IRC logs of #shogun for Monday, 2016-05-30

« Sunday, 2016-05-29 Index Tuesday, 2016-05-31 »
--- Log opened Mon May 30 00:00:17 2016
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has quit [Quit: Leaving.]00:11
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has joined #shogun00:12
shogun-buildbotbuild #12 of FC23 - libshogun - aarch64 is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/12  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>00:35
shogun-buildbotbuild #2884 of bsd1 - libshogun is complete: Failure [failed configure]  Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2884  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>00:39
shogun-buildbotbuild #22 of xenial - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/22  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>00:39
shogun-buildbotbuild #13 of FC23 - libshogun - aarch64 is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/13  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>01:17
-!- besser82 [~besser82@fedora/besser82] has joined #shogun01:56
-!- mode/#shogun [+o besser82] by ChanServ01:56
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 246 seconds]02:11
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has quit [Remote host closed the connection]02:13
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has joined #shogun02:15
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has quit [Remote host closed the connection]02:27
shogun-buildbotbuild #9 of clang - thread analysis is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang%20-%20thread%20analysis/builds/9  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>, OXPHOS <engelzora@gmail.com>03:44
shogun-buildbotbuild #8 of clang - undefined behaviour analysis is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang%20-%20undefined%20behaviour%20analysis/builds/8  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>, OXPHOS <engelzora@gmail.com>03:48
shogun-buildbotbuild #9 of memleak - valgrind is complete: Failure [failed memory check]  Build details are at http://buildbot.shogun-toolbox.org/builders/memleak%20-%20valgrind/builds/9  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>, OXPHOS <engelzora@gmail.com>06:15
shogun-buildbotbuild #1011 of nightly_none is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_none/builds/1011  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>, OXPHOS <engelzora@gmail.com>06:25
shogun-buildbotbuild #1140 of nightly_default is complete: Failure [failed test notebooks]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/1140  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Ariane Paola Gomes <arianepaola@users.noreply.github.com>, OXPHOS <engelzora@gmail.com>07:41
shogun-buildbotbuild #289 of deb1 - libshogun - PR is complete: Failure [failed git]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun%20-%20PR/builds/289  blamelist: OXPHOS07:41
shogun-buildbotbuild #290 of deb1 - libshogun - PR is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun%20-%20PR/builds/29007:42
-!- sanuj [~sanuj@117.204.246.43] has joined #shogun08:02
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has joined #shogun09:44
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has quit [Client Quit]09:47
-!- sanuj [~sanuj@117.204.246.43] has quit [Ping timeout: 246 seconds]10:37
-!- sanuj [~sanuj@117.204.246.43] has joined #shogun10:58
-!- besser82 [~besser82@fedora/besser82] has joined #shogun11:43
-!- mode/#shogun [+o besser82] by ChanServ11:43
sanujlisitsyn, hey11:52
lisitsynsanuj: hi11:53
sanujlisitsyn, getting a weird error with BaseTag11:53
sanujit is std::unordered_map<BaseTag, Any>11:54
sanujand not std::unordered_map<BaseTag*, Any>11:54
sanuj?11:54
lisitsynyes sure, no pointers11:54
sanuji'll share the error with you11:54
sanujlisitsyn, http://pastebin.com/CXG0Ck2U11:56
lisitsynwell you've got to implement hash for basetag11:58
lisitsyncheck out how hashes are implemented for custom types11:59
sanujlisitsyn, okay, i'll see12:00
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has joined #shogun12:08
-!- mode/#shogun [+o HeikoS] by ChanServ12:08
@HeikoSSaurabh7:  jo12:30
shogun-buildbotbuild #23 of xenial - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/23  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Saurabh7 <saurabh.mahindre@gmail.com>12:32
shogun-buildbotbuild #2885 of bsd1 - libshogun is complete: Failure [failed configure]  Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2885  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Saurabh7 <saurabh.mahindre@gmail.com>12:33
sanujHeikoS, hi12:33
@HeikoSsanuj: hi there12:34
sanujHeikoS, if you are fine with the code example for svr, i can send the integration data PR12:34
@HeikoSjust looking at yout cookbook12:34
sanuji saw your comments12:34
@HeikoSyeah I think this is good12:35
@HeikoSso go for it and update12:35
Saurabh7HeikoS: hi!12:35
@HeikoSSaurabh7: hi12:35
@HeikoSSaurabh7: saw your comment on x-validation12:35
Saurabh7HeikoS: about the get_features ?12:36
@HeikoSSaurabh7: yeah for now, just clone the features12:36
Saurabh7ok12:36
@HeikoSthe labels are stored in CMachine base class, so you can use a getter for them12:36
Saurabh7double clone :)12:36
@HeikoSSaurabh7: not nocessarily12:36
@HeikoSbut we should check that later12:36
@HeikoSso add to some todo list12:36
@HeikoSBut more important to get something working now12:37
@HeikoSand then have it tested a bit12:37
Saurabh7oh yeah only if passed via kernel12:37
@HeikoSyeah  but as I said, lets postpone this  for a bit12:37
@HeikoScan avoid these cases later12:37
@HeikoSfor now lets not care about memory too much12:38
@HeikoSpoint is to speed xvalidation up12:38
Saurabh7HeikoS: okay12:38
Saurabh7how would i test ?12:38
Saurabh7imean comapr outputs for simple cases ?12:38
Saurabh7I have checked crudely but if we ahve to add a unit test12:39
@HeikoSSaurabh7: first step:12:41
@HeikoSwrite a unit test for one particular model12:41
@HeikoSand compare single thread vs multithread results12:41
@HeikoSfor say LibSVM12:41
@HeikoSthen valgrind it12:41
@HeikoSthen put all that cleaned up in the patch12:41
@HeikoSthen we will go from there12:41
Saurabh7ok got it12:42
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 264 seconds]12:48
@HeikoSSaurabh7: let me know how it goes, it should be really fast to get this working btw, not more than a day12:50
@HeikoS, let me know if anything blocks you12:50
@HeikoSSaurabh7, sanuj, arianepaola keep in mind the weekly email update on Mondays12:54
Saurabh7HeikoS: i will push this today itself12:54
@HeikoSSaurabh7: cool!12:54
Saurabh7HeikoS: what should we plan ideally for this week12:54
Saurabh7other than getting this work ofc12:54
@HeikoSSaurabh7: yeah good point!12:55
@HeikoSthe xvalidation wasn't really planned ;)12:55
Saurabh7ye i know12:55
Saurabh7:)12:55
@HeikoSso first of all: let us finish this by tomorrow12:55
@HeikoStoday: send patch and unit test that is memory clean12:55
@HeikoStomorrow: do a few generalisations and more testing12:55
@HeikoSpreferably all of this today12:55
Saurabh7HeikoS: ok12:55
@HeikoSand then, lets have a look at the plan12:55
@HeikoScan you share a a link12:56
@HeikoSSaurabh7: because we should pick an algo12:57
@HeikoSSaurabh7: I am quite open towards things, what would you prefer?12:57
@HeikoSSaurabh7: for that algo we do12:57
@HeikoS1) benchmark to ryans platform12:57
@HeikoS2) optimise12:58
Saurabh7ok i will trying to pick one12:59
@HeikoSSaurabh7: can you send me the googledoc with the plan12:59
Saurabh7ok sec12:59
Saurabh7HeikoS: https://docs.google.com/document/d/1RtHundU7jVLMKRdVhBDgJzOSzxn1Y_QIDEKlNidxZ6Y/edit13:01
@HeikoSSaurabh7: what about LARS?13:02
@HeikoSSaurabh7: not much we can do about KRR13:02
@HeikoS(there is a guy working on approximate KRR though)13:02
@HeikoSso LARS should be a good one, important algorithm13:03
Saurabh7HeikoS: ye I saw13:03
@HeikoSstart by comparing this to scikit learn?13:03
@HeikoSand sending a patch to the benchmark framework?13:03
@HeikoSLARS is also not yet eigen3-ised13:03
Saurabh7LARS i already in i believe13:03
@HeikoSso thats some potential13:03
@HeikoSyeah I took it from your first week :D13:03
Saurabh7oh ok :D13:04
Saurabh7alright I will profiel it havent done yet using perf13:04
@HeikoSyeah cool13:04
@HeikoSbut lets first finish the x-valiudation13:04
@HeikoSand only start lars after merge13:05
Saurabh7LARS13:05
@HeikoSalso, I will try to merge the kmeans stuff soon.13:05
@HeikoSneeds a rebase though13:05
Saurabh7HeikoS: yes i was resolving those conflicts right now13:05
@HeikoScool13:05
@HeikoSgood13:05
Saurabh7they are overwritten anyways13:05
Saurabh7beacuse its moved to KMeansBase13:05
@HeikoSI see13:06
@HeikoSI updated some of the REQUIRES13:06
@HeikoSthat is really it13:06
@HeikoSoh and method name change13:06
@HeikoSruby friendly13:06
@HeikoSSaurabh7: btw one more thing13:08
@HeikoSSaurabh7: at the end of gsoc, we want to have a summary for this project: what you improved and how things compare in benchmarkls13:08
@HeikoSso we should have like a table13:08
@HeikoSwith before/after gsoc13:08
Saurabh7ah so i need to keep before ones13:09
@HeikoSand also all things you improved should be "proven in benchmark", so there should be a link to the new benchmarlk13:09
Saurabh7yes13:09
@HeikoSSaurabh7: not really, we can just use ryans benchmark system13:09
@HeikoSand execute it with a revision that is pre-gsoc13:09
@HeikoSso integrating changes into this system is key13:09
@HeikoSthen we can automatically generate numbers13:09
Saurabh7ok13:09
Saurabh7will put everything i touch in it13:10
@HeikoSbut also keep timings you do13:10
@HeikoSgreat13:10
Saurabh7most are already in13:10
@HeikoSthen we will have a great output of the gsoc :)13:10
Saurabh7except new ones like decision trees13:10
@HeikoSkk13:10
@HeikoSLARS is a good candidate13:10
@HeikoSvery very important algorithm13:10
@HeikoSah multi-core liblinear will be cool as well13:11
shogun-buildbotbuild #14 of FC23 - libshogun - aarch64 is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/14  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Saurabh7 <saurabh.mahindre@gmail.com>13:11
shogun-buildbotbuild #24 of xenial - libshogun is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/24  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Sanuj <sanuj.sharma.in@gmail.com>13:12
shogun-buildbotbuild #2886 of bsd1 - libshogun is complete: Failure [failed configure]  Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2886  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Sanuj <sanuj.sharma.in@gmail.com>13:12
@HeikoSSaurabh7: ok I will check back in a bit13:16
sanujHeikoS, svr initializes the kernel?13:18
@HeikoSsanuj: yeah13:18
sanujit's not done in the previous examples13:18
@HeikoSI think the train method does that13:18
sanujhttps://github.com/shogun-toolbox/shogun/pull/3227/files#diff-80ad9ff4477fe8db88875cfa3cb0bf33L3213:18
@HeikoSfor testing, is has to be done by hand if a plain "apply()" is called13:18
@HeikoSobvioulsy13:18
@HeikoSor you call apply(test_features)13:18
@HeikoSthis does the same13:19
@HeikoSfor training, I think all kernel methods do this themselves13:19
@HeikoSif not, this is a bug13:19
@HeikoSso no need to do it from outside13:19
@HeikoSif you find that this is not true, feel free to send a mini patch to rectify that13:19
@HeikoSbut good that you checked :)13:20
sanujHeikoS, but one needs to give a kernel object to libsvr13:20
@HeikoSyeah13:20
@HeikoSand then svr->train(feats) does13:20
@HeikoSkernel->init(feats, feats)13:20
@HeikoSso the two are equivalent:13:21
@HeikoSkernel->init(feats, feats)13:21
@HeikoSsvr->train()13:21
@HeikoSand13:21
@HeikoSsvr->train(feats)13:21
sanujHeikoS, so i just pass width to the kernel object13:22
sanujno need to pass training features13:22
@HeikoSyes13:22
sanujHeikoS, the constructor requires cache size also13:23
@HeikoSyeah I know13:23
@HeikoSwanna send a patch that avoids that? :)13:23
@HeikoSI hate this13:23
sanujsure13:23
@HeikoScool :)13:23
@HeikoSsanuj: ill check back later today, gotta go now13:24
@HeikoSsee you :)13:24
@HeikoSsanuj: btw make sure to send the monday email soon13:24
sanujHeikoS, yes i will13:24
@HeikoSgreat, thanks13:24
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has quit [Ping timeout: 246 seconds]13:31
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has joined #shogun13:33
-!- mode/#shogun [+o HeikoS] by ChanServ13:33
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has quit [Quit: Leaving.]13:39
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has joined #shogun13:39
-!- mode/#shogun [+o HeikoS] by ChanServ13:39
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has quit [Client Quit]13:40
shogun-buildbotbuild #15 of FC23 - libshogun - aarch64 is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/15  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, Sanuj <sanuj.sharma.in@gmail.com>13:52
arianepaolahello everyone14:44
-!- mizari [~mizari@95-174-213-100.nts.su] has joined #shogun15:29
-!- sanuj [~sanuj@117.204.246.43] has quit [Ping timeout: 240 seconds]15:42
-!- GandalfTheWizard [~Eva@112.10.170.39] has joined #shogun15:51
-!- OXPHOS [9d8b131c@gateway/web/freenode/ip.157.139.19.28] has joined #shogun16:09
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has joined #shogun16:48
-!- mode/#shogun [+o HeikoS] by ChanServ16:49
@HeikoSOXPHOS: jo16:55
@HeikoSwiking: hihi16:55
OXPHOSHeikoS: Hey!16:56
@HeikoSOXPHOS: all good?16:56
OXPHOSHeikoS: yes. I have a linalg prototype just ready. I'm gonna bring cookbook back to normal now16:57
@HeikoSOXPHOS: great about the linalg prototype16:57
OXPHOSHeikoS: And I'll try ifdef SWIG later. I think it gave an error during build16:57
@HeikoSOXPHOS: this SWIG macro is really just for SWIG16:57
@HeikoSnot for the compiler, completely ignored by gcc16:58
@HeikoSonly swig uses it16:58
@HeikoSOXPHOS: btw the other issue got resolved with the data :)16:58
@HeikoSlucky you, somebody else merged something afterwards and updated ;)16:58
OXPHOSHeikoS: saw it. thx. Actually that change often pops up and everytime I just discard it16:59
@HeikoSOXPHOS: it is if you change anything in the "data" dir16:59
@HeikoSsometimes, running the tests change things16:59
@HeikoSbut it should say "dirty"16:59
@HeikoSif you haven't commited the changes in data17:00
OXPHOSI see17:00
@HeikoSOXPHOS: so about the cereal17:00
@HeikoSI saw your sgvector patch17:00
@HeikoSand I dont get why you copied all the content in there, can you explain?17:00
@HeikoSNo need that we have all methods I think, just the data structure itself is needed for a prototype, no?17:01
OXPHOSyou mean copied all 'SGVector'? I just didn't take time to screen for unnecessary ones..17:01
OXPHOSYes I can definitely remove them all @HeikoS17:02
@HeikoSI see17:02
@HeikoSokok, just was curios what was your thinking17:02
@HeikoSthats fine17:02
@HeikoSjust saving time17:02
OXPHOS:)17:03
@HeikoSOXPHOS: so that works?17:04
OXPHOSHeikoS yep17:05
@HeikoSOXPHOS: great news then, curious what wiking will say17:08
OXPHOSHeikoS sth went wrong..17:10
OXPHOSfor linalg..17:10
arianepaolaHeikoS: is there any specific naming convention for the cookbook pages?17:10
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has quit [Ping timeout: 244 seconds]17:13
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has joined #shogun17:32
-!- mode/#shogun [+o HeikoS] by ChanServ17:33
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has joined #shogun17:37
leagoetzhi all17:38
leagoetzOXPHOS:, arianepaola: can you please also send your short reports?17:38
leagoetzthanks!17:38
OXPHOSleagoetz: hi! sure. working on it.17:39
leagoetzgreat, thanks!17:39
leagoetzOXPHOS: all going well?17:40
OXPHOSleagoetz: yep. everything starts to work17:40
leagoetznice :)17:40
arianepaolaleagoetz: ok17:42
arianepaolaany idea how to convert DNA to meta language?17:43
arianepaolafor StringCharFeatures17:44
-!- yorkerlin [63e6385e@gateway/web/freenode/ip.99.230.56.94] has joined #shogun17:50
leagoetzerm ...17:51
@HeikoSarianepaola: hi17:51
@HeikoSarianepaola: I think the way to go is to load a dataset17:52
@HeikoSarianepaola: have a look in the toy data, there are a few examples17:52
arianepaolaFor the meta: StringCharFeatures do I still need DNA then?17:52
@HeikoSarianepaola: since you are correct, the meta language has no way to modify data itself17:52
@HeikoSnot sure i get what you mean17:53
arianepaolahttps://github.com/shogun-toolbox/shogun/blob/develop/examples/undocumented/python_modular/classifier_ssk_modular.py17:53
arianepaolaCSVFile f_feats_train("../../data/fm_train_dna.dat")17:53
arianepaolaStringCharFeatures features_train(f_feats_train, DNA)17:53
-!- yorkerlin [63e6385e@gateway/web/freenode/ip.99.230.56.94] has quit [Client Quit]17:54
@HeikoSah I see17:54
@HeikoSthat should be just an enum17:54
@HeikoSwhich is available17:54
@HeikoSsee the linear svm example17:55
@HeikoSis passes an enum for the solver type17:55
@HeikoSthe DNA should work similarly17:55
@HeikoSarianepaola: let me know if that works17:55
@HeikoSif not I can help17:55
arianepaolaok, thanks17:56
@HeikoSarianepaola: found it?18:02
arianepaolaHeikoS: should we use the abstract in IEEE referenced papers as reference documentation? e.g. for SSK http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=5375806&url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel5%2F5375738%2F5375739%2F05375806.pdf%3Farnumber%3D537580618:02
@HeikoSarianepaola: is that the one mentioned in the code?18:03
@HeikoSwhich class18:03
@HeikoS?18:03
OXPHOSHeikoS: help18:04
OXPHOSHeikoS:  for the linalg, in my branch, I did "fetch upstream", and "rebase upstream/feature/linalg_refactor"18:04
OXPHOSand end up with tons of commits in the history18:04
arianepaolaHeikoS: it uses libsvm18:04
arianepaolahttps://github.com/shogun-toolbox/shogun/blob/develop/examples/undocumented/python_modular/classifier_ssk_modular.py18:04
@HeikoScheck the class for SubsequenceStringKernel18:05
@HeikoSthere is docs which contain a reference18:05
@HeikoSOXPHOS: ok lets see18:05
@HeikoSso you had a branch which was forked from develop18:05
@HeikoSand then did some commits18:05
OXPHOSyes yes18:06
@HeikoSand then did rebase upstream feature/linalg_refactor?18:06
OXPHOSyes18:06
@HeikoSand now?18:06
OXPHOSAll the old commits popped up in my commits history?18:06
@HeikoSarianepaola: I think this example can go under kernels, what do you thkink?18:06
@HeikoSwhich old commits?18:06
@HeikoSby whim18:06
@HeikoSwhom18:06
arianepaolaHeikoS: yes18:07
OXPHOSI just closed a PR. https://github.com/shogun-toolbox/shogun/pull/322918:07
@HeikoSOXPHOS: ok step 1: make a backup copy18:07
arianepaolaHeikoS: I found it in shogun/kernel/string/SubsequenceStringKernel.h18:07
arianepaolaso, I can use the class documentation as reference documentation for the meta, right?18:08
@HeikoSOXPHOS: step 2: reset your local feature branch to the upstream feature/linalg_refactor, by doing:ggit reset --hard upstream/feature/linalg_refactor18:08
@HeikoSstep 3: copy in your changes18:08
@HeikoSstep 3 commit and send PR18:08
@HeikoSOXPHOS: but BACKUP first :)18:08
@HeikoSarianepaola: no no18:08
@HeikoSarianepaola: you should just use the sgclass macro18:08
@HeikoSthat links to the class docs we have18:08
@HeikoSarianepaola: but you should also put in the bibtex for the research paper mentioned18:09
@HeikoSinto the cookbook page18:09
@HeikoSarianepaola: see what I mean?18:09
arianepaolaHeikoS: yes18:09
@HeikoSarianepaola: cool looking forward to finally have a string kernel example in there18:10
arianepaolaHeikoS: if you check the ssk source, there is a documentation about ssk18:10
@HeikoSarianepaola: in the header you mean?18:10
arianepaolayes18:11
@HeikoS@brief18:11
@HeikoSyes18:11
@HeikoSyou can take the reference from there18:11
@HeikoSi.e. convert it into a bibtex18:11
@HeikoS(easy via google scholar, search for the paper name, change settings to include bibtex, download bibtex, polish it and put in the references.bib)18:11
@HeikoSarianepaola: you can recycle the math18:12
@HeikoSin the cookbook page18:12
arianepaolathere is no way to extract @brief and reuse it in the cookbook?18:12
arianepaolawith a macro18:12
OXPHOSHeikoS: worked. Thanks! curious why though. rebase worked for me last time18:12
@HeikoSOXPHOS: maybe had to be git pull --rebase?18:13
@HeikoSbut if you confuse git, it will strike back and confuse you in turn ;)18:13
OXPHOSHeikoS: this is so true..18:13
@HeikoSOXPHOS: I got confused many times in the past ;)18:15
@HeikoSarianepaola: it is not really the point18:15
@HeikoSarianepaola: @brief is an interface documentation of code18:15
@HeikoSarianepaola: cookbook is illustration how to use it18:15
@HeikoSso cookbook should be shorter18:15
arianepaolaHeikoS: I understand that. But @brief has more than just an interface documentation, as it explains the concept.18:16
arianepaolaHeikoS: just wondered if it was possible, otherwise you have the same documentation snippet in several places.18:16
@HeikoSarianepaola: mmh this is a very good point actually18:16
@HeikoSI understand now what you mean18:16
arianepaola:-)18:16
@HeikoSarianepaola: ok lets think18:17
@HeikoSthere is still a few differences:18:17
OXPHOSHeikoS: It was claimed that you don't have to know coding to use github. This is bs. github is a way harder language.18:17
@HeikoSsome class docs say things like "you can overload this method to do this"18:17
@HeikoSarianepaola: that shouldnt be in cookbook18:17
arianepaolayes18:17
@HeikoSarianepaola: so not sure how we would link them18:17
@HeikoSbut would be cool if the cookbook and @brief were in sync18:17
@HeikoSremoves some redundancy18:17
@HeikoSarianepaola: at least for intro of cookbook18:18
@HeikoSarianepaola: I guess we have to do this at a later point in time18:18
@HeikoSwiking, lisitsyn ^18:18
@HeikoSarianepaola: you want to avoid having to copy the math etc right?18:18
arianepaolaHeikoS: Another thing that I asked earlier was: HeikoS: is there any specific naming convention for the cookbook pages?18:19
@HeikoSit is a little nightmare to unify these things, but would be dead cool18:19
@HeikoSarianepaola: yeah18:19
@HeikoSarianepaola: follow the ones existing18:19
arianepaolae.g. short vs long names18:19
@HeikoSah yeah18:19
@HeikoSnot sure to be honest18:19
@HeikoSsuggestions?18:19
@HeikoSknn - k_nearest_neightboutrs18:20
@HeikoSneighbours18:20
arianepaolawould be better to use long names though18:20
@HeikoSssk - sub_sequence_kernel18:20
@HeikoSI agree18:20
@HeikoSplease do so18:20
arianepaolaand the modular part? classifier_ssk_modular.py18:20
@HeikoSI will update the existing ones18:20
@HeikoSno modular18:20
@HeikoSyours would be "kernel/sub_sequence_kernel"18:20
arianepaolaSo this one classifier_ssk_modular.py would go in: kernels/classifier_subsequence_kernel18:21
arianepaolaah kernel :-)18:21
arianepaolathanks!18:21
@HeikoSno classifier I think18:21
@HeikoSjust kernel18:21
arianepaolayeah, right :-)18:21
@HeikoSthink its better to keep this directory structure flat18:21
@HeikoSif we have like "classifier" for each folder, then it gets too messy18:22
@HeikoS(and the cmake hacks I wrote only support 1 sub-dir for now ;) )18:22
arianepaolaagree18:22
@HeikoSarianepaola: btw re your suggestion18:22
@HeikoSif you use the sgclass macro18:22
@HeikoSthere will at least be a link to the class docs18:22
@HeikoSwe can think about following that up a bit later18:22
@HeikoSreally good idea, I will think a bit more about it18:22
-!- sanuj [~sanuj@117.203.14.46] has joined #shogun18:25
arianepaola:-)18:26
sanujGandalfTheWizard, !!18:43
-!- leagoetz [~leagoetz@host-92-0-162-192.as43234.net] has quit []18:49
-!- mizari [~mizari@95-174-213-100.nts.su] has quit [Ping timeout: 246 seconds]18:50
@HeikoSarianepaola: did it work with the enum?18:53
arianepaolaHeikoS: I am getting there18:55
arianepaolaHeikoS: have just fixed an error with sphinx in cmake. the sphinx version from apt did not work, missing theme.conf and the version from pypi was not found by default, as it installs in /usr/local/bin18:56
@HeikoSarianepaola: cool, send a separate patch if you like18:57
arianepaolayes18:57
-!- OXPHOS [9d8b131c@gateway/web/freenode/ip.157.139.19.28] has quit [Ping timeout: 250 seconds]19:02
-!- OXPHOS [9d8b131c@gateway/web/freenode/ip.157.139.19.28] has joined #shogun19:05
arianepaolaHeikoS: https://github.com/shogun-toolbox/shogun/pull/323219:06
@HeikoSarianepaola: does this add the path to be searched19:08
@HeikoSor does it only search there?19:08
arianepaolait adds the path19:08
@HeikoSi.e. does it still work if there is nothing in local/bin19:08
@HeikoSok19:08
arianepaolacmake uses NO_DEFAULT_PATH19:08
arianepaolahttps://cmake.org/cmake/help/v3.0/command/find_program.html19:08
@HeikoSyep reading that19:08
arianepaolak19:08
@HeikoSthanks19:09
arianepaolanp19:09
@HeikoSill leave it to wiking to merge19:09
arianepaolaok19:09
@HeikoSOXPHOS: btw there is a preview for PRs19:13
@HeikoSwhere you can see the changed files19:13
@HeikoSgood to check that before sending -- all of us get an email for every PR19:13
arianepaolaHeikoS: I am looking at meta/generator Python code19:15
@HeikoSyep and?19:15
arianepaolaI ran types/get_type_list.sh to check that there is StringCharFeatures19:16
arianepaolathe only string reference in the Python code is StringLiteral19:16
arianepaolawhich does not work19:16
@HeikoSarianepaola: ah that is more of a hack19:16
@HeikoSyou can use the exact same class names as in the python example19:16
@HeikoSall classes should be available19:16
OXPHOSHeikoS: thx I pulled from Desktop. Because I did remove that file?! anyway I'll pull again.19:17
@HeikoSOXPHOS: the new PR is fine and clean19:17
@HeikoSnice work btw19:17
@HeikoSarianepaola: which example in python are you again looking at?19:17
arianepaolaHeikoS: I get: Exception: Failed to obtain include path for StringCharFeatures or CStringCharFeatures or StringCharFeatures or CStringCharFeatures19:17
OXPHOSHeikoS thanks19:17
@HeikoSarianepaola: can you paste the gist19:18
@HeikoSof the sg19:18
@HeikoSI can check then19:18
arianepaolahttps://github.com/shogun-toolbox/shogun/blob/develop/examples/undocumented/python_modular/classifier_ssk_modular.py19:18
@HeikoSand your .sg file19:19
shogun-buildbotbuild #2887 of bsd1 - libshogun is complete: Failure [failed configure]  Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2887  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, OXPHOS <engelzora@gmail.com>19:19
@HeikoSarianepaola: we need to find a way to load the string features from file using the shogun interface19:19
@HeikoSthe python example uses this "load_dna" thing19:20
@HeikoScant do that19:20
@HeikoSarianepaola: template<class ST> CStringFeatures<ST>::CStringFeatures(CFile* loader, EAlphabet alpha)19:20
@HeikoSthere is this19:21
@HeikoSso that can load directly from file19:21
@HeikoSbut your error is caused by something else19:21
@HeikoSbut need to check the .sg listing19:21
shogun-buildbotbuild #25 of xenial - libshogun is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/25  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, OXPHOS <engelzora@gmail.com>19:22
@HeikoSarianepaola: checking it out19:26
arianepaolaok thanks19:26
@HeikoSarianepaola: ok this is easy to solve19:29
@HeikoSIll send a fix19:29
@HeikoSbut then you'll run into another problem ;)19:29
arianepaolalol19:29
@HeikoSarianepaola: pushed to develop19:31
@HeikoSsee the change19:31
arianepaolaok, thanks19:31
@HeikoSill explain what was going on19:31
@HeikoSmeta examples have two cases:19:31
@HeikoS1.) swig interface (all languages but cpp)19:31
@HeikoS2.) c++ interface19:31
@HeikoSin 2.) some class names are different to 1.)19:32
@HeikoSas swig renames a few classes19:32
@HeikoSsee src/interface/modular/Features.i19:32
arianepaolaso you overwrite the naming and parameter, right?19:32
@HeikoSthis is why some classes when used for swig are called differently19:32
arianepaolainstead of default19:32
@HeikoSyes, so for the cpp meta language (that doesnt see shogun through swig), we need to use the real names19:32
@HeikoSand since there is no (easy) way to tell the c++ class name from the swig class name, we need to keep a list19:33
@HeikoSand put them in explicitly19:33
@HeikoSbut luckily this will be few cases only19:33
arianepaolagood19:33
@HeikoSarianepaola: so now you get some other errors, but they should be possible to fix19:33
arianepaolaok19:34
arianepaolathank you HeikoS19:34
-!- sanuj [~sanuj@117.203.14.46] has quit [Quit: Leaving]19:35
arianepaolaHeikoS: hmm, I don't get any errors?!?19:36
@HeikoSwhen compiling meta examples?19:36
arianepaolayes19:36
@HeikoSwell then19:38
@HeikoShappy days :)19:38
@HeikoSsend the PR19:38
arianepaolasure, just fixing some math formatting errors19:38
@HeikoScool19:38
@HeikoSyou will also have to send a data patch19:38
@HeikoSsee the cookbook readme page19:39
@HeikoSfor integration testing19:39
@HeikoSwe can merge the page once travis executed the corresponding integration test for your example19:39
@HeikoS(and the page looks good)19:39
@HeikoS(first meta example takes longest, should be copy/paste aftrer)19:39
arianepaolayeah19:39
arianepaolaI thought the data was already there, the files that I referenced from the example19:40
@HeikoSyeah those are19:41
@HeikoSthat is the training data19:41
@HeikoSbut the integration testing data, we generate newly atm for every cookbook page19:42
@HeikoSit is just the output of the executable19:42
@HeikoSand then we compare future commit's output against that reference that you added19:42
@HeikoSand the cool thing with the meta examples is: that we can do that in all target languages, so we are increasing test coverage quite a bit and implicitly testing the swig typemaps19:42
@HeikoSsince cpp is output without swig19:43
arianepaolathanks great19:43
arianepaola* that's great19:48
shogun-buildbotbuild #2888 of bsd1 - libshogun is complete: Failure [failed configure]  Build details are at http://buildbot.shogun-toolbox.org/builders/bsd1%20-%20libshogun/builds/2888  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>19:57
OXPHOSHey HeikoS, so for the name,19:59
shogun-buildbotbuild #26 of xenial - libshogun is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/xenial%20-%20libshogun/builds/26  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>19:59
OXPHOSQuadratic_discriminant_analysis19:59
OXPHOSor  quadratic_discriminant_analysis19:59
@HeikoSOXPHOS: yeah sorry about not mentioning this earlier, but I only just realised with arianepaola19:59
@HeikoSlower case19:59
@HeikoSthe second19:59
OXPHOSokay np19:59
@HeikoSthanks!19:59
@HeikoSunforutnately19:59
@HeikoSthe data also needs to be changed19:59
@HeikoSargh ;)20:00
@HeikoSmy bad20:00
OXPHOSbut hierarchical and kmeans are fine ;)20:00
@HeikoSOXPHOS: yes they are20:02
@HeikoSphew20:02
shogun-buildbotbuild #16 of FC23 - libshogun - aarch64 is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/16  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>, OXPHOS <engelzora@gmail.com>20:03
@HeikoSarianepaola: still no error in the cookbook?20:13
arianepaolanope20:13
arianepaolaHeikoS: I think it is useful to add links to the research papers, when they are available online20:20
arianepaolaIn the bibtex I add:   url={http://www.jmlr.org/papers/volume2/lodhi02a/lodhi02a.pdf}20:20
arianepaolaThis generates: [LSST+02]H. Lodhi, C. Saunders, J. Shawe-Taylor, C. Cristianini, and C. Wathins. Text classification using string kernels. The Journal of Machine Learning Research, 2:419–444, 2002. URL: http://www.jmlr.org/papers/volume2/lodhi02a/lodhi02a.pdf.20:21
@HeikoSarianepaola: I dont like that too much too be honest20:21
@HeikoSthese links tend to outdate20:21
@HeikoSrather one can just use Google20:21
arianepaolahmm, ok20:21
@HeikoSalso it clutters the bibtex20:21
arianepaolaI'll remove it then20:21
@HeikoScool thanks20:21
shogun-buildbotbuild #17 of FC23 - libshogun - aarch64 is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/FC23%20-%20libshogun%20-%20aarch64/builds/17  blamelist: Heiko Strathmann <heiko.strathmann@gmail.com>20:41
arianepaolaHeikoS: Do you mean overflow for strlen. It should be >= 0 in https://github.com/shogun-toolbox/shogun/pull/321520:47
@HeikoSyes20:47
@HeikoSso if it is very large20:48
@HeikoSand then casted to signed20:48
@HeikoSthe result might be negative20:48
@HeikoSarianepaola: see what I mean?20:49
arianepaolayes20:50
arianepaolaI suggested to change the type of vlen, but wiking and lisitsyn didn't like it20:50
@HeikoSno20:51
@HeikoSthat will break tons of things20:51
@HeikoSyou can just do what I said20:51
@HeikoSi.e. cast vlen to unsigned20:51
@HeikoSthat is safe20:51
@HeikoSbut you have to asser >=0 first20:52
@HeikoSI mean it is just a small detail really, but since I spotted it, we should do it properly20:52
arianepaolayes20:53
@HeikoSarianepaola: just saw your cookbook page21:06
@HeikoSarianepaola: let me know if my comments make sense to you21:07
arianepaolaHeikoS: thanks, will check21:08
@HeikoSarianepaola: btw how are the debian packages going?21:12
@HeikoSwiking: holidays?21:23
@HeikoS :)21:23
-!- HeikoS [~heiko@host-92-0-162-192.as43234.net] has quit [Quit: Leaving.]22:00
-!- OXPHOS [9d8b131c@gateway/web/freenode/ip.157.139.19.28] has quit [Quit: Page closed]22:29
--- Log closed Tue May 31 00:00:19 2016
« Sunday, 2016-05-29 Index Tuesday, 2016-05-31 »

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