74c599feb58c8212f30e0d1f14:08
shogun-notifier-shogun: Merge pull request #1288 from iglesias/docfix14:08
shogun-notifier-shogun:14:08
shogun-notifier-shogun: Fix MulticlassLabels doc, returned SGVector needs not be cleaned up14:08
thoralf(ss >> n).fail() <-- Whats the return value of (ss >> n)?14:09
lisitsynthoralf: stream14:11
lisitsynwith some state14:11
lisitsynany << or >> return stream14:12
lisitsynto allow stuff like14:12
lisitsyniss << 1 << 2 << 3;14:12
thoralfI see, yes.14:13
thoralfthis is int32_max + 1 as a string: "2147483648"14:14
thoralfiss >> number; casts to 214748364714:14
thoralfNo error.14:14
thoralfNothing helps.14:14
shogun-buildbotbuild #746 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/746  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>14:15
-!- travis-ci [~travis-ci@ec2-54-224-203-225.compute-1.amazonaws.com] has joined #shogun14:16
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/942845414:16
-!- travis-ci [~travis-ci@ec2-54-224-203-225.compute-1.amazonaws.com] has left #shogun []14:16
thoralflisitsyn: I can't check on "ss", I really need to call what you told: (ss >> n).fail()14:17
lisitsynthoralf: that's different stream14:19
lisitsynss >> n produces a new stream14:19
lisitsynits state differs14:19
@iglesiasgwhat is the rpm build? red hat packet?14:19
lisitsyniglesiasg: ye14:19
lisitsyns14:19
lisitsynno14:19
lisitsyn:D14:19
@iglesiasg?14:20
lisitsynI thought you are asking what is rpm14:20
lisitsyn:D14:20
@iglesiasghaha14:20
@iglesiasgI mean in our buildbot14:20
lisitsyniglesiasg: that's redhat build but I think there is no package14:20
@iglesiasglisitsyn: aham got it14:21
-!- s3b4 [59e3a2e9@gateway/web/freenode/ip.89.227.162.233] has joined #shogun14:21
@iglesiasghi s3b4!14:21
s3b4hi fernando14:21
s3b4right, i'd like to help you reproduce this seg fault i'm seeing14:22
@iglesiasgs3b4: so how does the example fail?14:22
s3b4azureuser@test-py:~/shogun-develop/examples/documented/python_modular$ python st ructure_discrete_hmsvm_bmrm.py Discrete HMSVM BMRM Segmentation fault (core dumped)14:22
s3b4and that's it14:23
s3b4i'm not sure where to find the core dump on this machine, if that would help14:23
shogun-buildbotbuild #1263 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1263  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>14:23
@iglesiasgs3b4: it would help it we could use gdb, valgrind or similar14:24
@iglesiasgs3b4: but since you are running it from python that may make things a bit harder, do you think we can try running the C++ example?14:24
s3b4never used them. i can spend say an hour on this now. any pointers for me to start?14:25
@iglesiasgs3b4: if you are using python, your machine should be ready as well to run the C++ example14:25
@iglesiasgs3b4: one second, let me try it in my machine first14:25
s3b4so how does that go?14:25
@iglesiasgs3b4:  cd to shogun/examples/undocumented/libshogun14:26
@iglesiasgs3b4: g++ -lshogun so_multiclass_BMRM.cpp14:27
shogun-buildbotbuild #1525 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1525  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>14:27
shogun-buildbotbuild #1059 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1059  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>14:27
s3b4that would be too easy :-) the compilation log is full of undefined references; is there no make for the undocumented examples?14:28
s3b4there is, and i'm runnign it now14:29
@iglesiasgs3b4: yeah, you can do make-check-examples14:29
@iglesiasgall right14:29
@iglesiasgs3b4: in the meantime let's try to gdb the python program too14:30
@iglesiasgs3b4: gdb python and then in the gdb console run structure_discrete_hmsvm_bmrm.py14:31
-!- foulwall [~user@2001:da8:215:503:f4eb:f5fe:de2f:75f7] has quit [Remote host closed the connection]14:54
shogun-buildbotbuild #1382 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/1382  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>15:07
-!- nube [~rho@49.244.94.101] has quit [Ping timeout: 264 seconds]15:19
-!- travis-ci [~travis-ci@ec2-23-20-210-220.compute-1.amazonaws.com] has joined #shogun15:23
travis-ci[travis-ci] it's Fernando Iglesias'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/943099215:23
-!- travis-ci [~travis-ci@ec2-23-20-210-220.compute-1.amazonaws.com] has left #shogun []15:23
-!- foulwall [~user@2001:da8:215:c252:d56f:94fe:80b7:bdac] has joined #shogun16:04
-!- foulwall [~user@2001:da8:215:c252:d56f:94fe:80b7:bdac] has quit [Ping timeout: 264 seconds]16:15
-!- lambday [67157c36@gateway/web/freenode/ip.103.21.124.54] has joined #shogun16:32
lambdayHeikoS: hi16:33
lambdayHeikoS: checked your comments16:33
lambdayHeikoS: so, we'll be having 2 failing unit-tests if I remove those16:33
@iglesiasglisitsyn, sonney2k, wiking, HeikoS: around?16:43
lisitsyn iglesiasg: yes for a few more minutes16:44
@iglesiasglisitsyn: good, take a look if you can at this a moment please16:45
@iglesiasghttps://gist.github.com/iglesias/607125316:45
@iglesiasgso I am running in my machine the same program that s3b4 is doing16:45
@iglesiasgfrom the same commit in github16:45
@iglesiasghere it runs completely fine, but it crashes in his machine16:46
@iglesiasgthat is gdb output of the crash (sorry for the bad formatting)16:46
@iglesiasglisitsyn: any idea why can it crash there and run fine here if it is the same version of Shogun?16:46
lisitsyniglesiasg: no can't say much about it16:47
@iglesiasglisitsyn: I am clueless about it. Any idea how to gather more useful info?16:49
@iglesiasgI valgrind the example here and there was no invalid read/write16:49
lisitsyniglesiasg: valgrind would be more useful probably16:50
@iglesiasglisitsyn: yeah, let's valgrind the example in his machine too16:51
lisitsynwill get back quite soon!16:52
@iglesiasgall right16:52
-!- s3b4 [59e3a2e9@gateway/web/freenode/ip.89.227.162.233] has quit [Quit: Page closed]17:06
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]17:08
@iglesiasgsee you later guys17:10
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]17:10
-!- lambday [67157c36@gateway/web/freenode/ip.103.21.124.54] has quit [Ping timeout: 250 seconds]17:12
-!- lambday [67157c36@gateway/web/freenode/ip.103.21.124.54] has joined #shogun17:34
lambdayHeikoS: sonney2k lisitsyn : why its necessary for all template classes to pass for all ptypes in class_list.cpp?17:35
lambdayfor many classes, say LinearOperator, we won't be needing bool, char, etc... but still we gotta instantiate them :(17:36
lambdayresults in huge compile time :(17:37
lambdayis it feasible that we check what explicit instantiations are declared in corresponding h/cpp.. and check only for those?17:38
-!- lambday [67157c36@gateway/web/freenode/ip.103.21.124.54] has quit [Ping timeout: 250 seconds]17:56
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun18:09
@HeikoSlisitsyn: around?18:10
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun18:14
shogun-notifier-shogun: Fernando Iglesias :develop * 13c90ba / src/ (7 files): https://github.com/shogun-toolbox/shogun/commit/13c90baca9fabe01a5ae009cca67c8cf1517f3e818:14
shogun-notifier-shogun: Modifications so LMNN can be used from interfaces.18:14
shogun-notifier-shogun: Fernando Iglesias :develop * 5e950b1 / src/shogun/distance/CustomMahalanobisDistance.cpp: https://github.com/shogun-toolbox/shogun/commit/5e950b17455c55b15e8682982c692b1bea399a4718:14
shogun-notifier-shogun: Use static_cast instead of dynamic_cast to save runtime in custom Mahalanobis distance18:14
shogun-notifier-shogun: Fernando Iglesias :develop * 04e7a81 / src/ (8 files): https://github.com/shogun-toolbox/shogun/commit/04e7a816b03c1252fa54ce6988b8a84f64c43c7b18:14
shogun-notifier-shogun: Merge pull request #1289 from iglesias/feature/lmnn18:14
shogun-notifier-shogun:18:14
shogun-notifier-shogun: Feature/lmnn18:14
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has joined #shogun18:19
-!- mode/#shogun [+o iglesiasg] by ChanServ18:19
shogun-buildbotbuild #747 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/747  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>18:21
@iglesiasghi guys18:21
shogun-buildbotbuild #1060 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1060  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>18:25
@iglesiasgHeikoS, do you know something about this error? http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/747/steps/compile/logs/stdio18:26
@iglesiasgit seems to be in the logdet framework18:26
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has joined #shogun18:27
@iglesiasganother user got it before too in ubuntu 12.0418:27
lambdayHeikoS: hi18:27
@HeikoSiglesiasg: yeah thats lambday s latest patch18:27
@iglesiasglambday, or maybe you know about it even better :)18:27
@HeikoSlambday:  could you re-include the templates again?18:27
lambdaywhich templates?18:27
@HeikoSlambday: also the unit test fails for the faster runtime thing sometimes18:27
@HeikoSlambday: the ones you removed in the last patch18:27
@HeikoSadd them again for now18:27
@HeikoSwe first need a machanism to avoid testing them18:28
lambdayokay but18:28
lambdayalright18:28
lambdayshit I wrote many more things but I didn't realize I got dc18:28
@HeikoSlambday: I know compile time, but that can be fixed later18:28
lambdayanyway here they are18:28
@HeikoScurrent it breaks the tests18:28
lambdaysecondly, say a child class does not define a parent abstract method and stays abstract, but class_list can't find that :( we gotta keep writing that pure virtual signature in each one of them.. we can create a separate list of abstract classes and attach which method(s) makes it abstract, and check if a classs' [21:25] <lambday> parent belong to that list and the current .h doesn't define that, we add it to it [21:25] <lambday> iter18:28
lambdayHeikoS: alright18:28
lambdaybut why does it break!18:29
lambday:-/18:29
shogun-buildbotbuild #1264 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1264  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>18:29
@HeikoSlambday: if you want something to be virtual and also be recognised as virtual, you have to keep the =0 signature18:29
lambdayyes I did that18:29
@HeikoSlambday: just add them once again, we can figure that later ;)18:29
@HeikoSlambday:  http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/747/steps/compile/logs/stdio18:29
lambdayHeikoS: alright18:29
lambdayHeikoS: its not because of that18:31
lambdayIts because I split the template thing into two lines :(18:31
lambdayI think18:31
@HeikoSlambday: ah18:32
@HeikoSok then18:32
@HeikoSwhat cool18:32
@HeikoSjust make sure the build works again :)18:32
lambdayHeikoS: sending the patch18:32
@HeikoSlambday: thanks a lot!18:32
lambdayHeikoS: alright18:32
lambdayHeikoS: no problem :)18:32
shogun-buildbotbuild #1527 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1527  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>18:33
-!- FSCV [~FSCV@187.210.54.166] has quit [Quit: Leaving]18:37
lambdayHeikoS: please have a look18:38
@HeikoSlambday:  no PR here yet18:39
lambdayumm18:39
lambdayHeikoS: https://github.com/shogun-toolbox/shogun/pull/129118:40
@HeikoSlambday:  the unit test also sometimes fails18:40
@HeikoScould you run the thing locally a large number of times and then set the error to the maximum error obtained?18:40
shogun-notifier-shogun: lambday :develop * a726e43 / src/shogun/mathematics/logdet/ (2 files): https://github.com/shogun-toolbox/shogun/commit/a726e43a4e8e95622e877374757b9ef0c5d4a7e218:40
shogun-notifier-shogun: fixed multi line template declaration error18:40
shogun-notifier-shogun: Heiko Strathmann :develop * 85d86a4 / src/shogun/mathematics/logdet/ (2 files): https://github.com/shogun-toolbox/shogun/commit/85d86a4c0caa6afd6619ab2cd5e9c5e6a6032fa918:40
shogun-notifier-shogun: Merge pull request #1291 from lambday/feature/log_determinant18:40
shogun-notifier-shogun:18:40
shogun-notifier-shogun: fixed multi line template declaration error18:40
lambdayHeikoS: yes.. I'll fix a lower accuracy.. running 10k times as you said18:41
@HeikoSlambday: but not in the test though :)18:41
lambdayHeikoS: lol yes :D18:41
@HeikoSso set the test to n iterations18:41
@HeikoSthen you locally run 100000*n and choose the largest error you got18:41
@HeikoS(+ some small offset)18:41
lambdayHeikoS: alright18:41
lambdayHeikoS: shall I remove SNP and WeightedDegree from the clone test too? they failed in the clone test18:43
@HeikoSlambday: please rather try to fix them :)18:43
@HeikoSlambday: or let me do it18:44
@HeikoSlambday: but this blacklist is for other things (framework issues)18:44
@HeikoSlambday: these classes can currently not be tested automatically18:44
lambdayHeikoS: oh alright18:44
@HeikoSdue to issues in the way we treat generics18:44
lambdaygot it18:44
-!- travis-ci [~travis-ci@ec2-23-20-210-220.compute-1.amazonaws.com] has joined #shogun18:46
travis-ci[travis-ci] it's Fernando Iglesias'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/943961018:46
-!- travis-ci [~travis-ci@ec2-23-20-210-220.compute-1.amazonaws.com] has left #shogun []18:46
lambdayHeikoS: I removed those and added a comment18:47
gsomix_good evening18:55
-!- gsomix_ is now known as gsomix18:55
lambdayHeikoS: SNP clone gives segfaults18:55
naywhayareanybody here going to the GSoC mentor summit in October?18:56
@HeikoSnaywhayare: hi yes18:56
@HeikoSlambday: let me check18:56
naywhayareHeikoS: I will be there also, perhaps we will run into each other :)18:57
@HeikoSnaywhayare: cool! yeah lets meet then18:57
@HeikoSnaywhayare: which organisation were you again?18:57
naywhayareHeikoS: mlpack18:57
@HeikoSah yes18:57
naywhayare:)18:57
@HeikoSthe speed kmeans paper :)18:58
naywhayareI ran into a guy who works on Shark at ICML, I think he said he would be at the summit too18:58
naywhayareI completely forgot his name but I remember what he looks like...18:58
@HeikoSI am really looking forward to meet everyone18:58
naywhayare:)18:59
naywhayareI don't think any scikit-learn guys will be there; scikit-learn wasn't in GSoC this year -- I'm not sure why19:00
lisitsynI am curious if there is a chance to get there too19:00
lisitsynnaywhayare: they are but under PSF I think19:00
@HeikoSnaywhayare: dont know, I think one of our student also applied to them ...19:00
@HeikoSlisitsyn: would be cool if you joined us19:01
@HeikoSlisitsyn: but this time you will have to talk more :D19:01
lisitsynHeikoS: oh cooome ooon19:01
@HeikoSnaywhayare: we had our workshop recently, where everyone met, this was really nice19:01
naywhayareah yeah, there scikit is under PSF19:01
lisitsynI have been hearing that basically every day since our workshop19:01
@HeikoSlisitsyn: haha19:02
@HeikoSyou deserve it ;)19:02
lisitsynoh well then I would not like to join that meeting19:02
@HeikoSlisitsyn:  ah come on, I am just joking19:02
@HeikoSof course you should join us19:02
shogun-buildbotbuild #1529 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1529  blamelist: lambday <heavensdevil6909@gmail.com>19:02
@HeikoSwill be nice19:02
@HeikoSI am really afraid of my computer melting with all these shogun compiles19:03
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun19:03
@HeikoSvotjakovr: hi!19:03
@HeikoSlambday: what is the exact name of the class that fails?19:04
lambdaySNPStringKernel19:04
lambdayand another one19:04
@HeikoSthx19:04
votjakovrHeikoS: hi! i've finished posterior approximation stuff and tested with probit likelihood19:05
@HeikoSlambday: yep there is uninit memory19:05
@HeikoSvotjakovr: nice!19:05
@HeikoSvotjakovr: could you do logit likelihood next for it?19:05
lambdayWeightedDegreePositionStringKernel19:05
shogun-buildbotbuild #1383 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/1383  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>19:07
@HeikoSvotjakovr: send the pr! :)19:12
votjakovrHeikoS: i've just sent19:14
lambdayHeikoS: sending the patch with the changes for clone-test blacklist.. will add the log-det reduced accuracy after testing19:14
gsomixit seems "thunderstorm" is the most frequent news from my village19:19
gsomixso, thundrestorm!19:19
shogun-buildbotbuild #1530 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1530  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>19:19
@HeikoSlambday: thanks! I just fixed the memory errors19:21
@HeikoSlambday:  let me know if you find more19:21
lambdayHeikoS: okay :) currently checking the log-det thing19:21
shogun-notifier-shogun: Heiko Strathmann :develop * 11b9afe / src/shogun/kernel/string/ (2 files): https://github.com/shogun-toolbox/shogun/commit/11b9afed867ed10aaf8f47c54d7a1782e90b224819:22
shogun-notifier-shogun: fixed two uninitialised memory bugs19:22
shogun-notifier-shogun: Heiko Strathmann :develop * 97cd7af / src/shogun/kernel/string/ (2 files): https://github.com/shogun-toolbox/shogun/commit/97cd7af19655332aa91db5a7ea2dc08185478b0519:22
shogun-notifier-shogun: Merge pull request #1295 from karlnapf/develop19:22
shogun-notifier-shogun:19:22
shogun-notifier-shogun: fixed two uninitialised memory bugs19:22
@HeikoSvotjakovr: I put some comments in the PR19:26
@HeikoSvotjakovr: I will merge anyway, but could you change them before doing other things?19:27
@HeikoSthanks :)19:27
@HeikoSvotjakovr: oh you already answered, sorry didnt yee19:27
shogun-notifier-shogun: lambday :develop * af2cab3 / tests/unit/base/clone_unittest.cc.py: https://github.com/shogun-toolbox/shogun/commit/af2cab3d8212519dc1dcb0e5b1dc30bef51d6ac219:28
shogun-notifier-shogun: removed buggy classes from clone-test19:28
shogun-notifier-shogun: Heiko Strathmann :develop * 6cfc86f / tests/unit/base/clone_unittest.cc.py: https://github.com/shogun-toolbox/shogun/commit/6cfc86fd0b72082f7aa1177c8fb607e11495a41f19:28
shogun-notifier-shogun: Merge pull request #1294 from lambday/feature/log_determinant19:28
shogun-notifier-shogun:19:28
shogun-notifier-shogun: removed buggy classes from clone-test19:28
shogun-buildbotbuild #1061 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1061  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>19:32
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 264 seconds]19:32
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has joined #shogun19:33
-!- mode/#shogun [+o iglesiasg] by ChanServ19:33
shogun-buildbotbuild #748 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/748  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>19:33
@HeikoSvotjakovr: do you know btw why this is:19:34
@HeikoSLaplacianInferenceMethod.cpp19:34
@HeikoSif (eigen_W.minCoeff() < 0)19:34
@HeikoSline 38919:34
@HeikoSthis block where the matrix is inverted directly19:34
@HeikoSwhy is that done19:35
@HeikoSeigen_L then stores an inverted matrix rather than a cholesky19:35
shogun-buildbotbuild #749 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/749  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, lambday <heavensdevil6909@gmail.com>19:37
votjakovrHeikoS: hmm, it was there before...19:38
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 257 seconds]19:38
@HeikoSvotjakovr: yeah I know, just wondering for what that is19:38
@HeikoSvotjakovr: thanks for the patch btw19:39
@HeikoSvotjakovr: this might be used in our current NIPS paper here :)19:39
votjakovrHeikoS: btw gpml uses same trick19:39
@HeikoSvotjakovr: for the inversion?19:39
@HeikoSor for the thing I commented?19:40
votjakovrHeikoS: yep, there is switch between Cholesky and LU decomposition mode in GPML19:40
@HeikoSvotjakovr: I see, ok then leave it, maybe comment it and say that its done as in GPML and maybe even copy the comment the guys made in there19:41
@HeikoSvotjakovr: how about the logit classifier? you said its done?19:41
shogun-buildbotbuild #1062 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1062  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, lambday <heavensdevil6909@gmail.com>19:41
shogun-buildbotbuild #1265 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1265  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>19:41
-!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.]19:42
shogun-buildbotbuild #1531 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1531  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>19:42
votjakovrHeikoS: yep, but i need to debug a little bit and i'll send a PR with numerical integration and logit tomorrow19:43
@HeikoSvotjakovr: ok cool!19:43
@HeikoSvotjakovr: btw did you already have a look at the ipython notebooks?19:43
@HeikoSlots of possibilities there for GP stuff19:43
@HeikoSlots of potential19:43
@HeikoSscikit learn GPs are not as cool as ours, so if we manage to document things well, they will probably be used by many people19:44
@HeikoSvotjakovr: in fact, a collegue here just started using the regression framework and some things you did19:44
votjakovrHeikoS: cool :) i'll have a look at it19:46
votjakovrHeikoS: about approximation stuff, sorry, i was in hurry a little bit... I'll do like you suggested19:48
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has quit [Quit: Leaving.]19:48
@HeikoSvotjakovr: dont worry, I think it is a bit faster when you dont do the identity stuff19:48
@HeikoSvotjakovr: I have another suggestion btw:19:49
@HeikoSvotjakovr: so currently,  the likelihood evaluation gets a vectors of latent functions19:49
@HeikoSvotjakovr: but sometimes, one wants to evaluate the likelihood for some fixed y for many many fs19:50
@HeikoSso it would be cool to have a inline C method for that19:50
@HeikoSso that one doesnt need to call it multiple times19:50
@HeikoSsee what I mean?19:50
shogun-buildbotbuild #1266 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1266  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, lambday <heavensdevil6909@gmail.com>19:52
votjakovrHeikoS: yep :)19:52
@HeikoSvotjakovr: would you add that (returns a matrix rather than a vector19:53
@HeikoS(and unit test that results are the same)19:53
@HeikoSvotjakovr: I think it only needs to be done in base likelihood class19:53
votjakovrHeikoS: but currently we can't call method for evaluating the likelihood from there, since it is pure virtual19:56
@HeikoSah damn....19:56
@HeikoSmmh19:56
@HeikoSany idea how to solve this? would be annoying to re-write it everytime19:56
shogun-buildbotbuild #1384 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/1384  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, lambday <heavensdevil6909@gmail.com>19:56
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has joined #shogun19:57
-!- mode/#shogun [+o iglesiasg] by ChanServ19:57
votjakovrHeikoS: hmm, make it not pure virtual :)20:03
@HeikoSvotjakovr: sounds good, go for it, maybe just add SG_NOTIMPLEMENTED then20:12
votjakovrHeikoS: ok20:14
@iglesiasghi pickle2720:19
@iglesiasgpickle27, I am taking a look at the PR20:19
pickle27okay!20:19
@iglesiasgpickle27, is there already some unit test for Jade?20:20
pickle27yeah there is one already20:21
@iglesiasgpickle27, cool20:22
pickle27iglesiasg: the unit test was being wierd before but I think it should be good now, otherwise I'll take a look into it20:23
pickle27but then again there were bugs in Jade before so20:23
@iglesiasgI see20:23
pickle27iglesiasg: is Travis up?20:24
pickle27we can wait and see what it says20:24
@iglesiasgpickle27, It didn't start in your pull request for some reason indeed20:25
votjakovrpickle27: hi! I've run valgrind on the unit tests and it reported me memory leaks in Jade_unittest.cc and SOBI_unittest.cc, please check it.20:28
pickle27hmmm20:28
pickle27okay20:28
pickle27will do20:28
pickle27can you pastebin the output20:28
pickle27votjakovr: ^20:28
votjakovrpickle27: ok, please wait a minute20:29
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 264 seconds]20:36
pickle27iglesiasg: how do I print a matrix with SGDebug?20:36
votjakovrpickle27: http://pastebin.com/Nu1EtfUA and http://pastebin.com/USsdpLMm20:42
pickle27votjakovr: thanks20:42
@iglesiasgpickle27, sorry I missed your question20:44
@iglesiasgpickle27, did you write it before I commented about that in github or?20:44
pickle27before, so its all good20:44
pickle27Those were the debug prints I was using to get it working and I just thought why delete these?20:45
pickle27so I just ifdef'd them out20:45
pickle27but they aren't key so if you're okay with them as is I say we leave them otherwise I'll just take them out20:45
@iglesiasgpickle27, as you prefer20:47
pickle27iglesiasg: okay, have you finished going through? I'll push those changes20:47
@iglesiasgpickle27, I guess between DEBUG_JADE is better than nothing20:47
@iglesiasgpickle27, it might help people that wants to understand the code20:48
@iglesiasgpickle27, sure, go ahead20:48
pickle27iglesiasg: done20:49
@iglesiasgpickle27, no idea what's up with travis20:51
@iglesiasgpickle27, we will have to merge without it20:51
@iglesiasgpickle27, did you run unit tests? for what interfaces are you compiling, libshogun and python_modular maybe?20:51
pickle27yeah those 2, I haven't ran the unit tests but I can20:55
pickle27iglesiasg: unit tests in general aren't running on my machine right now20:59
pickle27I think I need to make clean them20:59
@iglesiasgpickle27, aren't running as they fail?21:01
@iglesiasgpickle27, just try the ones that are related to your patch if so21:01
@iglesiasgsomething like make to build all of them21:01
@iglesiasgwhen they start running you ctrl+c it21:01
@iglesiasgand then21:01
@iglesiasg./shogun-unit-test --filter_test=Jade*21:02
@iglesiasgI getting the name of the flag wrong probably, check it in the help of the executable if so21:02
-!- hushell [~hushell@8-92.ptpg.oregonstate.edu] has joined #shogun21:03
@iglesiasghi hushell21:03
hushellhi iglesiasg21:03
pickle27kk21:03
@iglesiasghushell, I am merging the PR. Let's see if everything is fine, otherwise there is no problem, we modify the test later :)21:03
@iglesiasghushell, for some reason travis is not firing in the pull requests now21:03
hushelliglesiasg: I see, I am waiting for travis compiling the PR21:04
@iglesiasghushell, but it didn't even start, right?21:04
@iglesiasghushell, I don't see anything in here21:04
hushellyes, I committed 40 mins ago21:05
shogun-notifier-shogun: hushell :develop * 41c0b5f / / (12 files): https://github.com/shogun-toolbox/shogun/commit/41c0b5f3aa1a8a1a8ebdaa29733cbc2eb540fdbc21:05
shogun-notifier-shogun: Factor Graph21:06
shogun-notifier-shogun: Fernando Iglesias :develop * 532ab10 / / (12 files): https://github.com/shogun-toolbox/shogun/commit/532ab10a2fd4a2c04b9402123264400ab917021521:06
shogun-notifier-shogun: Merge pull request #1224 from hushell/develop21:06
shogun-notifier-shogun:21:06
shogun-notifier-shogun: Factor Graph21:06
hushelliglesiasg: you just merged it?21:06
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving]21:08
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun21:08
hushella short question, for a vector I want to return in the argument, is it better to use SGVector<>& ? I am not sure the case pointer is inside a class21:08
@iglesiasghushell, yep21:09
@iglesiasghushell, use SGVector better please21:09
hushelliglesiasg: but sometimes it doesn't work21:09
@iglesiasghushell, an SGVector is basically an int32_t for the length and a T* for the data21:09
pickle27iglesiasg: what was the filter line again? I ctrl L irc and the logs are behind lol21:09
@iglesiasgpickle27, can you do ./shogun-unit-test -h?21:10
@iglesiasgpickle27, it is one of the first options, --filter sth21:10
pickle27k21:10
@iglesiasgpickle27, I never compiled test in the computer I am now so I cannot check it, sorry21:11
pickle27iglesiasg: np21:11
@iglesiasghushell, how that it doesn't work?21:11
@iglesiasgpickle27, did you find out?21:11
pickle27yeah I see the option21:12
hushelliglesiasg: e.g. CFactor::compute_gradients() I have to use SGVector<float64_t>& parameter_gradient21:12
pickle27just trying to get it to work now21:12
@iglesiasgpickle27, what's the name of your test?21:12
@iglesiasghushell, the problem is that SGVector<float64_t>& won't work in target interfaces, just C++21:12
hushelliglesiasg: otherwise nothing returned, this probably this parameter_gradient is passed to anther function inside21:12
pickle27iglesiasg: the file name?21:13
@iglesiasgpickle27, yeah that's enough21:13
hushellstructure/Factor.cpp21:13
pickle27Jade_unittest.cc21:13
@iglesiasgpickle27, then something like21:13
@iglesiasgpickle27, ./shogun-unit-test --filter=Jade*21:13
@iglesiasgshould make it21:13
pickle27yeah I just got it21:14
@iglesiasgwhere --filter is the option you have looked up :)21:14
pickle27iglesiasg: it ran21:14
pickle27wait21:14
pickle27it ran but filter didn't work yet21:14
@iglesiasg?21:14
hushell:D I got confused talking together with pickle27 at the same time21:14
@iglesiasghushell, hehe yeah sorry :S21:15
hushelliglesiasg: not a problem :D21:15
pickle27iglesiasg: k unit test needs some work21:16
shogun-buildbotbuild #1063 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1063  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>21:16
hushelliglesiasg: so you mean a function with argument SGVector& will not working for other languages21:16
@iglesiasghushell, exactly21:17
@iglesiasghushell, I came to know this the hard way a few months ago21:17
hushelliglesiasg: next PR will be smaller :)21:17
shogun-buildbotbuild #750 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/750  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>21:18
votjakovrpickle27: i ran Jade unit-test with valgrind --leak-check=full ./shogun-unit-test --gtest_filter=CJade*. It might be useful21:18
@iglesiasghushell, :)21:18
hushelliglesiasg: okay then I need to figure out how to deal with the return arugment problem, not using &21:18
pickle27votjakovr: thanks, I looked at your valgrind logs and it didn't help much21:19
@iglesiasghushell, It should be totally fine to return SGVector21:19
@iglesiasgman I love cyg1 buildbot saying "Error: no C++ compiler detected - cannot do anything"21:19
gsomixiglesiasg, "won't work in target interfaces, just C++" << strange21:23
@iglesiasggsomix, really?21:24
@iglesiasggsomix, I had some problems already with methods accepting things with &21:24
@iglesiasggsomix, they didn't work in python_modular21:24
gsomixswig just should transform reference to pointer in generated code21:24
gsomixmaybe we just need another typemap for this case21:25
gsomixI'll check.21:25
@iglesiasggsomix, yes, a typemap can fix it maybe. At the moment it didn't work though21:25
pickle27iglesiasg: I fixed the unit test and pushed21:26
shogun-buildbotbuild #1267 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1267  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>21:26
@iglesiasgpickle27, all good now then?21:27
pickle27iglesiasg: I think so21:27
shogun-buildbotbuild #1534 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1534  blamelist: hushell <hushell@hushell-U510.(none)>21:27
pickle27there is the possible leak the votjakovr found but21:27
@iglesiasggsomix, I remember I changed some SGMatrix const & to SGMatrix const, because it didn't work with the reference21:28
pickle27it all runs and passes the test21:28
@HeikoSvotjakovr: there are some issues with swig :(21:28
@HeikoSvotjakovr: I think we will have to move all GP stuff to one directory21:28
@iglesiasggsomix, I am not sure though if it is just with SGVector/SGMatrix or also with other types like int and so21:28
votjakovrHeikoS: yep, i know21:28
@HeikoSvotjakovr: ah well, annoying, since it was just all moved21:29
gsomixiglesiasg, but I have one question. what's going with reference counting on while passing SGVector<>&?21:29
@HeikoSvotjakovr: I suggest move everything to machine/gp?21:29
gsomixiglesiasg, ok.21:29
@HeikoSvotjakovr: could you do that?21:29
@iglesiasggsomix, I never thought of that actually21:29
@iglesiasgpickle27, you didn't see that leak in your valgrind?21:30
@HeikoSvotjakovr: and then add all classes in interfaces/modular/Machine.i and Machine_includes.i and remove them in the other files?21:30
pickle27iglesiasg: I do but there isn't really any other output from valgrind so Im confused21:30
pickle27Im swithcing something with my SGNDArray21:30
pickle27I think that might be it21:31
@HeikoSvotjakovr: I think this should solve all problems21:32
@HeikoSvotjakovr: or should I quickly do it?21:32
votjakovrHeikoS: hmm... why don't move to gp/?21:32
@HeikoSvotjakovr: since its machine based and we need things in swig from machine (I think at least)21:33
votjakovrHeikoS: ok21:34
votjakovrHeikoS: btw i've updated my PR21:34
lambdayHeikoS: regarding log-det unit-test, with 500 Gaussian samples after running 100,000 times, max difference I get is 1.9944846621, shall I just put it 2.0 ?21:35
@iglesiasgall right, I am off now. See you later if so21:36
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]21:36
@sonney2kHeikoS, lambday any clues why this fails http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/750/steps/compile/logs/stdio ?21:37
lambdaysonney2k: checking21:37
@sonney2klambday, thx21:38
lambdaysonney2k: I just fixed it and sent a patch.. merged already21:39
lambdaysonney2k: still it gives this error?21:39
@HeikoSlambday:  yep21:40
@HeikoSsonney2k: hey do you have a suggestion how to do GPs with swig?21:41
@HeikoSlike in which folder to put stuff21:41
@HeikoScurrently it is located in many folders some regression some classification etc21:41
pickle27votjakovr: I found the leak but I don't understand it21:41
@HeikoSsonney2k: but we seem to get issues with swig21:41
pickle27its the call to delete my SGNDArray21:41
@HeikoSsonney2k: and there are also these things in machine.i21:41
@HeikoSand GPs are inheriting from machine21:41
@HeikoSsonney2k: so where to put them, does it matter, doesnt it matter? etc21:42
@sonney2kman you guys have been chatty!21:45
@sonney2kI can hardly read all the logs PR comments21:45
-!- travis-ci [~travis-ci@ec2-75-101-178-60.compute-1.amazonaws.com] has joined #shogun21:46
travis-ci[travis-ci] it's Heiko Strathmann'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/944194621:46
-!- travis-ci [~travis-ci@ec2-75-101-178-60.compute-1.amazonaws.com] has left #shogun []21:46
@sonney2kHeikoS, yeah makes sense to put the regression part in regression and classification in classification21:47
@HeikoSvotjakovr: ok so then lets leave it as it is and figure out the swig issues :)21:47
@HeikoSvotjakovr: so ill try to do that21:48
gsomixsonney2k, so, LineReader class + Parsers, right?21:48
gsomixsonney2k, good evening21:48
votjakovrHeikoS: Ok :)21:48
votjakovrHeikoS: btw, please, check my PR, i've updated it21:49
shogun-notifier-shogun: Roman Votyakov :develop * a3e4f75 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/a3e4f75c554321adb6bd9593e5e09067223acd9d21:49
shogun-notifier-shogun: add posterior approximation mean and covariance for Laplacian inference method21:49
shogun-notifier-shogun: Heiko Strathmann :develop * 1099384 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/10993840452cc46f1ebbf51291a4af2febd598f121:49
shogun-notifier-shogun: Merge pull request #1293 from votjakovr/feature/gp_refactoring21:49
shogun-notifier-shogun:21:49
shogun-notifier-shogun: add posterior approximation mean and covariance for Laplacian inference method21:49
@HeikoS:)21:49
votjakovr:)21:49
@sonney2kHeikoS, what swig issues?21:49
@HeikoSsonney2k: classes not found etc21:49
@sonney2kgsomix, yes Parser would be your old Reader class21:49
@HeikoSits a bit painfull since its spread in machine/regression/classifier21:50
@sonney2kHeikoS, w/ swig the order of includes matters21:50
gsomixsonney2k, ok, that's right. :)21:50
gsomixsonney2k, preparing PR, need test some stuff.21:50
@sonney2kHeikoS, you need to include classes that are dependencies first!21:50
@sonney2klambday, what is this template<class T, class ST=T> ?22:00
@sonney2kin shogun/mathematics/logdet/IterativeShiftedLinearFamilySolver.h22:00
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has quit [Ping timeout: 250 seconds]22:03
shogun-buildbotbuild #1536 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1536  blamelist: Roman Votyakov <votjakovr@gmail.com>22:07
shogun-buildbotbuild #1385 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/1385  blamelist: Fernando Iglesias <fernando.iglesiasg@gmail.com>22:07
-!- travis-ci [~travis-ci@ec2-75-101-178-60.compute-1.amazonaws.com] has joined #shogun22:09
travis-ci[travis-ci] it's Heiko Strathmann'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/944209522:09
-!- travis-ci [~travis-ci@ec2-75-101-178-60.compute-1.amazonaws.com] has left #shogun []22:09
-!- zxtx [~zv@rrcs-76-79-81-162.west.biz.rr.com] has joined #shogun22:22
shogun-buildbotbuild #1537 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1537  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>22:24
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has joined #shogun22:34
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has left #shogun ["Fallen asleep!"]23:07
shogun-notifier-shogun: Soeren Sonnenburg :develop * a059a1f / src/shogun/mathematics/logdet/ (2 files): https://github.com/shogun-toolbox/shogun/commit/a059a1ffb9e4052bdd52c01b84ed301a33d44a7623:18
shogun-notifier-shogun: drop default type to hopefully fix compile error23:18
shogun-notifier-shogun: van51 :develop * 63ae07c / / (3 files): https://github.com/shogun-toolbox/shogun/commit/63ae07cbe4b0ffb05de7d029e969d29e1e21cdf823:20
shogun-notifier-shogun: Quadratic support in HashedDenseFeatures23:20
shogun-notifier-shogun: Soeren Sonnenburg :develop * f0af8bf / / (3 files): https://github.com/shogun-toolbox/shogun/commit/f0af8bf45c00b983abb88ecb153d1afef81cb09923:20
shogun-notifier-shogun: Merge pull request #1292 from van51/feature/quadratic23:20
shogun-notifier-shogun:23:20
shogun-notifier-shogun: Quadratic support in HashedDenseFeatures23:20
van51ah cool :D23:21
van51sonney2k: do you have in mind another collection I could test the hasheddocfeatures on?23:22
@sonney2kvan51, collection?23:22
@sonney2kvan51, dataset you mean?23:22
van51sonney2k: yes23:23
van51sonney2k: document collection I meant to say23:23
@sonney2kvan51, I still would want you to do the language detection thing23:23
@sonney2kvan51, but for this you need to do some wikipeidia harvesting23:23
van51sonney2k: that would be to get documents from wikipedia right?23:23
@sonney2kvan51, yes23:23
van51sonney2k: I can get the documents from another language, right?23:24
@sonney2kso for each supported language download say 10000 text documents23:24
@sonney2krandomly23:24
van51programming language*23:24
@sonney2kvan51, sure some wget / python whatever23:24
van51sonney2k: ok then :)23:25
@sonney2kI would prepare the final data similar to webspam then label with 0...<n_langs>23:25
@sonney2kand 0 terminated documents23:25
@sonney2kvan51, btw did you do finish the evaluations for webspam?23:26
van51sonney2k: should I strip the html completely?23:26
@sonney2kvan51, yeah23:26
@sonney2kjust plain text23:26
@sonney2kvan51, I guess n-grams will work best again23:26
van51sonney2k: yea I finished it23:26
van51there was a link in the email23:26
van51let me fetch it again23:26
@sonney2kvan51, svmocas with eps 1e-2?23:26
@sonney2kvan51, let me check23:26
@sonney2kvan51, please also run it with liblinear eps 1e-223:27
van51sonney2k: here : https://dl.dropboxusercontent.com/u/23851310/webspam_results.pdf23:27
@sonney2kvan51, found it reading...23:27
van51sonney2k: ok I'll leave it running tonight23:27
@sonney2ksonney2k, liblinear might be faster23:30
@sonney2kvan51, T^23:30
shogun-buildbotbuild #751 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/751  blamelist: van51 <vangelis_51@hotmail.com>23:30
@sonney2kvan51, interesting that C=1 gave even better results23:31
@sonney2kvan51, I can only guess that the train/test data split that you used gives different results auROC wise... 99.9 vs 99.6 seems quite a bit though23:32
@sonney2khope it is not train error :)23:32
van51sonney2k: nope it's test error23:33
van51sonney2k: do you want to chck the example?23:33
van51sonney2k: I'm worried I may have missed something23:33
@sonney2kvan51, I read the example code already23:33
@sonney2kdidn't see a mistake23:33
van51well the language detection will also be a nice indicator23:34
@sonney2kvan51, but it is worrying that for 5000 examples with hashing you are as good as the spectrum kernel on 100k examples23:34
shogun-buildbotbuild #752 of rpm1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/rpm1%20-%20libshogun/builds/752  blamelist: Soeren Sonnenburg <sonne@debian.org>23:34
@sonney2kvan51, true23:35
shogun-buildbotbuild #1064 of cyg1 - libshogun is complete: Failure [failed compile]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1064  blamelist: van51 <vangelis_51@hotmail.com>23:36
shogun-buildbotbuild #1065 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1065  blamelist: Soeren Sonnenburg <sonne@debian.org>23:37
shogun-buildbotbuild #1538 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/1538  blamelist: Soeren Sonnenburg <sonne@debian.org>23:37
shogun-buildbotbuild #1268 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1268  blamelist: van51 <vangelis_51@hotmail.com>23:39
shogun-notifier-shogun: Soeren Sonnenburg :develop * 73cb5a2 / src/shogun/machine/gp/InferenceMethod.h: https://github.com/shogun-toolbox/shogun/commit/73cb5a238f58e10afa8fa5694af530103c04dea023:39
shogun-notifier-shogun: fix doxygen doc23:39
shogun-notifier-shogun: Soeren Sonnenburg :develop * 07e8f7c / src/shogun/mathematics/Math.h: https://github.com/shogun-toolbox/shogun/commit/07e8f7ce467a61b565b3f6694fb976465cb2704723:39
shogun-notifier-shogun: drop std cmath compile workaround23:39
shogun-notifier-shogun: Soeren Sonnenburg :develop * ad88090 / src/configure: https://github.com/shogun-toolbox/shogun/commit/ad8809008204fb70cd82a6ed137b30b771edd09c23:43
shogun-notifier-shogun: drop C warning option -Werror-implicit-function-declaration from compile flags23:43
gsomixsonney2k, hey, can you help me?23:45
@sonney2kgsomix, wasup?23:48
gsomixsonney2k, I have strange errors in unit-tests. and just curious, is it only in my repo? http://pastebin.com/u2DyVu3523:49
shogun-buildbotbuild #1269 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1269  blamelist: Soeren Sonnenburg <sonne@debian.org>23:49
-!- travis-ci [~travis-ci@ec2-75-101-178-60.compute-1.amazonaws.com] has joined #shogun23:51
travis-ci[travis-ci] it's Fernando Iglesias'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/944518323:51
-!- travis-ci [~travis-ci@ec2-75-101-178-60.compute-1.amazonaws.com] has left #shogun []23:51
shogun-buildbotbuild #1386 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/1386  blamelist: Soeren Sonnenburg <sonne@debian.org>, van51 <vangelis_51@hotmail.com>23:51
gsomixtrying to reconfigure/recompile23:51
-!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has joined #shogun23:58
-!- mode/#shogun [+o iglesiasg] by ChanServ23:58
@sonney2kgsomix, hmmhh then just run your test with the filter thing23:58
@iglesiasghushell, it seems the evaluate_energy_param_data test is still failing on travis23:59
--- Log closed Thu Jul 25 00:00:46 2013