IRC logs of #shogun for Monday, 2013-05-27

« Sunday, 2013-05-26 Index Tuesday, 2013-05-28 »
--- Log opened Mon May 27 00:00:19 2013
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun00:40
shogun-notifier-shogun: Kevin Hughes :develop * 7d14502 / src/shogun/multiclass/QDA.cpp: https://github.com/shogun-toolbox/shogun/commit/7d14502376e2cee7d0b2fdcdc0605b51f480214500:40
shogun-notifier-shogun: fixed crash in QDA covar calculation00:40
shogun-notifier-shogun: Soeren Sonnenburg :develop * bfae181 / src/shogun/multiclass/QDA.cpp: https://github.com/shogun-toolbox/shogun/commit/bfae18199f8dd2f1e609a4c7faf30a6aa92f02ab00:40
shogun-notifier-shogun: Merge pull request #1131 from pickle27/qda00:40
shogun-notifier-shogun:00:40
shogun-notifier-shogun: fixed crash in QDA covar calculation00:40
-!- travis-ci [~travis-ci@ec2-107-22-28-98.compute-1.amazonaws.com] has joined #shogun01:01
travis-ci[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/751955801:01
-!- travis-ci [~travis-ci@ec2-107-22-28-98.compute-1.amazonaws.com] has left #shogun []01:01
shogun-buildbotbuild #1166 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1166  blamelist: Kevin Hughes <kevin@kevin-M3920.(none)>01:21
shogun-buildbotbuild #1167 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1167  blamelist: Soeren Sonnenburg <sonne@debian.org>01:49
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]03:40
shogun-buildbotbuild #408 of nightly_default is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/40803:44
-!- nube [~rho@49.244.8.172] has quit [Quit: Leaving.]03:44
-!- nube [~rho@116.90.239.3] has joined #shogun05:51
-!- nube [~rho@116.90.239.3] has quit [Ping timeout: 240 seconds]06:57
-!- nube [~rho@116.90.239.3] has joined #shogun07:02
gsomixgood morning07:04
gsomixsonney2k, good. tnx07:05
-!- foulwall [~foulwall@li379-21.members.linode.com] has joined #shogun07:52
-!- gsomix [~gsomix@188.168.2.227] has quit [Quit: Leaving]08:17
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun08:37
shogun-notifier-shogun-data: Soeren Sonnenburg :master * ca1d0b6 / testsuite/tests/ (2 files): https://github.com/shogun-toolbox/shogun-data/commit/ca1d0b696150224cc280f9daff612124852d502508:37
shogun-notifier-shogun-data: update qda data08:37
shogun-notifier-shogun: Soeren Sonnenburg :develop * 363d6c2 / data,tests/integration/python_modular/generator.py: https://github.com/shogun-toolbox/shogun/commit/363d6c273b60f538948589e9ae9b6f9130983f7008:38
shogun-notifier-shogun: update data and add director examples to blacklist08:38
@sonney2kfoulwall, good morning!08:39
shogun-notifier-shogun: Soeren Sonnenburg :develop * 93df87f / src/shogun/statistics/MMDKernelSelectionComb.cpp: https://github.com/shogun-toolbox/shogun/commit/93df87f1f531ab6cb03107092e7d9b740b60595b08:39
shogun-notifier-shogun: fix typo in warning msg08:39
foulwallsonney2k: morning, I'll have my first final this afternoon.08:42
@sonney2kfoulwall, oh good luck then and lets talk again tomorrow once google sent out the notifications!08:43
@sonney2kpickle27, I've regenerated the qda integration test data08:44
foulwallsonney2k: thanks~08:44
@sonney2kso finally tests should all pass08:44
@sonney2khushell, done with nips paper :D08:44
@sonney2koops gtg08:45
shogun-buildbotbuild #876 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/876  blamelist: Soeren Sonnenburg <sonne@debian.org>08:58
-!- travis-ci [~travis-ci@ec2-184-72-84-14.compute-1.amazonaws.com] has joined #shogun08:59
travis-ci[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/752568508:59
-!- travis-ci [~travis-ci@ec2-184-72-84-14.compute-1.amazonaws.com] has left #shogun []08:59
-!- vgorbati [5f6fea63@gateway/web/freenode/ip.95.111.234.99] has joined #shogun09:05
-!- sonne|work [~sonnenbu@sams-office-nat.tomtomgroup.com] has joined #shogun09:13
sonne|workRe09:13
hushellsonney2k: :D not as quick as I expected, but all results are obtained, I am writing introduction ...09:17
sonne|workhehe09:17
-!- travis-ci [~travis-ci@ec2-184-72-84-14.compute-1.amazonaws.com] has joined #shogun09:18
travis-ci[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/752571309:18
-!- travis-ci [~travis-ci@ec2-184-72-84-14.compute-1.amazonaws.com] has left #shogun []09:18
hushellsonney2k: gsoc starts next week?09:18
hushellwriting is painful09:19
sonne|workhushell: today will be notifications09:19
sonne|workof acceptance/reject09:19
sonne|workfrom then on everyone can start but official start is slightly later09:19
hushell!!!exciting09:20
hushellalso nervous!09:21
shogun-buildbotbuild #1168 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/116809:21
sonne|workjune 17 is the official start - but it sure won't hurt to start before09:22
hushellI am expecting for the gsoc for a long time, writing is so boring09:23
-!- nube [~rho@116.90.239.3] has quit [Ping timeout: 245 seconds]09:23
-!- nube [~rho@116.90.239.3] has joined #shogun09:24
-!- iglesiasg [d58f3247@gateway/web/freenode/ip.213.143.50.71] has joined #shogun09:32
-!- mode/#shogun [+o iglesiasg] by ChanServ09:32
-!- iglesiasg [d58f3247@gateway/web/freenode/ip.213.143.50.71] has quit [Client Quit]09:36
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has joined #shogun09:41
shogun-buildbotbuild #1169 of deb3 - modular_interfaces is complete: Failure [failed test python_modular]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1169  blamelist: Soeren Sonnenburg <sonne@debian.org>09:44
-!- lisitsyn [~lisitsin@mxs.kg.ru] has joined #shogun09:44
-!- mode/#shogun [+o lisitsyn] by ChanServ09:44
@lisitsynha-ha09:44
@lisitsynsonne|work: I am back from no-internet-connection-land!09:45
sonne|worklisitsyn: siberia?09:45
sonne|workhad fun playing with your grizzly bears?09:45
sonne|workor whatever you call them up there?09:45
shogun-notifier-shogun: Soeren Sonnenburg :develop * d6cb133 / tests/integration/python_modular/generator.py: https://github.com/shogun-toolbox/shogun/commit/d6cb133556c7e82ec4f1641f2fab4dcb5e77cca109:49
shogun-notifier-shogun: fix typo in blacklist09:49
sonne|worklisitsyn: anyways just in time to watch the test to go red -> green09:49
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has joined #shogun09:52
sonne|workvan51: welcome back09:52
van51sonne|work: hello09:53
@lisitsynsonne|work: no electricity due to some fire :)10:07
-!- travis-ci [~travis-ci@ec2-107-22-28-98.compute-1.amazonaws.com] has joined #shogun10:11
travis-ci[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/752706410:11
-!- travis-ci [~travis-ci@ec2-107-22-28-98.compute-1.amazonaws.com] has left #shogun []10:11
shogun-buildbotbuild #877 of cyg1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/87710:16
-!- thoralf [~thoralf@enki.zib.de] has joined #shogun10:25
thoralfGood morning10:25
shogun-buildbotbuild #1170 of deb3 - modular_interfaces is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/117010:34
-!- vgorbati [5f6fea63@gateway/web/freenode/ip.95.111.234.99] has quit [Ping timeout: 250 seconds]10:35
@lisitsynsonne|work: did you know ray dassen?10:41
sonne|workno10:41
sonne|workwho was/is he10:41
sonne|work?10:42
@lisitsyndebian developer10:42
@lisitsynpassed away yesterday10:42
@lisitsynoh not yesterday10:42
@lisitsyn23th10:42
@lisitsynsonne|work: http://www.debian.org/News/2013/2013052310:42
sonne|workwe all die...10:43
sonne|workwow around for 19 years in debian10:44
@lisitsynyes quite a long period10:44
@lisitsynthat's I am curious whether he was famous around debianeers10:44
-!- HeikoS [~heiko@nat-191-4.internal.eduroam.ucl.ac.uk] has joined #shogun10:51
-!- mode/#shogun [+o HeikoS] by ChanServ10:51
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has left #shogun ["QUIT :Leaving."]10:58
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun11:09
-!- mode/#shogun [+o iglesiasg] by ChanServ11:10
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has quit [Ping timeout: 250 seconds]11:13
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.55] has joined #shogun11:14
lambdayHeikoS: hi11:15
@HeikoSlambday: hi!11:15
lambdaygot your mail11:15
lambdayI did't understand what you meant by an extension of shogun framework11:15
lambdayfor integrating and testing new algorithms in shogun?11:16
lambdayas in, in future?11:16
@HeikoSlambday: A way of generalising independent computation tasks might be very useful for other parts11:16
@HeikoStake for example CMachine model-selection via grod-search11:16
@HeikoSthis is a stupid algorithm, but it is used by many people11:16
@HeikoSyou basically train/test machines with different parameters11:17
@HeikoSall these tasks are independent11:17
lambdayokay11:17
lambdayhmm11:17
@HeikoSyou we could formulise them as a Task as in the class diagram11:17
@lisitsynhah stupid == working HeikoS ;)11:17
@HeikoSthen pass them to an engine that solves them in some way11:17
-!- lambday [67157d37@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.125.55] has quit [Client Quit]11:17
@HeikoSlisitsyn: what?11:17
@HeikoSah11:18
@HeikoSwell11:18
@lisitsynHeikoS: grid search is stupid but it is a working algorithm11:18
@lisitsyn:)11:18
@HeikoSlisitsyn: it is a ridiculus thing in my eyes, but what can you do :)11:18
@HeikoSpeople just like simple algorithms11:18
@HeikoSbrute force stuff :D11:19
@lisitsynHeikoS: but how would you approach it11:19
@lisitsynI mean how do you search C then?11:19
@HeikoSthere are many way to do this, all described early 2000s11:19
@HeikoSsimilar to the way its done with GPs11:19
@HeikoSgradient descent on some upper bound of the error11:19
@lisitsynwell11:19
@lisitsynyou'd have to have a great API to make that common11:20
@lisitsynHeikoS: it is easy to reproduce brute solution11:20
@lisitsynand they always work for you11:21
@HeikoSall true11:21
@HeikoSbut we have this for GPs11:21
@HeikoSjust need to implement some of these error functions11:21
-!- lambday [67157e4e@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.126.78] has joined #shogun11:23
@HeikoSlisitsyn: sorry I dont want to dizz this stuff, just my thoughts11:23
@lisitsynHeikoS: what do you mean? :)11:23
@HeikoSits already pretty good that we have grid-serach for arbritary machines11:23
@HeikoSI mean its fine as it is11:23
@lisitsynHeikoS: I just wanted to note for stupid engineers like me it is working solution11:24
lambdaysorry I got disconnected :(11:24
@HeikoSlambday: no worries11:24
@lisitsynI don't give a heck how to approach it else :)11:24
@lisitsynprobably being stupid is a problem of mine though :D11:24
@HeikoShaha, I wouldnt agree ;)11:25
@lisitsynHeikoS: btw recalling that syntax I was pushing11:25
@lisitsynare we still up for it?11:26
@HeikoSyes totally11:26
@lisitsynhmm who could do that then11:26
@HeikoSlisitsyn:  btw there is also a way of speeding the std-grid search up quite a bit through some statistical tricks11:27
@HeikoSsonney2k:  pointed this out at some point11:27
@HeikoSmight be nice to have this11:27
@lisitsynHeikoS: like MCMC or so?11:27
@HeikoSno, MCMC and the word speed never go together :D11:27
-!- van51 [~van51@195.251.235.71] has joined #shogun11:27
@lisitsynHeikoS: you can't touch this11:27
@HeikoShaha :)11:28
-!- gsomix [~gsomix@37.61.179.240] has joined #shogun11:28
gsomixhello!11:28
@lisitsyncan't touch this11:28
@lisitsynHeikoS: agaaain11:28
@HeikoShttp://blog.mikiobraun.de/2011/12/fast-cross-validation.html11:28
@lisitsynwhoop11:28
@lisitsynahhh11:28
@lisitsynHeikoS: I recall some core-sh*t from NIPS11:28
@lisitsynsonne|work told us about it11:28
@lisitsyncore sets or so11:28
@lisitsynI don't recall11:29
@HeikoSyou drop parameters which are statistically significant loosers against others11:29
@lisitsynsounds genetic11:29
@lisitsynHeikoS: can't touch this11:29
@HeikoSnot really11:29
@lisitsynjk11:29
@lisitsyn:)11:29
@HeikoSanyway, we should have a look at this at some point11:30
@HeikoSproblem is that its not really parallelisable anymore11:30
sonne|workgsomix: finaly a function that is readable :)11:31
@HeikoSor that to that level11:31
@HeikoSbut other things first11:31
@lisitsynHeikoS: I would go for stupid but parallel solution actually11:31
@HeikoSlisitsyn: having both would be nice11:31
@HeikoSif you have many cores, go for the  easy, if not, go for the clever11:31
@lisitsynour CPUs are fast and our RAM is going infinite :)11:31
gsomixsonne|work, horray! I also got yours comments.11:31
@lisitsynHeikoS: you can't touch this11:33
@lisitsyn:D11:33
@lisitsynsrry11:33
@HeikoShaha :D11:33
@lisitsynHeikoS: this should be an anthem of shogun11:34
@lisitsynHeikoS: when you are entering an website11:34
@lisitsynyou should hearr that11:34
@HeikoShehe11:35
@HeikoSlambday: btw the Task.sove method should better be named .compute()11:35
@lisitsynHeikoS: are you going to push that mcmc hammer to shogun?11:35
lambdayHeikoS: okay, changing11:36
lambday:)11:36
@HeikoSlisitsyn: probably11:36
@HeikoSwe got a name clash unfortunately :)11:36
@HeikoShave to rename it11:36
@lisitsynHeikoS: I am submitting tapkee paper *today* so I am going to join your game with tasks at some point11:36
@lisitsynHeikoS: no way, whyyy11:36
@HeikoSgoogle for it11:36
@HeikoSmcmc hammer11:36
@lisitsynhahah11:36
@HeikoSlisitsyn: ah shit, I never gave you my comments did I?11:36
@HeikoSlisitsyn: cool, that would help a lot11:36
@lisitsynHeikoS: hmm you did I think11:37
@lisitsynI don't remember11:37
@HeikoSreally, good :)11:37
@lisitsynanyway nevermind I can't wait much longer :)11:37
@HeikoSme neither11:37
@HeikoSok11:37
sonne|workgsomix: you should write a test for this function though11:37
@lisitsynI've been working on it for years11:37
@lisitsynand now I am 62 and no submission still yet11:38
sonne|workgsomix: we have enough ascii files to try :D11:38
@lisitsynI had one ascii file somewhere11:38
@lisitsynHeikoS: https://twitter.com/Rick_Moyer/status/278641295285698560 LOL11:38
@HeikoSlambday: could you write a seperate diagram for all the computation task stuff (without the implementations, just base classes)11:38
@HeikoSlisitsyn:  thats it11:39
@HeikoSlambday: will help discussions with Sergey11:39
lambdayHeikoS: okay...you mean, computation task and computation task result classes?11:39
@HeikoSyeah just those for now, will be a small diagram and you can do this with copy paste11:40
@lisitsyncan't touch this diagram11:40
lambdayHeikoS: okay... mailing11:40
gsomixsonne|work, really? I need one ~#N GB sized file. :)11:40
@lisitsynohohohoho11:40
@lisitsynI recalled11:40
@lisitsynwiking: you made random() const!11:41
@lisitsynbad bad bad boy11:41
sonne|work*lol*11:41
sonne|workgsomix: you could use data/asp/Human.dat.bz211:42
sonne|workit is ~80M11:42
@lisitsynlet us have11:42
@lisitsynuint64_t a = random64();11:42
@lisitsynuint64_t b = random64();11:42
sonne|workerr 100M11:42
@lisitsynthe question is11:42
@lisitsynwhether a==b11:42
@lisitsynI'd say -O3 makes it *true*11:42
gsomixsonne|work, ok.11:42
@lisitsynokay not really sure but it could11:42
@lisitsynsonne|work: ^11:43
sonne|worklisitsyn: sorry btu why11:43
@lisitsynsonne|work: because it looks like a pure function if it is const11:43
@lisitsynwhy to recompute its value twice if it has no side-effects11:43
@lisitsynwhich cause changing its value11:43
@lisitsynthat's the meaning of const11:43
sonne|workbut only inside the CRandom class11:43
@lisitsynyes but we have to check11:44
@lisitsyncompiler is magic11:44
lambdayHeikoS: just these two base classes? but these contains almost nothing here, their subclasses do11:44
@lisitsynrandom function is not const11:44
sonne|worklisitsyn: well tests work so you are wrong11:44
@lisitsynit is as mutable as it can be11:44
sonne|workbut if tests are compiled with -O0 then :)11:44
@lisitsynsonne|work: vgorbati told me we have an issue with calling normal only once11:44
@lisitsynI didn't check due to blackout11:44
@lisitsynand now recalled11:45
@lisitsynwill check tonight11:45
@HeikoSlambday: maybe an example of a class that has generate_tasks / aggregate results11:45
sonne|worklisitsyn: normal?11:45
@lisitsynnormal_random I mean11:45
@lisitsynsonne|work: anyway it is not const11:45
sonne|workhmmhh, we have to move the other gaussian generator from CMath11:45
@HeikoSlambday: I am sorry, that was asked too quickly. Just leave it for now11:45
lambdayHeikoS: okay..11:46
@HeikoSlambday: we will use the full diagram for discussions11:46
@HeikoSits clearer, you are right11:46
@lisitsynconst tells me calling this function sequentially twice will return the same value11:46
lambdayHeikoS: yes its pretty clear when we see this whole picture, and log-det as an example of that11:47
lambdaythis task idea was awesome!! :)11:47
@HeikoSlambday: lets hope it is possible ;)11:48
lambdayHeikoS: implementation?11:49
@HeikoSlambday: as lisitsyn pointed out, there are always difficulties if one wants to write those things in a general way11:49
@lisitsyncan't touch this11:50
lambdayHeikoS: but its good to have this general framework, no?11:50
@HeikoSlambday: yes, it would be very useful11:50
@lisitsynlambday: I just mean we can't foresee all the stuff11:50
lambdaylisitsyn: true that11:51
@lisitsynthere is a wall right forward actually11:51
@lisitsynwe are moving to11:51
@lisitsyn:D11:51
@HeikoSlisitsyn, lambday: sonne|work pointed out that we can even have a qsub-based implementaton for the engine11:51
lambdaylisitsyn: all in all its just another brick in the wall, then :P11:51
@lisitsynonce we crashed into the wall we start refactoring11:51
lambdayqsub-based implementation?11:52
@lisitsyncuring our pain in the ass :D11:52
@HeikoSlambday: yes for qsub based clusters11:52
lambdaywow!11:53
@lisitsynHeikoS: we will become graphlab you have to live with that11:53
@lisitsyn:D11:53
@HeikoSlisitsyn: we will be fine! all independent11:53
lambdayHeikoS: i'll shortly be back.. starving.. gotta eat something... about the work, the picture is quite clear now... what should be the next step? adding all the implementations?11:58
lambdayHeikoS: and we don't need the linear system class then11:58
@HeikoSlambday: indeed, only the linear solver, which takes a linear operator and a vector11:59
lambdayHeikoS: okay..11:59
@HeikoSlambday: once everything is clear, I would write all the base classes11:59
@HeikoSto get a feeling whether things actually work11:59
@HeikoSmake sure to start a new feature branch with git-flow12:00
@HeikoSif those are done, refine class diagram12:00
@HeikoSthen some simple implementations12:00
lambdayHeikoS: okay..12:01
lambdayso, it may be good to have a diagram with the base classes only12:02
@HeikoSlambday: and could you send the current class diagram to lisitsyn?12:02
@HeikoSlambday: yes, totally12:02
@HeikoSalso without any overhead in the diagram12:02
@HeikoSjust plain classes12:02
@HeikoSit gets too big otherwise12:02
lambdayHeikoS: okay..12:03
lambdayHeikoS: oh I almost forgot... the CRationalApproximationTaskResults inherits from both CIndependentComputationTaskResults and CIndependentComputationTask... I didn't understand this12:06
@lisitsynnaming reminds me java12:07
@lisitsyn:D12:07
lambdaylisitsyn: :D12:07
lambdaybut sounds good and clear12:07
@lisitsynlet me find the longest class name in my current proj12:07
@lisitsynhmm I am not a java programmer12:08
@lisitsynthe longest is AgentMessagesFactory :(12:08
sonne|worklambday: I don't understand what this class does though so the name is not that helpful12:08
sonne|worklisitsyn: but I know what you mean :D12:09
lambdaysonne|work: umm....12:09
@lisitsynsonne|work: AbstractBusinessLayerCompositorDecompositorBuilderMasterCollaboratorFactory12:09
@HeikoSlambday: ehm, that might have been my mistake12:09
lambdaylisitsyn: whoa! :o12:09
lambdayHeikoS: hmmm... okay then removing it12:10
@HeikoSresults doesnt inherit from task12:10
sonne|workHeikoS: couldn't we use 'Job' ?12:10
@lisitsynyes Job is better12:10
lambdayyes I thought so but wanted to make sure12:10
@HeikoSsonne|work: yes totally12:10
sonne|workit is short12:10
@HeikoSmake it job12:10
@lisitsynhandjob12:10
@lisitsynscnr12:10
@HeikoStztz12:10
lambdaylisitsyn: lol man :D12:10
sonne|worklisitsyn: a few hours without power got you in the mood heh?12:11
@HeikoSlambday: CIndependentJob, but the subclasses dont get the "Independent"12:11
lambdayHeikoS: okay12:11
lambdayHeikoS: I'll change it after lunch and mail it back soon.. also adding sample_without_averaging method you told about12:12
@lisitsynsonne|work: yes sure!12:12
@HeikoSlambday: yes, maybe there is also a better name for this :)12:12
@HeikoSlambday: enjoy your lunch12:13
lambdaybe right back12:13
-!- nube [~rho@116.90.239.3] has quit [Quit: Leaving.]12:14
@lisitsynHeikoS: why to have such a class like independent job?12:22
@HeikoSalgorithms can generate them under a unified interface12:23
@HeikoScomputation class can call compute() on the interface12:23
@HeikoScomputation class does not need to worry whats going under underneath12:23
@lisitsynHeikoS: may be just a list<Job> dependencies?12:23
@lisitsynif it is empty you have no dependencies12:24
@HeikoSI would like to go without dependencies12:24
@HeikoSthe implementation of the job can still operate on say shared memory12:24
@HeikoSor serialized files12:25
@HeikoSbut this can be done inside12:25
@HeikoSfor every job12:25
@HeikoSI have the fear that adding this to the framework might make things complicated12:25
@HeikoSand it is task-dependent anyways12:25
@HeikoSlisitsyn: whats your opinion on this?12:28
@HeikoShow would you formulate dependencies and what would they be?12:28
@lisitsynHeikoS: that's the question of this day12:28
@lisitsyn:D12:28
@lisitsynI don't know tbh12:28
@HeikoSlisitsyn: so an example12:28
@HeikoSa matrix12:29
@HeikoSlarge one12:29
@HeikoSall jobs need this readonly12:29
@HeikoSwhat one could do is to add a pointer to the data to each job12:29
@lisitsynHeikoS: I'd go for something generic step by step like read only proxies for these matrices12:29
@HeikoSwhy?12:30
@lisitsynwell that's safe12:30
@HeikoSif we want to go multiple-machine, we *have* to serialize it anyway12:30
@HeikoSso every job will load this matrix from disc12:30
@HeikoSor network12:30
@lisitsynoh I can't touch this12:30
@lisitsyn:D12:30
@HeikoSand in multicore case, these things have to be read-only12:30
@lisitsynsounds like implementing hadoop12:30
@HeikoSlisitsyn: this is not meant for millions of nodes12:31
@HeikoSjust a few hundred max12:31
@lisitsynMR in particular12:31
@HeikoSso we can rely on the nfs12:31
@lisitsynuh uh I don't like doing that at all12:31
@HeikoSthis is not large-scale stuff12:31
@HeikoSwhats the problem (in our context)?12:32
@lisitsynI am afraid of NIH12:33
@HeikoSwhats that?12:33
@lisitsynnever invented here syndrome ;)12:33
@HeikoSI am not following you12:34
@lisitsynI mean we'd rather stick to something supported by anyone else12:34
@lisitsynthan writing that thing12:34
@HeikoSstill not following ;)12:34
@lisitsynHeikoS: I mean doing it with our own serialization etc12:34
@lisitsynreproduces what map reduce does12:35
@HeikoSwe are much simpler12:35
@HeikoSbut yes its similar12:36
@HeikoSso whats the alternative then?12:36
@lisitsynmay be employ something supported12:37
@lisitsynHeikoS: what I am most worried is that we have to support all the stuff12:37
@lisitsynwe already have no power/will to support all the code we have12:37
@lisitsynsuch IO routines12:37
@lisitsynand other infrastructure sounds painful12:37
@HeikoSagreed, but thats a general problem12:37
@HeikoSlisitsyn: so the multi-machine stuff is far away12:38
@HeikoSif this worked for multi-core, it would already be very helpful12:38
@HeikoSeven if the engine is just a single core loop its helpful12:38
@HeikoSto cast algorithms in terms of these jobs12:38
@HeikoSsince it makes transition easier12:38
@HeikoSfor the log-dets for example12:39
@HeikoSthere is no point in even doing this without this job stuff12:39
@lisitsynHeikoS: I can imagine we do that engines stuff12:39
@lisitsynbut we have to take it really seriously12:40
@HeikoSsure12:40
@lisitsynbasically it would become a supported core while algorithms get experimental12:40
@HeikoSI wouldnt change the existing algorithms for now12:40
@HeikoSjust try to write new ones in that style12:40
@HeikoSif it works, fine, maybe change existing ones12:41
@HeikoSif it doesnt (too hard in whatever way), we can just plug in the sequential engine and no problems12:41
@lisitsynHeikoS: I have a suggestion12:41
@lisitsynonce you get with some solution on log-det12:41
@lisitsynlet lambday change some existing algorithm to new 'paradigm'12:42
@HeikoSlisitsyn: yes good idea12:42
@lisitsynand we will see something12:42
@HeikoSgrid-search maybe12:42
@HeikoSor even simpler things12:42
@HeikoSkernel matrix12:42
@HeikoSone job for a block12:42
@HeikoSthen we wouldnt have to do this ultra messy pthread stuff anymore12:43
@HeikoSpure horror that thing (at least to me :)12:43
@lisitsynHeikoS: well jobs may be pthread powered still12:43
@HeikoSlisitsyn: yes, but with a central implementation12:43
@lisitsynyes12:43
@HeikoSrather than 100 local once12:43
@HeikoSones12:43
@lisitsynyes these pthread runners allthrough the code are scary12:44
@lisitsynHeikoS: I am quite interested in developing it actually12:46
@HeikoSlisitsyn: very welcome to do this12:47
@lisitsynHeikoS: yeah lets see if I get enough time to join you12:48
@HeikoSlisitsyn: we dont have time pressure on this12:48
@HeikoSthe log-det will be written in that framework anyway12:49
@HeikoSengine implementation will be sequential at first12:49
@HeikoSand if someone has time, we can do cooler things12:49
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]12:49
@lisitsynHeikoS: btw are you done with your nips thing?12:50
@HeikoSworking on it12:50
lambdayback12:51
lambdayHeikoS: task looked better than job :(12:51
@HeikoSwhy?12:51
lambdayjob is fine though..12:52
lambdaychanged it.. added that method12:52
lambdaymailing to both of you12:52
@lisitsynexcept to blowjob collision job sounds better12:52
@lisitsynhah12:52
lambdaylisitsyn: I was just about to say - let's hope we don't blow this job :12:53
lambday:D12:53
lambdaylisitsyn: I don't have your email id12:53
lambday:(12:53
@lisitsynlambday: lisitsyn.s.o@gmail.com12:53
lambdaylisitsyn: ty12:55
@HeikoSlambday: looks good, maybe even call aggregate results "reduce_results" or something12:58
lambdayHeikoS: umm... yes that's more generic name12:59
@HeikoSlooking good apart from that, we could now start writing code for the base classes12:59
@HeikoSso one more thing that we have to be careful about12:59
@HeikoSthe generic linear operators and operator functions12:59
@HeikoSI dont really know how we should do this13:00
@HeikoSshogun does not have the type Complex13:00
@HeikoSlisitsyn: what do you think on this?13:00
lambdaystd::complex should be fine, no?13:00
@HeikoSlambday: still13:01
@HeikoSmight be a problem13:01
lambdayHeikoS: implement complex things for shogun then?13:02
lambdaybut we need complex for only a few parts here13:02
@HeikoSyes, internally this is fine13:02
@HeikoSbut if we start having complex variables, we might get into trouble13:03
@HeikoSso I am not sure whether to have a generic CLinearOperator13:03
@HeikoSor CDenseMatrix13:03
@HeikoSCComplexDenseMatrix13:03
@HeikoSetc13:03
@HeikoSwhere do the complex values appear?13:04
@HeikoSonly in the matrix of the shifted system right?13:04
lambdayin the weights and shifts only13:04
lambdayyes13:04
@HeikoSso then we could even completely hide this13:04
@HeikoSand say a linear operator just applies to a real vector13:05
lambdayand we'll never really use direct solver13:05
lambdayHeikoS: yes13:05
@HeikoSso then we could have an implementation CComplexSparseMatrixOperator13:05
@HeikoSwhich represents the shifted stuff13:06
@HeikoSand CSparseMatrixOperator for the other13:06
@HeikoSthen we dont need any generic stuff anywhere13:06
lambdayumm.. okay then13:06
@HeikoSsorry for all those changes, just remembered that I made a mental note of this13:07
lambdaybut having a base linear operator class would have been helpful, no?13:07
@HeikoSwe have to try to avoid that any complex numbers are in function interfaces13:07
@HeikoSlambday: yes, but not generic13:07
lambdayHeikoS: no problem.. I myself am forgetting so many things13:07
lambdayHeikoS: only for real, you mean13:07
lambdayno complex stuff13:08
@HeikoSlambday: ah sorry13:08
@HeikoSlet me think13:08
@HeikoSso the abstract base might be generic13:08
@HeikoSfloat64, float32 etc13:08
@HeikoSnot complex though13:08
lambdaycool13:09
@HeikoSand the sublcass fixes T=float64_t13:09
lambdayfloat32_t would be used later?13:09
lambdayyes... okay13:09
@HeikoSI dont know13:09
@HeikoSmaybe leave it as it is13:10
@HeikoSso CLinearOperator<T>13:10
@HeikoSCoperatorFunction<T>13:10
-!- lambday [67157e4e@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.126.78] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client]13:11
@HeikoSand then CRationalApproximation<T=float64_t>13:11
@HeikoSlisitsyn: what do you think?13:11
@HeikoSshould these things be made available for types other that float64?13:11
@HeikoSwe dont need others13:11
-!- lambday [67157e4e@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.126.78] has joined #shogun13:13
lambdayHeikoS: sorry I got dc again :( my net connection sucks!13:13
@HeikoSlambday: see mail, leave the abstract bases generic it should be fine13:13
@HeikoSlambday: just one more sublety, we should not use SGVector<Complex>13:13
@HeikoSrather have to SGVector<float64_t> for real and imaginary parts13:14
lambdayso, array then?13:14
lambdayokay13:14
lambdaybut that makes it bad to look at :-/13:14
@HeikoSthe Complex type is not supported by shogun, so nothing that can be seen from outside a class should have this type13:14
lambdayHeikoS: alright... yes13:14
@HeikoSand linear operators you use all have T=float64_T13:14
@HeikoSbecause this is the type of the vector it is applied to13:15
lambdayimplementations?13:15
@HeikoSah shit13:15
@HeikoSso in the shifted system13:15
@HeikoSwe compute the result of (C + i)^(-1) v and then use the imaginary part only or something right?13:15
lambdayHeikoS: yes13:16
lambdayafter computing the system, only the imaginary part13:16
lambdayI mean, solving systems13:16
@HeikoSso actually, the result of the linear operator applied to a vector is complex13:16
lambdayHeikoS: yep, but we'll take the imaginary part of that13:16
@HeikoSthis is bad13:17
lambday:(13:17
@HeikoSwe run in all sorts of issues with that13:17
lambdaylike?13:17
@HeikoSsee, first we set T=float64_t in CRationalApprox13:17
lambdayyes13:18
@HeikoSso the type of the base classes will also be like this13:18
lambdayyes13:18
@HeikoSwhich is fine13:18
@HeikoSsince its a real matrix13:18
@HeikoSok13:18
lambdayhmm13:18
@HeikoSsorry, this works :)13:18
@HeikoSnext thing is13:18
@HeikoSwhen the jobs are created13:18
@HeikoSwe create LinearOperators which have to return complex value13:19
@HeikoSs13:19
lambdayyes13:19
@HeikoSand we cannot hide this away13:19
lambdaycocg_m takes care of that13:19
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has joined #shogun13:19
@HeikoSbut then is has a field m_operator where T=Complex13:19
@HeikoSwhich we cannot do13:19
lambdaywe don't have to make operator complex13:20
@HeikoSif it is real valued though13:20
@HeikoSapply() will return a real valued vector13:20
@HeikoSam I correct?13:22
lambdayHeikoS: that's fine, I guess?13:23
lambdayhttps://github.com/lambday/KRYLSTAT/blob/master/cocg_m/eigen/eigen_cocg_m.h#L23413:23
@HeikoSwhats that?13:24
lambdaylinear operator applies on real, returns real vector.. in cocg_m13:25
@HeikoSah thats in COCG_M13:25
lambdaythat's where we need this apply(), no?13:26
@HeikoSso this solves a real system and then internally loops over the complex shifts13:26
lambdayhmm13:26
@HeikoSyes, should be fine in COCG_M then13:27
@HeikoSbut what about solving all systems seperately13:27
lambdayHeikoS: that's problematic :-/13:27
lambdaywill we ever use a complex direct solver? that was just for testing purpose, right?13:28
lambdaytesting whether the framework works or not13:28
lambdayconfused :(13:28
@HeikoSlambday yes, however, a) we still want this and b) also for cg, the matrix you multiply is compelx13:28
@HeikoSso you get a complex result13:29
@HeikoSlambday: btw complex direct solvers are no problem13:29
lambdayhmmm13:29
@HeikoSthe problem is how to store the matrices without adding a new type to shogun13:29
@HeikoSwe can currnetly only have non-complex types13:30
@HeikoSso the operator applied to a vector cannot return somthing complex13:30
@HeikoSI see two possible solutions:13:30
@HeikoS1) add complex type to shogun (extremely messy but possible)13:31
@HeikoS2) make the complex linear operator operator one vectors where the first half is for real and the second half is for imag. Internally, we can then construct the complex systems. Even more messy13:31
@HeikoSsonne|work, lisitsyn what do you think about adding a new type complex_t to shogun?13:32
@HeikoShow much work would that be?13:32
@HeikoStypemaps, serialisation, lots of lots of switch statements, what else?13:32
sonne|workthe typedef is just 3 lines. supporting it natively for typemaps etc means writing it for each interface13:33
sonne|workall doable13:33
@HeikoSsonne|work: and feasible? did you ever consider complex types?13:33
@HeikoSlambday: maybe we should do this first then13:34
sonne|workHeikoS: no never needed it but it is not hard to do. however problem is that we might want to limit ourselves to float64_t for complex13:34
sonne|workto not have even more types13:34
lambdayadding complex... hmm13:35
@HeikoSsonne|work: yes I would also go for complex64_t13:35
@HeikoSor complex128_t ? I dont know how I would call it13:35
lambdayHeikoS: by the way, only the complex shifts makes the system complex, right? those are taken care separately in COCG_M, no? apply only applies the real matrix part - or may be I misunderstood13:36
lambday:(13:36
sonne|workHeikoS: 64_t13:36
@HeikoSlambday: you are right13:36
@HeikoSfor COCG_M it is fine13:36
@HeikoSbut what if we want to solve all systems seperately?13:37
lambdayHeikoS: got your point13:37
@HeikoSthen we need a cg solver for a complex system13:37
lambdayyes13:37
@HeikoSlambday: but since I would like to see this included (one can precondition each system seperately and gain speed)13:37
@HeikoSwe have to think about it13:37
@HeikoSlambday: but lets start with the COCG_M13:38
@HeikoSthe other one only makes sense once we have a cluster implementation13:38
lambdayHeikoS: what have we decided regarding the complex then?13:38
lambdayadd to shogun, or use separate vectors for real and imag13:38
@HeikoSlambday: haha, sorry for all the confusion :)13:38
@HeikoSso we will add complex64_t to shogun13:38
@HeikoSbut later13:39
@HeikoSonce COCG_M works, we can work on this13:39
lambdayHeikoS: no problem :D13:39
lambdayHeikoS: no I mean, even for COCG_M, we need shift vectors13:39
lambdaywhich are complex13:39
@HeikoSlambday, right13:39
@HeikoShmm13:39
@HeikoSso: either two float64_t vectors and you construct the complex numbers internally13:40
@HeikoSor add the complex64_t now13:40
@HeikoSif we go the first way, we have to change it later13:40
@HeikoSbut it might be easier for now13:40
lambdaynow is better :D cause handling all real and imag in arithmatic operation will be deadly :(13:40
@HeikoSlambday: you can just construct a vector of std::complex internally13:41
@HeikoSjust hide it to the outside world13:41
@HeikoSso no problem13:41
@HeikoSand easy to change later13:41
lambdaylike the one I did in elliptic curve functions? or you mean, no typedef at all13:41
lambdayoh that's not for vector13:42
lambdayhere we need vectors - sorry :(13:42
@lisitsynHeikoS: uh a lot13:42
@HeikoSlambday: this needs to be changed later13:43
@HeikoSas said, we cannot have Real or Complex in any interface, only shogun types13:43
@HeikoSotherwise we need to hide those to the modular/static interfaces13:43
@HeikoSah they are high precision anyway13:44
@HeikoSso we need to hide them13:44
@HeikoSthey can stay as they are13:44
@HeikoSlambday: okay, lets add the complex64_t now. Then this discussion is over13:46
lambdayHeikoS: okay13:46
lambdaythen we can have SGVector<complex64_t>13:46
@HeikoSyep13:46
lambdaycool13:46
lambday:)13:46
@lisitsynHeikoS: recalling porn things your project is really deep penetration13:47
@HeikoSlisitsyn: good stuff :)13:47
@HeikoSlambday: have  a look into the type stuff13:47
@HeikoSits a bit messy but should be possible13:47
@lisitsynHeikoS: I didn't follow your conversation sorry - are there unanswered questions?13:47
lambdayHeikoS: okay, I'm checking13:47
@HeikoSlisitsyn: solution to all of them: we will add complex64_t to shogun,13:48
lambdaywhere do we define this types in shogun?13:51
-!- vgorbati [~anonymous@77.47.132.39] has joined #shogun13:52
@HeikoSlambday:  common.h13:52
lambdayso, we need to add typedef std::complex<float64_t> complex64_t13:53
@HeikoSlambday: and DataType.h13:53
@HeikoSlambday: yes first thing13:54
@HeikoSthen the new SG primitive tye13:54
@HeikoSwhich might cause problems13:54
@HeikoSadd to your todo list: serialization, modular typemaps, check all switch over ptype13:55
lambdayI need to understand how the existing ones work first :(13:55
@HeikoSlambday: yes, take your time13:55
@HeikoSso we have this shogun type object13:55
@HeikoSTSGDataType13:56
lambdayhmm13:56
lambdayadd it to primitive type, right?13:56
@HeikoScontains primitive type, container type(vector,matrix), struct type (string,sparse)13:56
@HeikoSyes13:56
lambdayhmmm13:56
@HeikoSTSGDataType is used by shogun parameter framework to represent things13:57
@HeikoSfor example for serialisation, model-selection, equals method13:57
@HeikoSone has to register parameters with the SG_ADD macro13:57
@HeikoSwhich calls m_parameters->add of the current object13:57
@HeikoSin there, we need new methods for complex13:57
@HeikoSshould be a few13:58
@HeikoSin the same style as the existing ones13:58
@HeikoSthen we have to check serialisation13:58
@HeikoSwhich writes types to files13:58
@HeikoSthen modular maps which map the type the the type of the language (eg python complex)13:59
@HeikoSlambday: browse through the code a bit, take your time13:59
lambdayHeikoS: yes... and by new methods, we'll be needing all the operators, right?13:59
@HeikoSI will have lunch now and have to do some work afterwards, but will be around during the day, so ask if anything uclear13:59
lambdayoverloaded13:59
@HeikoSlambday: what do you mean by new operator?13:59
-!- vgorbati [~anonymous@77.47.132.39] has quit [Ping timeout: 256 seconds]14:00
@HeikoS+ - * += etc?14:00
lambdayyes yes14:00
@HeikoSwe get that from std::complex dont we?14:00
lambdayyes14:00
lambdayokay so nothing to do for that in shogun14:00
@HeikoSno thats all fine14:00
@HeikoSits very easy to add the type14:01
@HeikoSa bit more work to support it properly :)14:01
@HeikoSman, I realise that I type way too much :)14:01
@HeikoSso will get lunch now, see you in a  bit14:01
lambdayHeikoS: hmm.. I'm checking.. :)14:02
lambdayHeikoS: lol I do that too14:02
lambdaynot here though cause most of the time I'm confused :D14:02
lambdayyes have lunch14:02
lambdaysee you :)14:02
lambdayi too am going for some tea14:03
lambdaybe right back14:03
@HeikoSchai :)14:03
lambdayHeikoS: yes garam chai :D14:05
lambdaygaram=hot14:05
lambdayciao :)14:05
-!- nube [~rho@49.244.111.24] has joined #shogun14:10
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Ping timeout: 250 seconds]14:12
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun14:13
-!- mode/#shogun [+o iglesiasg] by ChanServ14:13
-!- van51 [~van51@195.251.235.71] has quit [Quit: Leaving.]14:46
@HeikoSlambday:  back14:55
-!- foulwall [~foulwall@li379-21.members.linode.com] has quit [Ping timeout: 256 seconds]15:11
-!- foulwall_ [~foulwall@2001:da8:215:503:4907:15c7:758:7cf5] has joined #shogun15:17
-!- foulwall_ [~foulwall@2001:da8:215:503:4907:15c7:758:7cf5] has quit [Remote host closed the connection]15:17
-!- foulwall [~foulwall@li379-21.members.linode.com] has joined #shogun15:20
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Quit: Page closed]15:24
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has joined #shogun15:25
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun15:53
-!- mode/#shogun [+o iglesiasg] by ChanServ15:53
-!- zxtx [~zv@ool-457e751d.dyn.optonline.net] has quit [Ping timeout: 248 seconds]15:56
-!- zxtx [~zv@ool-457e751d.dyn.optonline.net] has joined #shogun15:59
-!- goldbug [~goldbug@global-2-1.nat.csx.cam.ac.uk] has joined #shogun16:07
lambdayHeikoS: back16:15
-!- goldbug [~goldbug@global-2-1.nat.csx.cam.ac.uk] has quit [Ping timeout: 240 seconds]16:25
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has quit [Quit: Leaving.]16:30
-!- gsomix [~gsomix@37.61.179.240] has quit [Quit: Leaving]16:32
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has joined #shogun16:36
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Ping timeout: 250 seconds]16:48
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun16:50
-!- foulwall [~foulwall@li379-21.members.linode.com] has quit [Remote host closed the connection]16:52
@sonney2kvan51, around?17:08
van51sonney2k: yea17:09
van51but on a skype call atm17:09
@sonney2kvan51, ahh k17:10
@sonney2kvan51, once you are free please say so :)17:10
@sonney2kshogun-buildbot, force build nightly_default17:11
shogun-buildbotbuild forced [ETA 53m10s]17:11
shogun-buildbotI'll give a shout when the build finishes17:11
@sonney2khushell, have a look at http://shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/1170/steps/compile%20python_modular/logs/warnings%20%2817%2917:12
-!- foulwall [~foulwall@2001:da8:215:c252:addf:5916:dba8:873b] has joined #shogun17:12
@sonney2khushell, there still are some warnings in MulticlassOneVsRestStrategy that you might want to fix17:12
@sonney2kfoulwall, so how did it go? superhero?17:12
pickle27sonney2k: I saw that the test failed thats the same result I got on my system, for some reason there are 2 differences in the classification with the Eigen3 QDA17:13
pickle27it passes the synthetic test though17:13
foulwallsonney2k: not bad~17:14
@sonney2kpickle27, which tests fail?17:17
@sonney2kfoulwall, congrats then!17:17
foulwallsonney2k: thanks, I can do some work tonight~~17:18
@sonney2kfoulwall, when do you have your next exam?17:18
pickle27the python modular one on the fm_train_real.dat has a slightly different output17:18
pickle27sonney2k: but the unit test I wrote passes17:18
@sonney2kpickle27, compared to?17:18
foulwallsonney2k: 1 month later, 3 exams left17:19
pickle27sonney2k: the unit test is synthetic perfectly separated data and it is tested on the training input17:19
@sonney2kpickle27, I don't see anything that fails https://travis-ci.org/shogun-toolbox/shogun17:19
@sonney2khttp://shogun-toolbox.org/buildbot/waterfall17:19
pickle27the python example is on real data and doesn't produce 100 percent results the old and new version just make 2 different mistakes17:20
@sonney2kfoulwall, ok nice!17:20
pickle27it was on the build bot17:20
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun17:20
shogun-notifier-shogun: Soeren Sonnenburg :develop * 819f7de / src/shogun/mathematics/Random.h: https://github.com/shogun-toolbox/shogun/commit/819f7de7bb608a8d057e9ac6a6567cba6a39206617:20
shogun-notifier-shogun: add some documentation to the random() functions17:20
pickle27sonney2k: http://shogun-toolbox.org/buildbot/builders/nightly_default/builds/408/steps/test/logs/stdio17:21
@sonney2kpickle27, ohh but this is from last night where everything was still kaboom17:21
@sonney2kpickle27, the current one is the python_modular one and it is happy17:21
pickle27sonney2k: huh okay great17:22
@sonney2kpickle27, so just do you performance tests - and let the buildbot do its work :D17:22
pickle27sonney2k: right you wanted a benchmark17:22
@sonney2kfoulwall, what I consider most important is to get the structure of the demo's right17:23
pickle27sonney2k: you want between lapack and the new version?17:23
@sonney2kfoulwall, I mean general enough such that we don't have much code duplication17:23
@sonney2kpickle27, yeah and in general17:23
pickle27sonney2k: okay!17:23
foulwallsonney2k: I'm also worrying about that17:23
@sonney2kpickle27, I think we should provide benchmarks for each and every example in shogun but hey17:24
pickle27sonney2k: yeah I'll put the results in a comment block17:24
@sonney2kfoulwall, so for data it is pretty clear: it can come from a number of sources like artificially generated (datagenerator class, could be loaded from a file or input by the user via just clicks)17:24
foulwallsonney2k: I'll have a try? seperate the demos into modules and reuse them17:24
@sonney2kthen we have the coordinate system for drawing points - that is shared for everything17:25
@sonney2kand then the task specific machines, e.g. regression machines, binary classifiers, multiclass classifiers17:26
@sonney2kclustering (each one is rather specific...)17:26
@sonney2kthen dimension reduction (aka converters - we can use lisitsyn's example there)17:26
van51sonney2k: ok, I'm free now17:27
@sonney2kand then some illustrations for kernels, distances, loss, preprocessors, evaluations17:27
@sonney2kfoulwall, if we get the structure right adding one more method is like writing 5-10 lines17:27
foulwallsonney2k: ok~~ I'll rewrite the structures for reuse.17:27
@sonney2kfoulwall, or make a plan and discuss it before doing the actual work17:28
@sonney2kvan51, we have to talk later I have to leave the train now17:28
@sonney2khope you will be around in the evening17:28
van51sonney2k: I'll be around for 2-3 more hours for sure17:29
van51and then by night again17:29
foulwallsonney2k: so just give me some time and let me come up with a plan.17:29
foulwallsonney2k: the exams are easy to deal, I'll start work for 6 hours/day from tomorrow if i'm accepted.17:31
shogun-buildbotbuild #878 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/878  blamelist: Soeren Sonnenburg <sonne@debian.org>17:36
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Ping timeout: 250 seconds]17:38
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has joined #shogun17:41
-!- mode/#shogun [+o iglesiasg] by ChanServ17:41
shogun-buildbotbuild #409 of nightly_default is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/40917:46
-!- lambday [67157e4e@gateway/web/cgi-irc/kiwiirc.com/ip.103.21.126.78] has quit [Quit: http://www.kiwiirc.com/ - A hand crafted IRC client]18:12
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has joined #shogun18:22
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has quit [Client Quit]18:25
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has joined #shogun18:25
-!- lisitsyn [~lisitsin@mxs.kg.ru] has left #shogun []18:25
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun18:27
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has quit [Quit: Leaving]18:34
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has joined #shogun18:34
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has quit [Client Quit]18:37
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has joined #shogun18:37
-!- vgorbati_ [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun19:28
-!- lisitsyn [~blackburn@188.168.13.232] has joined #shogun19:30
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds]19:30
lisitsynahaha forgot *both* my room keys and notebook AC adapter19:30
-!- vgorbati_ is now known as vgorbati19:31
lisitsynI am probably with you gsocers tonight19:31
lisitsynnot*19:31
lisitsynnot with you :D19:31
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds]19:46
@sonney2klisitsyn, poor guy!19:54
lisitsynsonney2k: yes I got a bad luck19:54
@sonney2kI am probably too dead to stay awake long :/19:54
lisitsynsonney2k: in an hour19:54
@sonney2kaccumulated lack of sleep19:54
@sonney2kno idea if I survive that long19:55
lisitsynmy battery won't survive for sure19:55
@sonney2kHeikoS, around?19:55
@HeikoSsonney2k:  yes19:55
@sonney2kHeikoS, it seems to be more problematic to organize rooms for the hands-on workshop at TU19:56
@HeikoSokay?19:56
@HeikoSTUB is so big, how can that be?19:56
@sonney2kmaybe we should try to sit at c-base19:56
lisitsynlets sit at tiergarten!19:57
lisitsynwith vodka!19:57
@sonney2kHeikoS, it is lots of stuff like they want money, each participant by name19:57
@sonney2ketc etc19:57
@HeikoSannoying19:57
@HeikoSand now?19:57
@HeikoSc-base?19:57
@HeikoSdid you ask them?19:57
lisitsyntiergarten!19:57
lisitsyn:D19:57
@HeikoShaha :D :)19:57
lisitsynI've never been to tiergarten19:58
lisitsynis it nice?19:58
lisitsynalright I've never been ANYWHER19:58
lisitsyn:D19:58
@HeikoSGermany? Europe? NotRussia(tm)? :D20:01
@sonney2kHeikoS, no I am planning to though. I guess the group will be downsized quite a bit anyway on the post-official workshop day20:01
lisitsynHeikoS: never been anywhere abroad20:01
@HeikoSsonney2k: in that case, should be fine20:01
@HeikoSif cbase agrees20:01
@iglesiasghow many applicants for the workshop this far?20:02
@sonney2kI should have done the shogun & gsoc thing back when I was in academia. then one could have done the post conference sprints20:02
@HeikoSwell thats life :)20:03
@sonney2kHeikoS, there are other suitable spots in berlin - star bucks in the worst case :D20:03
@HeikoShaha :d please not20:05
@sonney2kI will ask around20:05
@sonney2kHeikoS, I mean places like http://hackerspaces.org/wiki/Berlin20:06
@sonney2kRaumfahrtagentur would be something fancy too :D20:06
@HeikoSyes, that sounds cool20:06
-!- iglesiasg [c1934d16@gateway/web/freenode/ip.193.147.77.22] has quit [Quit: Page closed]20:07
-!- van51 [~van51@athedsl-320452.home.otenet.gr] has quit [Quit: Leaving.]20:07
@sonney2kHeikoS, the big issue really is that I am away on that day so that makes it difficult20:07
@HeikoSI see20:08
@HeikoSso we should set beforehand right20:08
lisitsynbatterryyyyy dyyyying20:08
@HeikoSthen ask cbase first and then we will see20:08
@HeikoSlisitsyn: bye :)20:08
lisitsynI am alive yeet20:08
lisitsynnoooo20:08
-!- lisitsyn [~blackburn@188.168.13.232] has quit [Quit: Leaving.]20:09
@sonney2kHeikoS, we should really think about spending some money on gsoc'ers maybe some can join20:10
@HeikoSthe new ones?20:10
@HeikoSwould be cool20:10
@HeikoSbut might be expensive20:10
@HeikoSfor the non-europeans20:10
@sonney2kyeah :/20:10
-!- goldbug [~goldbug@global-1-26.nat.csx.cam.ac.uk] has quit [Ping timeout: 245 seconds]20:14
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]20:20
-!- deerishi [c649b206@gateway/web/freenode/ip.198.73.178.6] has joined #shogun20:24
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has joined #shogun20:25
lambdayHeikoS: hi...20:25
@sonney2kohh man I really have to sleep now everyone relax. those of you that get accepted for gsoc - congrats. for those of you that didn't get in this year. don't take it personally we have 8 slots (which is a lot in gsoc terms) but had >50 students applying. if you are still interested in working with us we certainly would be more than happy to work with you too.20:39
@sonney2kand if you want to know more about our decision process (why you did / did not get in) ask me personally20:40
-!- gsomix [~gsomix@185.6.239.144] has joined #shogun20:47
gsomixgood evening20:47
hushellsonney2k: I should check these warning msg myself :( I'll fix it quickly20:56
-!- iglesiasg [d58f32c1@gateway/web/freenode/ip.213.143.50.193] has joined #shogun21:17
-!- mode/#shogun [+o iglesiasg] by ChanServ21:19
@iglesiasgcongrats to everyone accepted!21:19
gsomixsonney2k, sonne|work hey. just got mail. thanks!21:19
hushellcong!!! guys21:20
-!- gsomix is now known as lisitsyn21:22
lisitsyniglesiasg, hushell, votjakovr, lambday, foulwall, pickle27  congrats21:22
@iglesiasgthanks lisitsyn! :)21:23
pickle27lisitsyn: thanks!21:23
foulwallthanks lisitsyn21:23
votjakovrlisitsyn: thanks :)21:23
lisitsynyou all have done a great work pre-season21:23
hushellthank you! lisitsyn21:23
lisitsyniglesiasg, hushell, votjakovr, lambday, foulwall, pickle27 let us have a meeting later - we will let you know21:25
@iglesiasgsure, nice idea21:25
lisitsynif you are curious who is the 8th one it is van5121:26
lisitsynI am sorry to have no student this year21:26
lisitsynbut that's how we vote21:26
pickle27thats awesome, I'mm really looking forward to it!21:26
@iglesiasglisitsyn: I am sorry too :(21:27
pickle27lisitsyn: aren't I working with you?21:27
@iglesiasglisitsyn: but then you will have more time to co-mentor me :)21:27
lisitsynpickle27, yes sure I'll will review your code21:27
@iglesiasgand the others too21:27
lisitsynpickle27, but I would have to learn BSS just like you :)21:28
pickle27lisitsyn: ah right you were co-mentor or something for me too21:28
lisitsynI know about dimension reduction a bit more21:28
pickle27lisitsyn: me too!21:28
lisitsynwe really hope that the most of you are here to stay21:29
lisitsynthat's really important for us to have guys like iglesiasg21:29
pickle27lisitsyn: yeah I deffs hope to stick around!21:29
lisitsynanyway we are going to have a dense summer21:30
lisitsynwe will schedule a meeting one of next days21:30
hushellreally happy to working with you guys!21:30
@iglesiasgdefinetely, lot of new and cool stuff coming up21:31
pickle27yeah let me know about the meeting!21:31
-!- deerishi [c649b206@gateway/web/freenode/ip.198.73.178.6] has quit [Quit: Page closed]21:31
pickle27I am going to be at a conference for a few days so email might be more reliable but I'll try and be on IRC too21:31
lisitsynpickle27, yeah that21:32
lisitsynthat's for sure21:32
lisitsynguys about some rules21:32
lisitsynonce we start21:32
lisitsynyou will have to send a weekly report21:32
lisitsynsay monday or so21:32
lisitsynI'd like to note that we have to make it mandatory21:33
lisitsynthere you would have just to say what you have done last week and what do you plan to do next week21:33
lisitsynjust to clarify - if you had some issues and had no time - it is still muuuch better to send a report about that21:34
lisitsyniglesiasg, hushell, votjakovr, lambday, foulwall, pickle27 alright? :)21:34
@iglesiasglisitsyn: of course21:34
foulwalllisitsyn: gotcha21:35
hushellsure, that's the way I'd like to schedule myself21:35
lisitsynwe will have a few meetings through the summer21:35
lisitsynnear milestones21:35
lisitsynso we will have one in next days21:36
lisitsynthen once we started21:36
lisitsynthen mid-term21:36
lisitsynthen final meeting21:36
foulwalllisitsyn: ok21:36
lisitsynthere we can discuss what we are heading to21:36
lisitsyniglesiasg, hushell, votjakovr, lambday, foulwall, pickle27 do you guys have any questions?21:36
foulwalllisitsyn: ok21:36
hushellI am good21:37
@iglesiasglisitsyn: yep, are we going to stick to IRC for these meetings this year or use something more interactive like google hangout?21:37
pickle27yeah that sounds good!21:37
lisitsyniglesiasg, I don21:37
lisitsynt know21:37
lisitsynwhat others think21:37
@iglesiasgok21:37
pickle27might be cool to do one hangout but IRC is probably easier for regular meetings21:37
@iglesiasglet's see what sonne|work, sonney2k thinks too21:38
@iglesiasgI agree with pickle2721:38
lisitsynyeah he will have some opinion on that for sure21:38
lisitsynalright21:38
hushelllet everybody talk also a great idea21:38
lisitsynwill pass gsomix's notebook to gsomix back then :)21:38
lisitsyncongrats again guys21:39
-!- lisitsyn is now known as gsomix21:39
gsomixspirit of lisitsyn. huh21:40
gsomixso, congrats too21:40
pickle27thanks!21:41
foulwallgo sleeping now, it's 3:xx am here... too late, thanks to all my friends, cu21:42
pickle27alright guys I am out for now too, talk to you later!21:42
-!- pickle27 [~kevin@rcv3-lab-pc.ee.queensu.ca] has quit [Quit: Leaving]21:42
-!- goldbug [~goldbug@global-2-11.nat.csx.cam.ac.uk] has joined #shogun21:42
-!- goldbug [~goldbug@global-2-11.nat.csx.cam.ac.uk] has quit [Client Quit]21:42
-!- goldbug [~goldbug@global-2-11.nat.csx.cam.ac.uk] has joined #shogun21:43
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has joined #shogun21:43
hushellgood nite21:45
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has quit [Ping timeout: 250 seconds]21:58
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has joined #shogun22:04
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has quit [Client Quit]22:05
-!- goldbug [~goldbug@global-2-11.nat.csx.cam.ac.uk] has quit [Ping timeout: 252 seconds]22:33
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has joined #shogun22:34
lambdayguys, I had an urgent sysad job in my lab :( missed the chat22:34
lambdaychecking the log22:35
-!- goldbug [~goldbug@global-2-11.nat.csx.cam.ac.uk] has joined #shogun22:35
gsomixnite22:46
lambdaynight22:56
lambday:)22:56
-!- vgorbati [d4029f22@gateway/web/freenode/ip.212.2.159.34] has quit [Ping timeout: 250 seconds]23:03
-!- goldbug [~goldbug@global-2-11.nat.csx.cam.ac.uk] has quit [Quit: Leaving]23:14
-!- votjakovr [~votjakovr@host-46-241-3-209.bbcustomer.zsttk.net] has quit [Quit: Leaving]23:18
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has quit [Ping timeout: 250 seconds]23:32
--- Log closed Tue May 28 00:00:20 2013
« Sunday, 2013-05-26 Index Tuesday, 2013-05-28 »

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