IRC logs of #shogun for Monday, 2016-07-18

« Sunday, 2016-07-17 Index Tuesday, 2016-07-19 »
--- Log opened Mon Jul 18 00:00:27 2016
-!- OXPHOS [c68f0c0c@gateway/web/freenode/ip.198.143.12.12] has joined #shogun04:09
-!- OXPHOS [c68f0c0c@gateway/web/freenode/ip.198.143.12.12] has quit [Ping timeout: 250 seconds]04:14
-!- OXPHOS [c68f0c0c@gateway/web/freenode/ip.198.143.12.12] has joined #shogun04:49
-!- sanuj [~sanuj@117.220.53.59] has joined #shogun06:20
@wikingOXPHOS: how's the serialization going?06:26
@wikingOXPHOS: sanuj weekly update on the mailing list plz06:27
@wikingSaurabh7:06:27
@wiking^06:27
OXPHOSwiking: promising...06:29
OXPHOSclose..06:29
@wikingcan u push something somewhere06:33
@wikingjust to see what's happening?06:33
@wikinganything is fine06:33
OXPHOSwiking: I can pull it now. compiling looks fine but I didn't test it yet06:43
-!- sanuj [~sanuj@117.220.53.59] has quit [Read error: Connection reset by peer]06:43
@wikingkk06:48
@wikingjust got the mail06:48
@wikinglemme see06:48
@wikingmmm why to develop?06:49
@wikingand not the feature branch? :)06:49
OXPHOSwiking: mistake...06:49
@wikingclose and do it for the feature branch plz06:50
@wikingthnx06:50
OXPHOSwiking: kk06:50
@wikingthnx06:50
-!- sanuj [~sanuj@59.97.245.111] has joined #shogun06:59
-!- sanuj [~sanuj@59.97.245.111] has quit [Quit: Leaving]07:27
-!- HeikoS [~heiko@217.138.5.14] has joined #shogun11:58
-!- mode/#shogun [+o HeikoS] by ChanServ11:58
-!- c4goldsw [82e2e6c8@gateway/web/cgi-irc/kiwiirc.com/ip.130.226.230.200] has joined #shogun12:10
@HeikoSc4goldsw: jo!12:10
c4goldswping HeikoS wiking hey, how are you?12:10
@HeikoSsaw the recent linalg patches?12:10
@HeikoSc4goldsw: we can start using it more globally soon12:10
c4goldswNo, not yet.  Got back from the states two days ago.  Good.12:10
c4goldswHad no internet for two weeks.12:10
c4goldswHeikoS any other developments?12:12
@HeikoSc4goldsw: we did the major part of the refactoring12:12
@HeikoSwhat we need now is some cmake changes12:12
@HeikoSand then porting the existing methods12:12
c4goldswporting to?12:12
@HeikoSother changes: tags was merged12:12
@HeikoSgsoc is going slowly12:12
@HeikoSthe new linalg12:13
@HeikoScheck the feature branch12:13
@HeikoSfeature/linalg_refactor12:13
c4goldswchecking12:13
@HeikoSit is now all runtime based12:13
@HeikoSno more compile time things12:13
@HeikoSand easier syntaxc12:13
c4goldswSo, the linalg backend (e.g. eigen / LAPACL / ViennaCL) is selected at run time?12:14
@HeikoSyes12:14
@HeikoSand we have one for cpu and one for gpu12:14
@HeikoSand then there is an explicit transfer of vector to gpu12:15
@HeikoSunit tests illustrate12:15
c4goldswI'll look at that then.  Brb in 20 minutes, eating lunch.12:16
c4goldswafk12:16
@wikingHeikoS: wazza12:21
@HeikoSwiking: jooooooooo12:21
@HeikoShow are things?12:21
@wikingmmm slow12:21
@wiking:)12:21
@HeikoSwiking:  yeah12:21
@HeikoSGSoC wise:12:21
@HeikoSlisitsyn is not there, so we should help sanuj a bit more12:22
@HeikoSespecially with plugins12:22
@HeikoSgive hints, review patches12:22
@wikingk12:22
@wikingyeah i saw that he is gone for a week now12:22
@HeikoSwiking: then there is OXPHOS cereal, this really needs some drafting from our side12:22
@HeikoSotherwise its gonna be too slow12:22
@wikingmmm it's going actually12:22
@wikingshe already has a PR12:22
@wikingfor tags12:22
@HeikoSyeah I know, just wanna make more of the time12:22
@HeikoSnot saying its not good12:22
@HeikoSjust wanna make sure it works in a few weeks ;)12:22
@wikingyeah12:23
@HeikoSwiking: then we can also remove the cmake flags for linalg in feature branch12:23
@HeikoSand then merge that thing soon12:23
@wikingyeah12:23
@wikingi mean now that thing is in cereal feature branch12:23
@wikingas soon as we have that fixed12:23
@wikingwe should merge first that12:23
@wikingimo12:23
@HeikoSyep12:23
@HeikoSand then remove Parameter.g12:23
@HeikoS.h12:23
@HeikoSwhich will break all old serialization12:24
@HeikoSmaybe we can keep the old until the new one works12:24
@HeikoSbut there are some dependencies12:24
@HeikoSlike SG_ADD12:24
@HeikoSreviewing some PRs now12:24
@HeikoSwiking: why is there an explicit save thing for vector?12:25
@HeikoSdo we have to implement save for all vector/matrix and base types?12:26
@HeikoSand then iterate over parameter map and call those?12:26
@HeikoSor how does it work?12:26
@HeikoSOXPHOS: ^12:26
@wikingHeikoS: yeah12:26
@wikingHeikoS: because vector and matrix12:26
@wikingis not SGObject12:26
@HeikoSok12:26
OXPHOSSGObject and SGObject independent stuff?12:26
@HeikoSso the way we do it:12:27
@HeikoS1.) we iterate through the parameter map12:27
OXPHOSI can update the PR now. I think it's something I'm not familiar with about map iteration12:27
OXPHOSgiving me weird result12:27
@HeikoSand do case distinction. a) base type b) SG* type c) sgobject12:27
@HeikoSwiking, OXPHOS can I ask about this:https://github.com/shogun-toolbox/shogun/pull/3375/files#diff-8ea96286d95b52029d31636117e0fe55R22512:27
@HeikoSthis means we expicitly need to switch over all combinations of SG* and base type?12:28
@HeikoSseems a bit crazy since we have 20 base types for each vector and matrix12:28
OXPHOSyes12:28
OXPHOSor like parameter, we use several lables？12:28
@HeikoSthe currently way in shogun12:29
@HeikoSdoes that in a nested way12:29
@HeikoSso first checks base/SG*/SGObject12:29
@HeikoSand then inside there does more checks12:29
@HeikoSbut it is ok as well12:29
@HeikoSOXPHOS: so for now you have: matrix and vector12:30
@HeikoSwe also need some recursion for SGOBject12:30
@HeikoSbut can we maybe start with vector only?12:30
@HeikoSthat is, serialization of a vector member works?12:30
@HeikoSif that works, we can add matrix12:30
@HeikoSand then sgobject12:30
OXPHOSvector works12:31
OXPHOSis there any scenario we need to serialize a SGObject as a whole?12:31
@HeikoSyeah of course12:31
@wikingyeah12:31
@HeikoSOXPHOS: e.g. SVM has a kernel12:31
@HeikoSSVM.save()12:31
@HeikoSshould call kernel.save() inside12:31
@wikingOXPHOS: it's recursive12:32
@HeikoSwiking: do we really need this enum stuff?12:32
@HeikoSwiking: type2enum?12:32
@wikingmmm12:32
OXPHOSThe problem I think is12:33
@HeikoShttps://github.com/shogun-toolbox/shogun/pull/3375/files#diff-8ea96286d95b52029d31636117e0fe55R27312:33
@HeikoSthis list will be long12:33
@wikingyeah i know which one12:33
@HeikoScurrently we have this nested12:33
@wikingthe thing is that we need to know the type12:33
@wikingright?12:33
@HeikoSi.e. we have one enum for vector/matrix12:33
@HeikoSyes we do12:33
@wikingi mean we have this for features12:33
@HeikoSTYPE_SGVECTOR_INT, TYPE_SGVECTOR_FLOAT, TYPE_SGVECTOR_SHORT, TYPE_SGVECTOR_ ....12:34
@HeikoSthis seems weitd12:34
OXPHOSokay I'm not sure how cereal handles this but I guess fine. just archive(kernel) in SVM.save() and it will call kernel.save()12:34
@HeikoSweird12:34
@HeikoScant we have12:34
@HeikoSEnumSGType { SGVECTOR, SGMATRIX }12:34
@HeikoSand then EnumBaseType {INT, FLOAT, DOUBLE }12:34
@HeikoSand then combine them?12:35
@HeikoSalso need to think about SGObject as member12:35
@HeikoSor even worse:12:35
@HeikoSwhat if a member if DynamicObjectArray12:35
@HeikoSthis has an array of CSGOBject12:35
@HeikoSOXPHOS: yeah that works12:36
@HeikoSOXPHOS: lets just continue to develop12:36
@wikingHeikoS: yeah we can combine of course12:36
@wikingif they are proper bits that we could &12:36
@wiking:)12:36
@HeikoSto messy ;)12:36
@wikingand then define a flag12:36
@wikingno not really12:36
@wikingstandard thing12:36
@HeikoSyeah12:37
@HeikoSmmh12:37
OXPHOSEverything recompiles even if I just changes a tiny bit stuff in SGObject or Any12:37
@wikingINT 0x112:37
@HeikoSis that necessary?12:37
@wikingetc12:37
OXPHOSit's extremely slooooooow12:37
@HeikoSOXPHOS: what is slow?12:37
@wikingjust make sure that when you &12:37
@wikingthen they are basically not having 2 inputs into the same value12:37
@HeikoSwiking: yeah I get it12:37
@wikingso its easy12:37
@wikingbut we need a flag that says12:37
OXPHOSHeikoS: I mean if I change anything in SGObject or Any everything will recompile from start12:37
@wikingthis is an SGvector with this type serialized12:38
@HeikoSOXPHOS: sure it is the base class12:38
@wikingso that when we deserialize we  know what to create12:38
@HeikoSwiking: so I think we should just have two enums12:38
@HeikoSeasy12:38
@wikingyeah12:38
@wikingthat's cool12:38
@HeikoSOXPHOS: so can you make two enums?12:39
@HeikoSone to distinguish between vector and matrix12:39
@HeikoSand one for the base type12:39
OXPHOSHeikoS: sure12:39
@HeikoSbut fore now, lets focus on serializing a vector member12:40
@HeikoSyou can create a simple class in the unit tests, just like sanuj did for the tags12:40
@HeikoSwhere you register a vector member12:40
@HeikoSand then serailize the class12:40
@HeikoSthen we can extend that class slowly to have matrix members, sgobject members ,etc12:40
@HeikoSuntil we covered all cases12:40
c4goldswwiking Hey, how are you?12:40
@HeikoSbut the PRs should always focus on one of the cases so that they are easy to review12:40
@HeikoSOXPHOS: can put a unit test and the class in the PR?12:41
OXPHOSHeikoS: the current one is not what you'r expecting?12:41
@wikingc4goldsw: heheh bit busy but good thnx12:41
@HeikoSOXPHOS: it is!12:41
@HeikoStwo things need change (apart from minors)12:41
@HeikoSOXPHOS: 1.) make two enums rather than one12:41
c4goldswwiking: Did you have time to debug my code?12:41
c4goldswIt's fine if you didn't.12:42
@HeikoSOXPHOS: 2.) put a unit test where you serialize a class with a vector member, (create a fake class somewhere in unit tests)12:42
OXPHOSHeikoS: oh you mean serializing vector from sgobject?12:42
@HeikoSOXPHOS: i mean serialize the SGObject (that has a vector member) itself12:42
@wikingc4goldsw: :( no not really12:42
OXPHOSHeikoS: i see12:42
OXPHOSHeikoS: okay one bigger problem here is, let me create a gist12:43
c4goldswwiking: Alright, I can try doing it myself then.  I may need a little direction from you though.12:43
@HeikoSOXPHOS: sure12:43
OXPHOSHeikoS: I'm trying to iterate unordered_map and archive every single Any entry as in cereal_save.cpp: https://gist.github.com/OXPHOS/521fba4f635590f8e7782b634874dc6612:45
OXPHOSHeikoS: but apparently this iteration is not working as I expected..12:45
@HeikoSOXPHOS: checking12:46
@HeikoSOXPHOS: so whats the problem ?12:47
@HeikoSnot sure I get it12:47
OXPHOSHeikoS: sry..I just got this result and didn't think through... on a second thought, not big problem for now. Just the enum type mapping is not working well12:48
@wikingOXPHOS: it->second_type12:49
@HeikoSOXPHOS: let me know if it sticks on being a problem12:49
@wikingshould give u the type actually12:49
OXPHOSHeikoS: okay so i'll do two enum and just check sgvector as the first step12:50
@HeikoSOXPHOS: before you do the two enum12:50
OXPHOSwiking: really? so there's sth. wrong..12:50
@HeikoSdo a class that is serilalised12:50
@HeikoSthat is12:50
@HeikoSinstance.save_serializable()12:50
@HeikoSinstance.load_serializable()12:50
@HeikoSshould work12:51
@HeikoSyou can flatten out the enums after that12:51
@HeikoSsince it is cosmetic, but the class serialization is what really counts12:52
OXPHOSkk12:52
OXPHOSwiking: what's the it->second_type is for? it's gonna be type Any right?12:53
@wikingno12:54
@wikinghttp://en.cppreference.com/w/cpp/utility/pair12:54
OXPHOSwiking: I still think it12:57
OXPHOSs any..12:57
OXPHOStypedef std::unordered_map<BaseTag, Any> ParametersMap;12:57
OXPHOSwiking: also can I add #define CEREAL_SAVE_FUNCTION_NAME cereal_save pair to common.h?12:58
@wikingOXPHOS: no need13:03
@wikingi mean yeah do it13:03
@wikingbut on the end before merging13:03
@wikingwe'll remove that part of the commit :)13:03
@wikingbecause we'll just have one load/save13:03
OXPHOSkk13:04
-!- c4goldsw [82e2e6c8@gateway/web/cgi-irc/kiwiirc.com/ip.130.226.230.200] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client]13:05
-!- HeikoS1 [~heiko@82-132-212-120.dab.02.net] has joined #shogun13:07
-!- HeikoS [~heiko@217.138.5.14] has quit [Ping timeout: 260 seconds]13:07
-!- HeikoS1 [~heiko@82-132-212-120.dab.02.net] has quit [Quit: Leaving.]13:22
-!- sanuj [~sanuj@59.91.129.169] has joined #shogun13:35
-!- c4goldsw [503e75df@gateway/web/cgi-irc/kiwiirc.com/ip.80.62.117.223] has joined #shogun13:35
-!- c4goldsw [503e75df@gateway/web/cgi-irc/kiwiirc.com/ip.80.62.117.223] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client]13:54
-!- c4goldsw [503e75df@gateway/web/cgi-irc/kiwiirc.com/ip.80.62.117.223] has joined #shogun14:04
-!- c4goldsw [503e75df@gateway/web/cgi-irc/kiwiirc.com/ip.80.62.117.223] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client]14:41
-!- OXPHOS [c68f0c0c@gateway/web/freenode/ip.198.143.12.12] has quit [Ping timeout: 250 seconds]14:54
-!- OXPHOS [c68f0c0c@gateway/web/freenode/ip.198.143.12.12] has joined #shogun15:51
-!- HeikoS [~heiko@nat-180-145.internal.eduroam.ucl.ac.uk] has joined #shogun15:54
-!- mode/#shogun [+o HeikoS] by ChanServ15:54
-!- HeikoS [~heiko@nat-180-145.internal.eduroam.ucl.ac.uk] has quit [Client Quit]15:56
-!- HeikoS [~heiko@nat-180-145.internal.eduroam.ucl.ac.uk] has joined #shogun16:01
-!- mode/#shogun [+o HeikoS] by ChanServ16:01
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun16:12
shogun-notifier-shogun-data: Sanuj :master * f9963d2 / testsuite/meta/distance/euclidean.dat: https://github.com/shogun-toolbox/shogun-data/commit/f9963d20a43ef6fd37c9d6544450b780eb73b98116:12
shogun-notifier-shogun-data: add data for euclidean distance cookbook16:12
shogun-notifier-shogun-data: Heiko Strathmann :master * 8b22045 / testsuite/meta/distance/euclidean.dat: https://github.com/shogun-toolbox/shogun-data/commit/8b22045a11b5444a609d82d8475c58e8d94a93ea16:12
shogun-notifier-shogun-data: Merge pull request #118 from sanuj/euclidean_dist16:12
shogun-notifier-shogun-data:16:12
shogun-notifier-shogun-data: add data for euclidean distance cookbook16:12
@HeikoSOXPHOS: jo16:13
@HeikoSOXPHOS: for the linalg branch, can you also start to send patches that remove all cmake options for linalg? (since now we are runtime)16:13
-!- lambday [8028b10a@gateway/web/freenode/ip.128.40.177.10] has joined #shogun16:22
-!- mode/#shogun [+o lambday] by ChanServ16:22
@HeikoSlambday: jo, know how we deal with vectors in ruby?16:23
@wikingyes16:24
@wikingnarray16:24
@lambdayHeikoS: no idea! :( never worked on ruby16:24
@wikingshitty16:24
@wikingprobably in the meanwhile there came a better matrix library16:24
@lambdayHeikoS: are you at the talk?16:24
@HeikoSyeah16:24
@HeikoSwiking: Narray16:24
@HeikoSand NMatrix16:24
@HeikoSok16:24
@HeikoSjust working out meta examples16:24
@HeikoSwiking: I think I wanna visit Yosemite16:25
@HeikoSare you up for that?16:25
@wikingprobabily this would be a better ide to be used16:25
@wikinghttps://github.com/SciRuby/nmatrix16:25
@wikingHeikoS: wanna climb or just hike?16:25
@wikingeither is fine for me16:26
@HeikoSwiking: you climb?16:26
@HeikoSI would be up for that, but fine to hike as well16:26
@wikingyeah i climb16:26
@HeikoSwiking: cool, single pitch, multi pitch?16:27
@HeikoSwiking: yeah I talked to the nmatrix guy a few gsoc mentor summits agp16:36
@HeikoSfor now,I will make the old one work16:36
@wikingwell lately i just fucking do bouldering16:36
@wikingbecause theres' nothing around me16:36
@wikingnearby16:36
-!- c4goldsw [503e75df@gateway/web/cgi-irc/kiwiirc.com/ip.80.62.117.223] has joined #shogun16:36
OXPHOSHeikoS: kk16:36
@HeikoSwiking: lets see what goes then, we can go for a couple of days?16:37
@wikingsure16:37
@HeikoSOXPHOS, wiking, lambday, sanuj, Saurabh7 meeting in 1.5 hrs!!16:37
@wikingoh really shit16:38
@wiking:)16:38
sanujcool16:40
OXPHOSwiking HeikoS: i updated the patch https://github.com/shogun-toolbox/shogun/pull/3375. I can't make it work...16:43
OXPHOSI have a derived class of SGObject, CCerealObject, which registers a sgvector16:43
OXPHOSI can serialize a sgvector with Any.cereal_save directly. However, with CSGObject.cereal_save, I got empty vector16:44
@HeikoSOXPHOS: cool will check16:44
@HeikoSlets see16:44
OXPHOSnot only empty vector, I was told it's empty, as the enumtype = 016:45
shogun-notifier-shogun: Heiko Strathmann :feature/meta_vector_matrix * e096903 / examples/meta/generator/targets/ruby.json: https://github.com/shogun-toolbox/shogun/commit/e0969033b0ff4680ab7b1fea86a888c2bfc9258416:51
shogun-notifier-shogun: add matrix and vector instantiations16:51
shogun-notifier-shogun: Heiko Strathmann :feature/meta_vector_matrix * 8f93dd1 / examples/meta/generator/targets/ruby.json: https://github.com/shogun-toolbox/shogun/commit/8f93dd18c0d8fcaf00795011543c29478d998e1f16:51
shogun-notifier-shogun: fix typos16:51
shogun-notifier-shogun: Heiko Strathmann :feature/meta_vector_matrix * fa44349 / examples/meta/generator/targets/ruby.json: https://github.com/shogun-toolbox/shogun/commit/fa4434916426b8b5aa3bf47f51bb49c3d2bb321c16:51
shogun-notifier-shogun: working ruby meta example tests for vector and matrix16:51
@HeikoSsanuj: ^16:53
@HeikoSsee what I am doing there16:53
@HeikoSmaking the meta examples work with all vectors etc16:53
@HeikoSand these tests also serve as examples what the language allows16:53
sanujHeikoS, okay16:54
sanujlet me see16:54
@HeikoSsanuj: also opened /updated the issues16:59
shogun-notifier-shogun: lambday :feature/bigtest * 3b29d82 / examples/meta/generator/targets/cpp.json: https://github.com/shogun-toolbox/shogun/commit/3b29d8261bcd383c26d8b2b3c174321f063a997b17:02
shogun-notifier-shogun: fixed enum class generation in meta examples17:02
shogun-notifier-shogun: Heiko Strathmann :feature/meta_vector_matrix * a02ea55 / examples/meta/generator/targets/java.json: https://github.com/shogun-toolbox/shogun/commit/a02ea55b81a7f6fa4c4472ce1fa29b028839992817:07
shogun-notifier-shogun: (non-working) draft for java matrix and vector types17:07
sanujHeikoS, i seee17:08
shogun-notifier-shogun: Heiko Strathmann :feature/meta_vector_matrix * 2bae200 / examples/meta/generator/targets/java.json: https://github.com/shogun-toolbox/shogun/commit/2bae20091a1c7e5739066f536c8d13ae9a7717c517:10
shogun-notifier-shogun: typo fix17:10
sanuji'm going to have dinner17:13
sanujwill be back before meeting17:13
-!- HeikoS [~heiko@nat-180-145.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.]17:24
-!- sanuj [~sanuj@59.91.129.169] has quit [Ping timeout: 258 seconds]17:27
-!- HeikoS [~heiko@nat-180-145.internal.eduroam.ucl.ac.uk] has joined #shogun17:28
-!- mode/#shogun [+o HeikoS] by ChanServ17:28
shogun-notifier-shogun: Heiko Strathmann :feature/meta_vector_matrix * 9e099d8 / examples/meta/generator/targets/octave.json: https://github.com/shogun-toolbox/shogun/commit/9e099d890a907be20d834c72d27540cf030ce1b517:32
shogun-notifier-shogun: make octave meta vector/matrix tests work17:32
c4goldswHeikoS I've made revisions to my PR and I've resolved a SWIG interface error that I was facing before I left for vacation.  Do you have some time before the meeting to discus the remainder of it?17:38
-!- leagoetz [~leagoetz@nat-246-164.internal.eduroam.ucl.ac.uk] has joined #shogun17:41
@HeikoSc4goldsw: cool!17:42
@HeikoSI am currently preparing the meeting17:43
@HeikoSwill try to check after17:43
@HeikoSif not tomorrow17:43
c4goldswSounds good.17:43
-!- sanuj [~sanuj@59.91.129.169] has joined #shogun17:51
-!- yorkerlin [63e6385e@gateway/web/freenode/ip.99.230.56.94] has joined #shogun18:01
-!- Saurabh7_ [~Saurabh7@1.39.96.52] has joined #shogun18:02
Saurabh7_asdfaf18:02
@HeikoSSa18:02
Saurabh7_hi18:02
@HeikoSsasdasdasd18:02
@HeikoShi :)18:02
@HeikoSwe can start soon18:03
Saurabh7_:)18:03
sanujlisitsyn, are you here?18:03
Saurabh7_tf is this lag :(18:03
Saurabh7_brb18:04
-!- Saurabh7_ [~Saurabh7@1.39.96.52] has quit [Client Quit]18:04
-!- Saurabh7_ [~Saurabh7@1.39.96.52] has joined #shogun18:04
OXPHOSping18:05
@HeikoSpong18:05
@HeikoSsanuj: there?18:05
sanujyes18:05
sanujsergey is not here though18:05
sanujHeikoS, \o/18:06
@HeikoSthats fine18:06
sanujcool18:06
@HeikoSok everyone ready?18:06
leagoetzI am :)18:06
sanujhaha18:06
sanujyeah18:06
@HeikoSlambday, wiking, leagoetz, lisitsyn, OXPHOS, sanuj Saurabh718:06
@HeikoSSaurabh7_: ?18:07
yorkerlinhi18:08
@HeikoSyorkerlin: ah hi18:08
Saurabh7_yes here18:08
@lambdayhello18:08
@HeikoSyeah good if you join the meeting as well, yorkerlin18:08
@HeikoSok then18:08
@HeikoSGSoC final stages meeting18:08
@HeikoSWelcome to our last "during GSoC" meeting.18:08
@HeikoSFirst of all: Well done OXPHOS, sanuj, Saurabh7_!18:08
@HeikoSAll of you have successfully passed the midterm evaluation, and we are very happy with the work you have been doing so far.18:08
@HeikoSThis year's GSoC projects were much more challenging than in previous years and we are glad things did work out.18:08
@HeikoSThe work on linalg refactoring, algorithm speedups and tags is essential for Shogun's modernization, and will have a significant impact on making Shogun's inside development simpler, end-users, and attracting new developers.18:08
@HeikoSso cheers for everyone18:09
@HeikoS*cheer*18:09
leagoetzwohooooo!18:09
yorkerlincheer!18:09
c4goldsw*clap*18:09
sanuj:D18:09
@HeikoSThe goal of the meeting today is to state the expectations for the last few weeks, and organise some necessary admin.18:09
@HeikoSMajor points:18:09
@HeikoS-Status updates18:09
@HeikoS-The last weeks of GSoC (deliverables, organisation, timing)18:09
@HeikoS-Post GSoC18:09
@HeikoSQuestions/remarks?18:09
@HeikoSAs said in email, If everyone could give a brief update of their project?18:10
@HeikoSwho wants to start?18:10
sanuji will18:10
@HeikoS(everyone else, prepare text)18:10
@HeikoSsanuj: ok then go for it :)18:10
sanujInitial plan for my project was to implement tags-based parameters and plugin architecture and migrate old code to this new architecture.18:10
sanujI think foundation work is done, and implementation of tags-based parameters and plugin architecture should be complete soon. Huge chunk of tags-based parameters has been merged. A PR has been started for plugins implementation.18:10
sanujMigration of old code has not started still. Although I have tested a couple of things to make sure that the new architecture supports old code transformation. But work still needs to be done here.18:11
sanujBy the end of this project, the new architecture should be completely implemented and migration should be started but not completed.18:11
sanujApart from this, I was able to do 11 cookbooks.18:11
@HeikoS*cheer*18:11
leagoetzthat's great18:11
sanujPost GSoC, I can continue with the migration work of old code.18:12
sanujThanks :D18:12
@HeikoSsanuj: thanks! :)18:12
sanujDifficult things are:18:12
sanuj- to decide base classes for shogun-base.18:12
sanuj- starting changing classes of tags parameters.18:12
sanuj- writing plugins for shogun classes.18:12
sanujthat's it for now :)18:12
@HeikoSsanuj: key thing is to structure the next weeks well, so have some realistic goals and concrete steps towards it18:12
@HeikoSthanks!18:12
@HeikoSSaurabh7_, OXPHOS who wants next?18:13
sanujHeikoS, yes18:13
OXPHOSI can go18:13
@HeikoSyep!18:13
OXPHOSMy projects were: linalg refactor, smart pointer, and serialization refactor.18:13
OXPHOSFor the linalg part, We had the new linalg framework ready, integrating cpu and gpu vector in SGVector and SGMatrix.18:13
OXPHOSI am now migrating the operations to the new system. And I will remove the old ones soon.18:13
OXPHOSI just started the serialization with cereal18:14
OXPHOSand the serialization of sgobject should be ready this week18:14
OXPHOSthen I'll also test the current sgobjects and remove the old ones18:14
OXPHOSI don't think I'll have time for more18:14
OXPHOSbut these two parts should be able to be done18:15
OXPHOSalso I had some cookbooks - I didn't count18:15
OXPHOSbasically that's it18:15
@HeikoSOXPHOS: yeah dont worry about the smart pointers  :)18:15
@HeikoSthe stuff was quite involved, and if cereal works that already is *so* useful18:15
OXPHOSyeaaaah18:15
@HeikoSOXPHOS: cheer!! :)18:16
@HeikoSok thanks18:16
@HeikoSSaurabh7_:  you want to give a quick summary?18:16
Saurabh7_yep18:16
Saurabh7_Updated implementations lars, kmeans, cartree, random forest, crossvalidation, liblinear , also add thread safety for some generic cases while doing this and additional test coverage,18:16
Saurabh7_next I am updating nearest neighbours algorithm with tree implementations, more tests and an approximate fast version.18:16
Saurabh7_I added cookbooks for most of this algos too. I have plan to optimize other clustering alogs later too like heirarchical clustering which is slow right now and finish pending benchmarks.18:16
-!- travis-ci [~travis-ci@ec2-107-22-44-110.compute-1.amazonaws.com] has joined #shogun18:17
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/14556134418:17
-!- travis-ci [~travis-ci@ec2-107-22-44-110.compute-1.amazonaws.com] has left #shogun []18:17
@HeikoSouch ^ ;)18:17
Saurabh7_done18:17
Saurabh7_thanks travis18:17
OXPHOSlol18:17
Saurabh7_:P18:17
sanuj:D18:17
@HeikoSSaurabh7_ nice, all algos are now at competetive speed to other libs ,which is great18:17
@HeikoSok cool, lets move on then18:17
@HeikoSThe last weeks of GSoC:18:17
@HeikoSThe program ends on 23 August 19:00 UTC.18:18
@HeikoSThis is the *hard* deadline for students to finish everything they need.18:18
@HeikoSWe have an internal deadline, 16 August, for you guys, so that we have a week of buffer.18:18
@HeikoSBy 16 August, we expect a number of deliverables from everyone.18:18
@HeikoSThis first and foremore is everything you specified in your revised timelines (if you haven't updated or discussed that with your mentors yet, please do that ASAP).18:18
@HeikoSleagoetz: check revised timelines18:18
@HeikoS;)18:18
sanujokay, I'll update the timeline :D18:18
@HeikoSAll patches should be merged (into develop is optimal, but we accept feature branches as well).18:18
leagoetzyep18:18
@HeikoSIn order to achieve that, please stop adding new features *at least two weeks* before end of GSoC.18:18
@HeikoSThat is, from the first week of August, your efforts should be polishing and merging PRs.18:19
@HeikoS(and docs, which I will talk about next)18:19
@HeikoSEveryone clear on that?18:19
sanujyeah18:19
@HeikoSReally take those last weeks as something different to the coding weeks18:19
@HeikoSAdditional deliverables: (Please plan a week of full-time for this)18:19
@HeikoS-Documentation of your project:18:19
@HeikoS * OXHPOS/sanuj18:19
@HeikoS- a polished README that describes the parts of your developed framework.18:20
@HeikoSThis is aiming at new developers so that they can understand what the high-level picture is.18:20
@HeikoSThere is already a README_linalg that OXPHOS can moify.18:20
@HeikoSsanuj can orient himself on this as well.18:20
@HeikoSGive high-level ideas, design decisions and rationale, examples how to use, how to extend18:20
@HeikoSExample: OXPHOS, imagine a GSoC student next year wants to extend linalg as an entrance task, or want to understand serialization.18:20
@HeikoSExample: sanuj, imagine a GSoC student wants to continue polishing Shogun's API using base classes only), or write a plugin.18:20
@HeikoSBoth: be creative and really give this some love, as it is the exposure of your project to the world.18:20
@HeikoS Questions?18:20
OXPHOSthe Additional deliverables18:21
OXPHOSshould be planned in the 16-23rd or earlier?18:21
@HeikoSno the week before18:21
OXPHOSkk18:21
@HeikoSOXPHOS: last week should be kept *free*18:21
@HeikoSso plan to finish this at the 16th18:21
@HeikoSthat is, start writing it on the 8th or so18:21
@HeikoSMake sure to spend some time on it18:22
@HeikoSI can't emphasize enough how important such a well polished doc is18:22
sanujokay18:22
@HeikoSimagine when you first got involved in Shogun, imagine you had a well written doc for the things you touched ;)18:22
sanujemphasis on docs and readmes18:22
@HeikoSleagoetz: anozther thing for checklist :)18:22
@HeikoSsanuj: yes exactly18:22
@HeikoSok, Saurabh7_18:22
@HeikoS * Saurabh7 - for you, we expect a website/blog-post (i.e. on github, or can be using the benchmarking framework) where you describe all the improvements you did.18:22
@HeikoSThis should contain: reproducible benchmark timings (i.e. merged benchmarks in zoq), clear "before/after" comparisons of runtimes.18:23
@HeikoS This would be something you could give to a person that asks you how efficient Shogun is, or which algorithms you speeded up during GSoC.18:23
sanujyeah, i had a tough time understanding linalg and then HeikoS made lambday write a readme :D18:23
@HeikoS Ideally, this would be a nice website some graphs. wiking can help you setting up zoq on the buildbot.18:23
@HeikoS But we also want to you write a little summary page yourself.18:23
@HeikoS Questions?18:23
@HeikoSsanuj: I did not make him, he did that at his own will ;)18:23
sanujyeah :D18:24
Saurabh7_ok18:24
@HeikoSSaurabh7_ clear on that? both website with description, and some automated stats?18:24
@HeikoSSaurabh7_: in your case, this will take more time than writing a readme18:24
Saurabh7_HeikoS, yes clear18:24
@HeikoSso make sure to stop optimizing and start doing that in time18:24
@HeikoSit is also something cool for you to show off with (at job interviews etc)18:24
@HeikoSso make it really publishable :)18:24
@HeikoS * OXPHOS and sanuj. Saurabh's final deliverable is more work than yours.18:25
@HeikoSFor you guys, we want you to take time to go through each others code/docs -- a peer review18:25
@HeikoSThe mindset you should have is: Imagine you were a student new to Shogun and wanted to continue working on the framework.18:25
@HeikoSAre all questions answered? Is everything documented? Does everything work? Is the README well written18:25
@HeikoSWe would like you to have a public conversation on the mailing list, and then work on improvements18:25
@HeikoSYou can do this while you are working on your own18:25
@HeikoSagain, this is why we want you to stop coding that last week, so you can fully focus on that18:25
OXPHOStoo bad I already read the tags file ;P18:26
sanuji see18:26
sanuj:P18:26
@HeikoSOXPHOS: read the docs sanuj writes18:26
@HeikoStry to tell him what is confusing18:26
OXPHOSsure18:26
@HeikoSagain, imagine you saw things the first time, what would make it better explained18:26
@HeikoSthis is really for the next generation of students, so they dont have to suffer the same pain you did ;)18:26
sanujyeah, docs are importatn18:27
@HeikoSsanuj, Saurabh7_, OXPHOS Again, please spend the last week before 16 August on these things *only*. no more coding18:27
OXPHOSnow i hear the ticking clock..18:27
@HeikoSyeah it goes fast, doenst it? :)18:28
@HeikoSok18:28
@HeikoSFinally, thank you for your great work all. We look forward to the last few weeks18:28
@HeikoSPlease stay around after GSoC, we really enjoyed working with you guys18:28
@HeikoSand we hope you learned a lot and had some fun on the way18:28
sanujit was fun to work with you all18:28
sanujespecially with the active irc18:28
@HeikoSyeah we enjoy GSoC times as well because of that :)18:28
OXPHOSthanks for being patient with me ;)18:28
@HeikoSlikewise!18:29
@HeikoSMeeting Summary:18:29
@HeikoS-Hard deadline on August 2318:29
@HeikoS-Internal deadline on August 16,18:29
@HeikoS-Start working on final deliverables in first week of August.18:29
@HeikoS-Deliverables: all patches merged, documentation (sanuj/OXHPOS: README, Saurabh7: website), peer review (sanuj/OXPHOS)18:29
@HeikoSAny final remarks/questions?18:29
@HeikoSok then. Meeting over! Thanks everyone18:29
sanujThank You18:29
sanuj:D18:30
OXPHOSThanks!18:30
Saurabh7_thanks!18:30
@lambdaythanks HeikoS18:30
leagoetzThanks everyone!18:30
yorkerlinHi Heiko, I will do another round of refactroing for the OPT framework.  Coordinate decent  solver will be added in the OPT framework.  It will be on a feature branch. BTW, the GP framework will have new features such as structured GP.18:30
yorkerlinI will use and test the linalg lib.18:31
@HeikoSyorkerlin: 1 sec18:33
@HeikoSbrb18:33
-!- leagoetz [~leagoetz@nat-246-164.internal.eduroam.ucl.ac.uk] has quit []18:34
@HeikoSyorkerlin: cool18:35
@HeikoSreally wanna do that in feature brnach for now?18:35
@HeikoSokok18:35
yorkerlinany idea why travis is down?  https://github.com/shogun-toolbox/shogun/pull/338018:36
OXPHOSyorkerlin: I think it's running - in the background https://travis-ci.org/shogun-toolbox/shogun/pull_requests18:39
yorkerlinthanks!18:40
@HeikoSyorkerlin: so about c++1118:41
@HeikoSjust use it in feature branch18:41
@HeikoSyorkerlin: did you have a chance to talk to wiking about the design?18:41
-!- c4goldsw [503e75df@gateway/web/cgi-irc/kiwiirc.com/ip.80.62.117.223] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client]18:41
@HeikoSyorkerlin: nice with michaelis email eh?18:41
@HeikoSyorkerlin: oh btw, you want to write a variational GP cookbook?18:41
@HeikoSthat would be cool!18:41
@HeikoSotherwise, you can also point it out to some of the students18:41
@HeikoSsince people are running out of notebooks18:42
yorkerlinyes. let's get this PR https://github.com/shogun-toolbox/shogun/pull/3380 merged. I will send him another email.18:42
Saurabh7_I would like to look at GP sometime , looks interesting18:42
Saurabh7_maybe later :)18:43
yorkerlinI will talk to wiking about the design for the OPT framework18:43
@HeikoSSaurabh7_:  you can do a variational GP notebook if you want18:43
@HeikoSsparse GP regression18:43
@HeikoSyorkerlin: can help18:43
yorkerlinyes.18:44
@HeikoSSaurabh7_: maybe you can do that cookbook as your next one!18:44
@HeikoSwe have a notebook with everything you need18:44
@HeikoSyorkerlin: do you have a gist of the notebook?18:44
@HeikoSalso btw for michalis, you could reply in the email18:44
Saurabh7_ok sounds cool18:44
-!- OXPHOS [c68f0c0c@gateway/web/freenode/ip.198.143.12.12] has quit [Quit: Page closed]18:44
Saurabh7_I will ping you yorkerlin18:45
@HeikoSbecause it is currently not on website18:45
yorkerlinhttp://www.shogun-toolbox.org/static/notebook/current/variational_classifier.html18:45
@HeikoSSaurabh7_: yorkerlin is rarely in irc, make sure to ask many things now :)18:45
Saurabh7_i see :)18:46
yorkerlinonce PR 3380  is merged, the notebook and memory leak bugs will disappear.18:46
@HeikoSgreat!18:46
@HeikoSok I gotta run now18:46
@HeikoSwill do other things tomorrow18:46
@HeikoSsee you guys!18:46
yorkerlinok18:46
Saurabh7_cya18:47
Saurabh7_yorkerlin, can you point me to the sparse gp thing in the notebook18:48
Saurabh7_just the section18:48
yorkerlin@Saurabh7_  some reference for you if you want to write a notebook for GP.   http://www.shogun-toolbox.org/static/notebook/current/variational_classifier.html https://github.com/yorkerlin/plots-for-GP/blob/master/Untitled1.ipynb18:48
yorkerlinfor regression, please check out this http://www.shogun-toolbox.org/static/notebook/current/gaussian_processes.html. (FITCInference and VarDTCInference)18:49
@HeikoSyorkerlin: it is cookbook, so very minimal, no math18:49
@HeikoSbut good reference18:49
-!- HeikoS [~heiko@nat-180-145.internal.eduroam.ucl.ac.uk] has quit [Quit: Leaving.]18:50
yorkerlinfor GP, you can write a meta notebook for  sparse GP regression, and sparse GP classification.18:51
yorkerlinsparse GP regression: please see the last session at http://www.shogun-toolbox.org/static/notebook/current/gaussian_processes.html.18:54
yorkerlinsparse GP classification: please see http://www.shogun-toolbox.org/static/notebook/current/variational_classifier.html#sparseGP18:55
yorkerlinok I gotta leave now. Please send me an email if you have any question about GP. :)18:55
-!- yorkerlin [63e6385e@gateway/web/freenode/ip.99.230.56.94] has quit [Quit: Page closed]18:56
-!- Saurabh7_ [~Saurabh7@1.39.96.52] has quit [Quit: Leaving]18:59
-!- travis-ci [~travis-ci@ec2-54-197-156-82.compute-1.amazonaws.com] has joined #shogun19:00
travis-ciit's lambday'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/14556434719:00
-!- travis-ci [~travis-ci@ec2-54-197-156-82.compute-1.amazonaws.com] has left #shogun []19:00
-!- travis-ci [~travis-ci@ec2-54-197-156-82.compute-1.amazonaws.com] has joined #shogun19:34
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/14556580719:34
-!- travis-ci [~travis-ci@ec2-54-197-156-82.compute-1.amazonaws.com] has left #shogun []19:34
-!- travis-ci [~travis-ci@ec2-107-22-44-110.compute-1.amazonaws.com] has joined #shogun20:15
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/14556656920:15
-!- travis-ci [~travis-ci@ec2-107-22-44-110.compute-1.amazonaws.com] has left #shogun []20:15
-!- lambday [8028b10a@gateway/web/freenode/ip.128.40.177.10] has quit [Ping timeout: 250 seconds]20:18
-!- sanuj [~sanuj@59.91.129.169] has quit [Remote host closed the connection]20:27
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]20:32
-!- c4goldsw [5da420e6@gateway/web/cgi-irc/kiwiirc.com/ip.93.164.32.230] has joined #shogun22:17
-!- c4goldsw_1 [5da420e6@gateway/web/cgi-irc/kiwiirc.com/ip.93.164.32.230] has joined #shogun22:20
-!- c4goldsw_1 [5da420e6@gateway/web/cgi-irc/kiwiirc.com/ip.93.164.32.230] has quit [Client Quit]22:20
-!- c4goldsw [5da420e6@gateway/web/cgi-irc/kiwiirc.com/ip.93.164.32.230] has quit [Client Quit]22:20
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has joined #shogun23:25
-!- mode/#shogun [+o HeikoS] by ChanServ23:25
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun23:32
shogun-notifier-shogun: OXPHOS :feature/linalg_refactor * 9f79e0a / / (7 files): https://github.com/shogun-toolbox/shogun/commit/9f79e0a8d02313dbda9897f316c30481ef57be7623:32
shogun-notifier-shogun: linalg-sgvector and sgmatrix-sum23:32
shogun-notifier-shogun: Heiko Strathmann :feature/linalg_refactor * 3bdc19f / / (7 files): https://github.com/shogun-toolbox/shogun/commit/3bdc19f9bdb4efc6399912234abe1eaf240d2aae23:32
shogun-notifier-shogun: Merge pull request #3344 from OXPHOS/linalg_sum23:32
shogun-notifier-shogun:23:32
shogun-notifier-shogun: LinalgRefactor-SGVector&SGMatrix-sum23:32
shogun-notifier-shogun: Heiko Strathmann :feature/meta_vector_matrix * 39003f0 / examples/meta/ (3 files): https://github.com/shogun-toolbox/shogun/commit/39003f073ab5ffb1991419c1dfe41b09e96c34c123:44
shogun-notifier-shogun: make matrix vector meta example tests work for csharp23:44
--- Log closed Tue Jul 19 00:00:28 2016
« Sunday, 2016-07-17 Index Tuesday, 2016-07-19 »

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