Open in new window / Try shogun cloud
--- Log opened Thu Jun 22 00:00:23 2017
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/381200:30
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3812 merged by vigsterkr00:30
@sukey[https://github.com/shogun-toolbox/shogun] vigsterkr pushed 2 commits:00:30
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/41f2ca4995d517c0729363e0f854bddb03af997c00:30
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/2def19500536f1408914d7cb5a2247d54720415700:30
-!- sukey [~nodebot@ks312251.kimsufi.com] has quit [Remote host closed the connection]00:46
-!- sukey [~nodebot@ks312251.kimsufi.com] has joined #shogun00:47
-!- mode/#shogun [+o sukey] by ChanServ00:47
-!- shogun-buildbot [~shogun-bu@7nn.de] has quit [Quit: buildmaster reconfigured: bot disconnecting]01:51
-!- shogun-buildbot [~shogun-bu@7nn.de] has joined #shogun01:51
shogun-buildbotbuild #70 of nightly jessie deb is complete: Failure [failed dput]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly%20jessie%20deb/builds/70  blamelist: Giovanni De Toni <giovanni.det@gmail.com>, Viktor Gal <vigsterkr@gmail.com>, MikeLing <sabergeass@gmail.com>, Viktor Gal <viktor.gal@maeth.com>03:01
-!- mikeling [uid89706@gateway/web/irccloud.com/x-oewvncwnqrkuhecv] has joined #shogun03:17
@wikingshogun-buildbot: force build --branch=develop 'trusty - libshogun - viennacl'03:32
shogun-buildbotbuild #286 forced03:32
shogun-buildbotI'll give a shout when the build finishes03:32
shogun-buildbotbuild #286 of trusty - libshogun - viennacl is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/trusty%20-%20libshogun%20-%20viennacl/builds/28603:35
@sukey[https://github.com/shogun-toolbox/shogun] vigsterkr pushed 2 commits:04:11
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/6e0fe4ca2f929806fbbeb54b14b00c7bf351612504:11
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/dc5c4f2af33e7d5f9dd7ed413dc4e05af26d0dc504:11
shogun-buildbotbuild #287 of trusty - libshogun - viennacl is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/trusty%20-%20libshogun%20-%20viennacl/builds/287  blamelist: Viktor Gal <viktor.gal@maeth.com>04:22
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/7f0ec8896951cd870a03aca057435276c5bb0d4f by vigsterkr04:23
shogun-buildbotbuild #289 of trusty - libshogun - viennacl is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/trusty%20-%20libshogun%20-%20viennacl/builds/28904:30
-!- travis-ci [~travis-ci@ec2-54-221-162-143.compute-1.amazonaws.com] has joined #shogun05:02
travis-ciit's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/24561362705:02
-!- travis-ci [~travis-ci@ec2-54-221-162-143.compute-1.amazonaws.com] has left #shogun []05:02
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/a86dfda35b5d7d178fb3fbe55e409ab7afd914cb by vigsterkr06:08
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/df72b5f7f095f7c6ef9278e4ba22c91b9791db1b by vigsterkr07:26
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/d027609dbb4a19b121a441636c913a6720ffaf45 by vigsterkr07:35
-!- travis-ci [~travis-ci@ec2-54-82-70-90.compute-1.amazonaws.com] has joined #shogun08:03
travis-ciit's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/24564487508:03
-!- travis-ci [~travis-ci@ec2-54-82-70-90.compute-1.amazonaws.com] has left #shogun []08:03
-!- TingMiao [uid229534@gateway/web/irccloud.com/x-ekdazixzuyuobllp] has joined #shogun08:17
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has joined #shogun09:09
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has quit [Ping timeout: 260 seconds]09:16
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has joined #shogun09:17
-!- geektoni_ [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has joined #shogun09:21
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has quit [Ping timeout: 260 seconds]09:22
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3848 synchronized by geektoni09:49
shogun-buildbotbuild #218 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb3%20-%20modular_interfaces/builds/21810:46
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun10:53
-!- mode/#shogun [+o HeikoS] by ChanServ10:53
micmnwiking: ping11:11
@wikingpong11:11
@wikingmicmn, rebase rebase rebase :)11:11
micmnyeah I'm on it11:11
@wikingthnx11:11
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3711 synchronized by MikeLing11:11
micmnsome changes in class_list.py that I had to merge11:12
micmnand the global fixture thing too11:12
micmnI was looking at the SVMOcas unittest11:12
micmnhttps://github.com/MikeLing/shogun/blob/41f2ca4995d517c0729363e0f854bddb03af997c/tests/unit/classifier/svm/SVMOcas_unittest.cc11:12
micmnsame labels used for train and test?11:12
@wikingmikeling, ^11:13
mikelingmicmn: yes11:13
@wikinghttps://github.com/MikeLing/shogun/blob/41f2ca4995d517c0729363e0f854bddb03af997c/tests/unit/classifier/svm/SVMOcas_unittest.cc#L2311:14
@wikingisn't this a typo?11:14
@wikingCBinaryLabels* ground_truth = (CBinaryLabels*)mockData->get_labels_train()11:14
@wiking?11:14
@wikingor?11:14
micmnI guess it's the same, train and test of the same size and labels split 50/50? but...11:15
@wikingah but still i guess it's a bit misleading11:16
@wikingthe naming11:16
mikelingwiking: micmn They are the same11:16
@wikingso for the ground trught11:16
mikelingthe train and test label11:16
@wiking*truth semantically it should be _train11:16
micmnyeah11:16
@wikingmikeling, basically the order of the examples are the same that's what you are saying11:17
@wikingthey are not the same examples11:17
@wikingbut thye are having the same labels...11:17
@wikingit'd be still good to have it fixed11:18
@wikingCBinaryLabels* ground_truth = (CBinaryLabels*)mockData->get_labels_train()11:18
@wikingso that semantically one gets11:18
@wikingwhat is this about11:18
@wikinghttps://github.com/MikeLing/shogun/blob/41f2ca4995d517c0729363e0f854bddb03af997c/tests/unit/classifier/svm/SVMOcas_unittest.cc#L3611:18
@wikingshould compare with get_labels_test()11:18
@wikingeven if it's the same (as the order of it is the same)11:18
@wikingjust to keep the semantics correct11:18
mikeling wiking ok, I will fix it in next push11:19
@wikingk thnx11:19
mikelingnp11:19
micmnwiking, mikeling: mmm what should I use for RegressionLabels?11:26
@wikingcreate one :)11:26
mikelingmicmn: we don''t have that fixture for now :)11:27
micmnok thx :)11:27
micmnwiking, mikeling: sorry to bother you again12:21
micmnhttps://github.com/MikeLing/shogun/blob/41f2ca4995d517c0729363e0f854bddb03af997c/tests/unit/classifier/svm/SVMOcas_unittest.cc#L4012:21
micmnaren't the data shared by all tests?12:21
@wikingyeah12:22
@wikingthat should not be there12:22
@wiking:)12:22
mikelingmicmn: yes, I notice that is not necessary12:22
@wikingthnx for the review12:22
@wiking:)))12:22
@wikingfeel free to fix12:22
micmnin fact I think that SG_REF is needed12:22
micmnotherwise the machine destructor will unref those12:23
mikelingwiking: btw, I'm still not understand how sg_ref works12:23
@wikingmicmn, yep12:23
-!- geektoni_ [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has quit [Quit: Page closed]12:23
@wikingmicmn, but that should be somewhere within the creation12:23
@wikingof those objects12:23
@wikingi mean fixtures12:23
micmnmaybe in get_features_train() and similar?12:24
@wikingyes12:24
@wikingthat is an adequate place12:24
@wikingalthough12:24
@wikingyou only want to have that once ++12:24
@wikingnot every time you get12:24
@wikingas in case12:24
@wikingCSVMOcas* ocas = new CSVMOcas(1.0, train_feats, ground_truth);12:24
@wikingyou'll have a ++ on ref anyways12:24
@wikingmicmn, see what i mean?12:24
micmnah right12:24
micmnso in the environment setup12:25
@wikingyes12:25
@wikingthat'd be the best12:25
@wikingHeikoS, ping?12:27
@HeikoSwiking: ponmg12:27
@wikingHeikoS,there were couple of things... you should check some of the discussions we had with sergey and fernando yesterday12:28
@wikingonce you are alie12:28
@wiking*alive12:28
@HeikoSyeah I kind of am12:28
@HeikoSwhat were they about so I can find them?12:28
@wikingHeikoS, just check the last rant on #gsoc12:28
@HeikoSin slack?12:29
@wikingyep12:29
@wikinglovely -rwxr-xr-x  1 root  admin   103M Jun 22 15:40 /usr/local/lib/python2.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so12:31
@wiking:)12:31
@HeikoSwiking: you gotta summaries things for me, I am not following the gsoc channel :D12:36
@wikingro:)12:36
@wikinglife is a shit12:37
@wiking:)12:37
@wikingthat's in summary12:37
@HeikoSi see12:40
@HeikoSthanks for the update :D12:40
@wikingno worries :)12:41
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3856 opened by micmn12:57
-!- iglesiasg [~iglesiasg@217.119.234.214] has joined #shogun13:05
-!- mode/#shogun [+o iglesiasg] by ChanServ13:05
-!- iglesiasg [~iglesiasg@217.119.234.214] has quit [Client Quit]13:06
-!- iglesiasg [~iglesiasg@217.119.234.214] has joined #shogun13:06
-!- mode/#shogun [+o iglesiasg] by ChanServ13:06
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 240 seconds]13:27
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3751 synchronized by micmn13:34
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3856 merged by vigsterkr13:44
@sukey[https://github.com/shogun-toolbox/shogun] vigsterkr pushed 2 commits:13:44
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/479a93625f7845626d470d9a2fe6412b9392b3d213:44
@sukeyhttps://github.com/shogun-toolbox/shogun/commit/b6657907cd5aaefa3af47bd4aefa4be320b76bd013:44
@wikingok13:45
@wikingmerged good lets see the other one13:45
@wikingmicmn, why are those ignores in the macines?13:46
@wikinghttps://github.com/shogun-toolbox/shogun/pull/3751/files#diff-856555dd8e1fe665a77a0707b995f058R8013:46
micmnwiking: machines that fail for various reasons, mostly because they need some inizialization13:55
@wikingmicmn, ok so those are the ones to be fixed13:58
@wiking:)13:58
@wikingright?13:58
@wikingmicmn, what was your fix on the end for the benchmarks cript?14:00
@wikingmicmn, just because currently mlpack people are running benchmark over shogun14:01
@wikingand getting a lot of timeout14:01
@wikingi reckon it's the same reason :)14:01
@wikinghttp://masterblaster.mlpack.org/job/benchmark%20-%20shogun/47/consoleFull14:02
micmnyeah14:05
@wikingso yeah we need a stupid fix14:05
@wikingwhich is stupid14:05
@wikingbecause the only way to fix it14:05
@wikingis to disable parallel processing14:05
@wiking:D14:05
micmnhttps://gist.github.com/micmn/2b2d61f08b8e303d8d7767f3cddcab3f14:06
micmnall shogun's methods should be run in a different process14:06
@wikingoh you mean no shoung should be run14:07
@wikingprior the fork14:07
micmnyeah14:07
micmnpreviously accuracy ecc where computed in the main process14:07
micmn*were14:07
micmnhttps://gist.github.com/micmn/2b2d61f08b8e303d8d7767f3cddcab3f#file-lda-py-L13414:08
micmnbut this has to be done in every script14:08
@wikingmicmn, ok i've copy pasted this to the mlpack people14:08
@wiking:)14:09
@wikinghope they have the resource for it14:09
@wiking:)14:09
@wikingmicmn, ok so unit tests are passing on travis14:15
@wikingmicmn, since there's no jinja2 on appveyor i should be merging this14:16
@wikingonly request i have14:16
@wikingactually two14:16
@wikingb) can we close this https://github.com/shogun-toolbox/shogun/pull/375014:16
@wiking?14:16
micmnwe'll add regression etc later?14:16
micmnyeah if we merge the one I updated today14:17
@wikinga) i mean14:17
@wikingok fuck14:17
@wikingso i mean14:17
@wikingthere has been quite a lot of comment14:17
@wikingby Heiko there14:17
@wikingso to use ctags14:17
@wikingso let's think about that then14:18
@wiking+ yeah we can add the other type of machines14:18
@wikingin some other PRs14:18
@wikingonly thing apart from this14:18
@wikingthere should be an issue for every machine that has problems with serialization14:19
micmnok14:19
micmnanyway most of the ignored machines fail because they need some custom inizialization for which the defaiult 'ctor isn't enough14:21
micmn*default14:21
micmnnot directly by serialization problems (...we don't know)14:22
@wikingk14:23
@wikinglet's then address Heiko's comments14:23
@wikingin the PR you've updated14:23
@wikingand close the other one14:23
@wikingand let's start merging14:23
@wikingi was trying to get the fucking msvc story fixed14:23
@wikingbut still couldn't find a good way about the heap story14:23
@wikingbut yeah those PRs needs rebasing as well14:24
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has joined #shogun14:42
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3711 synchronized by MikeLing14:43
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3711 synchronized by MikeLing14:51
mikelingwiking: micmn The label name has been fixed in https://github.com/shogun-toolbox/shogun/pull/3711/files#diff-78679d0fd2c2058165e04034b113e7dcR23 :)14:52
mikelingsorry for the duplicate push...14:53
@wikingmmm14:53
@wikingcoll14:53
@wikingthnx14:53
@wikingadded some comments14:55
mikelingwiking: Thank you.14:56
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3711 synchronized by MikeLing15:01
@wikingmikeling, fix the license info15:04
@wikingbut no need for more test like this only convert other unit test that uses some custom data generation15:05
@wikingif possible15:05
@wikingas otherwise the test that micmn is working on15:05
@wikingis actually gonna cover this15:05
mikelingwiking: copy and past this part of comment https://github.com/shogun-toolbox/shogun/pull/3812/files#diff-9739eb22131152d61edc04fe3812921bR2 as license, right?15:10
mikelinglike this data generator https://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/classifier/svm/LibLinear_unittest.cc#L23, do we need to create a new fixture for it or just make it use the GaussianCheckerboard fixture we have15:12
micmnwiking: should I rebase the open PRs now? it will clutter travis...15:17
@wikingyes indee15:17
@wikingd15:17
micmnyes = I rebase? :p15:18
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3848 synchronized by geektoni15:19
geektoniwiking: I've upgraded #3848 and I've also tested it inside a Shogun docker container, but I wasn't able to reproduce Travis error :((15:23
@wikinggeektoni, well15:26
@wikinggeektoni, allowed to fail? :)15:26
@wikingdoes gtest has such flag?15:27
@wikingmicmn, fuck travis :)15:27
@wikinglet's kill it15:27
geektonimmh I could disable the tests for now15:27
lisitsyntravis is gone already, wiking15:28
geektoniand investigate on this problem later on.15:28
lisitsyn:P15:28
@wikinglisitsyn, :D kallaaaakaalaaaaaaaa kaaaaaaaaaaaaaaaalaaaa15:28
@wikingNYICKI15:29
@wikingkonjiec15:29
@wikingtravis15:29
@wiking:D15:29
@wikinggeektoni, SG_PAUSE_COMP15:34
@wiking;)15:34
@wikingnico15:34
@wikingi mean15:34
@wikingnice-o machine15:34
@wiking:)15:34
@wikinglets see15:34
@wikingworst case u can disable15:34
@wikingif it still fails on travisO15:34
geektoniwiking: mmh it will take a loong time before my PR will get tested by Travis :P15:36
geektonithe queue is a bit crowded ;)15:36
@wikingmmm15:37
@wikinglalal15:37
@wikingwhat should i kill?15:37
@wiking:)15:37
@wikinglisitsyn, pingu15:37
lisitsynwiking: pingo15:38
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun15:38
-!- mode/#shogun [+o HeikoS] by ChanServ15:38
@wikinglisitsyn, read and comment15:38
@wikingplz15:38
@wikingtired of this lalal15:38
lisitsynkill travis?15:38
lisitsynbut he is already fired15:38
lisitsynok ok ok15:39
lisitsynreally? kill travis?15:39
@wikingkilled15:39
@wikinggeektoni, ok so only one before your stuff :)15:39
geektoniLOL15:40
geektonithanks15:40
lisitsynyou killed travis? :)15:40
@wikingwe need a fucking way to speed up tests15:40
@wikingthis is getting again annoying15:40
@wikingshould we add ccache to the docker img?15:40
@wiking:)15:40
lisitsynmake tests annoying again?15:40
@wikinglisitsyn, well i mean travis is super slow15:40
@wikingtakes again a long time to get things tests15:41
@wiking*tested15:41
@wikingwhat we could do is to add ccache to the docker img15:41
lisitsynhow does that help15:41
@wikingwell15:41
@wikingwe init it :)15:41
lisitsynhahaha15:41
@wikingand that's how we push it to the docker repo15:42
@wiking:)15:42
lisitsynit would get useless in a month15:42
@wikingand then after every release15:42
@wikingwe update it15:42
@wikingyeah15:42
@wikingalthough it would still be better as it is now15:42
@wikingi mean15:42
@wikingok15:42
@wikingthen question15:42
@wikinganybody KNOWS how the fuck we could have something similar15:42
@wikinglike an EBS? :)15:42
@wikingon aws15:42
@wikingfor travis15:42
@wikingwe do that with the buildbot builders15:42
lisitsyns3fs15:42
@wikinglisitsyn, speed?15:42
lisitsynI didn't benchmark15:43
lisitsynshouldn't be that slow15:43
lisitsynok idea15:43
@wikingok so that could work15:43
lisitsynyes15:43
@wikingor just have15:43
@wikinga tar.xz15:43
lisitsynmount a volume using fuse that is s3fs15:43
lisitsyncaches are append-only anyway15:44
@wikingappend-only? :D15:44
@wikingif you cannot read15:44
@wikingwhat's the point15:44
@wiking:D15:44
@wiking(ok i'm tired)15:44
lisitsynI mean you just add a file that's MD515:44
lisitsynhaha ok I was ambiguous15:45
lisitsynI mean there is no issue with parallel write15:45
lisitsynto the same file15:45
lisitsynbut there could be some crazy waits if we go parallel15:45
lisitsyndue to posix stuff15:45
@wikinghehe15:45
@wikingyeah i mean this is shitty actually15:45
@wikingwith those 10 jobs15:45
@wikingon travis15:46
@wikingmikeling, yes exactly like these needs to be replaced https://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/classifier/svm/LibLinear_unittest.cc#L2316:17
@wikingno i mean16:17
@wikingthe only thing we want16:18
@wikingthat if we can use the fixture16:18
@wikingthen let's use that16:18
@wikingso this is like a 2 gaussians each being dim = 216:18
@wikingwith 50 samples16:18
mikelingyes, so we don't need new fixture right? Just replace the generate_gaussians with the fixture we have right now16:19
mikelinggot it16:19
@wikingyeah16:19
@wikingbut you need to take care of some details16:19
@wikingright?16:19
mikelingyes16:19
@wikinghttps://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/classifier/svm/LibLinear_unittest.cc#L4916:19
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 276 seconds]16:32
-!- iglesiasg [~iglesiasg@217.119.234.214] has quit [Quit: leaving]16:41
geektoniwiking: https://travis-ci.org/shogun-toolbox/shogun/jobs/245784419#L198316:51
@wikingdead?16:51
geektoniall interfaces now16:52
geektoniLOL16:52
geektoniapparently16:52
@wiking:)16:52
geektoniSWIG generate wrong code when dealing with rxcpp::subscriber16:52
@wikinggeektoni, make_subscriber16:52
@wikinggeektoni, you can disable that code section from16:52
@wikingswig16:52
@wikingdo you need to expose it to SWIG interfaces16:52
@wiking?16:52
@wikingi dont think so16:53
geektoninope16:53
@wikingno the actual handlers16:53
geektonimmh, I don't think16:53
geektoniso16:53
@wikingthere's a macro16:56
@wiking#ifndef SWIG16:57
geektoniyeah I've already seen them around the code16:58
geektonibut16:58
geektoniI'm not quite sure where is the code that causes SWIG to fail16:58
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun16:58
-!- mode/#shogun [+o HeikoS] by ChanServ16:58
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3857 opened by MikeLing17:00
geektoniwiking: at least gcc now works lol https://travis-ci.org/shogun-toolbox/shogun/jobs/24578441717:01
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3848 synchronized by geektoni17:25
geektoniwiking: ^ interfaces should be okay now17:26
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3711 synchronized by MikeLing17:36
@HeikoSwiking, mikeling btw it is dangerous to test libsvm with gaussians as just by chance, there will be errors.17:37
@HeikoSwiking, mikeling we have some datasets with linearly separable data as well that could be used instead17:37
@HeikoSwiking:  in general, I think we should somehow distinguish between unit tests and tests for statistical correctness (e.g. in a random sense)17:37
mikelingHeikoS: you mean we make unit test depend on dataset in data/ folder ?17:38
@wikingno17:38
@HeikoSmikeling: thats an option, though I know it is not optimal17:38
@wikingwe do not want to use17:38
@wikingany file17:38
@wikingfrom data17:38
@HeikoSjust saying:17:38
@wikingin a unit tests17:38
@HeikoSif you use gaussian blobs17:38
@wikingthat's not a unit test17:39
@HeikoSthen your test will sporadically fail17:39
@HeikoSunless you fix the seed17:39
-!- geektoni [c1cdd252@gateway/web/freenode/ip.193.205.210.82] has quit [Ping timeout: 260 seconds]17:39
@HeikoSwhich is implicitly fixing a dataset, which is also not good17:39
@HeikoSso having something that generates linearly separable data, and then checking 100% accuracy is a better idea than gaussians17:39
@HeikoSgaussians with compact support for example17:39
mikelingHeikoS: ok, maybe CMath::ini_random(10) for now?17:39
mikeling* init_random(10)17:40
@HeikoSmikeling: thats not really nice either17:41
@HeikoSwhat data are you using?17:41
@HeikoSor fixture?17:41
@HeikoSwhat does it generate?17:41
@HeikoSmikeling: ?17:42
@HeikoSmikeling: you know how to restrict the support of Gaussians? This way you could ensure that they dont overlap, i.e. your linear classifier can perfectly distinguish the classes17:43
mikelingHeikoS: mmm, no, i don't know how to restrict the Gaussians17:43
@HeikoSmikeling: what class/code are you using?17:43
mikelingHeikoS: in https://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/environments/GaussianCheckerboard.h17:44
@HeikoSmikeling: uuuh, that code is very hard to read17:45
mikelingHeikoS: mmm, which part?17:45
@HeikoSwiking: shouldnt have been merged17:46
@HeikoSfor (index_t i = 0, j = 0; i < data.num_cols; ++i)17:46
@HeikoStest_idx[j++] = i;17:46
@HeikoS17:46
@HeikoS17:46
@HeikoSlabels[i / 2] = (i < data.num_cols / 2) ? 1.0 : -1.0;17:46
@HeikoSi mean are you kidding me?17:46
@wiking?17:47
@HeikoSthats horrible17:47
@HeikoSto read code17:47
@HeikoSall this can be done nicely and readable17:47
@wikingsure17:47
mikelingHeikoS: basically, I'm doing the same thing in https://github.com/shogun-toolbox/shogun/pull/3812/files#diff-78679d0fd2c2058165e04034b113e7dcL2017:47
@wikingthe PR was open for more than 2 weeks17:47
@HeikoSmikeling: sure I have seen things like this elsewhere17:48
@HeikoSdoesnt change the fact that one can also write this nicely17:48
@HeikoSok anyways17:49
@HeikoSas for the gaussians17:49
@wikingany suggestion for improvement is more than welcome17:49
@wikingbefore or after merge17:49
@HeikoSmikeling: you have the PR somewhere?17:49
@HeikoSI can comment there17:50
@wikingthis is not breaking yet any CI17:50
mikelingsure, just a second17:50
@wikingapart from aestetics17:50
@HeikoShard to maintain, puts people off17:50
mikelingHeikoS: here is the pr https://github.com/shogun-toolbox/shogun/pull/381217:50
@wikingsure17:50
@wikingas said17:50
@HeikoSwill hardly b touched17:50
@wikingHeikoS, it was open for a long time17:50
@wikingat least 2-3 weeks17:51
@HeikoSI have read that17:51
@wikingso i mean17:51
@HeikoSI know I was gone for a bit, but pls next time ping me in the PR and I will comment17:51
@HeikoSyou started the whole thing about having multiple dev +1 before merging, remember?17:52
@wikingHeikoS, look man17:52
@wikingi'm fine doing this here17:52
@wikingyou were not around17:52
@wikingfor the last couple of weeks17:52
@wikingi'm trying to maintain 3 students input17:52
@wikingwhere one of them is struggling with some very basic things17:53
@wikingapart from this i'm trying to do my job17:53
@wikingif you think you could have done this better17:53
@wikingi'm happy to hear about your input17:53
@wikinghow i could have done better things17:53
@wikingbut sorry17:53
@HeikoSI will comment17:53
@wikingnone of the merged PRs17:53
@wikinghave broken17:53
@wikingANY17:53
@wikingCI17:53
@wikingwhereas my comment of17:53
@wiking[23:49]  <HeikoS> you started the whole thing about having multiple dev +1 before merging, remember?17:54
@wikingwas due to the fact17:54
@wikingthat there was17:54
@wikingNUMEROUS cases17:54
@wikingwhere PRs were merged17:54
@wikingthat broke CIs17:54
@wikingbut again17:54
@wikingif this thing is hurting anybody's aeastetics17:54
@wikingmea culpa17:54
@wikingthis is all i could manage17:54
@wikingi can stop it17:55
lisitsynhey18:03
lisitsynI'd vote for merging working stuff as well18:03
mikelingsorry....I should do it better18:03
lisitsynpolishing is easier18:03
@HeikoS(and never happened in the past)18:04
lisitsynwhat never happened?18:13
lisitsynI think it has happened quite a few times before18:13
@HeikoSmikeling: dont worry, I sent a list of things that could be improved in the PR18:41
mikelingHeikoS: Thank you! I will address it right away18:42
@HeikoSmikeling: I think some of the code I complained about was not written by you so I should blame the person who wrote it instead (which is likely to be myself :D )18:45
@HeikoSwiking: ^18:45
@wikingdunno who wrote the ocas test18:45
@wikingi guess a blame could tell18:46
@HeikoSI think this stuff comes form the gaussian blobs18:46
@HeikoSsomehow got copy pasted in there18:46
@HeikoSand slightly modified18:46
@HeikoSfrom a messy start18:46
@HeikoSbut it doesnt matter18:46
@HeikoSI put some suggestions in18:46
@HeikoSwiking: what are your thoughts about separating unit tests from correctness tests?18:47
@wikingmicmn, i dont know if i already wrote this or not.... but lemme be sure that i did: so would be great if you could fire an issue against the benchamrks of mlpack about how to fix the shogun openmp problem... just so that this doesn't disappear into the void18:48
@wikingthnx heaps18:48
@HeikoSwiking: sorry for the waterfall, but what happened to the std::vector thing in the end?18:48
@wikingi reported it to zoq (main developer of benchmarks) but just in case having an issue that marks the things would be great18:48
@wikingHeikoS, ^ this is a fucking shit btw18:49
@HeikoSwhats going on there?18:49
@wikingopenmp + gnu fork18:49
@wikingit's deadlock18:49
@wikingthere's a workaround18:49
@wikingbut18:49
@wikingstill18:49
@wikingthis is a shit18:49
@HeikoSah18:49
@wikingso if you for example wanna use shogun18:49
@wikingand then fork some processes18:50
@wikingand use some shogun more18:50
@wikingyou are fucked18:50
@wikingbecause the openmp will get into some hanging state18:50
micmnwiking: sure, will do18:50
@wikingand basically this is what they do in benchmarks of mlpack18:50
@wikingmicmn, thnx heaps18:50
@HeikoSha18:50
@HeikoSis that reported?18:50
@wikingand now they ran18:50
@HeikoSas in "shogun crashes" ?18:50
@wikingagainst new shogun18:51
@HeikoSin their papers? :D18:51
@HeikoScute18:51
@wikingand tests timeouts18:51
@wikinglike a lot18:51
@wikingbecuase actually the openmp code18:51
@wikingis happening18:51
@wikingin eigen level18:51
@wiking:)18:51
@wikingmostly18:51
@wikingso we cannot even control anything18:51
@wiking...18:51
@wikinganyhow i told it to zoq so it'll get handled18:51
@wikingand then they rerun stuff18:51
@wikingbt this we should mark somewhere18:52
@wikingbecause it's a major issue18:52
@wikingwe again dont have a FAQ18:52
@wikingright?18:52
@wikingcoz this would be worth to be mentioned anywhere there18:52
@wikingif you fork how to fork18:52
@wikingas gnu fork with gnu openmp18:52
@wikingwill fuck you18:52
@wikingHeikoS, doesn't happen with icc18:53
@wikingof course...18:53
@HeikoSok18:53
@wikingmicmn, and the other reason why it'd be cool because you had all the logs of bugs etc... but i saw that you've logged thos in your daily18:53
@wikingso that's really great18:53
@HeikoSyeah an issue is really important18:53
@wikingwe can use that later for the faq or something18:53
@HeikoSboth in shogun and in zoq as well18:53
@wikingHeikoS, zoq said he'll port the fix18:54
@wikingduring th weekend18:54
@wikingis18:54
@wiking*ish18:54
@wikingcoz basically this is all linalg :)18:54
@wikingso yeah this should be somewhere easily found info18:54
@wikingand we cannot do anything about this18:54
@wikingi.e. not an issue18:55
@wikingthis is a feature of gnu openm18:55
@wiking*openmp18:55
@wikingi'm wondering if this shit comes up with clang's openmp18:55
@wikingdoes it use the same gomp?18:55
@HeikoSno idea18:55
@wikingi guess no because libiomp518:56
@wikingbuuuut18:56
@wikingwould be great to figure these shit out...18:56
@wikingHeikoS, mmmm correctness...18:57
@wikingHeikoS, you mean similar to serializable?18:57
@wikingi mean in one way and the ohter18:58
@HeikoSwiking: sorry what do you mean?18:58
@wikingtrain -> serialize18:58
@HeikoSah18:58
@wikingtest18:58
@HeikoSyeah I mean18:58
@wikingwould be similar18:58
@wikingthan the 'correctness' what you mean18:58
@HeikoSunit tests are kind of tests for modules working fine18:58
@wikinghere18:58
@wikingright?18:58
@HeikoSindepdendent of random data18:58
@HeikoSlike a+a = 2*a18:58
@HeikoSbut we have lots of test that are like18:58
@HeikoS"linearly separates linearly separable data"18:58
@HeikoSwhich seems a bit different to me18:59
@HeikoSserialization is more unit test like18:59
@wikingmicmn, is just refurbishing this https://github.com/shogun-toolbox/shogun/pull/375118:59
@HeikoSbut different because it is auto generated and covers all classes18:59
@wikingso i mean19:00
@wikingit is in a way similar19:00
@wikingbut of course19:00
@wikingthere any fucking numb data would do it19:00
@wikingfor the purpose of testing19:00
@wikingserialization19:00
@wikingmicmn, so there we do the refactor still of the python scripto?19:00
@wiking :)19:00
micmnyeah, you mean Heiko's comments on ctags?19:01
@wikingyes19:02
@wikingcool19:02
@wikingok i'm gonna check wtf is there19:02
@wikingwith the https://github.com/shogun-toolbox/shogun/pull/384319:02
@wikingas for some reason19:02
@wikingtravis hiccuped19:02
@wikingaaah19:03
@wikingi know19:03
@wikingstupid formatting19:03
@wiking:D19:03
micmnof course19:03
* wiking thiking about removing either msvc 14 or 15 from the appveyor... as 1 pr test takes 1 hour19:04
@wikingany objections^19:04
@wikingand ther's no ccache for msvc19:05
@wikinglisitsyn, fucking appveyor has cache dir support  :)19:05
@wikinglisitsyn, Caches lets Travis CI store directories between builds, which is useful for storing dependencies that take longer to compile or download.19:06
@wikingargh19:06
@HeikoSwiking: no objections from my side19:06
@wikingfuck19:07
@wikinghttps://docs.travis-ci.com/user/caching/#ccache-cache19:07
@wikingsince when ?:)19:07
@wikinglisitsyn, ^ :D19:07
@wikingwhat do you think? :)19:07
@wikingkill with fire? :D19:07
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 246 seconds]19:10
@sukey[https://github.com/shogun-toolbox/shogun] New branch travis-ccache created19:18
@sukey[https://github.com/shogun-toolbox/shogun] New commit https://github.com/shogun-toolbox/shogun/commit/9029307e64c47c3b32573c884782d19ecc31b0e9 by vigsterkr19:18
lisitsynwiking: oh19:18
lisitsynthat's good19:18
@wikinglets see19:19
@wikingif it works19:19
@wikingi guess i'll know it in the morning19:22
@wikingas i'm off19:22
@wikinggnite all19:22
rcurtinHeikoS: wiking: I don't think we've published any benchmarks with these failures on our site, I think it has been a long time since any new shogun benchmarks have been done19:48
rcurtindefinitely we wouldn't want to publish incorrect benchmarks, so I think marcus will update using the patch we got today and then we can re-run19:49
rcurtinthe benchmarking system has not received the love it should have :(  we are trying to change that...19:51
-!- travis-ci [~travis-ci@ec2-23-22-171-32.compute-1.amazonaws.com] has joined #shogun21:07
travis-ciit's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: https://travis-ci.org/shogun-toolbox/shogun/builds/24587907821:07
-!- travis-ci [~travis-ci@ec2-23-22-171-32.compute-1.amazonaws.com] has left #shogun []21:07
-!- mikeling [uid89706@gateway/web/irccloud.com/x-oewvncwnqrkuhecv] has quit [Quit: Connection closed for inactivity]22:06
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun22:08
-!- mode/#shogun [+o HeikoS] by ChanServ22:09
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 260 seconds]22:42
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun22:52
-!- mode/#shogun [+o HeikoS] by ChanServ22:52
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Client Quit]22:56
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun23:17
-!- mode/#shogun [+o HeikoS] by ChanServ23:17
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 240 seconds]23:26
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3826 synchronized by micmn23:46
@sukey[https://github.com/shogun-toolbox/shogun] Pull Request https://github.com/shogun-toolbox/shogun/pull/3842 synchronized by micmn23:46
--- Log closed Fri Jun 23 00:00:25 2017