travis-ci shogun-buildbot shogun-buildbot --- Log opened Thu Sep 19 00:00:05 2013 -!- FSCV [~FSCV@66.187.72.219] has quit [Quit: Leaving] 00:08 -!- travis-ci [~travis-ci@ec2-54-211-149-66.compute-1.amazonaws.com] has joined #shogun 00:20 [travis-ci] it's Soeren Sonnenburg's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11524217 00:20 -!- travis-ci [~travis-ci@ec2-54-211-149-66.compute-1.amazonaws.com] has left #shogun [] 00:20 -!- zxtx [~zv@ip-64-134-197-118.public.wayport.net] has joined #shogun 00:28 build #14 of clang34 - static analysis is complete: Failure [failed analyse]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20static%20analysis/builds/14  blamelist: Soeren Sonnenburg 00:31 -!- Heiko__ [05453f70@gateway/web/freenode/ip.5.69.63.112] has quit [Quit: Page closed] 00:44 -!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] 01:08 -!- zxtx [~zv@ip-64-134-197-118.public.wayport.net] has quit [Ping timeout: 260 seconds] 01:12 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has joined #shogun 01:45 -!- van51 [~van51@athedsl-409794.home.otenet.gr] has quit [Quit: Leaving.] 02:48 -!- sonne|osx_ [~sonne@f053037175.adsl.alicedsl.de] has joined #shogun 03:05 build #471 of nightly_none is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_none/builds/471 03:06 -!- sonne|osx [~sonne@f053045233.adsl.alicedsl.de] has quit [Ping timeout: 240 seconds] 03:07 -!- sonne|osx_ is now known as sonne|osx 03:07 build #459 of nightly_all is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_all/builds/459 03:32 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has quit [Ping timeout: 248 seconds] 04:53 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has joined #shogun 04:58 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has quit [Ping timeout: 248 seconds] 05:29 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has joined #shogun 05:44 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has quit [Ping timeout: 245 seconds] 06:01 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has joined #shogun 06:04 -!- foulwall [~zhengyang@114.255.40.22] has joined #shogun 06:27 -!- sonne|osx [~sonne@f053037175.adsl.alicedsl.de] has quit [Quit: sonne|osx] 06:27 -!- zxtx [~zv@ip-64-134-197-230.public.wayport.net] has quit [Ping timeout: 248 seconds] 07:11 -!- zxtx [~zv@149.165.244.2] has joined #shogun 07:23 -!- besser82 [~besser82@fedora/besser82] has joined #shogun 07:38 -!- lisitsyn [~lisitsyn@fb2-lo1.global63.net] has quit [Quit: Leaving.] 08:07 -!- hushell [~hushell@c-98-232-178-161.hsd1.or.comcast.net] has joined #shogun 08:13 -!- sonne|osx [~sonne@82.113.98.161] has joined #shogun 08:17 shogun-buildbot: force build --branch=develop "rpm1 - libshogun" 08:21 build #1214 forced 08:21 I'll give a shout when the build finishes 08:21 shogun-buildbot: force build --branch=develop "precise - libshogun" 08:22 build forced [ETA 22m07s] 08:22 I'll give a shout when the build finishes 08:22 -!- sonne|osx [~sonne@82.113.98.161] has quit [Quit: sonne|osx] 08:35 build #100 of precise - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/precise%20-%20libshogun/builds/100 08:38 mmm we have a lot of problems with the 2 nightly targets :S 08:38 build #1214 of rpm1 - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/rpm1%20-%20libshogun/builds/1214 08:53 wiking: 08:57 (09:35:38 PM) sonney2k: wiking, you might consider turning of integration tests on travis IMHO it is sufficient that we run this stuff on the buildbots 08:57 (09:36:21 PM) sonney2k: wiking, I mean travis is good to get a rough picture if sth works and IMHO that is all we want at this stage leaving detailed analysis to our own BB 08:57 ^ any comments? 08:58 wiking: /home/buildslave/nightly_none/build/tests/unit/multiclass/BaggingMachine_unittest.cc:25:45: error: 'class testing::NiceMock' has no member named 'ref' 08:59 on nightly_none 08:59 yesh i know 08:59 sonne|work: we can disable integration on travis 08:59 that's no problem 08:59 currently i'm fixing leaking errors 09:01 i think parametermap or such is leaking... 09:15 -!- iglesiasg [~iglesias@2001:6b0:1:1da0:247a:a4fc:e52c:8133] has joined #shogun 09:53 -!- mode/#shogun [+o iglesiasg] by ChanServ 09:53 good morning! 09:54 -!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun 09:59 iglesiasg: yo! need help 10:06 iglesiasg: StructuredLabels unit test is indicating a lot of leak :( 10:06 wiking, shit, that's not good 10:06 wiking, do you have the trace? 10:06 iglesiasg: i can copy paste 10:06 just a sec 10:06 iglesiasg: http://pastebin.com/38xthspY 10:08 wiking, there are also leaks coming from init_shogun_with_defaults 10:09 wiking, that looks weird to me 10:09 iglesiasg: dont bother about that now 10:09 iglesiasg: the lot of parameterMap leak is just beecause there's an objec that is not being freed 10:10 wiking, it leaks because the test is wrong I think 10:12 wiking, https://github.com/shogun-toolbox/shogun/blob/develop/tests/unit/labels/StructuredLabels_unittest.cc#L49 10:12 get_label returns an object, but there is no SG_UNREF for them 10:12 iglesiasg: fixit plz 10:13 if u can 10:13 thoralf, ^ 10:13 wiking, yep. I can do that right away 10:13 cool thanx heaps 10:13 wiking, will need to compile in debug to double check so will take a few minutes.. 10:16 yeps... thnx 10:16 i'm fixing some other shit in the meanwhile.. 10:16 wiking, it should be ready, https://github.com/shogun-toolbox/shogun/pull/1627 10:29 ok i'll pull it 10:30 as soon as i have my fixes ready as well 10:30 all right 10:30 thnx a lot 10:30 no fixes were left here 10:30 so actually the ones in init_shogun_with_defaults were also a consequence of these ones 10:31 sorry 10:31 no leaks were left 10:31 wiking: I have a hard time debugging cyg1 and other crashers. Could we when a test fails run it with the gdb script in .src again to get a proper bt? 10:41 sonne|work: i'll try to see if i can do something about that 10:42 speaking of which, cygwin is still the only way to install Shogun in Windows, right? 10:47 wiking, sonne|work, ^ 10:47 iglesiasg: yes 10:48 just wondering since, IIRC, something was mentioned about native in Windows when CMake came 10:48 iglesiasg: yeah sth to work on for later 10:48 I see 10:48 iglesiasg: i've applied some changes 10:49 iglesiasg: but more changes are required 10:49 as msvc is not the most standard compiler 10:49 aham I understand 10:50 ok finally i've found a valgrind error that is actually not a unit test leak but an actual leak in the library 10:50 wiking, StructuredLabels still? 10:53 iglesiasg: nono 10:53 when's the meeting today? 10:56 in one hour approx 10:56 iglesiasg: we need some massive thing about __declspec(dllexport) 10:58 lisitsyn, for windows stuff? 10:59 iglesiasg: yeah 10:59 lisitsyn, I see 10:59 iglesiasg: otherwise all the classes are not visible from .dll 10:59 I am not really interested in it myself (windows sucks :P) but I know a guy who might be interested 10:59 I don't know if there is other way 10:59 although it is true that this could help us reach more people 11:00 no we definitely need it 11:00 I don't want to hear about windows sucks ;) 11:00 lisitsyn, lol why not? 11:00 just like most of the people are idiots but we have to live with that 11:01 it is the truth!! 11:01 hahaha 11:01 the majority is windows so we have to live with that too 11:01 I like your pragmatism 11:01 lisitsyn: MALSAR_unittest is leaking like hell :) 11:03 lisitsyn, anyway, our users are developers as well. Do you still think windows is majority in that too? 11:03 lisitsyn, I would say people using mac os or linux for development are more 11:04 iglesiasg: well remember that detection thing I develop 11:04 yep 11:04 windows people the ones who ask you to do that? 11:04 iglesiasg: it has to be windows 11:05 but run on windows or actually developed in windows? 11:05 wiking: I am deeply saddened ;) 11:05 iglesiasg: well run on windows 11:05 lisitsyn, that's another thing :) 11:05 iglesiasg: of course I can train on linux 11:05 lisitsyn: plz fix 11:05 wiking: yeah let me try 11:05 ok here's the first batch of leakfixes 11:10 -!- shogun-notifier- [~irker@7nn.de] has joined #shogun 11:11 shogun: Viktor Gal :develop * c6f4ca7 / tests/unit/ (8 files): https://github.com/shogun-toolbox/shogun/commit/c6f4ca70c5cc278e15218f326355f8bf5d264b8b 11:11 shogun: Fix a bunch of leaking unit tests 11:11 shogun: and disable ref calls when reference counting is disabled 11:11 shogun-buildbot: force build --branch=develop 'nightly_none' 11:12 build #472 forced 11:12 I'll give a shout when the build finishes 11:12 i'd be great to have an xslt for this: http://buildbot.shogun-toolbox.org/memcheck/20130918-2140/DynamicAnalysis.xml 11:14 -!- foulwall [~zhengyang@114.255.40.22] has quit [Quit: leaving] 11:14 build #472 of nightly_none is complete: Failure [failed compile]  Build details are at http://buildbot.shogun-toolbox.org/builders/nightly_none/builds/472 11:15 -!- foulwall [~AndChat63@219.141.159.37] has joined #shogun 11:16 iglesiasg: Thanks for fixing. 11:20 shogun: Fernando Iglesias :develop * d952d8d / tests/unit/labels/StructuredLabels_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/d952d8de4e0901f974a0c7ec3f3e84107cffe853 11:20 shogun: Fix memory leaks and indentation in structured labels unit test 11:20 shogun: Viktor Gal :develop * 8567050 / tests/unit/labels/StructuredLabels_unittest.cc: https://github.com/shogun-toolbox/shogun/commit/85670508cde6cae94502602773e1d0415818753c 11:20 shogun: Merge pull request #1627 from iglesias/fix/leak/structured_labels_unittest 11:20 shogun: 11:20 shogun: Fix memory leaks and indentation in structured labels unit test 11:20 wiking: https://github.com/lisitsyn/tapkee/blob/master/test/valgrind_tests_transformer.xslt I had sth like that 11:20 iglesiasg: After having a working regression test, I "forgot" to care about SG_UNREF ;) 11:20 thoralf, no problem 11:20 hehe 11:20 -!- foulwall [~AndChat63@219.141.159.37] has quit [Read error: Connection reset by peer] 11:23 build #2213 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2213  blamelist: Viktor Gal , Fernando Iglesias 11:24 wiking: 11:27 -- Looking for clapack_dpotrf in ATLAS_LAPACK-NOTFOUND 11:27 CMake Error: The following variables are used in this project, but they are set to NOTFOUND. 11:27 Please set them or make sure they are set and tested correctly in the CMake files: 11:27 ATLAS_LAPACK 11:27 linked by target "cmTryCompileExec2966321832" in directory /home/blackburn/shogun/build/CMakeFiles/CMakeTmp 11:27 wiking: have you seen that before? 11:27 no not really 11:36 shogun-buildbot: force build --branch=develop 'deb1 - libshogun' 11:36 build #2214 forced 11:36 I'll give a shout when the build finishes 11:36 build #2214 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2214 11:38 -!- az_de [82954e22@gateway/web/freenode/ip.130.149.78.34] has joined #shogun 11:38 -!- foulwall [~AndChat63@219.141.159.55] has joined #shogun 11:39 -!- thoralf [~thoralf@enki.zib.de] has quit [Quit: Konversation terminated!] 11:46 hey 12:00 yo 12:00 anybody? :D 12:01 sonne|work: around? 12:01 Hello everyone 12:01 meeting time? 12:01 yeah 12:01 ok 12:01 what's the agenda for today? 12:02 hello hello 12:02 okay so agenda would be gsoc + release 12:02 it sounds good 12:02 az_de: around? 12:02 * az_de yes, hi all 12:03 good 12:03 it is not good time for kevin so good you are here 12:03 gsomix: foulwall: Soeren guys around? 12:03 lisitsyn: hi there 12:03 oh sorry I forgot to send a reminder 12:03 -!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun 12:04 okay nice votjakovr here 12:04 -!- thoralf [~thoralf@enki.zib.de] has joined #shogun 12:04 lets wait for a bit, Heiko should be here iirc 12:04 -!- Cheng [73406f11@gateway/web/freenode/ip.115.64.111.17] has joined #shogun 12:04 hey Cheng thanks for joining us 12:04 Hi all 12:04 -!- HeikoS [05453f70@gateway/web/freenode/ip.5.69.63.112] has joined #shogun 12:05 hi Cheng! 12:05 hello! sorry for the delay 12:05 hi 12:05 HeikoS: no problem we didn't start anyway 12:05 :) 12:05 hi everyone! 12:05 so who's missing? Georg said he will be a bit late 12:05 and Kevin probably can't make it 12:05 HeikoS: lambday is out too, right? 12:06 lisitsyn: I dont know, we did not say anything 12:06 let me check my mail 12:06 apologies but I will be offline for ~10 mins - please proceed without me 12:06 sonne|work: alright 12:06 gsomix: foulwall: 2nd ping ;) 12:07 lisitsyn, pong. 12:07 good 12:07 pong 12:07 okay then lets start 12:07 forgot about the meeting. 12:07 I didn't prepare detailed agenda actually 12:07 but you all know that we have pencil down approacing 12:08 -!- lambday [67157e37@gateway/web/freenode/ip.103.21.126.55] has joined #shogun 12:08 and actually soft pencil down is here 12:08 oh lambday good time to join us 12:08 ;) 12:08 lisitsyn: sorry I am late 12:08 lambday: hi! 12:08 HeikoS: hey :) 12:08 np lambday thanks for joining us 12:08 lets just talk about each project one by one 12:09 and discuss what have been done 12:09 and what's you think is still left 12:09 are you ok with that? 12:09 sure 12:09 and may be anybody would like to start? 12:09 we can 12:10 lambday: ? 12:10 just one note 12:10 please put an url with ipython notebook 12:10 all the parts of my project has been implemented... 12:10 build #1497 of cyg1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/cyg1%20-%20libshogun/builds/1497  blamelist: Viktor Gal 12:10 ok let me go ahead 12:10 (for some reason, I have very large  lag, so I might take time to answer -- sorry for that) 12:11 so Rahul started with the notebook, the current draft is here 12:11 http://nbviewer.ipython.org/6591994 12:11 Still some things to improve and polish, but its getting there 12:11 currently there is an issue wie complex typemaps in swig, which he is at 12:11 python examples not there yet but are easy since its all subparts from the notebook 12:11 lambday: what about the code-review? did you have a look on that? 12:12 HeikoS: I just sent an email to lambday about the peer review 12:12 HeikoS: hushell mailed me regarding this... I am thinking of doing it in the weekends 12:12 lambday: hi! 12:12 Will something in the dev meeting be non-gsoc related? 12:12 hushell: hey man! just replied to your mail.. :) 12:13 thoralf: on the end 12:13 thoralf: yeah lets talk about it in the end 12:13 thoralf: so take a coffee ;) 12:13 lambday: good! :) 12:13 -!- georg_zeller [~zeller@embln.embl.de] has joined #shogun 12:13 ok so code review will be done last minute then, guys, please start this before Saturday and take some time 12:13 hey georg_zeller 12:13 hi and sorry for being late 12:13 Okay, please ping me.  My IRC client will wake me up then. :) 12:13 HeikoS: do we have review pairs anywhere around? 12:14 lambday is reviewing ..? 12:14 HeikoS: alright... 12:14 people should know, if not I will find them 12:14 lisitsyn: hushell's 12:14 good 12:14 -!- sonne|osx [~sonne@89.204.135.239] has joined #shogun 12:14 foulwall, you are my couple :P 12:14 okay so lambday 12:14 van51 - gsomix hushell - lambday pickle27 - votjakovr iglesias - foulwall 12:14 what you might need to focus is polish ipython notebook 12:15 I don't think we need python examples right now 12:15 lambday: and ping me on the end of the meeting plz 12:15 ehm the split is not the minus but the " " 12:15 lisitsyn: yes alright 12:15 wiking: sure :) 12:15 yeah notebook is more important than python examples 12:15 notebook seems more important 12:15 notebook++ 12:15 but writing examples takes just one day both libshogun and notebook, 12:15 plus python examples won't take more than an hour once the notebook is complete 12:16 well 12:16 you need to resolve that issue 12:16 @all In general, notebooks should explain more things, keep in mind that people reading them might not know what you are doing there :) 12:16 and code-review 12:16 yep agreed! 12:16 lisitsyn: yeah swig is giving me real pain :'( 12:16 lambday: just ask here 12:17 lambday: if this takes too long, maybe postpone the complex stuff and finish the rest first 12:17 lambday: we've fixed the other day the bug we had with r modular... 12:17 lambday: its just a small subset of your work 12:17 wiking: alright let's talk about the swig thing in the end.. 12:17 okay so lambday's project is ok and we have no real issues 12:17 HeikoS: yeah, so kept that part incomplete 12:17 HeikoS: want to continue with votjakovr? 12:17 wiking: I saw, it was renaming afterall :) 12:17 lambday: just tell us what you need / where you have issues 12:17 lisitsyn: okay 12:18 lambday: we will get it to work. 12:18 lambday: indeed... but lets get back to this on the end... anyhow great work! 12:18 sonne|osx: I need to make our complex64_t (std::complex) with numpy complex64 12:18 wiking: thanks man :) 12:18 votjakovr: wanna give us the update? 12:18 lambday: should be easy to do :D 12:18 lambday: but later 12:19 alright 12:19 :) 12:19 oh okay so while votjakovr is afk 12:19 may be anyone else? 12:19 votjakovr: sleeping? :) 12:19 ok so let me do this 12:19 HeikoS: oops :) 12:19 -!- travis-ci [~travis-ci@ec2-54-211-149-66.compute-1.amazonaws.com] has joined #shogun 12:19 [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11543121 12:19 -!- travis-ci [~travis-ci@ec2-54-211-149-66.compute-1.amazonaws.com] has left #shogun [] 12:19 ah 12:19 okay we have some issues here right? 12:20 HeikoS: unfortunately my project is not ready yet :( 12:20 model selection part is missing 12:20 votjakovr: and apart from that? 12:20 what can we do about it? 12:21 votjakovr: if you cannot finish this today or latest tomorrow, we will have to postpone it, it is very important that the notebook and some examples are there 12:21 i'll finish it today 12:21 I mean, already you can write about a lot of things 12:21 HeikoS, votjakovr but the rest is 'done' ? 12:21 sonne|osx: yeah if this is merged, the we can polish 12:21 lets just decide what is missing to consider the project done 12:22 sonne|osx: wanted to do multiclass originally, but that is unrealistic 12:22 lisitsyn: model-selection, notebook, examples 12:22 can we postpone examples already? 12:22 lisitsyn: they are last in the list, but I think they are important 12:22 but the notebook is more important 12:23 HeikoS: can you write some GP notebooks? IIRC you already have them? 12:23 I mean we are in difficult timing 12:23 votjakovr: in fact I cal help you a bit with the notebook if that helps 12:23 then votjakovr can do the MS 12:23 sonne|osx: yeah, I will start one, good idea 12:23 votjakovr: do you agree? then you can focus on the modelselection 12:23 votjakovr: but please hurry up a lot! 12:23 and code-review 12:23 votjakovr: but the agreement would be that you finish it then 12:24 -!- georg_zeller [~zeller@embln.embl.de] has quit [Quit: georg_zeller] 12:24 actually i've started both of them (model-selection, notebook, examples) 12:24 -!- georg_zeller [~zeller@embln.embl.de] has joined #shogun 12:24 votjakovr: pls send your current state then and I will push them a bit ok? 12:24 votjakovr: lets just assure both HeikoS and you mean the same 12:24 votjakovr: so what do you think? When would you finish? 12:24 HeikoS: i'll try to finish model-selection today 12:25 what's the worst estimate of that part? 12:25 votjakovr: that would be good! 12:25 examples toaday/tomorrow 12:25 HeikoS: notebook takes some time 12:25 HeikoS: I would suggest to see how it goes. If MS is done by today or tomorrow everything will work out nicely. 12:25 votjakovr: notebook first! Then examples 12:26 examples are just demonstration of API 12:26 votjakovr: so do notebooks a little later and make sure to get some help from HeikoS 12:26 notebook is about what is going on behind the scences and what you have done 12:26 yep, votjakovr pls send the current state or even send a PR with it, then I can update 12:26 as soon as you have *something* for the notebook, send it 12:27 HeikoS: ok, i'll send little bit later 12:27 alright so I guess that's the plan? 12:27 looks like 12:27 votjakovr: please keep HeikoS updated 12:27 lisitsyn: ok 12:27 so if you are stucked or whatever just let him know 12:28 lisitsyn: next? 12:28 I am ready! 12:28 iglesiasg: georg_zeller please continue then 12:28 (it is already 12:28) 12:28 There is this notebook http://nbviewer.ipython.org/6576096 that currently contains a first section explaining what LMNN is, and a second section applying LMNN to a real data set illustrating feature selection. 12:28 I have to update this second part to use Shogun's cross validation, instead of how it is done at the moment. 12:28 apologies, but I have to be AFK for a while, checking back later for non-gsoc stuff 12:28 sonne|osx: yeah lets speed up 12:28 Then, I will add another data set in the notebook where I am comparing kNN and LMNN in a multiclass classification task. I will normalize and whiten the data for this task as we have discussed in the mailing list. 12:28 your notebook looks quite nice 12:29 lisitsyn, thanks! :) I think it is pretty standard actually 12:29 georg_zeller: iglesiasg: so what do you think is left here? 12:29 oh! And I forgot, I have to update the first part too, showing some neighbourhood graphs 12:29 we have discussed a few updates on the notebook but those shouldn't take long 12:30 so basically updating the notebook with these things 12:30 ah yes 12:30 apart from that? 12:30 maybe 2-3 days I'd guess? 12:30 yep 12:30 georg_zeller, yes, I think so 12:30 then I have to do the code review 12:30 so nothing apart from that right? 12:30 coding is done as far as I'm concerned 12:30 yeah and the code review 12:30 but I hope to get that done in the weekend 12:30 what about examples? 12:30 iglesiasg: I wouldn't mind if you could help us get the release polished 12:30 okay it looks healthy 12:30 georg_zeller, there are libshogun, octave and python examples already 12:31 iglesiasg: you know website, bug and warning fixes etc 12:31 just simple stuff 12:31 helping out others... 12:31 then fine from my side 12:31 sonne|osx: none gsoc related stuff on the end of this meeting :) 12:31 sonne|osx, sure, once GSoC period is over I get to this 12:31 okay thanks a lot georg_zeller iglesiasg your project is totally ok :) 12:31 sonne|osx: unfortunately lisitsyn mixed up monthly dev meeting with gsoc meeting... 12:31 next is sonne|work foulwall Cheng please 12:31 haha 12:31 we talked about a few interesting applications but better for the time after GSoC 12:32 well 12:32 georg_zeller: yeah it would be nice if you continue 12:32 wiking: no no that is not the issue 12:32 would be great for me too 12:32 wiking: lets say I combined meetings 12:32 wiking: we need help to get the release done so anyone who thinks his project looks OK please help us! 12:32 otherwise I seem stupid hahah 12:33 lisitsyn: no u fucked it up ;) 12:33 wiking: tss 12:33 okey continue 12:33 sonne|osx: foulwall bitte 12:33 Cheng foulwall sonne|osx 12:33 speed it up 12:33 the demo site just looks great 12:33 foulwall: ? 12:33 gsomix? 12:33 yep 12:33 sonne|osx: gsomix bitte 12:34 gsomix: let's talk 12:34 OK. I'm finishing protobuf support for shogun right now. For now we have simple IO system (buffered read, simple parsers), csv and libsvm readers. 12:34 schnelle 12:34 :D 12:34 After protobuf related works I'll start writing ipython notebook and nice examples. 12:34 Unfortunately support of matlab m-files and some exotic' formats is out of schedule (*bad guy). But I'll stay with shogun for this work. 12:34 So whoever has a nice idea for a demo ask foulwall 12:34 Maybe Soeren want to say sth? 12:34 what's left with protobuf? 12:34 e.g. van51's language detection demo is now in applications 12:34 gsomix: yeah those io stuff are great you've implemented... looking forward for the protobuf stuff 12:34 http://foulwall.com:8000/application/language_detect/ 12:35 if you can think of sth like this for your project and need help from foulwall now would be the time 12:35 sonne|osx: what's left? 12:35 well the splice site recognizer/arts detector 12:35 but it really is us that need to do sth here IMHO 12:36 that's out, right? 12:36 lisitsyn, today I'll send PR with skeleton. And then we need and protobuf support for many data types... 12:36 *need to add 12:36 gsomix: ok fine! 12:36 I will try to do the CR quickly 12:36 * gsomix *sigh* 12:36 I guess everyone here hopes that you can continue with .mat files 12:36 Hi, all . sorry for the network  latency 12:36 even psot gsoc 12:37 post 12:37 sonne|osx, of course. 12:37 gsomix: cvmatio... have you seen that project? 12:37 wiking, nope. we plan use some code from octave. 12:38 gsomix: it's binding opencv stuff to .mat files: github.com/hbristow/cvmatio 12:38 but OK, I'll see. thanks! 12:38 gsomix: it's pretty small and easy to port stuff that you need from there 12:38 gsomix: that would work as well. but it is matlab 5 only not the new 7.3 format 12:38 -!- foulwall [~AndChat63@219.141.159.55] has quit [Quit: Bye] 12:38 wiking: makes sense to use it! 12:38 sonne|osx: works fine with newer matlab stuff 12:38 cool! 12:38 sonne|osx: i mean the one i've sent just now 12:38 wiking: most certainly not! 12:39 -!- foulwall [~AndChat63@219.141.159.55] has joined #shogun 12:39 sonne|osx: ? 12:39 sonne|osx: i'm using matlab 2012 and i can read those .mat files with this :) 12:39 wiking: ohh you are right 12:39 man I had an old link 12:39 https://github.com/hbristow/cvmatio 12:40 sonne|osx: CVMATIO implements a Matlab v7 MAT file parser in C++, giving users the 12:40 gsomix: this will be of great help then 12:40 ability to interact with binary MAT files in their own projects. 12:40 ;) 12:40 Hi all, sorry  for the  disconnect. 12:40 so go ahead and have fun with it... 12:40 gsomix: so it would be quite easy to do 12:40 sonney2k, wiking is super hero for me now. :) 12:40 alright then gsomix finish protobuf and then you can try this out :) 12:40 foulwall: we are listening 12:40 :> 12:40 foulwall: great work with the demo site! 12:41 gsomix: alright then that's about it 12:41 next ? 12:41 or foulwall ? 12:41 Hi , I'm vsry happy to implement your ideas 12:41 sorry back 12:41 what's left in foulwall's project? 12:42 wiking: I still think it cannot do v7.3 (or4) formats though. But matlab 5 to 7.2 will work 12:43 gsomix: https://github.com/hbristow/cvmatio/blob/master/src/MatlabIO.cpp - really very little code 12:43 I'll make a more detailed docs, and make a python interface for the mldata hdf5 downloader 12:43 sonne|osx: it has some extra stuff in the .hpp... but yeah it's about 2000 LoC 12:43 tops 12:43 I'll make them before  the hard deadline 12:43 sonne|osx: well it's better than nothing... :P 12:44 foulwall: I think the only missing stuff is arts/asp ? 12:44 HeikoS: do we have any example which we want to work interactively still? 12:45 Oh. forget to merge that, arts and asp pr will  send today 12:45 sonne|osx: HeikoS is away atm 12:45 foulwall: ahh and the tree clustering but you have the tree already there so all that is needed is to call the hierarchical clustering example 12:46 wiking: we should put foulwalls demos live on fatbot 12:46 wiking: can we use your image for that? 12:46 sonne|osx: why on fatbot? 12:46 sonne|osx: but of course we can use that image... 12:46 sonne|osx: i'll setup up the startup script for that image today... along with the ipython notebook server 12:47 okay so 12:47 and then i can add as well the demo site 12:47 wiking: fast/big machine! 12:47 sonne|osx: okok no worries 12:47 can we proceed? 12:47 lisitsyn: yes who's next? 12:47 az_de: pickle27 and me 12:47 okey 12:47 maybe I can continue 12:48 az_de: around 12:48 ? 12:48 ok,  I just repeat Kevins e-mail, that 12:48 az_de: yeah thanks 12:48 "all of his code for GSoC has been pushed to GitHub and merged! " 12:48 az_de: we don't have anything left in the project, right? 12:48 foulwall: send a PR with arts/asp and then I show you how it works with hierarchical clustering (tomorrow?) 12:48 az_de: that says everything great work from kevin! :) 12:48 like his ipython notebook etc 12:49 and he is now waiting for feedback ... 12:49 az_de: in this case can you involve him to help us with the release? 12:49 ok sonne|work 12:49 yeah, The ipy-notebooks he created are really cool :-) 12:49 no he is not available I am afraid 12:49 sonne|osx: i think he doesn't have much time (see that preproc email he sent) 12:49 he wanted to go travelling once gsoc is done 12:49 az_de: it will take time to get feedback from users. not before 3.0 I think 12:49 ahh too bad 12:49 ok, I see 12:49 but anyhow kevin did a great job 12:50 he's gsoc project is totally ok 12:50 imo 12:50 All in all, we have a very successful GSoC project! 12:50 yes, great work from all of you! 12:50 There was a discussion to have some explicit whitening function in shogun, which is currently only implicitly done in some ICA methods. 12:50 az_de: somebody will take over... 12:50 Kevin has no time to do this now, but it should be straightforward . 12:50 thanx a lot again! 12:50 next? 12:50 yeah would be a nice preprocesor 12:50 lisitsyn: who's next? 12:51 so thanks az_de for doing a great job with mentoring :) 12:51 :-) 12:51 az_de: yeah really super job ! 12:51 re 12:51 HeikoS: great 12:51 wiking: thats only hushell left I think 12:51 hushell: heheeey 12:51 how's factor graph going along? 12:51 i see a lot of discussion about the notebook 12:52 what are the remaining issues? 12:52 ok, let me start it 12:52 not much to say, I have finished basic coding, the ipynb is here: http://nbviewer.ipython.org/6579472 12:52 build #15 of clang34 - static analysis is complete: Failure [failed analyse]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20static%20analysis/builds/15  blamelist: Viktor Gal 12:53 hushell: ok cool how's the state of the notebook 12:54 what left is some comparison between SOSVM solvers, I decided to postpone the BCFW solver after gsoc, before the hard deadline, I'll compare the primalMosek, dualBMRM and the coming subgradient descent one 12:54 when is it going to be finalized? 12:54 hushell: incredibly cool! 12:54 hushell, btw, did you implement SGD finally? 12:54 hushell: u should try various solvers of dualBMRM 12:54 hushell: as there are  4 of them 12:54 iglesiasg: I have finished the code of SGD 12:54 hushell, I thought you mentioned something about it in one of the discussions in github, but maybe I missed the PR 12:54 hushell: it's easy to do that you just give the right enum for that... 12:55 wiking: yeah, good idea, then there are more things to compare 12:55 hushell: so very well done 12:55 hushell: go with the dualBMRM variants first... as mosek for example is not for free.. 12:55 hushell: and as well we can see how the various dualBMRM behaves with a real world problem... 12:55 yeah all non-free code is not so useful :/ 12:56 and we cannot automatically test it 12:56 I found mosek stuff are not in documentation 12:56 so it will die out anyway 12:56 hushell: yeah as it's non-free.... as said go with the other SOSVM solvers... 12:56 sonne|osx: what was that with the interactive stuff? (later) 12:56 wiking: okay, then compare all BRRM solvers and the SGD 12:57 hushell: which is at the moment DUALBMRM... CCSOSVM needs more love from svmlight ;P 12:57 hushell: yeah that's good 12:57 I'll done this by Friday 12:57 ok any more issues? 12:57 or stuff you need help with? 12:57 yeah, one question about the dataset 12:57 HeikoS: anything we still want to run interactively as demo that would need extra work from foulwall? 12:57 hushell: shoot go ahead 12:58 -!- travis-ci [~travis-ci@ec2-54-211-149-66.compute-1.amazonaws.com] has joined #shogun 12:58 [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11543374 12:58 -!- travis-ci [~travis-ci@ec2-54-211-149-66.compute-1.amazonaws.com] has left #shogun [] 12:58 sonne|osx: ah ok, not really for GSoC, but I hope it is easy to do these things myself and its well documented 12:58 okay so gsoc-wise 12:58 hushell, if you can use matlab, basically load the .mat file and then use the function csvwrite 12:58 so if I have a cell array, is it possible to write it into csv, and read it from python? 12:58 sonne|osx: wanted to pimp the GP stuff a bit mote (regression heatmaps etc) but no time for this now 12:58 hushell: csv u can read from python w/o any problems... 12:58 HeikoS: well you should do it now - otherwise you will never realize if there are issues 12:58 hushell, what do you have in this cell array? 12:59 just a general thing - don't bother with code submission for now 12:59 build #8 of debian wheezy - memcheck is complete: Failure [failed memory check]  Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/8  blamelist: Viktor Gal 12:59 iglesiasg: just many matrix 12:59 HeikoS: I mean realize when it is too late 12:59 sonne|osx: the model-selection is not yet working, but it would be nice to have a button for that 12:59 hushell: I sse 12:59 * see 12:59 sonne|osx: thats why I say that there should be very good documentation of the framework 12:59 each cell is a matrix 12:59 hushell: maybe we just use a .mat for now 12:59 sonne|osx: if no-one can edit it after foulwall left, that is of no use 12:59 sonne|osx: hushell go with .mat 12:59 so with scipy.io is more convenient 13:00 sonne|osx: and i'll help out gsomix with .mat reader 13:00 I don't think it is easy to convert that to a .csv 13:00 it needs a custom format 13:00 sonne|osx: I will have a look later, its not possible before GSoC end since I want to help my guys a bit 13:00 hushell: don't waste time on this... 13:00 sonne|osx: yeah, because I don't want to do the train-test splitting in python 13:00 -!- thoralf [~thoralf@enki.zib.de] has quit [Ping timeout: 260 seconds] 13:00 okay 13:00 I'll try to improve the dataset later 13:00 hushell: that is the easiest. anyway just do it 13:00 HeikoS , I'll be very careful writting the docs of the framework 13:01 with .mat 13:01 just go ahead with .mat and forget the rest... 13:01 cool 13:01 hushell: thanks a lot for the work 13:01 great job! 13:01 so we are done with the summaries? 13:01 any of the gsoc students... if u get stuck plz mail shout here on irc for help 13:01 dont waste your time on things alone.. 13:01 Thanks for help! :) 13:02 *all* ^ 13:02 cool 13:02 sonne|osx: yeah it's done 13:02 So Sept 16 is suggested pencils down 13:02 err 23 13:02 no 13:02 no 23 is hard 13:02 sept 16 is suggested 13:02 23 is hard 13:02 (well the hard pencils down) 13:02 ok 13:02 we will then again have to do evaluations 13:02 and we should do them by Sept 25 13:02 students bug your mentors! 13:03 otherwise no money ;) 13:03 alright date's saved 13:03 yeah and HeikoS no summit :D 13:03 build #3 of clang34 - undefined behaviour analysis is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20undefined%20behaviour%20analysis/builds/3  blamelist: Viktor Gal 13:03 haha 13:03 HeikoS, bug mentors you too!! 13:03 :D 13:03 haha 13:04 can do :) 13:04 Alright - this cannot be repeated often enough. 13:04 We are very happy with the work you have all done. 13:04 And it would be very cool if you would stay post-gsoc and do cool stuff with shogun. 13:04 great work from everyone indeed! Impressive! 13:05 Yep, incredible results this GSoC. I am sure this will get a lot of attention 13:05 As you know we plan to do a shogun 3.0.0 release just once gsoc is over. 13:05 yeah plz hang around 13:05 So all your work will be included and it would be *great* if you could help us fixing bugs/warnings etc. 13:05 very exciting about the 3.0.0! 13:05 ookeeeeey 13:06 We probably will need a bugfix/cleanup release 3.1 later this year where we fix all problem that people found in 3.0 :) 13:06 I was in a similar situation last year, I cannot do other thing but strongly suggest to keep hanging around and collaborate 13:06 3.0.0 time! 13:06 it cannot be of course at the same pace as during GSoC, we all have other stuff to do 13:06 agenda for that 13:06 Indeed 3.0 means we doubled shoguns code base since the last release 13:06 1) fix fix fix 13:06 2) fix fix fix 13:06 :> 13:06 haa this means 13:06 okey so i need some help here 13:06 with 5.0 we will get unmaintanable software 13:06 :D 13:06 we are having a lot of leaks atm 13:06 and a lot of undefined behaviour and a lot of static analyzer errors 13:07 so 13:07 lisitsyn: you forget wikings botnet 13:07 now comes the prio list 13:07 wiking: maybe we should have a  priority list for the release? checking all issues for 3.0 13:07 yeah as wiking said 13:07 sonne|osx: well it could help to ddos NSA 13:07 but not to maintain ;) 13:07 so HeikoS it's just coming now 13:07 http://buildbot.shogun-toolbox.org/waterfall 13:07 so please pleaaaase 13:07 before doing anything try to do this first 13:07 fix bugs / failed tests 13:07 warnings 13:07 HeikoS: for you the list: 13:07 for example these http://buildbot.shogun-toolbox.org/static_analysis/2013-09-19-1/ 13:08 ok so first goal is to make all builds work I guess? 13:08 -!- lambday [67157e37@gateway/web/freenode/ip.103.21.126.55] has quit [Ping timeout: 250 seconds] 13:09 yeah we are trying this for a long time already... 13:09 but we are close (only cygwin fails basically) 13:09 HeikoS: https://github.com/shogun-toolbox/shogun/issues/1574 evaluation_cross_validation_multiclass base_load_file_parameters /base_load_all_file_parameters  aaadn all the errors at static analyzer for  /base /Parameter.cpp (http://buildbot.shogun-toolbox.org/static_analysis/2013-09-19-1/) 13:09 HeikoS: and there are some stuff about *MMD 13:09 and the rest is bugs from static analyzers 13:09 sonne|osx: yeah not always... 13:10 wiking, alright I will do those things first and then get back 13:10 sonne|osx: most of the time actually not but 13:10 even if it is 13:10 one can silence it with an assert or so 13:10 well stuff we should fix but it is not prime importance 13:10 iglesiasg: then I'll have another PR for the SGD on Friday, will you have time to have a short review? 13:10 HeikoS: especially start with https://github.com/shogun-toolbox/shogun/issues/1574 13:10 hushell, sure, thank you! 13:10 i need this fixed asap... 13:10 we need debian packages, besser82 is doing fedora packages 13:10 kk 13:10 and we need osx packages 13:11 etc 13:11 HeikoS: because it silences other warnings from other modules... 13:11 *all* of you 13:11 ah thats annoying 13:11 who are doing release: http://buildbot.shogun-toolbox.org/static_analysis/2013-09-19-1/ 13:11 sonne|osx: aren't you the DD? ;) 13:11 try to at least silence it 13:11 iglesiasg: as Patrick suggested, I should check the objective values, slack variables, should be the same as his matlab code, this will take a lot of time 13:11 lisitsyn: i gave you your malsar leaks... 13:11 lisitsyn: and fix the dead assigments here: http://buildbot.shogun-toolbox.org/static_analysis/2013-09-19-1/ 13:12 hushell, I understand 13:12 wiking: yeah ok 13:12 for all the tapkee related stuff 13:12 iglesiasg: but I'll try to fix this by Friday, we have code review in the weekend 13:12 besser82: sure sure but I need to sleep too :D 13:12 iglesiasg: thanks in advance! :) 13:12 hushell, but it will be so cool to have SGD working good 13:12 sonne|osx: sleep? Who needs that? :-P 13:12 current leak xml: http://buildbot.shogun-toolbox.org/memcheck/20130919-1012/DynamicAnalysis.xml 13:12 iglesiasg: I still hope you do the website so we have the notebooks displayed nicely 13:13 basically there's still tons of them there 13:13 iglesiasg: I hope it will be faster then cutting plane methods ;) 13:13 sonne|osx, yes 13:13 and HeikoS you wanted to change the README to be a .md and to have links between the READMEs and INSTALLs 13:13 okay, I'll go back to sleep, thank you guys! 13:13 sonne|osx: yes I still want to 13:13 hushell: vojtech's stuff is *very* fast so I am not so convinced 13:13 sonne|osx, after what wiking has done with the cloud notebooks thing it should be easy, right? 13:14 hushell: thanks everyone for attending 13:14 iglesiasg: unrelated 13:14 hushell: and good night 13:14 goood night :) 13:14 iglesiasg: we need to statically show the notebooks somewhere... 13:14 hushell: gnite 13:14 iglesiasg: ? 13:14 have a great day! guys! 13:14 ok, then I misunderstood the cloud notebooks thing 13:14 iglesiasg: indeed 13:14 wiking: ? 13:14 iglesiasg: cloud is for editing playing 13:14 wiking, oh yeah sure 13:15 well that is totally unrelated 13:15 iglesiasg: but we still need to be able to statically show somewhere the notebooks 13:15 fuck I am stupid 13:15 iglesiasg: we need to display examples on the website 13:15 iglesiasg: and it should be able to handle shogun versions 13:15 interactive demoing is sth else 13:15 and we have 2 parts here 13:15 the cloud thing is to offer like an ipython session with shogun installed 13:15 1) is foulwalls demos / apps 13:15 iglesiasg: say "notebooks generated" with shogun 3.0, and then latest and shogun 2.0 13:15 and such... 13:15 wiking, aham I see 13:15 2) is the notebooks ... 13:15 so there are a couple of parts here 13:16 one is to generate the notebooks periodically 13:16 yes 13:16 iglesiasg: that's done 13:16 save them 13:16 guys I have to get back to work 13:16 and put a link to them in the webpage 13:16 cu all 13:16 iglesiasg: it needs integration now to the webpage 13:16 -!- sonne|osx [~sonne@89.204.135.239] has quit [Quit: sonne|osx] 13:16 sonne|osx, bye! 13:16 iglesiasg: #1482 13:16 #1481 13:16 #1483 13:16 that's all the issues here 13:17 wiking, mm ok. Is it basically about putting links in the webpage then? 13:17 the notebook generation (nightly) is done 13:17 build #8 of clang34 - thread analysis is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20thread%20analysis/builds/8  blamelist: Viktor Gal 13:17 that sounds tooe asy 13:17 iglesiasg: well not only that... somehow we need to be able to switch between versions 13:17 ok 13:17 so like shogun 3.0 13:17 and latest 13:17 for the same notebook 13:17 but first things first probably 13:17 -!- lambday [67157c37@gateway/web/freenode/ip.103.21.124.55] has joined #shogun 13:17 first I will aim at showing them generated for one version, and then extend to several versions 13:18 wiking, what do you think? 13:18 lambday: i've told you about your mem leak.... please fix that before doing anything else... 13:18 wiking: working on that 13:18 iglesiasg: i'm fine with that 13:18 lambday: great thnx... 13:18 sorry I got dc.. power cut :( 13:18 HeikoS: i think you should just define -1 enums and then do a default: SG_ERROR(...) in the switch branch :P 13:18 wiking, where are they currently generated? the notebooks 13:19 iglesiasg: jsut a sec... 13:19 the idea is that they are in doc/notebooks/* without the output saved 13:19 -!- Cheng [73406f11@gateway/web/freenode/ip.115.64.111.17] has quit [Quit: Page closed] 13:19 so that must be executed somewhere and saved somewhere else with the output 13:20 wiking, sure 13:20 iglesiasg: the nightly_default does that 13:20 i mean execution + saving 13:20 ok 13:20 wiking, one question about the visualization 13:20 iglesiasg: http://buildbot.shogun-toolbox.org/builders/nightly_default/builds/528/steps/notebooks/logs/stdio 13:21 i'm just checking where's the output 13:21 wiking, are we aiming at basically providing links to nbviewer.ipython.org/**** 13:21 iglesiasg: no... we need to save it locally 13:21 i mean the full thing 13:21 or have it rendered somewhere else? 13:21 it should be fully rendered on our webpage imho 13:21 wiking, yep sure, the full thing saved locally 13:21 mmm I see rendered on our webpage 13:22 it is indeed possible because basically the notebook runs on the browser always 13:22 iglesiasg: on shogun website server the notebooks are here 13:23 /home/shogun/static/notebook 13:23 i dont know if this helps you at allk 13:23 iglesiasg: indeed it does: http://shogun-toolbox.org/static/notebook/ecg_sep.ipynb 13:23 cool! 13:24 that helps me for sure 13:24 iglesiasg: so all the notebooks are saved there 13:24 but these are the nightlies 13:24 wiking, but there is no output in there I think 13:25 iglesiasg: http://shogun-toolbox.org/static/notebook/gp_regression.ipynb 13:25 this one does for sure 13:25 if u think one is missing something 13:25 let me know 13:25 and i'll fix it 13:25 wiking, yep 13:25 that one has the output 13:25 so first I have to investigate how to render the notebook from those urls 13:26 lcoally 13:26 if u need anything for that let me know 13:26 i think we'll just do something like 13:26 static/notebook/{RELEASE_VERSION}/ 13:26 for the various verions of the notebooks 13:26 make sense 13:26 and then we'll do somehow the switching from the website... 13:27 sorry, what did you mean with locally just a minute ago? 13:27 iglesiasg: basically what we need is somehow list the availble notebooks on the website 13:27 and the somehow to be able to render it 13:27 agree 13:27 list is easy from django i guess 13:27 ls directory 13:28 :P 13:28 ok i have to take a break now 13:28 ppl r waiting for me on lunch 13:28 hehe yeah 13:28 bbl 13:28 let me know if u need something from me 13:28 I will work on the rendering thing 13:28 iglesiasg: afaik there's a python script for this ;) 13:28 iglesiasg: within ipython notebook 13:28 it makes total sense 13:28 I was going to say that 13:28 wiking: what is with videos? 13:28 that I bet there is something already to do it :) 13:29 workshop videos I mean? 13:29 iglesiasg: even more... i know that there is an option of ipython to generate a html 13:29 yeeah videoooooz 13:29 iglesiasg: please find out the command and i'll make sure that you get the htmls generated 13:29 sonne|work: will be ready by 3.0 13:29 :D 13:29 bbl ... ppl r really waiting for me.. 13:29 wiking: ohoh 13:29 what? 13:30 plz be fast 13:30 wiking, ok 13:30 go and have lunck now! 13:30 :D 13:30 iglesiasg: "These notebooks are normal files that can be shared with colleagues, converted to other formats such as HTML or PDF, etc" 13:30 so we can even put the pdf somewhere... 13:30 anyhow as u can see it's possible 13:30 it's just one ipython notebook command 13:30 yep 13:31 so the idea is to the nightly doing that as well, right? 13:31 generating the HTML 13:31 -!- thoralf [~thoralf@enki.zib.de] has joined #shogun 13:34 Hey. 13:34 lunch time for me too, see you later 13:35 -!- foulwall [~AndChat63@219.141.159.55] has quit [Remote host closed the connection] 13:35 -!- georg_zeller [~zeller@embln.embl.de] has left #shogun [] 13:37 -!- az_de [82954e22@gateway/web/freenode/ip.130.149.78.34] has quit [] 13:37 -!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: Good bye!] 13:53 back 13:56 wiking, found it out already 13:57 I just tested it 13:57 wiking, ipython nbconvert notebook.ipynb 13:57 iglesiasg: what's the command? 13:57 you get a nice notebook.html 13:57 iglesiasg: it'll be the same name? 13:57 pandoc, among maybe other stuff, is required 13:57 yes 13:57 iglesiasg: i guess it's not a selfcontained html 13:58 but there are like images saved next to it 13:58 or? 13:58 why not? 13:58 I just opened it in my browser 13:58 images and everything 13:58 yes, self contained 13:58 iglesiasg: well i thought that the images are not within the html 13:58 so they are in the html encoded base64? 13:58 wiking, they are encoded in the html 13:59 no idea what base 13:59 lol 13:59 ok lemme test it 13:59 hehehe 14:00 PandocMissing: The command 'pandoc -f markdown -t html --mathjax' returned an error: [Errno 2] No such file or directory. 14:00 told ya 14:00 wiking, http://pastebin.com/yVdJ7mb3 14:00 wiking, tell me the base :P 14:00 okey 14:00 base64 14:00 14:30 before comitting python examples you might have to update data 14:30 HeikoS: alright 14:30 checking 14:30 wiking: once changed it stops executing, right? 14:31 lisitsyn: yes it's like a breakpoint 14:32 lisitsyn: and it'll as well show the new and old value 14:32 and of course you have the whole backtrace 14:32 wiking: awesome 14:33 yeah it's basic 14:33 ;) 14:33 i just keep forgeting the syntax 14:33 and i just just really start using lldb 14:33 it's really fucking great in some aspects 14:33 build #2216 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2216  blamelist: Heiko Strathmann 14:34 wiking: Using lldb?  How?  Tell me. :) 14:38 thoralf: http://lldb.llvm.org/lldb-gdb.html 14:39 wiking: All the great things you found out. 14:39 shogun: Viktor Gal :develop * 12d3062 / src/shogun/machine/LinearMulticlassMachine.h: https://github.com/shogun-toolbox/shogun/commit/12d3062d2d7e9ba3ad334f74f952ce62b31c031b 14:39 ok here's another leakfix 14:39 shogun: Fix memory leak in LinearMulticlassMachine issue #1629 14:39 Oh no.  No read-the-fine-manual again. ;) 14:39 build #2217 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2217  blamelist: Viktor Gal 14:44 oh fucking libshogun-library_mldatahdf5 14:44 build #2215 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2215  blamelist: lambday 14:47 mmm 14:48 this seems to be an old PR 14:48 shogun-buildbot: force build --branch=develop 'deb1 - libshogun' 14:48 build #2218 forced 14:48 I'll give a shout when the build finishes 14:48 build #2218 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2218 14:49 HeikoS: awesome for that PR 14:49 HeikoS: can we merge? 14:50 wiking: I cannot run tests locally here since computer from 2006 14:50 or even older ;) 14:50 oookeey 14:50 so have a look at travis and feel free to merge if it passes, which I think it should 14:50 i'll see how the first 2 jobs are runnning on travis 14:50 if they pass i'll merge 14:50 lambday: your merged PR does not cause trouble on travis, its other people's stuff 14:55 HeikoS: yes... I guess earlier issue was because shogun-data PR was not merged till then 14:56 lambday: I see, good! 15:02 HeikoS: sending the PR for memory leak then 15:02 lambday: i'll check it now 15:03 wiking: thanks :) 15:05 shogun: lambday :develop * 72cbe0e / src/shogun/mathematics/linalg/linop/SparseMatrixOperator.h: https://github.com/shogun-toolbox/shogun/commit/72cbe0e60b1dc580c9f1103883b6dde5e0fa0562 15:05 shogun: SparseMatrixOperator memory leak fixed 15:05 shogun: Heiko Strathmann :develop * 6e8ab05 / src/shogun/mathematics/linalg/linop/SparseMatrixOperator.h: https://github.com/shogun-toolbox/shogun/commit/6e8ab05a25626208cb7aa53a9066082e2bc73795 15:05 shogun: Merge pull request #1632 from lambday/feature/log_determinant 15:05 shogun: 15:05 shogun: SparseMatrixOperator memory leak fixed 15:05 HeikoS: the fix was indeed weird 15:06 yeah, thanks for figuring that out! :) 15:06 still no idea why it was :-/ 15:06 yeah num_feature/num_vector thing was there but I was testing with a square matrix :-/ 15:07 it didn't cause anything 15:07 lambday: ok great thnx for the fix! 15:09 and the fast response 15:09 wiking: np :) 15:09 wiking: now you have to help me out with swig :( 15:10 ah cool 15:10 lemme check what could be done about that one 15:10 basically u need a numpy complex64 <-> complex64_t mapping? 15:10 right? 15:10 wiking: yes... I want to use ComplexSparseMatrixOperator, which is CSparseMatrixOperator 15:11 and SGVector 15:11 okeeey 15:11 wiking: I didn't add this template to modular interface yet, tried with a few things but didn't work 15:11 i'll check how it can be done 15:11 wiking: thanks man 15:12 where's complex64_t defined? 15:16 ok i see 15:17 -!- van51 [~van51@athedsl-409794.home.otenet.gr] has quit [Quit: Leaving.] 15:18 lambday: ok so you need numpy mapping for SGVector 15:20 and what else? 15:21 wiking: and numpy mapping for csc_matrix(complex64) for SGSparseMatrix 15:22 ok so u need SGSparseMatrix as well 15:23 yes 15:23 maybe SGMatrix ? 15:23 wiking: good to have, not urgent 15:23 for my examples I'll just be using sparse ones 15:24 ok i'm testing now 15:26 wiking: unit tests segfault. I have to look into this a bit later today 15:33 going now,see you! 15:33 wiking: thanks 15:36 HeikoS: I'll update notebook once complex things work.. see you 15:37 -!- HeikoS [05453f70@gateway/web/freenode/ip.5.69.63.112] has quit [Ping timeout: 250 seconds] 15:38 -!- travis-ci [~travis-ci@ec2-54-234-21-192.compute-1.amazonaws.com] has joined #shogun 15:51 [travis-ci] it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11549043 15:51 -!- travis-ci [~travis-ci@ec2-54-234-21-192.compute-1.amazonaws.com] has left #shogun [] 15:51 build #1498 of cyg1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/cyg1%20-%20libshogun/builds/1498  blamelist: lambday 15:55 -!- lambday_ [67157d37@gateway/web/freenode/ip.103.21.125.55] has joined #shogun 15:56 -!- lambday [67157c37@gateway/web/freenode/ip.103.21.124.55] has quit [Ping timeout: 250 seconds] 15:59 sonne|work: is there a reason why we dont use https://github.com/numpy/numpy/blob/master/doc/swig/numpy.i 16:00 ? 16:00 wiking: yes it does not support all the data types shogun has - however we started with this one 16:01 :( 16:01 sonne|work: i've just realised that we are actually copying back and forth values of an array 16:01 i.e. sgvector -> numpy 16:02 that's not a mapping but a copy 16:02 or am i reading it wrong? 16:02 wiking: yes we copy 16:02 in all interfaces! 16:02 sonne|work: isn't there a way to actually just map the memory? 16:02 or the memory is totally differnt 16:03 hence we cannot do it 16:03 i mean the structure 16:03 wiking: shogun objects destroy memory currently 16:03 free I mean 16:03 sonne|work: mmm you mean reference is not being hold on the modular interface side 16:03 right? 16:03 yes 16:05 and we free memory too 16:06 mmmm 16:06 how can a create a RealVector from numpy array? 16:06 you cannot 16:06 ohh no I think you can 16:06 wiking: x=RealVector() 16:07 x.set(numpy.array([1.,2,3])) 16:07 ok 16:08 doh 16:08 SGVector a; 16:09 a[0] should work 16:09 right? 16:09 wiking: no idea 16:10 lambday_: ? 16:10 wiking: with float64_t for sure 16:10 sonne|work:  yeah i know 16:11 In [34]: x = ComplexVector() 16:11 In [36]: a = np.array([1, 2, 3], dtype=complex) 16:11 In [37]: x.set(a) 16:11 this works 16:11 but Idon't see why not 16:11 but then i try 16:11 x[0] 16:11 In [38]: x[0] 16:11 --------------------------------------------------------------------------- 16:11 TypeError                                 Traceback (most recent call last) 16:11 in () 16:11 ----> 1 x[0] 16:11 TypeError: 'ComplexVector' object does not support indexing 16:12 ahh from python 16:12 yeah 16:12 no I guess it needs the inplace stuff gsomix was doing 16:12 but there's a lot of things like this in SGVector.cpp 16:12 define COMPLEX64_ERROR_NOARG(function) \ 16:12 SG_SERROR("SGVector::%s():: Not supported for complex64_t\n",\ 16:12 ... 16:12 wiking: yeah sure. some ops cannot be sanely defined with complex 16:13 ok i think i need more work on this 16:13 but then again 16:13 wiking: min/max/sort won't work with complex 16:13 thoralf: i've seen that 16:13 [40]: x.vector 16:13 Out[40]: *' at 0x113039500> 16:13 In [41]: x.vlen 16:13 Out[41]: 3 16:13 so it seems it is set... 16:14 i need extra stuff in the typemapping 16:14 i guess.. 16:14 but basic stuff is already ready : 16:14 :P 16:14 wiking: hey 16:15 lambday_: so you were working with SGVector right? 16:16 wiking: didn't try with vector, was trying with sparse matrix 16:16 but I need vector too 16:16 -!- travis-ci [~travis-ci@ec2-23-22-84-25.compute-1.amazonaws.com] has joined #shogun 16:36 [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11549375 16:36 -!- travis-ci [~travis-ci@ec2-23-22-84-25.compute-1.amazonaws.com] has left #shogun [] 16:36 build #1499 of cyg1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/cyg1%20-%20libshogun/builds/1499  blamelist: Heiko Strathmann 16:40 build #16 of clang34 - static analysis is complete: Failure [failed analyse]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20static%20analysis/builds/16  blamelist: Heiko Strathmann 16:44 -!- gsomix [~gsomix@188.168.3.89] has quit [Quit: Leaving] 16:52 build #4 of clang34 - undefined behaviour analysis is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20undefined%20behaviour%20analysis/builds/4  blamelist: Heiko Strathmann , lambday 16:55 lambday_, 6909 haha 16:55 build #9 of debian wheezy - memcheck is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/debian%20wheezy%20-%20memcheck/builds/9 16:56 lambday_, I had noticed heavensdevil, but didn't the numbers :P 16:56 -!- lambday_ [67157d37@gateway/web/freenode/ip.103.21.125.55] has quit [Ping timeout: 250 seconds] 17:00 -!- lambday [67157c37@gateway/web/freenode/ip.103.21.124.55] has joined #shogun 17:01 iglesiasg: :D 17:01 I am going for dinner.. will come back in a while 17:01 see you :) 17:01 -!- lambday [67157c37@gateway/web/freenode/ip.103.21.124.55] has quit [Client Quit] 17:01 -!- van51 [~van51@athedsl-409794.home.otenet.gr] has joined #shogun 17:02 -!- HeikoS [~heiko@nat-163-62.internal.eduroam.ucl.ac.uk] has joined #shogun 17:04 -!- mode/#shogun [+o HeikoS] by ChanServ 17:04 -!- lisitsyn [~lisitsin@mxs.kg.ru] has quit [Quit: Leaving.] 17:07 wiking: wow by changes screw up a lot of things ;) 17:09 build #9 of clang34 - thread analysis is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20thread%20analysis/builds/9  blamelist: Heiko Strathmann , lambday 17:10 wiking: fixed 17:29 tests pass locally 17:29 iglesiasg: your lmnn modular integration tests fail on my machine still 17:36 HeikoS, :O 17:36 HeikoS, latest develop? 17:37 iglesiasg: yes, have you tested? 17:37 yes, I did test yesterday when I pushed it 17:37 shogun: Heiko Strathmann :develop * c5dbcb4 / src/shogun/ (5 files): https://github.com/shogun-toolbox/shogun/commit/c5dbcb49d370180f24441d2e2367d0dccd74b9b9 17:37 shogun: -added UNDEFINED cases in data type enums to avoid unpredictable behaviour. 17:37 shogun: -Many switch statements did not handle default, fixed that 17:37 shogun: Heiko Strathmann :develop * 26f4434 / src/shogun/base/ParameterMap.cpp: https://github.com/shogun-toolbox/shogun/commit/26f443420c27a8b8a3d0d57e19ed7a5d3553e5d9 17:37 shogun: -added enums to is_empty check of SGParamInfo 17:37 shogun: -to_string method now doesnt throw an error when data types are undefined 17:37 shogun: Heiko Strathmann :develop * 1983e95 / src/shogun/lib/DataType.cpp: https://github.com/shogun-toolbox/shogun/commit/1983e95f954c8e2475fd60eaef6e12692d44b9b2 17:37 shogun: same as before 17:37 shogun: Heiko Strathmann :develop * 5a300a2 / src/shogun/ (5 files): https://github.com/shogun-toolbox/shogun/commit/5a300a210492f1f11152abb50e3158a0e5afe3c0 17:37 it was fine both here and in travis 17:37 shogun: Merge pull request #1631 from karlnapf/develop 17:37 shogun: 17:37 shogun: added undefined cases for data type enums 17:37 wiking: ^ 17:37 iglesiasg: mmm 17:37 iglesiasg: maybe my data is out of date, let me check 17:37 HeikoS,  ok, let me know please 17:41 iglesiasg: nope, passes, forry! 17:42 HeikoS, :) 17:42 -!- zxtx [~zv@149.165.244.2] has quit [Ping timeout: 268 seconds] 17:50 build #17 of clang34 - static analysis is complete: Failure [failed analyse]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20static%20analysis/builds/17  blamelist: lambday 17:54 HeikoS: you got my messages from yesterday? 17:55 besser82: nope sorry, computer crashed in the evening, 17:55 besser82: ah I read the cmake thing 17:55 besser82: you are going to refactor it? 17:55 whats the motivation? 17:55 -!- FSCV [~FSCV@66.187.72.219] has joined #shogun 17:56 HeikoS: some quirks on different dists with different modules 17:56 I see 17:56 besser82: cool! 17:56 HeikoS: your ATLAS-problem btw is caused by a mad atlas-lib your admin brewed :( 17:56 I like that  alot 17:56 besser82: I got around that finally 17:56 besser82: using your classic thing to compile shogun 17:56 HeikoS: allright 17:56 and virtual-env'ed a custom python 17:57 with all modules 17:57 HeikoS: :D 17:57 I need 17:57 annoying, but ok, at least I am independent now 17:57 besser82: btw stripping libshogun did not make eigen3's sparse module faster 17:59 They actually know that they are half as fast as choldmo 18:00 d 18:00 But installing cholmod on this cluster seems impossible, tried many things and failed. The admins also dont get it working 18:00 so I guess i have to rely on eigen3 18:00 good thing is: it now finally works! :LD 18:00 HeikoS: Sry, then. but was worth a try :) 18:00 build #2222 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2222  blamelist: Heiko Strathmann 18:01 build #2221 of deb1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2221  blamelist: Heiko Strathmann 18:02 -!- FSCV [~FSCV@66.187.72.219] has quit [Ping timeout: 264 seconds] 18:09 -!- foulwall [~zhengyang@114.255.40.22] has joined #shogun 18:17 ping iglesiasg 18:17 foulwall, hey 18:17 foulwall, how are you doing? 18:17 iglesiasg: :) 18:17 foulwall: hey! 18:19 Hey van51 :) 18:20 foulwall: could you take a look at : https://github.com/shogun-toolbox/shogun-demo/pull/24 18:20 foulwall: I'd like to know if you have a problem with disabling autoescaping 18:20 foulwall, so would it be good for you if I put all my code in another repository? 18:20 foulwall, I think it is probably the most comfortable option 18:20 foulwall, and you can just write comments in github 18:20 ok van51 18:21 foulwall, so I am checking how comments work in github 18:22 I have read hushell's suggestion, and it's ok without creating another repo. 18:22 foulwall, it seems it is not possible to comment in files directly but one can comment in commits 18:22 Ah, I forgot the comment, 18:23 foulwall, so I can do several commits 18:23 foulwall, one for the method itself, another one for some examples, etc 18:23 foulwall, and you can put particular stuff associated with a line or a method, etc in there 18:23 foulwall, and more general stuff in an issue 18:23 Thank you:) 18:23 That's pretty convenient 18:24 foulwall, awesome then 18:24 foulwall, are you ok doing the same for me to review your code? 18:24 I'll have a look at your repo. and for me, just shogun-demo? Maybe tommorrow after I update the asp/arts demo and hierarchical clustering example? 18:25 foulwall, aaah true you worked in another repo 18:26 foulwall, so basically all your work is in shogun-demo? 18:26 or should I also look at something in shogun-web? 18:26 iglesiasg: All my work are wiped out from shogun-web 18:27 foulwall, mmm ok. I understand I just have to look at shogun-demo then 18:28 -!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun 18:28 foulwall, I will prepare the repository then either tomorrow or on Saturday morning 18:30 foulwall, so that is like between 1 day or 1.5 days from now 18:30 foulwall, is that good for you? 18:30 -!- foulwall [~user@114.255.40.14] has joined #shogun 18:31 -!- thoralf [~thoralf@enki.zib.de] has quit [Quit: Konversation terminated!] 18:32 -!- foulwall [~zhengyang@114.255.40.22] has quit [Ping timeout: 245 seconds] 18:34 iglesiasg: okey:) and I've done a simple mldata.org downloader. That's quite a toy stuff. Here is a commit: https://github.com/shogun-toolbox/shogun/commit/cd9c28e592a8b660d48f8120f3ce6aa2456e2dcf 18:35 sorry for the network crashed again. 18:35 ok 18:36 no problem 18:36 -!- foulwall [~user@114.255.40.14] has quit [Remote host closed the connection] 18:37 -!- foulwall [~user@greymonkey.im] has joined #shogun 18:37 HeikoS: hi! build_parameter_dictionary() method of SGObject now add to dictionary only exact parameters (not SGObject). GradientEvaluation class builds dictionary of current differentiable function lazy. Is it ok? 18:37 votjakovr: hi! 18:37 votjakovr: is that in your fork? 18:38 votjakovr: can I have a look? 18:38 votjakovr: seems ok to me, but I am not sure whether I completely get it 18:38 HeikoS: sorry not yet 18:38 so let me understand that 18:38 before, the parameter dictionary contained SGOjects? 18:39 and now? 18:39 lazy dictionary building is OK 18:39 HeikoS: contains only exact paramters 18:39 what does that mean? 18:39 votjakovr: are there any problems that this might cause? 18:40 HeikoS: for example "sigma" of GaussianLikelihood 18:40 ah 18:40 yeah thats ok for now 18:40 votjakovr: I mean if it causes problems we have to think about it more carefully later 18:40 but now the focus is on finishing 18:40 so just go ahead with this 18:40 how much longer do you think it will take? 18:40 HeikoS: yeah, i'll try to send PR tonight, late tomorrow morning 18:41 votjakovr: ok 18:41 votjakovr: what about your notebook 18:41 van51: :) Thanks for the pr. Sorry for I forget considering escape:( And I'll comment to ask sonney2k to merge. 18:41 votjakovr: I will be away on saturday/sunday, so tomorrow is the last chance to get help from me 18:41 I can still review though 18:42 but not edit myself 18:42 HeikoS: ok, thank you very much :) 18:42 votjakovr: so when will you send your draft? 18:42 foulwall: ok, thanks! I'm glad that you approve :) 18:43 van51: Seen your notes, I'll add a section of the class/author link. 18:43 HeikoS: i'll try to do it tomorrow 18:43 foulwall: cool! although it doesn't have to be something separate, it can be linked directly from where they are mentioned in the about 18:45 van51: I'll to add some style on that, so plain there 18:46 iglesiasg: okey:) 18:47 foulwall: ok then, as you wish :) 18:47 foulwall, so it is set now, I hope it is not a pain to review my code :P 18:47 haha iglesiasg 18:49 -!- foulwall` [~user@greymonkey.im] has quit [Remote host closed the connection] 18:50 votjakovr: okay, thanks. Pls keep in mind that it is fine to send a PR with an incomplete draft. I will just extend it during the day tomorrow. 18:50 HeikoS:yes 18:57 HeikoS: did u fix it? 18:57 wiking: ehm yes what? 18:57 wiking: yes all fine now 18:57 HeikoS: ok cool 18:57 wiking: waiting for travis, but I checked everything locally and merged so that you can go on 18:57 ah travis says green 18:57 for libshogun 18:57 modular red 18:57 let's see 18:58 but these are other people errors 18:58 shogun-buildbot: force build --branch=develop 'deb1 - libshogun' 18:58 build #2223 forced 18:58 I'll give a shout when the build finishes 18:58 iglesiasg: lmnn failed, but I think thats fine since he fixed it 18:58 wiking: quite tricky to change the stuff 18:58 as I did some very nasty things with these uninitialised values 18:58 HeikoS, what's up with LMNN? Why does it fail still? 18:58 but now better, just had to handle them in switch 18:59 iglesiasg: I dont know, it passed locally here 18:59 HeikoS: now i'll be interested in the undefined behaviour bot's output 18:59 build #2223 of deb1 - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb1%20-%20libshogun/builds/2223 18:59 shogun-buildbot: force build --branch=develop 'clang34 - undefined behaviour analysis' 19:00 build #5 forced 19:00 I'll give a shout when the build finishes 19:00 gsomix ? 19:00 -!- travis-ci [~travis-ci@ec2-54-234-21-192.compute-1.amazonaws.com] has joined #shogun 19:07 [travis-ci] it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11550387 19:07 -!- travis-ci [~travis-ci@ec2-54-234-21-192.compute-1.amazonaws.com] has left #shogun [] 19:07 build #5 of clang34 - undefined behaviour analysis is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20undefined%20behaviour%20analysis/builds/5 19:09 ok i think i have ComplexVector working :) 19:10 HeikoS: do u know what else do we need? 19:10 iglesiasg: you have some weird stuff going on in lmnn 19:10 although that's more like an eigen problem as well 19:11 but worth to investigate 19:11 wiking, mmm 19:11 wiking, there is nothing random AFAIK 19:11 wiking, what does it happen with the test? It fails sometimes and sometimes not? 19:11 nono 19:11 iglesiasg: http://buildbot.shogun-toolbox.org/builders/clang34%20-%20undefined%20behaviour%20analysis/builds/5/steps/test/logs/stdio 19:11 search for LMNNImpl.find_impostors_exact 19:12 and you'll see 19:12 HeikoS: around? 19:12 wiking, wow 19:18 wiking, wtf is that 19:18 iglesiasg: hehehe ... heavy shit 19:18 iglesiasg: you are not the only one who's making this mistake 19:18 but it'd be great to find out how we could fix it 19:19 wiking, but the test does not fail anyway? 19:19 funny 19:19 iglesiasg: well it's an alignment issue 19:19 so most of the time i think it's fine 19:19 but i guess there's a machine + os + compiler combo on the world 19:19 wiking, I am writing an issue for me so I don't forget to have a look 19:19 where this would result in a segfault 19:19 iglesiasg: yeah go ahead... i'm just checking now the other issues 19:20 but I guess I will have a look in some time 19:20 /home/buildslave/clang34_-_undefined_behaviour_analysis/build/build/src/shogun/libshogun.so.14:0x47c3706: runtime error: value -inf is outside the range of representable values of type 'int' 19:20 :) 19:20 so there's some weird shit going on 19:20 HeikoS: May i temporally move to MS_NOT_AVALIABLE m_compact parameter of GaussianKernel and combined_kernel_weight of Kernel? 19:24 HeikoS: there is no code to compute derivatives wrt these parameters 19:26 HeikoS: and i'd like to temporally disable them for model selection 19:27 -!- iglesiasg [~iglesias@2001:6b0:1:1da0:247a:a4fc:e52c:8133] has quit [Ping timeout: 259 seconds] 19:37 votjakovr: no please dont, can you do another solution? 19:40 votjakovr: like some return value that tells you that there are no gradients available? 19:40 -!- iglesiasg [~iglesias@2001:6b0:1:1da0:247a:a4fc:e52c:8133] has joined #shogun 19:41 -!- mode/#shogun [+o iglesiasg] by ChanServ 19:41 HeikoS: ok 19:41 HeikoS: so as i was saying i'm preparing now the complex64_t mapping for lambday. do you know what he exactly would need? 19:42 HeikoS: i have now sgvector sgmatrix sgsparsevector and sgsparsematrix 19:42 wiking: not really, but basically he needs shogun's complex64 to numpy complex 19:42 and this for SGVector, SGSparseMatrix 19:42 ok 19:43 let's hope that this is what he needs 19:43 and that it works 19:43 sparse vector is a good idea, but I dont htink he uses that 19:43 wiking: yeah :) 19:43 i've tested sgvector 19:43 that works nicely 19:43 for the others i need some examples to test 19:44 HeikoS: i think it's not really good 19:44 votjakovr: but disabling parameters just for GPs is neither 19:45 people wont be able to do grid-search for them 19:45 votjakovr: also, these are discrete parameters so no derivative will ever be able to be computed 19:45 votjakovr: and for this, we cannot kick them out 19:45 HeikoS: yeah unfortunately 19:46 votjakovr: so maybe a call by reference paramter which is a boolean flag which tells you whether its possible to compute the derivate? 19:47 HeikoS: but we could add something like GRID_SEARCH_AVALIABLE ? 19:47 votjakovr: and its false by default, only true if there is an implementation 19:47 yeah something like this 19:47 but I would rather go for GRADIENT available 19:48 since grid-search is more general 19:48 HeikoS: oops yeah 19:48 HeikoS: so they should be registered like: SG_ADD(..., GRADIENT_SEARCH_AVALIABLE), right? 19:49 votjakovr: mmmh 19:50 votjakovr: I dont like explicit things 19:50 since then people have to register them with gradient AND implement gradient function 19:50 rather have a gradient function that tells you NO, and then devs might overload it to say YES, HERE IS YOUR GRADIENT 19:51 HeikoS: ok 19:54 shogun: Viktor Gal :develop * 5592220 / / (4 files): https://github.com/shogun-toolbox/shogun/commit/5592220bd062571ddfb02817bf3023aada143e86 19:55 shogun: WIP: adding complex64_t <-> numpy complex mapping (issue #1493) 19:55 HeikoS: this should be a member of TParameter, right? 20:00 all right guys, see you later 20:10 -!- iglesiasg [~iglesias@2001:6b0:1:1da0:247a:a4fc:e52c:8133] has quit [Quit: Ex-Chat] 20:10 votjakovr: no, this should be in the class that offers the derivative 20:22 HeikoS: mmmh, GaussianKernel offers derivative wrt width. I need to choose width from all of m_model_selection_parameters of GaussianKernel only width 20:27 votjakovr: I dont get that, could you explain again? 20:28 HeikoS: for example i get GaussianKernel and i'd like to build parameter dictionary, which contains only the parameters which we can use, when we want to compute derivative ("width" parameter in this case). How is it better to do this? 20:30 build #104 of precise - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/precise%20-%20libshogun/builds/104  blamelist: Viktor Gal 20:34 votjakovr: I see 20:35 votjakovr: so there are two problems: 20:35 1) is the parameter continous 20:35 2) if yet, is the derivative implemented 20:35 if 1) is false, you dont ask for derivative 20:35 if you ask for derivative 2), then it might not be implemented 20:36 votjakovr: so what about saying that every float parameter can have derivatives 20:36 votjakovr: this is the only thing that makes sense, 1) is solved 20:36 for 2) we need some mechanism that a class knows which derivatives it offers 20:36 but I would like to not do this via the SG_ADD macro 20:37 but rather by a local map in CSGObject 20:37 or even better 20:37 something that automagically says YES if a method is implemented 20:37 so that in order to add a derivate, you only have to implement/overload a method 20:38 HeikoS: we have just one method 20:39 HeikoS: which gets parameter as argument 20:39 HeikoS: and returns derivative wrt given parameter 20:39 if there is no derivatives wrt given parameter it shows an error 20:40 votjakovr: what about a flag in that method then? 20:40 votjakovr: std implementation returns "empty derivative" and set the flag to false 20:40 if a dev overloads, it returns the real derivate and sets flag to true 20:40 -!- travis-ci [~travis-ci@ec2-54-242-103-128.compute-1.amazonaws.com] has joined #shogun 20:41 [travis-ci] it's Heiko Strathmann's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11557013 20:41 -!- travis-ci [~travis-ci@ec2-54-242-103-128.compute-1.amazonaws.com] has left #shogun [] 20:41 this way, you can just check whether the derivative exists via trying to compute it 20:41 the only downside is that you always have to compute the derivative if you just want to know whether thats possible. But that should not be a problem 20:41 HeikoS: i actually don't need to check 20:41 HeikoS: i need the parameters 20:41 HeikoS: to build a dictionary 20:42 votjakovr: ah so you just need to know the parameters that offer a derivative? 20:42 HeikoS: yeah 20:42 votjakovr: mmh, so this cannot be solver via overloading the method then 20:42 votjakovr: needs to be set by hand 20:43 so a subset of the modelselection parameters would have also to be derivative parameters 20:43 HeikoS: yeah 20:43 votjakovr: what about adding another Parameter instance to CSGObject then? 20:43 votjakovr: or in fact, maybe even your flag in SG_ADD 20:43 votjakovr: would there be any other cases? 20:44 HeikoS: that what i suggested :) 20:44 because we dont want to pollute the SG_ADD too much 20:44 most classes wont have derivatives 20:44 HeikoS: SG_ADD(..., GRADIENT_SEARCH_AVAILABLE) 20:44 votjakovr: I think another Parameter instance would be best 20:44 votjakovr: no, too much writing and most classes dont have this 20:44 mmmh 20:45 sonney2k: thoughts on this? 20:45 votjakovr: macros do not support default arguments right? 20:45 since I dont want t rouch all SG_ADD lines in shogun 20:46 HeikoS: unfortunately 20:46 only the ones where a derivative is there 20:46 sonne|work: thoughts on this? ^ 20:46 votjakovr: mmh tough decision 20:47 votjakovr: wait 20:47 if you just want to know which parameters are continous, you are just interested in those which potentially can have derivatives right? 20:48 *potentially* 20:48 but the impleentation still can throw an error 20:48 you just have to implement it if you want to do gradient model-selection 20:48 so all float* parameters could be added, 20:48 so maybe its possible to add a method to CSGObject that tells you the parameters that are continous? 20:49 votjakovr: and the othher possibility would be to actually let the user provide a parameter tree with the parameters 20:49 votjakovr: time is tight, maybe we should go with this solution for now, its ugly, but the other thing takes too long 20:49 votjakovr: your problem is solved if users explicitly provide parameters for gradient search right? 20:49 you can add the other thing later maybe? 20:50 HeikoS: i need to build dictionary anyway 20:50 HeikoS: since gradient result is using it 20:51 votjakovr: but cant you build it from a ModelSelectionParameter tree? 20:52 HeikoS: why not to add m_gradient_parameters to SGObject? 20:52 votjakovr: maybe you are right 20:53 HeikoS: add third option GRADIENT_SEARCH_AVAILABLE 20:53 votjakovr: but lets think about one thing first 20:53 HeikoS: and if it is enabled we add parameter to m_parameters m_model_selection_parameters and m_gradient_parameters 20:53 votjakovr: in fact, do it. Replace all SG_ADD calls in shogun with GRADIENT_SEARCH_AVAILABLE=false 20:54 and then set it to true for the ones you want 20:54 votjakovr: though a default argument would be better 20:54 votjakovr: i.e. SG_ADD(bla, MS_NOT_AVAILABLE) and SG_ADD(bla, MS_NOT_AVAILABLE, GRADIENT_NOT_AVAILABLE) being equivalent 20:54 so that one only has to explicitly say when one wants gradients 20:55 agreed? 20:55 HeikoS: yep 20:56 votjakovr: I dont know if its possible to overload macros, but try, if not we have to add it to all SG_ADD for now 20:56 and a third field in CSGObject 20:56 HeikoS: ok, i think this is temporal solution, right? I mean, will we do some improvement of parameter and modsel in future? 20:59 votjakovr: yeah, I still like to keep things as clean as possible since doing things later doesnt work most of the time ;) 20:59 votjakovr: but go ahead 20:59 this is a good solution in fact 20:59 we just should think about ways of overloading the macro 21:00 HeikoS: have a look: http://stackoverflow.com/questions/8886812/overload-c-macros 21:02 votjakovr: okay, if you can make this work, it would be great, but dont spend too mnuch time on macro magic ;) 21:03 build #1500 of cyg1 - libshogun is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/cyg1%20-%20libshogun/builds/1500  blamelist: Viktor Gal 21:05 votjakovr: ok, sounds like a good plan, I will go home now, see you tomorrow! looking forward to the PR, this is really going to be good stuff  :) 21:07 HeikoS: see you ;) 21:07 -!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: Good bye!] 21:21 -!- van51 [~van51@athedsl-409794.home.otenet.gr] has quit [Quit: Leaving.] 21:26 -!- lisitsyn [~lisitsyn@fb2-lo1.global63.net] has joined #shogun 21:28 -!- travis-ci [~travis-ci@ec2-54-234-21-192.compute-1.amazonaws.com] has joined #shogun 21:35 [travis-ci] it's Viktor Gal's turn to pay the next round of drinks for the massacre he caused in shogun-toolbox/shogun: http://travis-ci.org/shogun-toolbox/shogun/builds/11562257 21:35 -!- travis-ci [~travis-ci@ec2-54-234-21-192.compute-1.amazonaws.com] has left #shogun [] 21:35 -!- lisitsyn1 [~lisitsyn@fb2-lo1.global63.net] has joined #shogun 21:48 -!- lisitsyn [~lisitsyn@fb2-lo1.global63.net] has quit [Read error: Connection reset by peer] 21:48 build #18 of clang34 - static analysis is complete: Failure [failed analyse]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20static%20analysis/builds/18  blamelist: Viktor Gal 21:52 -!- lisitsyn [~lisitsyn@fb2-lo1.global63.net] has joined #shogun 21:56 -!- lisitsyn1 [~lisitsyn@fb2-lo1.global63.net] has quit [Read error: Connection reset by peer] 21:56 lool 22:03 why dont we ever try to integrate this: https://projects.coin-or.org/Osi 22:03 ? 22:03 -!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has joined #shogun 22:04 -!- mode/#shogun [+o iglesiasg] by ChanServ 22:04 it's like having a million of solvers 22:04 build #6 of clang34 - undefined behaviour analysis is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20undefined%20behaviour%20analysis/builds/6  blamelist: Viktor Gal 22:04 interfaced as one 22:05 iglesiasg: i have a question 22:05 I might have an answer 22:05 wiking, shoot 22:05 m_rescode = MSK_putavec(m_task, MSK_ACC_CON, con_idx, nnz+1, 22:05 asub.vector, aval.vector); 22:05 Replaces all elements in one row of A.? 22:05 ok it is a exercise of memory 22:05 * wiking is trying to update the interface for mosek 22:06 yes 22:06 it should do that 22:06 I am almost 90% confident it does that 22:06 ok cool 22:07 wiking, new Mosek version? 22:07 iglesiasg: yeah 7 22:08 i'm now trying to find an equivalent of MSK_putaveclist 22:09 I am so looking forward to have some time and substitute that for another solver 22:09 -!- lisitsyn [~lisitsyn@fb2-lo1.global63.net] has quit [Read error: Connection reset by peer] 22:09 -!- lisitsyn1 [~lisitsyn@fb2-lo1.global63.net] has joined #shogun 22:09 when I talked to Nico in the workshop 22:09 he said it should be possible to do subgradient descent and use it instead of the Mosek solver we are using right now 22:09 wiking, do you think it makes sense too? 22:10 and I mean, gradient descent it is not hard. Even if Mosek solver is faster, it probably pays off just to get rid of this shit 22:10 iglesiasg: Osi: https://projects.coin-or.org/Osi 22:11 iglesiasg: it is interfacing a lot of solvers 22:11 (mosek included) 22:11 wiking, LP 22:11 mmm fuck it only interfaces LP 22:12 no QP? 22:12 wiking, we are using mosek for a QP 22:12 they only mention LP so far 22:12 let me check 22:12 I think so, yeah 22:12 https://projects.coin-or.org/OptiML 22:13 A'x=0 22:13 http://www.coin-or.org/projects/ 22:14 it is like libqp 22:14 i'm just checking there what they have there 22:14 seems like heaps of stuff 22:14 for the SO-SVM we need A'x<=b 22:14 okok 22:14 -!- lisitsyn1 [~lisitsyn@fb2-lo1.global63.net] has quit [Quit: Leaving.] 22:14 wiking, still. I think it might be good just to code up a solver for that 22:16 -!- besser82 [~besser82@fedora/besser82] has quit [Quit: freedom, friends, features, first ---> fedoraproject.org] 22:18 This routine finds a KT point for the Quadratic LCP (QP) problem 22:18 19 22:18 20c       minimize    f(x) = ct.x + xt.G.x/2 22:18 21 22:18 22c       subject to  l <= [I : A]t.x <= u 22:18 oh come oon man 22:18 this guy if fucking crazy 22:18 he wrote this solver in fortran 22:18 build #10 of clang34 - thread analysis is complete: Failure [failed test]  Build details are at http://buildbot.shogun-toolbox.org/builders/clang34%20-%20thread%20analysis/builds/10  blamelist: Viktor Gal 22:18 haha 22:19 wiking,  I never did fortran. Is it more hardcore than C? 22:19 I guess not 22:19 it is fucking hardcore 22:19 I mean, it is not assembler, right? 22:19 wiking, have you seen our BMRM? 22:19 that is hardcore I think hehe 22:19 iglesiasg: hehe no its cool 22:20 i've even fixed some shit in it 22:20 and then did some extension of it in libncbm 22:20 :P 22:20 nice 22:21 ok so i wonder what could be the replacement of that MSK_putaveclist 22:22 check for the methods to add constraints 22:22 it is basically that 22:23 linear inequality constraint 22:23 that's how they were called in Mosek's doc IIRC 22:23 Changes the bounds of a list of constraints. 22:23 ? 22:23 could it be this? 22:23 not really 22:23 A x <= b 22:23 that one you mention is to change the b, I think 22:24 yes 22:24 and u change with aveclist? 22:24 x? 22:24 i mean A's 22:24 no no 22:24 row 22:24 yes yes 22:24 a row of A 22:24 wiking, do you have the doc page? 22:25 ah and it replaces n number of elements 22:25 I can try to help to find it 22:25 for the new api? 22:25 mm n number of elements 22:25 http://docs.mosek.com/7.0/capi/Changes_and_new_features_in_MOSEK.html 22:26 so that method was done to dine a row of A 22:26 but you do it in a sparse way 22:26 so you give 22:26 1) the number of values in a row you want to change 22:26 2) the indices within the row, i.e. col idxs 22:26 3) the actual values 22:26 wiking,  http://docs.mosek.com/7.0/capi/MSK_putarow_.html? 22:27 wiking, to me it looks like it must be that one 22:28 wiking, now you have m_rescode = MSK_putavec(m_task, MSK_ACC_CON, con_idx, nnz+1, 22:28 asub.vector, aval.vector); 22:28 I think it must be pretty much like 22:29 m_rescode = MSK_putarow(m_task, con_idx, nnz+1,  asub.vector, aval.vector); 22:29 not sure right now why the +1 nnz+1 but with any luck I wrote a comment for that 22:30 -!- van51 [~van51@athedsl-409794.home.otenet.gr] has joined #shogun 22:39 -!- lisitsyn [~lisitsyn@fb2-lo1.global63.net] has joined #shogun 22:49 -!- iglesiasg_ [~iglesias@s83-179-44-135.cust.tele2.se] has joined #shogun 22:51 -!- iglesiasg [~iglesias@s83-179-44-135.cust.tele2.se] has quit [Quit: Ex-Chat] 22:51 -!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout] 22:55 ? 23:00 wiking, is that for me? 23:00 -!- iglesiasg_ is now known as iglesiasg 23:00 -!- mode/#shogun [+o iglesiasg] by ChanServ 23:01 i was looking for replacing MSK_putaveclist 23:02 not MSK_putavec 23:02 well then MSK_putarowlist 23:02 that'll be good ! thnx 23:07 ok only one more is missing MSK_getsolutionstatus 23:09 ok this should be it 23:11 MSK_getprosta 23:11 welcome 23:12 oh damn 23:12 wiking, btw we cannot use any of coin-or due to licence conflicts 23:12 http://docs.mosek.com/7.0/capi/Problem_status_keys.html 23:12 they are toootaly different 23:13 why they use what? 23:13 wiking, they used to use CPL 23:15 I forgot what they switched to 23:16 but it was again in conflict to GPL 23:16 -!- shogun-notifier- [~irker@7nn.de] has joined #shogun 23:21 shogun: Soeren Sonnenburg :develop * 6197bb4 / examples/undocumented/python_modular/ (6 files): https://github.com/shogun-toolbox/shogun/commit/6197bb4ab279aec6cb3d84c6c5d6fbe87ccf22a8 23:21 shogun: convert more examples to CSVFile 23:21 shogun-demo: van51 :master * 2bd9761 / / (3 files): https://github.com/shogun-toolbox/shogun-demo/commit/2bd9761c84f52dfae94fdce22bd1c504b62b0bd1 23:23 shogun-demo: Changed prompt and about in language detection demo 23:23 shogun-demo: Soeren Sonnenburg :master * 6569227 / / (3 files): https://github.com/shogun-toolbox/shogun-demo/commit/6569227662c69ce5645d298cd82eeb6676f2701d 23:23 shogun-demo: Merge pull request #24 from van51/master 23:23 shogun-demo: 23:23 shogun-demo: Changed prompt and about in language detection demo 23:23 build #105 of precise - libshogun is complete: Success [build successful]  Build details are at http://buildbot.shogun-toolbox.org/builders/precise%20-%20libshogun/builds/105 23:49 build #1849 of deb3 - modular_interfaces is complete: Failure [failed test python modular]  Build details are at http://buildbot.shogun-toolbox.org/builders/deb3%20-%20modular_interfaces/builds/1849  blamelist: Soeren Sonnenburg 23:56 --- Log closed Fri Sep 20 00:00:06 2013