IRC logs of #shogun for Wednesday, 2016-09-28

« Tuesday, 2016-09-27 Index Thursday, 2016-09-29 »
--- Log opened Wed Sep 28 00:00:09 2016
CaBalisitsyn: :)00:36
-!- witness_ [uid10044@gateway/web/irccloud.com/x-twrhktfvvndhjwpb] has joined #shogun01:17
-!- travis-ci [~travis-ci@ec2-54-158-182-34.compute-1.amazonaws.com] has joined #shogun02:14
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/16326936502:15
-!- travis-ci [~travis-ci@ec2-54-158-182-34.compute-1.amazonaws.com] has left #shogun []02:15
-!- witness_ [uid10044@gateway/web/irccloud.com/x-twrhktfvvndhjwpb] has quit [Quit: Connection closed for inactivity]03:20
-!- dennyjooo [~dennyjooo@60.170.224.6] has joined #shogun05:10
-!- dennyjooo [~dennyjooo@60.170.224.6] has quit []05:16
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun14:28
-!- mode/#shogun [+o HeikoS] by ChanServ14:28
@wikingHeikoS: man14:38
@wikingHeikoS: you have never answered my question14:38
@wikingregarding the invoice14:38
@HeikoSwiking: jojo14:39
@wikingand we should send that asap14:39
@HeikoSsure14:39
@HeikoSremind me :)14:39
@wikingdoing it now :)14:39
@HeikoSI looked at it and it looked fine14:39
@HeikoSwhat was the question14:39
@wikingok had some questions there14:39
@wikingdont remember what :D14:39
@HeikoShaha :D14:39
@wikingbut it's in the email i've sent to u14:39
@HeikoSchecking14:40
@HeikoSwiking: cant find any email with a question14:41
@HeikoSapart from "need to book hotel" ;)14:41
@wikingahaha14:41
@wikingok lemme check14:41
@HeikoSah here we go14:41
@wikingoh ok14:41
@HeikoSI replied to one14:42
@HeikoSjust use fernandos name14:42
@HeikoSthat is fine14:42
@wikingok14:42
@wikingthen i'm gonna fill out the spreadsheet -> pdf14:42
@wikingand upload14:42
@HeikoSyep good!14:43
@HeikoSthey will complain if something is not good14:43
@HeikoS(that might have happened before ;)14:43
@wikingok done14:48
@wikingsubmitted14:48
@wikingHeikoS: ok so i started to put together a patch for big input14:52
@HeikoSwiking: nice one thanks for taking care14:52
@HeikoSwiking: big input as in huge sparse?14:52
@wikingman we are mixing int64_t and int32_t all the time14:52
@HeikoSwiking: I know14:52
@HeikoSI think index_t should be 64bit14:53
@wikingso yeah i'm now switching things to index_t14:53
@wikingand then i'm compiling with -Woverflow14:53
@HeikoSwiking: needs some systematic refactoring14:53
@HeikoSah nice14:53
@wikingand basically going one by one fixing14:53
@HeikoSwiking: ok cool14:54
@HeikoSdid this fix this guys problem?14:54
@wikingnot yeti mean i'm far from finishing that patch14:54
@wiking:))))14:54
@wikingi almost have to change every fucking class14:54
@wikingin the code :)14:54
@HeikoShaha14:55
@wikinganyhow14:55
@HeikoSI guess this can be done incrementally?14:55
@HeikoSor does it break anything?14:55
@wikingi cannot get it compiled14:55
@wiking:)))14:55
@wikingbecause you need to change the interface etc14:55
@wikingso yeah it's gonna be a per directory patch14:55
@wiking:)14:55
@wikingbut that's all14:56
@wikingbut i leave it in the feature/BigInput branch14:56
@HeikoSkk14:56
@HeikoSok we can merge that in SF14:56
@wikinghave it tested by him once it's ready14:56
@wikingetc14:56
@HeikoScool14:56
@wikingbecause i dont have a machine where14:56
@HeikoSI am on some meta language stuff atm14:56
@wikingi could test this14:56
@HeikoSto merge a few more cookbooks14:56
@HeikoSbtw what about java script?14:56
@HeikoSshall I get these cookbooks working?14:56
@wikingah js14:56
@wikingyeah there i'm working14:56
@wikingon typemapping14:56
@HeikoSbecause the patches that are merged in that feature branch are not really dealing with the full thing14:57
@wikingat least for v8+nodejs14:57
@wikingyeah not yet14:57
@HeikoSlet me know when I shall do that, then we get a lot of tests for free14:57
@wikingyep14:57
@wikingbtw14:57
@wikingso i was checking the14:57
@wikinghttps://www.tensorflow.org/versions/r0.10/how_tos/distributed/index.html14:57
@wikingand it's super heavy weight14:57
@HeikoSif I managed to figure out some generic template stuff, we also get implicit tests for the typemaps14:57
@wikingi would rather use this https://actor-framework.org/14:57
@HeikoSthere was some weird stuff going on in java14:57
@HeikoSyeah distributed14:58
@HeikoSI am not too interested in that for now14:58
@HeikoSso I leave it to you :)14:58
@HeikoSI care more about the autodiff14:58
@HeikoSand having an API to build computation graphs14:58
@HeikoSbut would be good14:58
@HeikoSto get some multicore stuff easy at least14:58
@wikingyeah but caf is the same as computation graph14:58
@HeikoSsure14:58
@wikinganyhow14:59
@wikingi'll see when i have time14:59
@wikingwanna fix first the bug in crossval multithreading14:59
@wiking(dunno yet wtf is going on)14:59
@HeikoSah yeah14:59
@wikingso we can merge the windows branch14:59
@wikingthen this big input14:59
@HeikoSyes that'd be huge14:59
@wikingand then javascript14:59
@wikingi mean this is the only fucking thing14:59
@wikingthat crossval multithread failing14:59
@wiking:S14:59
@HeikoSI want to fix things to merge OXPOHS patches so that she remains interested in shogun and comes baack!15:00
@wikinghahaha yeah15:00
@wikingthe serialization15:00
@HeikoSwiking: thats probably related to Saurabh7's fix15:00
@wikingand the linalg15:00
@HeikoSyeah15:00
@HeikoSIll merge things bit by bit for this stuff15:00
@wikingyeah those things should get in line soon15:00
@HeikoSyep15:00
@wikingso we can merge those into develop15:00
@HeikoSI have a bit of time before SG15:00
@HeikoSSF15:00
@HeikoSso can prepare15:00
@wikingbtw we are no aiming 5.0.0 right?15:00
@HeikoSyes15:02
@HeikoSdefinitely!15:02
@HeikoSI made a milestone in fact15:02
@HeikoS;)15:02
@HeikoSwith one issue15:02
@wiking:)15:03
@HeikoSwiking: we should also migrate the parameters15:03
@HeikoSand the cereal stuff15:03
@HeikoSmaybe even before 5.015:03
@HeikoSI also want to clean up the test directory a bit15:03
@HeikoSget rid of all but the meta ones, any thoughts on that?15:04
@HeikoSremove static interfaces as well15:04
@HeikoSlots of old code, lots of questions on that often15:04
@HeikoSremove "_modular" and just support swig interfaces15:04
@wikingyep15:04
@HeikoScleaning up15:04
@HeikoSyou know the website status?15:05
@HeikoScan I edit it already on github? :)15:05
@wikingyep15:05
@wikingi mean static dropping woudl be great15:05
@wikingedit what?15:05
@HeikoSthe text15:06
@HeikoScontent15:06
@wikingah15:06
@wikingno auto delivery yet15:06
@wikingbut that shoudl be easy15:06
@wikingquestion is15:06
@HeikoSI want to show it to some people and then edit based on their suggestions15:06
@wikinglisitsyn: wtf do we miss?15:06
@wikingapart from autodeploy?15:06
@wikingHeikoS: as you can see this works15:07
@wikingbeta.shogun.ml15:07
@HeikoSah cool15:07
@wikingbut if you change content15:07
@HeikoSand deplyong?15:07
@wikingon github15:08
@wikingwill not get deployed15:08
@HeikoSkk15:08
@HeikoSis there a preview or so?15:08
@wikingcan look into that15:08
@HeikoScool15:08
@wikingjust wanna see what lisitsyn15:08
@wikingsays15:08
@HeikoSyeah ok15:08
@HeikoSand the master branch release mechanism?15:08
@wikingyep15:08
@HeikoShaha i like your tweet ! :)15:09
@wikingah that was in august15:09
@wikingwas a productive shogun month: )15:09
@HeikoSpitty I was gone15:09
@HeikoSrahul also has a massive feature branch that is ready for mergining15:10
@HeikoSmostly science code15:10
@HeikoSbut might get us a few citations15:10
@HeikoSsince its cool stuff15:10
@HeikoSand fast as fuck15:10
@HeikoSjust need to push him to document the API15:10
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 264 seconds]17:00
-!- HeikoS [~heiko@host31-53-162-24.range31-53.btcentralplus.com] has joined #shogun17:15
-!- mode/#shogun [+o HeikoS] by ChanServ17:15
@HeikoSwiking: still awake?17:26
@HeikoSwiking: seems like we dont have a typemap in csharp for bool[] and BoolVector17:27
-!- mizari [~mizari@95-174-213-100.nts.su] has joined #shogun17:44
-!- OXPHOS [92bd305b@gateway/web/freenode/ip.146.189.48.91] has joined #shogun17:49
@HeikoSOXPHOS: hihi17:50
@HeikoShow are things?17:50
@HeikoSOXPHOS: looking into this problem with the bool vectors, it is more tricky than I thought17:51
@HeikoSbut I might have a hack17:51
OXPHOS@HeikoS: hey hello! back to work17:51
@HeikoSOXPHOS: nice, all good over there?17:51
OXPHOS@HeikoS: I might disappear once in a while but keep going17:52
@HeikoSOXPHOS: yeah no worries17:52
OXPHOSyes just kinda busy in the mornings [now], so it's hard to reach to you guys17:52
@HeikoSOXPHOS: whats the best time to talk for you?17:52
@HeikoSin a few hours?17:52
@HeikoSI am not that far off from your timezone17:53
@HeikoSso my evenings (in a few hours) should work17:53
@HeikoSOXPHOS: Ill try to send a fix for the problem soon17:53
@HeikoSOXPHOS: shall we aim to get the linalg in as well over the next few weeks?17:53
@HeikoSOXPHOS: we are preparing a release and that should definitely be a feature of it17:53
OXPHOSso there is something wrong with the java vector?17:55
@HeikoSOXPHOS: there are two problems17:55
@HeikoSOXPHOS: and the java problem ;)17:56
OXPHOSusually 10 min from now till 2h10 min from now are better, 12pm - 2pm17:56
@HeikoSthe java problem I fixed17:56
@HeikoSOXPHOS: cool Ill try to catch you t hen17:56
OXPHOSthx17:56
@HeikoSOXPHOS: see my reply and listing how the caid_tree works for me17:56
@HeikoSthen there are problems with csharp17:56
@HeikoSit is quite nice how the meta example coverage reveals all these problems inside shogun, but it is also a bit annoying that there are so many ;)17:56
OXPHOShaha true17:57
OXPHOSI am checking17:57
OXPHOSand I don't know why meta examples start failing for me after the short break of last month17:58
OXPHOSit seems like when it runs into unknown symbols, but the symbols are just numbers this time17:58
OXPHOSand i updated ply, which didn't help17:59
@HeikoSremove the build dir17:59
@HeikoSthat should solve it18:00
OXPHOSI did18:00
@HeikoSand still?18:00
OXPHOSyes still the same problem18:00
@HeikoSI mean the build works18:00
@HeikoSdid you change your computer?18:00
@HeikoSre-run cmake?18:00
OXPHOSno i didn't18:00
OXPHOSbut I can try a different computer18:00
@HeikoSthen it doesnt make sense, so I guess it is a sneaky little setup issue18:01
@HeikoSyeah try that18:01
@HeikoSworks on mine18:01
OXPHOSI removed the build and repeated cmake18:01
@HeikoSOXPHOS: do you remember which meta example had a static call in it?18:01
OXPHOSi ll try a hard reset i guess some time18:02
OXPHOSno18:02
@HeikoShard reset?18:03
OXPHOSyeah just reset --hard upstream/develop and see what will happen18:03
@HeikoSkk18:04
@HeikoSOXPHOS:  I am trying to solve your vector problem like this now:18:05
@HeikoSBoolVector ft = BoolVector:const_vector(0, 2)18:05
@HeikoSwhere I added this method to SGVector18:05
@HeikoSwe need the typemap to do hand constructed bool[] ones18:05
@HeikoSsee my issue18:05
@HeikoSwiking: ^18:05
OXPHOS@HeikoS: not really..but seems like a big change18:07
@HeikoSargh there are other issues with this solution as well18:08
@HeikoSsigh18:08
@HeikoSso annoying that this mini thing blocks so much stuff18:08
OXPHOSand also ture for the linalg - you want me to move the codes to .cpp? no more header only? This is the only way I can see with the Macros18:09
@HeikoSOXPHOS: what do you mean by "no more header only18:11
@HeikoSsorry I dont have this on my radar right now, can you explain again?18:11
OXPHOS@HeikoS: so to apply linalg library shogun-wisely, I can include it in CMath.h, which is included by a lot of algos18:18
@HeikoSno we dont want that18:18
@HeikoSI want to include some linalg.h in the algorithm itself18:19
OXPHOSokay18:19
OXPHOSbut if we include linalg.h in too many files, the compiling is still going to be slow, because for every single file included linalg.h, it needs to go through all the types specified by macros18:20
@HeikoS"it needs to go" means what here?18:20
@HeikoSdont we guard against multiple including?18:21
@HeikoSnot sure I understand the proiblem18:21
OXPHOSyeah that's the thing. So just assume we include linalg.h in cmath.h, so linalg.h should only be included once right, because of the guard18:24
OXPHOSbut actually, if there's warnings jump out of "DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_MAX, SGVector)", you can see the warnings poped out for every single file that included cmath.h18:24
@HeikoSthats not good18:27
@HeikoSmmh are all the linalg header guarded against multiple inclusion?18:27
@HeikoS#ifndef LINALG_H__18:28
@HeikoS#define LINALG_H__18:28
@HeikoS....18:28
@HeikoS#endif18:28
@HeikoS?18:28
@HeikoSI am offline for 20 now, let's discuss when I am back18:28
OXPHOSkk18:30
-!- HeikoS [~heiko@host31-53-162-24.range31-53.btcentralplus.com] has quit [Ping timeout: 276 seconds]18:33
@wiking?18:41
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun18:49
-!- mode/#shogun [+o HeikoS] by ChanServ18:49
OXPHOScheck19:03
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 272 seconds]19:04
-!- mizari [~mizari@95-174-213-100.nts.su] has quit [Ping timeout: 264 seconds]19:29
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has joined #shogun19:50
-!- mode/#shogun [+o HeikoS] by ChanServ19:50
@HeikoSOXPHOS: still around=19:51
@HeikoS?19:51
OXPHOS@HeikoS: yes19:55
@HeikoSOXPHOS: so is the linalg stuff guarded in the way I mentioned ?19:55
OXPHOSthe LinalgNameSpace.h is actually missing the guards but unlikely it is causing the trouble because it's more downstream. But I am going to fix it and try19:56
@HeikoSI just wonder why you will get thing multiple times -- that is not the way the c++ preprocessor works19:56
OXPHOSI asked a friend a while ago and he told me the macros in .h files will be re-built everytime it is included19:57
@HeikoSah really19:57
@HeikoSmmh19:57
@HeikoSI have a hard time imagining that19:57
@HeikoSif you guard it, then it should remain hidden19:58
OXPHOSI tried to google but maybe didn't hit the appropriate keywords for searching..so didn't find any useful info about it20:01
-!- HeikoS [~heiko@host-92-0-178-129.as43234.net] has quit [Ping timeout: 276 seconds]20:03
-!- OXPHOS [92bd305b@gateway/web/freenode/ip.146.189.48.91] has quit [Quit: Page closed]22:33
--- Log closed Thu Sep 29 00:00:10 2016
« Tuesday, 2016-09-27 Index Thursday, 2016-09-29 »

Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!