Open in new window / Try shogun cloud
--- Log opened Mon Feb 20 00:00:32 2017
@sukeyPull Request #3399 "LinalgRefactor - Add sum_symmetric and Rm old sum methods" - https://github.com/shogun-toolbox/shogun/pull/339900:06
@wikingCaBa, yes if you want to have proper clone-able object you need to have the params registered01:31
-!- goksinen_ [~goksinen@2604:2000:c591:8400:3420:31fb:5538:73fd] has joined #shogun01:48
@sukeyIssue #3644 "New Linalg library cannot compile with Eigen 3.3.x" opened by OXPHOS - https://github.com/shogun-toolbox/shogun/issues/364402:46
-!- IRCFrEAK [~gk.1wm.su@2a03:4a80:3:43d:43d:23b:2e38:2839] has joined #shogun03:30
-!- IRCFrEAK [~gk.1wm.su@2a03:4a80:3:43d:43d:23b:2e38:2839] has left #shogun []03:30
@sukeyPull Request #3641 "add tests for KNN and fix an error in KDTree solver"  synchronized by MikeLing - https://github.com/shogun-toolbox/shogun/pull/364104:42
-!- goksinen_ [~goksinen@2604:2000:c591:8400:3420:31fb:5538:73fd] has quit [Remote host closed the connection]05:00
@sukeyPull Request #3641 "add tests for KNN and fix an error in KDTree solver"  synchronized by MikeLing - https://github.com/shogun-toolbox/shogun/pull/364105:34
@sukeyPull Request #3635 "LinalgRefactor - Memory Transfer Mutex"  synchronized by OXPHOS - https://github.com/shogun-toolbox/shogun/pull/363506:21
-!- madgoat [~gk.1wm.su@2a03:4a80:5:41e:41e:efdd:c10d:1c12] has joined #shogun06:36
-!- madgoat [~gk.1wm.su@2a03:4a80:5:41e:41e:efdd:c10d:1c12] has left #shogun []06:36
shogun-buildbotbuild #53 of nightly_default is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/53  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Soumyajit De <heavensdevil6909@gmail.com>, Tiramisu 1993 <sabergeass@gmail.com>08:38
shogun-buildbotbuild #31 of nightly_all is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_all/builds/31  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Soumyajit De <heavensdevil6909@gmail.com>, Tiramisu 1993 <sabergeass@gmail.com>09:02
@sukeyNew Commit "Update NEWS" to shogun-toolbox/shogun by lambday: https://github.com/shogun-toolbox/shogun/commit/142ba6b916e623f75c393799598b1d91332687b109:12
@sukeyIssue #3625 "Installation failed"- https://github.com/shogun-toolbox/shogun/issues/362509:14
-!- travis-ci [~travis-ci@ec2-54-91-162-115.compute-1.amazonaws.com] has joined #shogun10:06
travis-ciit's Soumyajit De'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/20335583510:06
-!- travis-ci [~travis-ci@ec2-54-91-162-115.compute-1.amazonaws.com] has left #shogun []10:06
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun10:30
-!- mode/#shogun [+o HeikoS] by ChanServ10:31
CaBaHeikoS: heya10:44
CaBawiking: thanks for the info10:44
CaBahttps://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/base/class_list.cpp.templ <- what dark magic is this?10:44
@wikingCaBa, generates a factory clas10:45
@wikingclass10:45
@wikingfor the swig interface10:45
CaBawiking: my class still fails to clone, apparently in there10:45
CaBawiking: do i have to "register" it somewhere?10:45
@wikingthen i suppose the way you defined the class10:45
@wikingis having some issue10:45
CaBawiking: well my class get_name() value is not in the class_list i suppose10:46
@HeikoSwiking: jojo10:46
@HeikoSCaBa: might well be some cache thing10:46
@HeikoSthere is build bugs when you add classes after the class_list has been generated10:47
CaBaHeikoS: my kernel is not in shogun. i have it in my application, which links against shogun10:47
@HeikoSthen you shouldnt need this10:48
CaBaHeikoS: well my kernel won't ::clone() otherwise10:48
@wikingHeikoS, so not true10:49
@wikingHeikoS, please update yourself with some of the latest development of10:49
@wikingshogun10:49
@wiking:)10:49
CaBalol10:49
@wikinglike CLock10:49
@wikingand this class_list.cpp10:49
@HeikoSwiking: ?10:50
@wikingi worked on this10:51
@wikingquite extensively10:51
@wiking2 weeknds ago10:51
@HeikoSworked on what?10:51
@wikingboth clock and class_list.cpp10:51
@HeikoSah i see10:51
@HeikoScool then10:51
CaBawiking: so what do i need to do for my outside-of-shogun kernel then? write my own clone()?10:51
@HeikoSwhat about this? https://github.com/shogun-toolbox/shogun/issues/293410:52
@wikingmmm but wait10:52
@wikingCaBa, you wanna write10:52
@wikinga class that is not part of the library10:52
@wikingsay a new kernel10:52
@wikingand then have clone? :)10:52
@wikingmmmmmm10:52
CaBayeah i actually have quite a few CKernel derivatives in my application10:53
@wikingit might work on some level10:53
@wikingbut for sure not for swig10:53
@wiking:(10:53
@wikingwhich i know is a shit10:53
CaBai use c++ only10:53
@wikingbut this is what is it10:53
@wikingCaBa, and what's your error?10:54
@wikingwhen you call kernel->clone()10:54
@wiking?10:54
CaBawiking: shogun::new_sgserializable returns NULL10:54
@wikingdoh10:54
@wiking:D10:54
CaBawiking: since it doesn't find my class name in it's list10:54
@wikingyeah10:54
@wikingso here we go10:54
@wikinganother shit10:54
@wiking:D10:54
@wikingso actually then10:55
@wikingit doesn't work at all10:55
@wikingif it's not part of the library10:55
@wikingwhich makes everything fucking crazy10:55
@wiking:)10:55
@wikingbecause to be able to extend your library10:55
@wikingi mean the library10:55
@wikingyou actually have to make your code10:55
@wikingpart of the library10:55
@wiking:(10:55
@wikingHeikoS, you wrote clone no?10:55
@HeikoSyep10:56
@HeikoSbut this sg_serializable thing not10:56
@wikingHeikoS, caef78db (Heiko Strathmann 2013-06-11 14:49:26 +0100 747) CSGObject* CSGObject::clone()10:56
@HeikoSthe clone is based on that10:56
@wikingok but the way you do it10:56
@wikingis to call the factory object10:56
@wiking?10:56
@wikingi.e. class_list.h's sg_ser?10:56
@HeikoSI would have to check, but I think yes10:56
@wikingye yep10:57
@wikingok so this is major shit10:57
@wiking:D10:57
@HeikoShttps://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/base/SGObject.cpp#L75210:57
@wikingi wonder how tags will help us here10:57
CaBawell... overriding ::clone() could be a workaround, right? the requirements aren't that complex, i just have to return a pointer to a copy... i can even call ::clone() for all upstream shogun member variables...10:57
@wikingyeah10:57
@wikinglemme see if you can actually override10:57
CaBayes its virtual10:58
@wikingvirtual CSGObject* clone();10:58
@wikingluckly10:58
@wiking;)10:58
@wikingok so then you can implement your own clone method10:58
@wikingbut10:58
@wikingstill this we need to fix urgently10:58
@wikingHeikoS, ^10:58
CaBai already override that before :P since i have non-SGobject members :P10:58
CaBabut then in the override i called SGObject::clone() for the rest10:58
@wikingi see10:58
@wikingwell then you cannor rely obviously10:58
@HeikoSwiking: how would you create the new instance?10:59
@wikingHeikoS, yeah i'm just wondering if we could create at least a helper method10:59
@wikingthat CaBa would need10:59
@HeikoSif we just used cereal, then all good10:59
@HeikoSserialize to memory10:59
@HeikoSde-serialize10:59
@wikingmemory?10:59
@wikinguffofff10:59
@wikingthat's very brute force10:59
@wikingwhat about just cutting out the part11:00
@wikingwhere we copy the registered params11:00
@wikinghttps://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/base/SGObject.cpp#L76111:00
@wikingand add that as11:00
@wikinga helper method11:00
@wiking(protected?)11:00
@wikingto the SGObj11:00
@HeikoSI see11:01
@HeikoSyeah thats an option11:01
@wikingand i wonder11:01
@wikinginstead of11:01
@wikingCSGObject* copy=new_sgserializable(get_name(), this->m_generic);11:01
@wikingwhy can't we just call11:01
@wikingthe default ctor of the obj?11:01
@wikingi mean that shit i think you can do11:01
@wikingwith type11:01
@wikingin c++1111:01
@wikingand once linalg is in11:01
@wikingc++11 is in11:01
@wikingright?11:01
@HeikoSyep11:02
@wikingbecause now it's a bit shit i woudl say11:02
@wikingor?11:02
@wikingand then i wonder11:02
@wikingif we could get rid of the whole class_list.cpp hack11:02
@wikinglet's see where else new_sgserializable is being used11:02
@wiking:)11:02
@wikingsrc/shogun/base/Parameter.cpp:  *param = new_sgserializable(sgserializable_name, generic);11:03
@wikingsrc/shogun/base/SGObject.cpp:   CSGObject* copy=new_sgserializable(get_name(), this->m_generic);11:03
@wikingtests/unit/base/DynamicObjectArray_unittest.cc.jinja2:          array1->push_back(new_sgserializable(class_name, PT_NOT_GENERIC));11:03
@wikingtests/unit/base/DynamicObjectArray_unittest.cc.jinja2:          array1->push_back(new_sgserializable(class_name, {{type}}));11:03
@wikingok11:03
@wikingthis is doable11:03
@wikingno?11:03
@wiking:)11:03
@HeikoSde-serialization11:04
@HeikoSalso needs default ctor objects11:04
@HeikoSbut thats all in fact11:04
@HeikoSso yeah much nicer11:04
@wikingi mean then we dont need11:04
@wikingthe classlist hack11:04
@wikingso let's see11:04
@wikingi'll make a feature branch11:04
@wikingand see if it's possible11:04
@wikingand make it like that11:04
@wikingthat if c++11 available11:04
@wikingit'll rather use that11:04
@wikingthan class_list11:04
@wikingso later it is easy to cut out the code11:05
@HeikoSyes11:05
@HeikoSgood11:05
@HeikoSwiking: just saw you changed location of shogun.unit-test11:06
@HeikoSchanged any others?11:06
@wiking?11:06
@wikingyou mean bin?11:07
@wikingthat has moved a long time ago11:07
@wiking(with windows feature)11:07
@HeikoSkk11:07
@HeikoSany other thing? just realised readme is outdated ...11:07
@wikingmmm11:07
@wikingyeah cmake readme i'll have to update11:07
@wikingbefore release11:07
@wikingbecause now every feature has11:07
@wikingENABLE_<package> flag11:08
@wikingso even if it's detected11:08
@wikingyou can turn it off11:08
@wikingsay11:08
@wiking-DENABLE_ZLIB=OFF11:08
@wikingetc.11:08
@wikingand yeah the strings to change linalg backend11:08
@wikingchanged as well11:08
@wikinghad to11:08
@wikingthe copy paste code11:08
@wikingwas a mess11:08
@HeikoSok then i wont update the readme submodule for now11:09
@HeikoSjust patched developing.md11:09
@wikingyeah that needs fixing11:09
@wikingbut lets get first things into releaseable state11:09
@wikingwhich we are quite off now from11:09
@wiking(many failing builds again)11:10
@sukeyNew Commit "disable failing linear time mmd unit tests" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/1560b3b39e214be4cd066c6ca4005496dc9a335411:11
@wikingis this the solution? :)11:12
@HeikoSno11:12
@wikingwhy are they failing?11:12
@HeikoSI have no idea11:12
@wikingok why dont we let Rahul fix it?11:12
@wikinghe said he'll fix it11:12
@HeikoSyeah11:12
@wikingso ?11:12
CaBahmmm... it's not so trivial to clone the CKernel components of my kernel, right? unfortunately, CKernel also doesn't offer a copy ctor...11:12
@HeikoSI want travis green11:13
@wikingfor?11:13
@wikingit's already red11:13
@wikingsee octave11:13
@HeikoSwiking: things not slipping through even more11:13
@HeikoSI know11:13
@HeikoSthese tests are for like weird parts anyways ...11:14
@HeikoSso they could even be deleted, they are old11:14
@HeikoSand testing non-interpretable stuff11:14
@HeikoSbut yeah Ill leave it11:14
@wikingok then why not delete it?11:14
@HeikoSrahul promised that hell do the guarding today11:14
@wikingi mean iether just chuck it out11:14
@wikingif it's really not necessary11:14
@wikingthe guarding is a totally different story11:14
@wikingi dont even know how that not raised any concerns11:14
CaBahow about11:15
CaBahttps://github.com/auchenberg/volkswagen11:15
CaBa?11:15
CaBa:-)11:15
@HeikoShahaha11:15
@HeikoSnic eone11:15
@wiking:)11:15
@wikingHeikoS, when's that course starting?11:18
@HeikoSsooo11:18
@HeikoSwe decided to not use cloud for the march one11:18
@HeikoSbut for the may one11:18
@HeikoSthere are two11:18
@wikingwhy not?11:18
@HeikoSeach year11:18
@wikingi mean the problem is11:18
@wikingthat in may11:18
@wikingprobably11:18
@HeikoSbecause we would have to develop the material now11:18
@wikingyou wont be able to use it anymore11:18
@wiking(the cloud)11:19
@wikingnote: the credits are out in may11:19
@HeikoSuh I see11:19
@HeikoSwe need to get new ones then :)11:19
@wikingand for being able to get new ones11:19
@wikingone would need to demonstrate11:19
@wikingthat the previously got credits were used in a constructive manner11:19
@wikingi suppose :)11:19
@HeikoSnah I am sure we can get more via the usual channels11:19
@wikingbut yeah11:19
@wikingif you wanna get more11:19
@HeikoSI mean I can sneak some things in11:19
@HeikoSin march11:20
@wikingmake sure you dont forget11:20
@wikingto remind gunnar about this11:20
@wikinglatest around end of march11:20
@HeikoSyeah we talk soon right11:20
@wikingas you remember it took a while11:20
@wikingto get those credits11:20
@HeikoSI can put some demo things in11:20
@HeikoSbut the course also needs re-writing11:20
@wikingYour promotion is valid for 12 months and will expire on May 31, 2017 or when credits are fully depleted, whatever comes first11:20
@wikingjust checked11:20
@HeikoSand arthur was hesitant to do that yet as the cloud is not yet workintg11:20
@HeikoSand the old material software works11:21
@wiking?11:21
@wikingbut for the cloud11:21
@wikingyou11:21
@wikingor who ever11:21
@wikingjust needs to create a docker image11:21
@wikingnothing else11:21
@wikingi mean if you have the notebooks?11:21
@wikingthen it's super easy11:21
@wikingif you dont have any notebook11:21
@wikingand you have to write the notebooks11:21
@wikingthen i understand11:21
@wiking"old material software" =?11:22
@wikingwritten in?11:22
@HeikoSwe will only start putting time into that once the thing is working on the aws11:22
@wikinglolface11:22
@HeikoSthat is how it goes here, unfortunately11:22
@wikingi mean lolface11:22
@wikingi'm sorry11:22
@wikingthis is way too mcuh man11:22
@HeikoSthere is many people involved I am not to decide11:22
@wikingok look11:22
@wikingi dont give a damn11:22
@wikingit's just very fucking disappointing11:23
@HeikoSold material is matlab stuff and SVMLight ;D11:23
@wikingwhen somebody works11:23
@wikinghours and hours11:23
@wikingand you get this shit11:23
@wikingbut hey11:23
@wikingyeah11:23
@wikingsure11:23
@HeikoSwe will use it in may11:23
@wikingi mean the question11:23
@wikingif we had an agreement11:23
@HeikoSits way before 31st11:23
@wikingthat we are gonna do this11:23
@HeikoSso all good with the credits11:23
@wikingin march11:23
@wikingand i held up to my end11:23
@wikingi wonder11:23
@wikingwhy do i get this now?11:24
@wiking"[18:20]  <HeikoS> we will only start putting time into that once the thing is working on the aws11:24
@wiking"11:24
@wikingit's a bit11:24
@wikinglet's say11:24
@wikingawkward11:24
@HeikoSI cant convince anyone in relying on the cloud thing before it works11:24
@wikingagain11:24
@HeikoSyou see?11:24
@wikingif we have11:24
@HeikoSI say: lets do it11:24
@wikingan agreement11:24
@wikingthat we do this11:25
@HeikoSthey say: well it doesnt work, lets wait until it works11:25
@wikingin march 1711:25
@wikingand i say let's do it11:25
@wikingthen ok i do my part11:25
@wikingyou gusy do your part11:25
@wikingif you say11:25
@HeikoSthere is more than just us11:25
@HeikoSarthur also has to re-work the course11:25
@wikingthat when there's cloud11:25
@wikingwe wil think about doing it11:25
@wikingok that's acceptable as well11:25
@wikingand all good11:25
@wikingbut that's another story11:25
@wikingsee what i mean?11:25
@wikingi was calculating with this being set in stone11:26
@wikingnow i'm being told11:26
@wiking'when ready we'll see'11:26
@wikingok11:26
@HeikoSnot we will see11:26
@HeikoSwe want to do it11:26
@wiking[18:22]  <HeikoS> they say: well it doesnt work, lets wait until it works11:26
@wikingit is we'll see once it's ready11:26
@wikingi mean ready = cloud working11:26
@wikingthen after that t11:27
@wikingthings gonna get rolling on your end11:27
@wikingand dont get me wrong11:27
@HeikoSyes arthur will do the course after:11:27
@HeikoS-the cloud works11:27
@wikingit's not your fault or anything11:27
@HeikoS-i wrote some coursework11:27
@wikingbut please next time11:27
@wikingbe more specific when you write11:27
@wiking'lets have march course thing'11:27
@HeikoSand it needs time11:27
@wikingbecause i was counting on this11:27
@wikingnow it's a disappointment from my side11:28
@wikingbecause i really wasted a lot of fucking time on this11:28
@wikingwhereas if you start11:28
@wikinglets do it in may11:28
@wikingthen i might have done it in different phase11:28
@wikingthat's all11:28
@HeikoSlets do it now11:28
@wikingyeah i told you11:28
@wikingprepare the notebooks11:28
@HeikoSI am11:28
@HeikoSthats why i merged bigtest11:29
@wikingthat's all is needed11:29
@wikingis it working from python?11:29
@wikingor what notebook you want?11:29
@wikingc++11:29
@wikingjava?11:29
@HeikoSnono python is ok11:29
@wikingR?11:29
@wikingok have you guys tested11:29
@HeikoSoctave maybe11:29
@wikingthe python interface11:29
@wikingof the bigtest?11:29
@HeikoSyes11:29
@wikingif there's c+11111:29
@wikingit works i guess11:29
@wikingok so then why dont you prepare the notebooks?11:30
@wikingand give me a location where i can download them11:30
@wikingsame for the data11:30
@wikingit requires11:30
@wikingand currently the way the authentication works for the notebooks is github11:30
@wikingbut it can be anything from this list11:30
@wikinghttps://github.com/jupyterhub/oauthenticator11:31
@wikingor you can develop your own authenticator if you wanna use something else11:31
@HeikoSkk11:32
@HeikoSso one more thing11:32
@HeikoSthe course in march will be on svm, the one in may on testing11:32
@HeikoSso if i write things for svm, that only makes sense if the thing works then11:33
@wikingbut man11:33
@HeikoSif I write for kernel testing, it is all a bit more relaxed as we want to change it to that anyways11:33
@wikingthis is ready11:33
@wikingthe cloud11:33
@wikingit is FUCKING READY11:33
@wikingi'm now just wanking11:33
@wikingto add extra support11:34
@wikingto be nice and easy11:34
@wikingwith dcos11:34
@wikingbut on the other hand functionally it was ready last week wednesday11:34
@wikingi just want to have it liekt hat11:34
@wikingthat i dont need to touch it11:34
@HeikoSif I had a link, I could go to arthur and say: here it is done, lets change the course, that would be best11:34
@wikingif suddenly11:34
@wiking100+ people registers11:34
@wikingand start using it11:34
CaBahttps://gitlab.unique-internet.de/snippets/11 <- can i have a quick opinion on the idea of this clone hack? i guess it leaks mem, but i'm not quit sure how to go around that...11:34
@wikingi.e. the whole shit autoscales11:34
@HeikoSwiking: is there something I can show?11:35
@HeikoSthat would help immensely to push stuff here11:35
@wikingplease11:35
@wikingfind a docker image11:35
@wikingthat has something in it11:35
@wikingdoes't fucking matter what now11:35
@wikingthat would help11:35
@wikingor prepare one11:36
@wikingdoesn't even have to be shogun11:36
@wikinghttps://hub.docker.com/r/jupyter/singleuser/11:36
@wikingi'm currenlty starting this11:36
@wikingbut it has no notebook example in it11:36
@wikinghttps://hub.docker.com/r/jupyterhub/singleuser/11:37
@wikingthis one11:37
@wikingi mean11:37
@wikingthere's two options11:37
@wikingif you can show him something now in the next 2 hours11:37
@wikingor 511:37
@wikingthen i can get it back manually11:37
@wikingand then you demo it11:37
@HeikoSdoesnt need to be in next hours11:38
@HeikoSjust this week11:38
-!- goksinen [~goksinen@cpe-74-71-4-175.nyc.res.rr.com] has joined #shogun11:38
@wikingif not then i want to finsh this whole thing nicely11:38
@wikingso that really we can reproduce11:38
@wikingthe whole environment11:38
@wikingany time11:38
@wikinganywhere11:38
@wikingwithout11:38
@wikingdoing fucking devops11:38
@wikingfor days11:38
@HeikoSyes ok11:38
@wikinghttp://en.cppreference.com/w/cpp/types#Runtime_type_identification11:39
@wikingit'll be at cloud.shogun.ml11:39
@wikingand github auth11:39
@HeikoSwiking: cool, so let me reproduce the coursework stuff then11:39
@wikingwhat i'm gonna be working on later11:39
@wikingis to prepare11:39
@wikingthe old cloud image11:39
@wikingi.e. all the notebooks in it11:39
@wikingwith the data11:40
@wikingthat's another docker image11:40
@wikingbut the nice thing with this jupytherhub is11:40
@wikingthat we can have multiple instance of it11:40
@wikingso one for cloud.shogun.m11:40
@wiking.ml11:40
@wikingand one for11:40
@wiking<whatever course>shogun.ml11:40
@wiking:)11:40
@wikingand it runs different instances11:40
@wikingwith differnet content in it11:40
@wikingHeikoS, note that since this is jupyterhub11:41
@wikingwe could even prepare a cloud.shogun.ml image11:41
@wikingsuch11:41
@wikingthat we chuck into it all the fucking metaexamples as well11:41
@wikingin all the languages11:41
@wikingand since it's jupytherhub11:41
@wikingi mean jupyter11:42
@wikingthey can play with any of those languages11:42
@sukeyIssue #3645 "replace `sg_newserializable` with c++11 magic" karlnapf added label: "Cleanups" - https://github.com/shogun-toolbox/shogun/issues/364511:42
@sukeyIssue #3645 "replace `sg_newserializable` with c++11 magic" opened by karlnapf - https://github.com/shogun-toolbox/shogun/issues/364511:42
@sukeyIssue #3645 "replace `sg_newserializable` with c++11 magic" karlnapf added label: "entrance" - https://github.com/shogun-toolbox/shogun/issues/364511:42
@HeikoShttps://github.com/shogun-toolbox/shogun/issues/364511:42
@HeikoSput any comments you have in there, maybe somebody picks it up as a nice entrance task11:42
-!- goksinen [~goksinen@cpe-74-71-4-175.nyc.res.rr.com] has quit [Ping timeout: 240 seconds]11:42
@HeikoSwiking: yeah thats great with the courses in the link11:42
@HeikoSwiking: and love the idea on putting all the meta stuff in11:43
@HeikoSwiking: would be nice to maybe at some point develop some meta notebooks with plots11:43
@wikingyeah11:43
@wikingthat is a bit more shitty11:43
@HeikoSwiking: thats a nice one to get funding for in fact11:43
@wikingwell11:43
@wikingwe could11:43
@wikingreach out to jupyter people11:43
@wikingno?11:43
@HeikoSits not really part of shogun11:43
@HeikoSyeah11:43
@wikingshow them the meta stuff we have now11:43
@wikingshow them the ipython notebooks11:43
@wikingthat are actually nice imo11:44
@wikingand then tell that that we wanna merge the 211:44
@sukeyIssue #3645 "replace `sg_newserializable` with c++11 magic"- https://github.com/shogun-toolbox/shogun/issues/364511:44
@wikingmeta-notebooks11:44
@HeikoSwiking: yep11:44
@wikingdoes Lea care to write up such a letter11:44
@wiking?11:44
@wikingand then we just add the link to the cloud.shogun.ml11:44
@wikingand once everything is in place11:44
@wikingwe send it to jupyter11:44
@wikingmaybe internally they are having some effort for this11:45
@wikingwho knows11:45
@wikingwould be good to know11:45
@wikingi totally fucking accidentally saw this story with juptyerhub11:45
@HeikoSwiking: yes I had that in mind already11:45
@HeikoSit would be great to use that to get some funding11:46
@HeikoSits kinds of a self contained project11:46
@HeikoSand maybe juypter can support as well11:46
@wikingbtw i dont think yo urealise11:46
@wikingwhat sorts of shit11:46
@wikingis being done for the cloud  ;P11:46
@HeikoSand then we can pay someone in summer to do that or something11:46
@wikingi think i should demo it to you once11:46
@HeikoSwiking: I guess you are right11:47
@wikingit is very nice11:47
@wikingi learnt from our past mistakes11:47
@wikingwith using custom hacks11:47
@wikingnow it's all easily reproducable11:47
@wikingeasily maintainable11:48
@sukeyIssue #3645 "replace `sg_newserializable` with c++11 magic"- https://github.com/shogun-toolbox/shogun/issues/364511:48
@HeikoSwiking: we should have some video session or so and you walk me through it11:48
@HeikoSas for now, you are the only one who knows whats going on there11:48
@wikingmm it's all code11:48
@wikingand will be publicly available11:48
@wikingjust the custom proprietary settings11:49
@HeikoSI know, but the things you were refering to earlier11:49
@wikingare gonna be in shogun-infra11:49
@HeikoSyou ever saw that before? https://travis-ci.org/shogun-toolbox/shogun/jobs/203383958#L630811:49
@wikingnever11:50
@wikingbut11:50
@wikingtravis is acting weird lately11:50
@wikingsometimes i see clang jobs11:50
@wikingexit 011:50
@wikingbut still travis marks them as failed11:50
@HeikoSyep saw that11:50
@wikingmmm fuck11:51
@wikinglet's see if this works11:51
@wikingfingers fucking crossed11:51
@sukeyPull Request #3399 "LinalgRefactor - Add sum_symmetric and Rm old sum methods"  merged by karlnapf - https://github.com/shogun-toolbox/shogun/pull/339911:51
@sukeyNew Commit "Merge pull request #3399 from OXPHOS/linalg_sum_remove11:51
@sukeyLinalgRefactor - Add sum_symmetric and Rm old sum methods" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/53150efb68b7d04c56d733a03e4e393221d3342111:51
@HeikoSwhooo random 10s ubuntu freeze without any cpu load11:53
-!- travis-ci [~travis-ci@ec2-54-91-162-115.compute-1.amazonaws.com] has joined #shogun12:04
travis-ciit's Heiko Strathmann'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/20338395712:04
-!- travis-ci [~travis-ci@ec2-54-91-162-115.compute-1.amazonaws.com] has left #shogun []12:04
@wikingCaBa, lemme check12:07
CaBawiking: it's crap12:07
CaBawiking: there's no way i can use CKernel::clone() for the CKernel members only...12:08
CaBai think12:08
@wikingCaBa, i seriously think12:08
@wikingthat adding that patch for you12:08
@wikingin sgobject12:08
@wikingwould solve your problem faster12:08
@wiking:)))12:08
CaBawiking: what patch?12:09
@wikingfor clone12:09
@wikingget ride of sg_newserializable12:09
@wikingsee this bug12:10
@wiking https://github.com/shogun-toolbox/shogun/issues/364512:10
@HeikoSwiking: even easier is the pulling the parameter cloning out12:11
@HeikoSso people can just instantiate themselves and then call the method12:11
@wikingHeikoS, yeah or that12:11
@wikingeither is fine12:11
@wikingand then you dont need these hacky stsuff12:11
@wiking:)12:11
@HeikoSCaBa: so if you just cut out the loop over the parameters, and put that into a protected helper method and call it from clone12:12
@wikingit's really not more than a couple of line of diffs12:12
@HeikoSCaBa: then you can call that in your class12:12
CaBaHeikoS: ok, lemme see12:13
@wikingHeikoS, yeah and the inner part of the clone could be a big12:13
@wikingifdef have_cxx1112:13
@wikingelse12:13
@wikingand once we are c++11 then we can get ride of class_listcpp12:13
@wikingCaBa, if you do it12:14
@wikingthen could you do it12:14
@wikingin a way12:14
@wikingthat if c++11 available12:14
@wikingyou use the tag12:14
@wikingsystem12:14
@wiking:D12:14
@wikingHeikoS, ^ ?12:14
@HeikoSwiking: yes, but thats not for CaBa to do, as tags doesnt work yet12:14
@wikingok12:14
@HeikoSotherwise agreed12:15
@wikingthen CaBa forget it12:15
@wikingjust do that cut put12:15
CaBa^^12:15
@wikingand that's all12:15
@wikingmove the for loop into a protected function12:15
@HeikoSCaBa: I can do it for you12:15
@wikingand call that from the clone12:15
@HeikoSgive me a min12:15
@wikingor HeikoS will do it for ya12:15
CaBaHeikoS: thank you!12:15
CaBaso that clone_params() or however you call it will get the new ptr passed?12:17
@HeikoSbool CSGObject::clone_parameters(CSGObject* other)12:32
@HeikoSthats the signature12:32
lisitsynuhm I don't think https://github.com/shogun-toolbox/shogun/issues/364512:32
-!- goksinen [~goksinen@2604:2000:c591:8400:3420:31fb:5538:73fd] has joined #shogun12:32
lisitsynHeikoS: it would be replaced by plugins12:32
lisitsynand stuff12:32
@HeikoS/* Iteratively clones all parameters of the provided instance into this instance.12:32
@HeikoS * This will fail if the objects have different sets of registered parameters,12:32
@HeikoS * or if they have a different type as defined by get_name().12:32
@HeikoS *12:32
@HeikoS * @param other other object whose parameters are to be cloned12:32
@HeikoS * @return True if cloning was successful12:32
@HeikoS */12:32
@HeikoSlisitsyn: what do you mean?12:33
lisitsynHeikoS: I just mean it is not an entrance task12:33
@HeikoSwhy not?12:33
@HeikoSbefore tags is used12:33
lisitsynwell defines a lot12:33
lisitsynlike architecture of shogun12:33
@HeikoSno it is just about creating empty instances12:34
lisitsynI am terrible but it is a task for me12:34
lisitsyn:D12:34
@HeikoSand the current way only allows this for class_list ojects12:34
@HeikoSlisitsyn: yeah do it :)12:34
lisitsynyeah ok :)12:35
-!- goksinen [~goksinen@2604:2000:c591:8400:3420:31fb:5538:73fd] has quit [Ping timeout: 240 seconds]12:37
@wikingyeah do it12:39
@wiking:)12:39
@wikingbecause CaBa needs it12:39
@wikingand i need it12:39
@wiking:)12:39
@wiking<312:39
@HeikoSyou know whats shiat12:40
@HeikoSthat I have to recompile all of shogun whenever I change docs of SGObject.h12:40
@HeikoSand it takes time12:40
@HeikoSand yes the ccache is on12:40
lisitsynuhm that's not going to be fixed12:41
CaBaHeikoS: do you push that clone_parameter() directly to develop?12:44
@HeikoSCaBa: I might12:44
@HeikoSCaBa: but maybe prefer a PR12:44
@HeikoSsince travis can have a look then12:44
@HeikoSso CaBa12:45
@HeikoSonce it is merged, you can just create an empty instance of your kernel12:45
@HeikoSand then call the method providing another one12:46
CaBacool12:48
@wikingHeikoS, yeah sgobject touches everything12:54
@wiking;)12:54
-!- travis-ci [~travis-ci@ec2-54-90-131-18.compute-1.amazonaws.com] has joined #shogun12:54
travis-ciit's Heiko Strathmann'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/20339483612:54
-!- travis-ci [~travis-ci@ec2-54-90-131-18.compute-1.amazonaws.com] has left #shogun []12:54
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  opened by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364612:55
@HeikoSCaBa: https://github.com/shogun-toolbox/shogun/pull/364612:55
@HeikoSlets see what travis says12:55
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364612:57
@HeikoSgnaa12:58
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364612:58
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364613:00
@HeikoSok lunch time13:00
CaBatjanks13:06
-!- CaBa [~Diu7saig@unaffiliated/caba] has left #shogun []13:09
-!- CaBa [~Diu7saig@unaffiliated/caba] has joined #shogun13:09
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364613:14
-!- goksinen [~goksinen@2604:2000:c591:8400:3420:31fb:5538:73fd] has joined #shogun13:27
-!- goksinen [~goksinen@2604:2000:c591:8400:3420:31fb:5538:73fd] has quit [Ping timeout: 240 seconds]13:31
CaBaHeikoS: seems broken unfortunately14:25
CaBajust a sec14:27
CaBamight be me... my feature class is also self baked :P14:30
@wikingHeikoS, here?15:15
@wikingso here's our infra for cloud15:19
@wikinglisitsyn, HeikoS https://github.com/shogun-toolbox/shogun-cloud/tree/master/backend/aws15:19
@wiking:)15:19
-!- goksinen [~goksinen@2604:2000:f64b:c700:3067:a1ec:ec57:e5c6] has joined #shogun15:45
-!- goksinen [~goksinen@2604:2000:f64b:c700:3067:a1ec:ec57:e5c6] has quit [Ping timeout: 240 seconds]15:49
@HeikoSwiking: now15:50
@HeikoSCaBa: no it is broken, fixing15:50
@wikingHeikoS, there's the cloud :)15:51
@HeikoSwiking: nice15:51
@HeikoSso I just talked to Arthur15:51
@HeikoSwe will use it15:51
@HeikoSin march already15:51
@HeikoSwiking: i need to actually check out what you did there15:52
@HeikoSthis is some framework import15:52
@wikingyes15:52
CaBaHeikoS: yeah i also ran into more trouble with it...15:56
CaBasome memory access... didnt really figure out where it comes from though15:56
@HeikoSCaBa: yeah I got the order of the copy wrong15:59
CaBaHeikoS: ah :)15:59
@HeikoSso copying uninitialised memory into the one that I wanted rather than the other way around15:59
CaBaHeikoS: i also commented some minor stuff on the diff16:00
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364616:01
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 240 seconds]16:06
@wikinggetting there https://hub.docker.com/r/shogun/jupyterhub/16:12
@wiking:)16:12
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364616:17
-!- HeikoS [~heiko@eduroam-int-pat-8-27.ucl.ac.uk] has joined #shogun16:18
-!- mode/#shogun [+o HeikoS] by ChanServ16:18
CaBaHeikoS: looks better now16:18
@HeikoSyeah it might work now16:18
@HeikoSwill still wait for travis ok?16:19
CaBaHeikoS: sure, i cherry picked to my own temp branch until then16:20
CaBawhich is based on develop from last october *duck* :P16:20
@HeikoSlol16:21
@HeikoSwhere are you based btw?16:21
CaBaHeikoS: berlin16:22
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  synchronized by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364616:23
@HeikoSwe should oeganise another workshop there16:24
@HeikoSdid we ever meet?16:24
@HeikoSwiking, lisitsyn we need to have some kind of systematic testing for our REQUIRE etc16:25
@HeikoSI bet there is lots of null pointer de-referencing, and other bad stuff16:25
@HeikoSits currently never tested that methods throw exceptions on certain inputs16:25
CaBaHeikoS: nope, at least not that i'm aware of :) another student from my grad school visited one of your workshops though iirc, matt16:26
@HeikoSah yeah16:26
@HeikoShe fixed R back then16:26
CaBayep16:27
@HeikoShail the berlin hackers ;)16:27
CaBa;)16:29
CaBahmhm... now i'm stuck with mem errors again. but this time it seems related to the parallel x-val code i took back in... in combination with my own kernel class16:29
@HeikoSthe parallel xval can be re-introduced I think16:30
@HeikoSas wiking fixed stuff that was causing problems16:30
@wikingyeah hopefully16:31
@wikingtry it16:31
CaBai'm trying it and i get bad_access here [1] when it reaches my own kernel...https://github.com/lkuchenb/shogun/blob/production/src/shogun/kernel/CombinedKernel.cpp#L24016:31
CaBai'm trying it and i get bad_access here [1] when it reaches my own kernel...16:31
CaBa[1] https://github.com/lkuchenb/shogun/blob/production/src/shogun/kernel/CombinedKernel.cpp#L24016:31
CaBaapparently the k pointer is no longer valid16:31
CaBa*sigh*16:32
CaBai think i have tortured shogun with too much hackery O.O16:32
@HeikoShehe ;)16:34
-!- goksinen [~goksinen@2604:2000:f64b:c700:3067:a1ec:ec57:e5c6] has joined #shogun16:39
-!- goksinen [~goksinen@2604:2000:f64b:c700:3067:a1ec:ec57:e5c6] has quit [Ping timeout: 240 seconds]16:43
-!- HeikoS [~heiko@eduroam-int-pat-8-27.ucl.ac.uk] has quit [Quit: Leaving.]16:57
-!- OXPHOS [401e476d@gateway/web/freenode/ip.64.30.71.109] has joined #shogun17:00
-!- goksinen [~goksinen@2604:2000:f64b:c700:3067:a1ec:ec57:e5c6] has joined #shogun17:02
-!- lambday [31cf349d@gateway/web/freenode/ip.49.207.52.157] has joined #shogun17:06
-!- mode/#shogun [+o lambday] by ChanServ17:06
-!- goksinen [~goksinen@2604:2000:f64b:c700:3067:a1ec:ec57:e5c6] has quit [Ping timeout: 240 seconds]17:06
CaBawiking: i guess i should move forward from my october 2016 shogun if i want to use the parallel x-val code? i'm running into mem issues, but those were fixed in the backend in the meantime you said?17:14
CaBahm, i'm even in december with my codebase. so maybe those are already included and not the cause :(17:16
@wikingmmm17:24
@wikingwhat do you mean meamleaks?17:24
@wikingand yeah there are may shit fixed17:24
@wikingin the last 1 month17:24
@wikingso rebase with upstream/develop is encouraged17:25
@wikingif u can17:25
CaBai'll work my way through the rebase first and check if it's still a problem17:42
@sukeyPull Request #3646 "pull cloning all parameterers out to a seperate method"  merged by karlnapf - https://github.com/shogun-toolbox/shogun/pull/364617:44
@sukeyNew Commit "Merge pull request #3646 from karlnapf/feature/clone_refactor17:44
@sukeypull cloning all parameterers out to a seperate method" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/ce652852c504edfad450a68e8b664702fcbd0b9c17:44
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun17:45
-!- mode/#shogun [+o HeikoS] by ChanServ17:45
@HeikoSCaBa: merged it17:45
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun17:48
-!- travis-ci [~travis-ci@ec2-54-90-131-18.compute-1.amazonaws.com] has joined #shogun18:40
travis-ciit's Heiko Strathmann'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/20351008818:40
-!- travis-ci [~travis-ci@ec2-54-90-131-18.compute-1.amazonaws.com] has left #shogun []18:40
CaBaHeikoS: neat19:10
CaBaHeikoS: shogun won't build, at least for me...19:10
CaBaIn file included from /Users/lk/RESEARCH/external/shogun/src/shogun/statistical_testing/MultiKernelQuadraticTimeMMD.cpp:41:0:19:11
CaBa/Users/lk/RESEARCH/external/shogun/src/shogun/statistical_testing/internals/mmd/ComputeMMD.h: In member function 'shogun::SGVector<double> shogun::internal::mmd::ComputeMMD::operator()(const shogun::internal::KernelManager&) const':19:11
CaBa/Users/lk/RESEARCH/external/shogun/src/shogun/statistical_testing/internals/mmd/ComputeMMD.h:132:10: error: ambiguous overload for 'operator[]' (operand types are 'shogun::SGVector<double>' and 'std::size_t {aka long unsigned int}') result[k]=compute(terms[k]);19:11
CaBadoes that ring a bell to either of you?19:11
CaBawiking: anything special, mac-wise lately?19:12
CaBabuilding with GCC 6.319:12
-!- lambday [31cf349d@gateway/web/freenode/ip.49.207.52.157] has quit [Ping timeout: 260 seconds]19:26
@HeikoSCaBa: thats a recent feature branch19:34
@HeikoSthat has been merged19:34
@HeikoSwe need to fix that19:34
@HeikoScan you post the error on an issue and poke @lamday19:34
@HeikoShe is fixing things for that atm19:35
-!- OXPHOS [401e476d@gateway/web/freenode/ip.64.30.71.109] has quit [Ping timeout: 260 seconds]19:45
CaBaHeikoS: what branch was that?19:53
@sukeyIssue #3647 "Build fails" opened by lkuchenb - https://github.com/shogun-toolbox/shogun/issues/364719:55
@sukeyIssue #3647 "Build fails"- https://github.com/shogun-toolbox/shogun/issues/364719:56
CaBaHeikoS: so how far do i have to go back to get a build through?19:57
@HeikoSbefore sunday20:01
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 260 seconds]20:58
CaBawiking: ping21:16
CaBawiking: do you build against colpack on mac os>?21:16
CaBawiking: you have killed my 'ENABLE_COLPACK' :P21:18
CaBawiking: actually it's protbuf that is causing my trouble. anyways - have all the options disappeard?21:30
CaBa*dissappeared?21:30
CaBa.oO(i'll give up spelling that correctly)21:30
--- Log closed Tue Feb 21 00:00:34 2017