IRC logs of #shogun for Thursday, 2017-02-16

« Wednesday, 2017-02-15 Index Friday, 2017-02-17 »
--- Log opened Thu Feb 16 00:00:27 2017
@wikinglisitsyn, you see, JIT https://haosdent.gitbooks.io/tensorflow-document/content/resources/xla_prerelease.html :)01:07
@sukeyPull Request #3608 "Clean up KNN"  synchronized by MikeLing - https://github.com/shogun-toolbox/shogun/pull/360802:04
@sukeyPull Request #3620 "Add time comparison for KNN_KDTREE"  synchronized by MikeLing - https://github.com/shogun-toolbox/shogun/pull/362004:18
-!- remote [~remote@portemanteau.ca] has left #shogun []04:26
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Remote host closed the connection]04:59
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun05:00
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]05:05
-!- Guest85308 [~hazzard@142.54.173.74] has joined #shogun06:01
-!- Guest85308 [~hazzard@142.54.173.74] has quit [K-Lined]06:01
@sukeyPull Request #3620 "Add time comparison for KNN_KDTREE"  synchronized by MikeLing - https://github.com/shogun-toolbox/shogun/pull/362006:02
-!- Guest78383 [~haha@223.13.66.34] has joined #shogun08:09
-!- Guest78383 [~haha@223.13.66.34] has quit [Ping timeout: 258 seconds]08:18
-!- Guest23603 [~ff@223.13.66.34] has joined #shogun09:35
-!- Guest23603 [~ff@223.13.66.34] has left #shogun []09:38
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun09:44
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]09:49
-!- abhinav_ [~abhinav@202.173.126.154] has joined #shogun10:09
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun10:15
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]10:20
-!- abhinav_ [~abhinav@202.173.126.154] has quit [Ping timeout: 260 seconds]10:34
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun10:46
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]10:51
-!- abhinav_ [~abhinav@202.173.126.154] has joined #shogun10:55
-!- abhinav_ [~abhinav@202.173.126.154] has quit [Quit: Leaving]11:10
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun11:18
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]11:23
-!- ShellcatZero [~ShellcatZ@cpe-76-93-136-67.san.res.rr.com] has quit [Ping timeout: 264 seconds]12:13
-!- ShellcatZero [~ShellcatZ@cpe-76-93-136-67.san.res.rr.com] has joined #shogun12:16
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has joined #shogun13:11
-!- mode/#shogun [+o HeikoS] by ChanServ13:11
@sukeyPull Request #3608 "Clean up KNN"  synchronized by MikeLing - https://github.com/shogun-toolbox/shogun/pull/360813:12
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun13:19
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]13:23
@sukeyWiki page: Google-Summer-of-Code-2017-Projects edited on shogun-toolbox/shogun by karlnapf13:24
@sukeyWiki page: Google-Summer-of-Code-2017-Projects edited on shogun-toolbox/shogun by karlnapf13:24
CaBaHeikoS: heya13:26
CaBaHeikoS: have a minute?13:27
@sukeyIssue #3629 "Polish class API documentation" opened by karlnapf - https://github.com/shogun-toolbox/shogun/issues/362913:28
@sukeyIssue #3629 "Polish class API documentation" karlnapf added label: "entrance" - https://github.com/shogun-toolbox/shogun/issues/362913:28
@sukeyIssue #3629 "Polish class API documentation" karlnapf added label: "Documentation" - https://github.com/shogun-toolbox/shogun/issues/362913:28
CaBaHeikoS: i'm writing that quick and dirty version of the customkernel that we were talking about the other day right now ;)13:28
@sukeyWiki page: GSoC_2017_low_hanging_fruit edited on shogun-toolbox/shogun by karlnapf13:28
@HeikoSCaBa: hi13:29
@HeikoSyes13:29
CaBaHeikoS: and i guess i'm running into problems that already came up with CustomKernel, so i thought to ask l)13:29
CaBaHeikoS: for one, what is the scenario where lhs != rhs in a kernel? i guess that cannot be covered with precalculated kernels, right?13:30
@HeikoSyes13:30
@HeikoSthis is when you want to apply the method to a new datapoint13:30
@HeikoSi.e. you trained on X and then want to apply to Y that you havent seen when computing the kernel matrix for X13:30
@HeikoSso in that case you need to precompute the kernel between X and Y before you can predict13:31
CaBaHeikoS: so how's that handled in the case of CCustomKernel? I saw some C++ code where ->set_kernel() was called with the testing precalc matrix...13:31
@HeikoSyou need to provide the kernel if you want to predict on new data13:31
@HeikoSas f(x) = \sum_i alpha_i k(x_new, x_i)13:32
@HeikoSwiking: jo around?13:32
@HeikoSI got back from gunnar! finally :)13:33
@wikingHeikoS, yes13:33
@wikingsup?13:33
@HeikoS- Feb 21, 2 or 3pm- Feb 22, 10 or 11am13:33
@HeikoSwhats better=?13:33
@HeikoSthe 11am I guess?13:33
@wiking11 yes13:33
@HeikoScool13:33
@HeikoSlet me reply13:33
@wikingbut i can do 3 if need be13:33
@HeikoS10 is better?13:33
@HeikoSbecause it is 10 or 1113:33
@wikingeither is fine13:34
@HeikoSok then 11 better for me13:34
@HeikoSwiking: cool13:35
@HeikoSIll write a little plan until then so we can present concrete things to gunnar and he just has to yes/no them13:35
@HeikoSwiking: saw this https://github.com/shogun-toolbox/shogun/pull/3534#issuecomment-28015476613:35
@HeikoSis that the one you warn about in cmake?13:36
@HeikoScould do that as well and disable the test for that version?13:36
@wikingdunno yet13:36
@wikingi have to finish somethng13:36
@HeikoSkk13:36
CaBaHeikoS: https://github.com/shogun-toolbox/shogun/blob/develop/examples/undocumented/libshogun/classifier_svmlight_string_features_precomputed_kernel.cpp#L11013:39
CaBaHeikoS: so that's the training / testing scheme for custom kernel?13:39
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun13:39
@sukeyNew Commit "remove message about unused variable warnings as they will just confuse users" to shogun-toolbox/shogun by karlnapf: https://github.com/shogun-toolbox/shogun/commit/6599b893c731e246e54bd140b4f6a8d858eb864f13:40
@HeikoSCaBa: kind of13:41
@HeikoSjust here all data is known at training time13:41
@HeikoSdo you can precompute the test kernel matri xas well13:41
@HeikoSbut yeah13:41
@HeikoSyou see below that the precomputed test kernel matrices are passed13:41
@HeikoSso it is that13:42
CaBaah, so one approach is to concat training + testing data, then subset for training and subset for testing?13:42
@HeikoSno13:44
@HeikoSthat is just one way to do it13:44
@HeikoSbut you could also imagine13:44
@HeikoSyou traing your model on all data13:44
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]13:44
@HeikoSand then, afterwards, you get new test data13:44
@HeikoSand then you compute the kernel between train and test (no subsetting)13:44
@HeikoSand then prediuct13:44
@HeikoShere it is done with the subsetting since all data is known from the start13:44
CaBayes. which will be the case for my usecase, too. sure, generally it is desirable to be able to work with new data at a later time13:48
CaBaHeikoS: is there a code snippet for "solution 1" somewhere?13:49
@HeikoSCaBa: no dont think so13:49
@HeikoSbut it is easy13:49
@HeikoSyou just need to create the km_test matrix in your snippet above manually13:50
@HeikoSwithout subsetting13:50
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun14:11
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]14:16
-!- travis-ci [~travis-ci@ec2-54-158-130-127.compute-1.amazonaws.com] has joined #shogun14:28
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/20198243614:28
-!- travis-ci [~travis-ci@ec2-54-158-130-127.compute-1.amazonaws.com] has left #shogun []14:28
-!- mikeling [uid89706@gateway/web/irccloud.com/x-heyxhvoscogkbdtr] has joined #shogun14:33
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun14:43
@sukeyPull Request #3608 "Clean up KNN"  synchronized by MikeLing - https://github.com/shogun-toolbox/shogun/pull/360814:46
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has quit [Ping timeout: 240 seconds]14:47
-!- travis-ci [~travis-ci@ec2-54-158-130-127.compute-1.amazonaws.com] has joined #shogun15:25
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/20223085315:25
-!- travis-ci [~travis-ci@ec2-54-158-130-127.compute-1.amazonaws.com] has left #shogun []15:25
-!- suhas2go [uid201652@gateway/web/irccloud.com/x-dozyivljzebrwbwh] has joined #shogun15:44
CaBaHeikoS: when CCustomKernel is initialized with a matrix, it creates CIndexFeatures somewhere internally?16:09
-!- goksinen [~goksinen@rrcs-50-75-193-138.nyc.biz.rr.com] has joined #shogun16:20
mikelinghi HeikoS: I found there are several sub-projects for Shogun detox 2, is that mean I need to pick some sub-projects from them if I want to work on Shogun detox 2 as GSoC project?16:24
@HeikoSCaBa: I think so16:25
@HeikoSmikeling: hi!16:25
CaBaHeikoS: i can't find that anywhere...16:25
@HeikoSmikeling: yeah we will probably not be able to do them all16:25
@HeikoSmikeling: this list will need to be made much more concrete16:25
mikelingSp, is Shogun detox 2 an independent project ? is that possible more than one students work on it?16:26
@HeikoSmikeling: it also depends on how much of the last years projects we manage to fully integrate until gsoc 2016 starts16:26
@HeikoSmikeling: I think it will be independent, but maybe a small subproject will be soved by another student, like the efficiency one or others16:26
@HeikoSCaBa: why?16:26
CaBaHeikoS: well i'm currently using 1-dimensional CDenseFeatures<uint32_t> for my own custom kernel hack, was wondering if CIndexFeatures would be more straight forward to use. so i was looking whether i can see them in action somewher16:28
@HeikoSCaBa: I currently dont have a good idea in my head what the IndexFeatures do, but I did not write that stuff initially16:29
CaBa^^16:29
mikelingHeikoS: great! Thank you for your kindly explanation :)16:30
CaBaHeikoS: https://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/kernel/CustomKernel.cpp#L10516:30
@HeikoSmikeling: no worries, looking forward to your application16:30
CaBaHeikoS: looks as if CDummyFeatures are used and not CIndexFeatures16:30
@HeikoSmikeling: definitely have a look at the cereal / tags / linalg branch from last years gsoc16:30
mikelingHeikoS: great! Thank you for your kindly explanation :)16:31
CaBaHeikoS: whereas i don't understand why this is initialized with 2 dummy features, one for kmatrix rows and one for cols... it should reflect the number of feature vectors, right? O.o16:33
@sukeyPull Request #3608 "Clean up KNN"  merged by vigsterkr - https://github.com/shogun-toolbox/shogun/pull/360816:50
@sukeyNew Commit "Merge pull request #3608 from MikeLing/clean_up_KNN16:50
@sukeyClean up KNN" to shogun-toolbox/shogun by vigsterkr: https://github.com/shogun-toolbox/shogun/commit/988fe1e4c3a39e1ca32f91d4a1933661bdbe2f0716:50
@wikingok this was a big one16:50
@wiking:)16:50
CaBasubsetting... is that a kernel thing or a feature thing?16:57
CaBafeature it seems. and seems like dummyfeatures don't really support it but indexfeatures do. in a rather expensive fashion O.o16:59
-!- travis-ci [~travis-ci@ec2-54-147-160-215.compute-1.amazonaws.com] has joined #shogun17:43
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/20229629217:43
-!- travis-ci [~travis-ci@ec2-54-147-160-215.compute-1.amazonaws.com] has left #shogun []17:43
@sukeyIssue #3630 "Shogun error in compilation at 68% " opened by ahmedmazari - https://github.com/shogun-toolbox/shogun/issues/363017:50
@sukeyIssue #3630 "Shogun error in compilation at 68% "- https://github.com/shogun-toolbox/shogun/issues/363018:13
@sukeyIssue #3630 "Shogun error in compilation at 68% "- https://github.com/shogun-toolbox/shogun/issues/363018:13
-!- mikeling [uid89706@gateway/web/irccloud.com/x-heyxhvoscogkbdtr] has quit [Quit: Connection closed for inactivity]18:39
@HeikoSCaBa: well both!18:59
@HeikoSCaBa: we are always happy to take patches :)19:08
CaBaHeikoS: hm, i know. and i'm sure you don't want what i'm cooking right now :D19:08
@HeikoShehe ok19:08
@sukeyPull Request #3617 "Port sgd svm example to meta"  closed by karlnapf - https://github.com/shogun-toolbox/shogun/pull/361719:10
lisitsynHeikoS: I'm trying to rebase feature/tags now19:33
lisitsynon top of develop19:33
lisitsynuhm19:35
lisitsynwhy19:35
lisitsynhahah19:35
lisitsynI totally forgot we merged it in19:36
lisitsynhttps://github.com/shogun-toolbox/shogun/blob/develop/src/shogun/loss/SmoothHingeLoss.h#L919:39
lisitsynTechnosmoothHingey19:39
lisitsynO_o19:40
-!- abhinav [~abhinav@103.211.19.116] has joined #shogun19:49
-!- HeikoS [~heiko@untrust-out.swc.ucl.ac.uk] has quit [Ping timeout: 240 seconds]20:06
@sukeyPull Request #3597 "Add Scala Interface"  synchronized by abhinavrai44 - https://github.com/shogun-toolbox/shogun/pull/359720:12
@sukeyIssue #3459 "Compilation error with g++ 6.1" closed by vigsterkr - https://github.com/shogun-toolbox/shogun/issues/345920:23
@sukeyIssue #3631 "How to import shogun in python ?" opened by ahmedmazari - https://github.com/shogun-toolbox/shogun/issues/363120:42
-!- abhinav [~abhinav@103.211.19.116] has quit [Ping timeout: 240 seconds]20:45
@sukeyPull Request #3599 "fix invalid use of pre-declared class (#3459)"  closed by besser82 - https://github.com/shogun-toolbox/shogun/pull/359921:21
--- Log closed Fri Feb 17 00:00:28 2017
« Wednesday, 2017-02-15 Index Friday, 2017-02-17 »

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