Open in new window / Try shogun cloud
--- Log opened Mon Jul 15 00:00:32 2013
thoralfBye.00:09
-!- thoralf [~thoralf@24-134-81-194-dynip.superkabel.de] has quit [Quit: Konversation terminated!]00:09
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun00:55
-!- mode/#shogun [+o iglesiasg] by ChanServ00:55
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]01:50
shogun-buildbotbuild #458 of nightly_default is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/45804:27
-!- gsomix [~gsomix@185.6.239.196] has quit [Quit: Leaving]05:22
-!- nube [~rho@116.90.239.3] has joined #shogun05:49
-!- nube [~rho@116.90.239.3] has quit [Ping timeout: 256 seconds]06:01
-!- nube [~rho@116.90.239.13] has joined #shogun06:15
-!- foulwall [~user@2001:da8:215:6100:c01b:d373:53b1:bd42] has joined #shogun06:17
-!- foulwall [~user@2001:da8:215:6100:c01b:d373:53b1:bd42] has quit [Ping timeout: 245 seconds]06:30
-!- foulwall [~user@2001:da8:215:6901:452b:8de3:b659:6495] has joined #shogun06:38
-!- foulwall` [~user@2001:da8:215:503:8587:b3a2:46ae:c7d6] has joined #shogun06:40
-!- foulwall [~user@2001:da8:215:6901:452b:8de3:b659:6495] has quit [Ping timeout: 245 seconds]06:42
-!- nube [~rho@116.90.239.13] has quit [Ping timeout: 260 seconds]08:04
-!- nube [~rho@116.90.239.3] has joined #shogun08:18
-!- nube [~rho@116.90.239.3] has quit [Ping timeout: 256 seconds]08:58
-!- nube [~rho@116.90.239.13] has joined #shogun09:17
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has joined #shogun09:18
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 264 seconds]09:21
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has joined #shogun09:35
-!- lambday [67157d36@gateway/web/freenode/ip.103.21.125.54] has quit []09:51
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has joined #shogun10:24
-!- foulwall` [~user@2001:da8:215:503:8587:b3a2:46ae:c7d6] has quit [Ping timeout: 245 seconds]10:45
-!- foulwall` [~user@2001:da8:215:503:801e:26a9:1a95:4b99] has joined #shogun10:58
@wikingsonney2k: ping11:04
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has joined #shogun11:10
@sonney2kwiking, hey there!11:16
@sonney2kfoulwall`, neat! so the whole machinery beneath is already working?11:18
@wikingsonney2k: can we fix buildbot nightly? :)11:18
@sonney2kfoulwall`, just the centering / extracting the region is missing and I guess you should also load the model just once for this demo if it takes too long11:19
@sonney2kwiking, what's wrong with it?11:19
@wikingsonney2k: i mean to generate the .tar.gz of the binaries?11:19
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has joined #shogun11:19
@sonney2kwiking, we just need a make target for that and then we can do it11:20
@wikingok11:20
@sonney2kwiking, did sergey come to the 3nd workshop day btw?11:20
@wikingbtw are we doing this for the r_modular: PKG_LIBS="example.cxx" R CMD SHLIB -o example example_wrap.cpp11:20
@wiking" R is also sensitive to the name of the file extension in C and C++ mode. In C++ mode, the file extension must be .cpp rather than .cxx for the R compile command to recognize it. If your C++ code is in a file using something other than a .cpp extension, then it may still work using PKG_LIBS:11:20
@wikingswig"11:20
@sonney2kwiking, no we are doing something totally custom11:21
@wikingsonney2k: oh i see11:21
@sonney2kwiking, I did a bit of R debugging in the last days11:22
@sonney2kwiking, at least my minimal example worked with ref/unref jsut fine11:22
@sonney2kwiking, now I am trying to get a minimal shogun class to work11:22
@sonney2ksay BinaryLables11:22
@sonney2k*Labels11:23
@sonney2kif this works then it is just some shogun thing11:23
@sonney2kbut it is complex...11:23
foulwall`ok, sonney2k, gotcha, will send a pr soon.11:35
-!- thoralf [~thoralf@enki.zib.de] has joined #shogun11:38
thoralfHey.11:38
thoralfSomeone could change the topic. ;)11:38
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has left #shogun ["PING 1373881241"]11:40
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has joined #shogun11:41
-!- ChanServ changed the topic of #shogun to: "Shogun Machine Learning Toolbox | Devs are usually around between 10am-10pm UTC | GSoC 2013 is running http://bit.ly/16LaK5Z | When asking questions please be patient or use the mailing list. We always answer but it might take a while. | Channel logs http://bit.ly/eQDk8Q"11:46
@sonney2kwiking, did you have the chance to check the videos of the workshop?11:48
lambdaysonney2k: hi11:51
@sonney2klambday, hey there!11:53
lambdaysonney2k: I need to add a different dense_dot in SGSparseVector, something that takes another template argument... say, when I need to compute sparse-matrix-vector product when matrix is complex64 and vector is float6411:54
lambdaysonney2k: would that be okay?11:55
@sonney2klambday, it makes sense to have these12:02
lambdaysonney2k: okay... so adding a few of them, complex64, float64, int etc....12:02
@sonney2klambday, I wanted to have this in CDotFeatures but IIRC deriving from a templated  class can be problem12:03
@sonney2ka12:03
thoralfsonney2k: I thought you'll in holiday now?12:05
thoralfyou're12:05
lambdaysonney2k: umm.. problem as in?12:05
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun12:09
-!- mode/#shogun [+o iglesiasg] by ChanServ12:09
thoralfHey iglesiasg12:09
@iglesiasggood morning!12:09
van51hey12:10
van51sonney2k: I've sent a PR, when you have time please review it :)12:10
van51sonney2k: I've started working on the streaming version of those12:10
van51sonney2k: if you want to discuss anything I'll be around12:11
van51lambday: I think it will be a bit messier to extend a templated class12:11
@iglesiasgthoralf: I will fix the SO multiclass example ASAP (I want to finish an LMNN thing first). But if you want to do it, feel free ;)12:11
van51lambday: I tried it two-three days ago with DenseFeatures (which are templated)12:11
thoralfiglesiasg: No worries.12:11
van51lambday: ofc I'm not a c++ expert (yet :P) so that may be we it troubled me12:11
van51why*12:12
lambdayvan51: yes templates are real pain :(12:12
lambdayvan51: well I am thinking of having something like this12:12
thoralfiglesiasg: I need multilabel instead of multiclass.  So I anyway will write my own example/classes.12:12
lambdaytemplate<class T1> class { template<class T2> void foo(T2 a); };12:13
van51lambday: yea I saw what you wanted to do, with the sparse product12:14
lambdayoops I forgot the class name itself :D12:14
@iglesiasgthoralf: Aham, I understand.12:14
lambdayvan51: yes12:14
van51lambday: i think it will be really useful :)12:14
lambdayvan51: I hope so.. I will be using those anyway12:14
thoralfiglesiasg: I spend a night reading your SO stuff and I think I need some modifications for multilabel.12:15
lambdayby the way, how did the workshop go, guys?12:15
lambdaysaw the pics :)12:15
-!- nube [~rho@116.90.239.13] has quit [Quit: Leaving.]12:16
van51and what about the videos? :D12:16
@iglesiasgthoralf: So, first of all you need a new model, right? In multilabel you want to model dependencies between all pairs of classes, or not?12:16
lambdayyes yes what about the video :D12:16
@iglesiasgall pairs of labels, better said probably12:16
lambdayumm... make unit-tests now treats all warnings as errors :-/12:17
@iglesiasglambday: the workshop was pretty cool12:18
@iglesiasglambday: I enjoyed a lot the days in there. We had a couple of coding days during the weekend12:18
lambdayiglesiasg: awesome, man!!12:18
@iglesiasgvan51: I think wiking or sonney2k are editing them and later they will be uploaded to youtube12:19
van51iglesiasg: ok cool! thanks for letting me know12:19
@iglesiasglambday, van51: did you guys follow the live stream?12:20
lambdayiglesiasg: no I couldn't :(12:20
lambdayiglesiasg: will the slides be available?12:20
@iglesiasglambday: I thought they already were12:26
@iglesiasglambday: have you checked the logs? I think there were some links to them pasted here12:26
@iglesiasglambday: anyway, yes. I think they will be uploaded somewhere in the website once the videos are up and so on.12:27
lambdayiglesiasg: alright.. thanks.. I'll check it out :)12:32
thoralfiglesiasg: No, only between some classes.12:54
thoralfiglesiasg: It's a taxonomy.12:54
thoralfiglesiasg: I don't even need to learn label-label interactions.12:55
@iglesiasgthoralf: Aham! Got it.12:58
@iglesiasgthoralf: So you could as well use directly multiclass classification too. Each class corresponding to a particular choice of the labels.12:59
thoralfiglesiasg: How do I model that a class belongs to class i and class j?13:00
thoralfiglesiasg: btw.: I have 7000 labels - so "label combinations" won't work. ;)13:01
@iglesiasgthoralf: yeah, indeed that's way too many classes for doing that13:02
@iglesiasgthoralf: so if I understand correctly, for your labels you can probably use the Sequence labels, and for the features probably MatrixFeatures work well. What needs to be new is the StructuredModel13:04
-!- nube [~rho@49.244.49.62] has joined #shogun13:18
-!- gsomix [~gsomix@83.234.54.35] has joined #shogun13:22
thoralfiglesiasg: Something like this, yes.13:41
-!- foulwall` [~user@2001:da8:215:503:801e:26a9:1a95:4b99] has quit [Ping timeout: 245 seconds]13:53
thoralfSomebody who could merge my PR?  https://github.com/shogun-toolbox/shogun/pull/124314:25
-!- flxb_ [~flxb@master.ml.tu-berlin.de] has quit [Quit: Lost terminal]14:39
@iglesiasgthoralf: it seems there is some conflict with it, right?14:56
thoralfiglesiasg: No, travis wasn't triggered.14:57
thoralfI don't know why.14:57
@iglesiasgthoralf: but the PR cannot be merged from the github's interface14:58
@iglesiasgthoralf: "We can't automatically merge this pull request. Use the command line to resolve conflicts before continuing."14:58
thoralfiglesiasg: That's weird.14:59
thoralfI'll have a look.14:59
@iglesiasgthoralf: I have not seen it before either. I would try to rebase to latest develop and see if I get any conflict14:59
thoralfiglesiasg: Okay, I did a rebase and pushed it again (there was indeed a conflicting change) - let's hope github is able to merge it.15:07
thoralfiglesiasg: Travis runs, a good sign.15:07
@iglesiasgthoralf: good15:11
@iglesiasgthoralf: I am no real Makefile expert though, so I am probably not the right one to merge this. Do you need to get this merged soon?15:13
thoralfiglesiasg: I guess this happened because HeikoS PR got merged before mine - and introduced a conflict, which wasn't there before. ;)15:13
@iglesiasgprobably15:13
thoralfiglesiasg: Not really, but later merges mean more potential conflicts. ;)15:13
thoralfiglesiasg: I can explain my changes - would that help?15:14
@iglesiasgthoralf: I understand what's going on. But I don't know if it can potentially arise conflict with sometihng15:18
@iglesiasglunch time now!15:18
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun16:02
@iglesiasgre16:06
gsomixtoday is monday...16:17
gsomixhello all16:18
@iglesiasghello hello16:18
@iglesiasggsomix: what's wrong with Monday man? :)16:18
gsomixiglesiasg, report time! :)16:18
@iglesiasgoh yeah!16:19
-!- van51 [~van51@athedsl-399972.home.otenet.gr] has quit [Quit: Leaving.]16:49
-!- gsomix [~gsomix@83.234.54.35] has quit [Ping timeout: 264 seconds]16:51
-!- foulwall [~user@2001:da8:215:c252:19b6:f7da:70f4:e333] has joined #shogun16:59
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has quit [Quit: Leaving]17:05
-!- vgorbati [c3ee5cb1@gateway/web/freenode/ip.195.238.92.177] has quit [Quit: Page closed]17:24
-!- gsomix [~gsomix@188.168.3.86] has joined #shogun17:42
-!- gsomix [~gsomix@188.168.3.86] has quit [Read error: Connection reset by peer]17:44
-!- gsomix [~gsomix@188.168.3.86] has joined #shogun17:44
-!- foulwall [~user@2001:da8:215:c252:19b6:f7da:70f4:e333] has quit [Ping timeout: 245 seconds]18:13
-!- foulwall [~user@2001:da8:215:c252:19b6:f7da:70f4:e333] has joined #shogun18:15
-!- foulwall [~user@2001:da8:215:c252:19b6:f7da:70f4:e333] has quit [Ping timeout: 245 seconds]18:48
-!- HeikoS [~heiko@f052017018.adsl.alicedsl.de] has joined #shogun18:50
-!- mode/#shogun [+o HeikoS] by ChanServ18:50
thoralfHallo Heiko.18:51
thoralfFalls Du kurz Zeit hast: https://github.com/shogun-toolbox/shogun/pull/124318:52
thoralfiglesiasg said, he's no Makefile expert.  But I claim you are. ;)18:53
@HeikoSthoralf: Hi!19:03
@HeikoSthoralf: I just looked at it, your claim is only 50% true though19:05
@HeikoSlooks good to me19:05
@HeikoSbut Id like to wait for others to confirm that ;)19:05
@HeikoSthoralf: checking out the pythonevn stuff, pretty awesome19:05
thoralfYeah.19:05
@HeikoSwiking: around=?19:13
-!- shogun-notifier- [~irker@7nn.de] has joined #shogun19:20
shogun-notifier-shogun: van51 :develop * 14cbb3c / / (3 files): https://github.com/shogun-toolbox/shogun/commit/14cbb3c564a8f72b67d41929881d404e924b00dc19:20
shogun-notifier-shogun: HashedSparseFeatures class19:20
shogun-notifier-shogun: Soeren Sonnenburg :develop * 37b3940 / / (3 files): https://github.com/shogun-toolbox/shogun/commit/37b39407fbd2d51e0080ffa0e5bc0b3e260a793719:20
shogun-notifier-shogun: Merge pull request #1245 from van51/feature/hashing19:20
shogun-notifier-shogun:19:20
shogun-notifier-shogun: HashedSparseFeatures class19:20
-!- hushell [~hushell@c-24-21-220-10.hsd1.or.comcast.net] has joined #shogun19:21
@HeikoSlambday: currently looking into the clone bug you found19:30
@HeikoSlambday: thanks for reporting!19:30
shogun-buildbotbuild #1254 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1254  blamelist: van51 <vangelis_51@hotmail.com>19:31
lambdayHeikoS: hey!19:33
@HeikoShi19:34
lambdayHeikoS: I changed the template thing19:34
lambdayHeikoS: now will have to make sure that the unit-tests work properly and then I'll send PR19:34
@HeikoSok19:34
@HeikoSsounds good19:34
lambdayHeikoS: amazing about the engine and job :D19:35
lambdayHeikoS: I am in :D19:35
@HeikoSlambday: good to hear :) we would be very happy if you stayed with us after gsoc19:35
@HeikoSlambday: any reasons why you did not initialise  the m_operator in SparseMatrixOperator?19:35
lambdayHeikoS: wait... I am checking19:36
@HeikoSlambday: I think this is one source of the bug19:36
@HeikoSlambday: dont worry, I will investigate a bit nowq19:37
lambdayHeikoS: alright... I'll check it too19:37
lambdayso, in init I should do something like m_operator=SGSparseMatrix<T>();19:38
@HeikoSlambday: I think this is any automatically isnt it?19:39
lambdayHeikoS: yes that's why I didn't do anythng explicitely :-/19:40
@HeikoSlambday: you are right with that, my fault, checking other things now19:40
lambdayHeikoS: I am not sure what do you mean by initializing this one :(19:40
lambdayHeikoS: okay19:40
@HeikoSlambday: found the bug, its clone19:41
shogun-buildbotbuild #1050 of cyg1 - libshogun is complete: Success [build successful]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/105019:42
lambdayHeikoS: what is it?19:42
shogun-buildbotbuild #1051 of cyg1 - libshogun is complete: Failure [failed configure]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/cyg1%20-%20libshogun/builds/1051  blamelist: Soeren Sonnenburg <sonne@debian.org>19:43
@HeikoSsome NULL arrays are not copied, but ignored so they stay uninitialised19:43
shogun-buildbotbuild #1255 of bsd1 - libshogun is complete: Failure [failed test]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/1255  blamelist: Soeren Sonnenburg <sonne@debian.org>19:43
lambdayokay.. I'll add a unit test for this when you fix it19:46
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving]19:47
-!- travis-ci [~travis-ci@ec2-107-22-152-223.compute-1.amazonaws.com] has joined #shogun19:48
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/910470919:48
-!- travis-ci [~travis-ci@ec2-107-22-152-223.compute-1.amazonaws.com] has left #shogun []19:48
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has joined #shogun19:50
@HeikoSlambday: still around?19:53
@HeikoSlambday: so you registered sparse vectors as CT_SCALAR and ST_SPARSE19:53
-!- lisitsyn [~lisitsyn@92-240-133-94.clients.tlt.100megabit.ru] has joined #shogun19:53
lisitsynohh19:53
@HeikoSbut sparse matrices as CT_MATRIX and CT_SPARSE19:54
@HeikoSlisitsyn: hey!19:54
lisitsynHeikoS: hey19:54
@HeikoSlisitsyn: where are you? :)19:54
lisitsynHeikoS: just got back home19:54
lisitsynI am like pheew19:54
@HeikoSlisitsyn: good trip?19:54
lisitsynHeikoS: the issue is that I feel a bit pheeew about russia now19:55
@HeikoSlisitsyn: ah, Berlin got you? :D19:55
lisitsynHeikoS: yes totally19:55
@HeikoSlisitsyn: many Russians here, you should come ;)19:55
pickle27from what i'19:55
lisitsynHeikoS: I am afraid I could do that one day19:55
@HeikoSstill enjoying the city, will depart tomorrow19:55
lisitsynHeikoS: have you been to einstein stammhaus?19:56
pickle27sorry from what I've heard Berlin would make a lot of places seem pheew19:56
@HeikoSlisitsyn: not yet19:56
@HeikoSpickle27: yeh its kinda nice here19:56
lisitsynHeikoS: you should try their apfelstrudel and espresso I think!19:56
@HeikoSthe only thing is: if you are here, you tend to not work but do other nice things ;)19:56
@HeikoSlisitsyn: haha :) apfelstrudel and espresso! :)19:56
lisitsynnot really cheap (these two ones would cost like around 10EUR)19:57
lisitsynpickle27: well as I have not been to any foreign country before19:57
pickle27ah19:58
lisitsyn;)19:58
@HeikoSlisitsyn: really glad you liked it! :)19:58
pickle27travelling is fun!19:58
lisitsynpickle27: well I started to earn some enough money so I get it now ;)19:59
@HeikoSlambday: something is weird in the Parameter::add things for example19:59
@HeikoSParameter::add_matrix(SGSparseVector<int32_t>** param,19:59
@HeikoS  index_t* length_y, index_t* length_x,19:59
@HeikoS  const char* name, const char* description) {19:59
@HeikoSTSGDataType type(CT_MATRIX, ST_SPARSE, PT_INT32, length_y,19:59
lisitsynHeikoS: I should be back and check some other cities too!19:59
pickle27awesome!19:59
@sonney2klisitsyn, what did you do on sunday?19:59
@HeikoSlisitsyn: you should, Berlin is among the nicest though!19:59
lisitsynsonney2k: hauptbahnhof shopping :D19:59
@sonney2kpfff19:59
lisitsynwell at least I have found something for my mother (she has birthday in a few days)20:00
lisitsynHeikoS: oh man I want to get back already20:00
lambdayHeikoS: add matrix?20:00
lambdayHeikoS: I didn't add that20:01
@HeikoSlambday: ah I see20:01
lambdayoh20:01
lambdaynono20:01
@HeikoSlambday: even so, is that correct, see email20:01
@HeikoSlisitsyn: hehe20:01
@HeikoSlisitsyn: you should come back, and next time hang out with us some more :)20:01
gsomixlisitsyn, hi!20:01
lambdayHeikoS: checking20:02
lambdayHeikoS:20:05
@HeikoSlambday: ah found some more bugs in clone :(20:05
lambdayParameter::add(SGSparseVector<complex64_t> has all CT_SCALARs20:05
@HeikoSlambday: I know20:05
@HeikoSsonney2k: how are sparse vectors represented20:06
@HeikoSCT_SCALAR ot CT_VECTOR?20:06
lambdayI mean Parameter::add(SGSparseVector<T> has all CT_SCALARs20:06
@HeikoScurrenlty confused20:06
@sonney2kHeikoS, scalars yes20:06
@HeikoSsonney how about sparse matrix?20:06
@HeikoSsonney2k: and when do we use CT_VECTOR with ST_SPARSE?20:07
@sonney2kHeikoS, dirk designed it that way20:07
@HeikoSsonney2k: its fine, just need to know, so CT_VECTOR is never used along with ST_SPARSE ?20:08
@sonney2kHeikoS, sparse matrix would be CT_SPARSE IMHO20:08
@HeikoSCT_SPARSE?20:08
@HeikoSwe only have scalar, vector, matrix20:08
@sonney2kwell we don't have that yet20:08
@HeikoSand ST_NONE and ST_SPARSE20:08
@HeikoSsonney2k: well yes there are combinations20:09
@sonney2kI am not up-to-date20:09
@sonney2kthe idea was that one could have vectors/strings in a matrix20:09
@HeikoSsonney I see, makes sense20:09
@HeikoSsonney2k: but how about sparse vectors20:10
@sonney2kIMHO it makes things too complex20:10
@HeikoSso first thing is: sparse vectors of scalars20:10
@sonney2kwe don't need that20:10
@HeikoSthis would be CT_SCALAR and ST_SPARSE20:10
@HeikoSand then sparse matrices, how do we do that: in the code we have: ST_SPARSE and CT_MATRIX20:10
@HeikoSbut thats weird20:10
@HeikoSnot consistent20:10
@sonney2kHeikoS, well we don't have matrices of strings either20:12
@HeikoSsonney2k: no its all just scalar20:12
@HeikoSsonney2k: how do we represent sparse matrices of floats?20:12
thoralfHeikoS: Lists of sparse vectors ;)20:13
@sonney2kwell I would be totally fine if we just had a container type scalar/vector/...sparse_matrix20:14
@sonney2kwith some primitive type20:14
@HeikoSsonney2k: I dont want to change things, just understand20:14
@sonney2kand rather sgstring as container type20:14
@sonney2kI am saying that this is how I would have liked it to be20:15
@HeikoSsonney2k: ok I agree on that, but for now I want to fix that bug20:15
@sonney2know we have for strings IIRC CT_VECTOR with ST_STRING20:15
@HeikoSand its related to sparse matrix things20:15
@sonney2kso that is why ST_SPARSE is the correct thing to do with this framework20:16
@HeikoSsonney2k: I am just wondering how to represent sparse matrices correct20:16
@sonney2kso you have a vector of sparse20:16
@HeikoSyes, thats CT_SCALAR and ST_SPARSE20:17
@sonney2kyeah20:17
@HeikoSand now I want sparse matrix, which seems to be CT_MATRIX and ST_SPARSE20:17
thoralfI like the scipy compressed sparse matrices.20:17
@sonney2kno and no20:17
@sonney2kshogun sparse vector === CT_SCALAR + ST_SPARSE20:18
@HeikoSsonney2k: I found this:20:18
@sonney2kshogun sparse matrix == CT_VECTOR + ST_SPARSE20:18
@HeikoSvoid Parameter::add(SGSparseMatrix<bool>* param,20:18
@HeikoSconst char* name, const char* description)20:18
@HeikoS{20:18
@HeikoSTSGDataType type(CT_SGMATRIX, ST_SPARSE, PT_BOOL, &param->num_vectors,20:18
@HeikoS&param->num_features);20:18
@HeikoSadd_type(&type, &param->sparse_matrix, name, description);20:18
@HeikoS}20:18
@sonney2kthoralf, heh they have 3 formats - so which one?20:19
@HeikoSwhats that?20:19
thoralfsonney2k: compressed row/columns20:19
@sonney2kLIL, CSC, CRS20:19
@HeikoSsonney2k: I see, whats this code snippet doing the?20:19
@sonney2kthoralf, they are static though...20:19
@HeikoSthere20:19
@HeikoSsonney2k: any ideas?20:19
@sonney2kthoralf, I mean no way to add a vector / insert20:20
thoralfsonney2k: Yeah, that's right.20:21
@sonney2kHeikoS, ohh I see an issue - with CT_VECTOR you cannot say how many dims the sparse matrix has20:21
@HeikoSsonney2k: so thats might be the reason why it is CT_MATRIX?20:22
thoralfsonney2k: Nobody wants to change sparse matrices... I always treat them as if they were read-only ;)20:22
@HeikoSsonney2k: that makes sense20:22
thoralfsonney2k: Btw: Appending vectors works well.20:22
@sonney2kthoralf, I mean for dense matrix types it totally makes sense to use just one big array (lapack & friends are optimized for that like hell) no idea if sparse libs have similiar optimiaztions20:22
thoralfsonney2k: At least for CSR20:22
@sonney2kHeikoS, yeah I guess but it is totally crap20:23
@HeikoSsonney2k: how would be the best way?20:23
@HeikoSjust have container types only I guess: SCALAR, VECTOR, MATRIX, SPARSE_VECTOR, SPARSE_MATRIX?20:23
@sonney2kHeikoS, doing it with CT_SGSPARSEMATRIX20:23
@HeikoSand drop ST?20:24
@sonney2kand CT_SGSPARSEVECTOR20:24
@sonney2kyeah I would love to drop ST_*20:24
@HeikoShow to do strings then?20:24
@HeikoSvectors of chars?20:24
@sonney2kbut I guess we can only drop it once strings are gone20:24
thoralfsonney2k: Hmm, I think sparse matrices are hard to optimize.  I'm just thinking of dot products.20:25
thoralfsonney2k: So many conditions.  I'd like to know how sparse a vector needs to be for that to pay.20:25
@sonney2kstrings would be CT_SGSTRINGLIST20:25
@HeikoSI see, so no strings of matrices etc?20:25
@sonney2kthoralf, I guess people like the one doing gotblas would do that20:25
@HeikoSonly SGObject20:25
@HeikoSsonney2k: man, this should have been done before clone/equals20:26
thoralfgotblas20:26
thoralfgotblas?20:26
@sonney2kHeikoS, I admittedly would prefer doing a new type for that when it happens20:26
@HeikoSsoooo much refactoring now20:26
@sonney2kthoralf, gotoblas20:26
thoralfOh, damn.  Gotta got.  See you!20:26
@HeikoSthoralf: bye!20:26
@sonney2kthoralf, cu!!20:26
@HeikoSsonney2k: parameter framework could be redone somehow then20:28
@HeikoSmost classes dont interact that much20:28
@sonney2kI suspect it woudl be much easier this way20:29
lambdayHeikoS: I'll be back later20:30
lambdaysee you20:30
@HeikoSsonney2k: mmh, so we would have to rewritte: adding/storing/registering and then load/save, and migrate(useless anyway in its current state)20:30
@HeikoSlambday: bye!20:30
-!- lambday [67157e4f@gateway/web/freenode/ip.103.21.126.79] has quit []20:30
@HeikoSsonney2k: equals/clone once more20:31
@sonney2kHeikoS, sorry is it really that much?20:31
@HeikoSsonney2k: dont know20:31
@HeikoStrying to find out :)20:31
@sonney2kHeikoS, I mean ST_* exisits only for NONE and STRING20:31
@sonney2kso it cannot be20:31
@sonney2kI think just string needs to be done differently20:31
@sonney2kthen all good20:31
@HeikoSsonney2k: save/load/migrate/equals/clone are all totally dependent on the structure CT/ST20:32
@HeikoSif we drop one, these basically have to be re-done20:32
@HeikoSand when we do this, we can also change the TSGDataType20:32
@HeikoSbecause its totally annoying20:32
@HeikoSand then we can also do things in a way that migration is easier20:32
@sonney2ksure just drop this parameter but then all the rest will still work20:32
@HeikoS(allocating paramters from scratch etc)20:32
@HeikoSsonney2k: dont know whether all that is feasible for someone doing other things ....20:33
@HeikoSmight be a good GSoC project20:33
@HeikoSbut ultra hard20:33
@sonney2kHeikoS, you are the expert with migration / modsel / parameters20:34
@sonney2kwho would know if not you20:34
@HeikoSsonney2k: you :D20:34
@sonney2kbut just removing SG_* seems very easy!20:34
@sonney2kST_*20:34
@HeikoSsonney2k: ok, maybe thats a good start20:34
@sonney2ksimply stuff later is a different problem20:35
@HeikoSsonney2k: what do you think about all those types that are blowing up Parameter.h?20:35
@HeikoSuint16_t?20:36
@HeikoSI never used that20:36
@HeikoSrather float64, float32, int32, int64, char20:36
@HeikoSnot even unsigned20:36
@HeikoSsonney2k: oh on another thing:20:37
@HeikoSsonney2k: the spectrum kernel currently is done via those obtain_from_* method of string features20:37
@HeikoSwhich then generates this sparse representation of the k-mer counts20:37
@HeikoSand those are given to a kernel20:37
@HeikoSsonney2k: so no model-selection of k-mer length20:38
@HeikoSsince one cannot do model-selection over features20:38
@sonney2kHeikoS, I am only using unsigned stuff like uint16_t / 32_t 64_t for strings20:38
@sonney2kHeikoS, I did some other spectrum features using all k=1...max_K20:39
@sonney2kwith weights20:39
@sonney2kforgot what the name was20:40
@sonney2kbut maybe it has the same problem20:40
@HeikoSsonney2k: can the kernel select the weights?20:40
@sonney2kdon't remember20:40
@HeikoSok20:40
@sonney2kIMHO it is not that much of a problem20:40
@HeikoSsonney2k: no problem, but no automated model selection20:41
lisitsynpickle27: hey I just checked andreas mail20:41
@HeikoSsonney2k: btw model selection framework might be a bit useless anyway20:42
pickle27lisitsyn: yeah I've been working on this today20:42
@HeikoSsonney2k: what counts more is cross-validation20:42
@HeikoSsince grid-search is just a loop20:42
pickle27lisitsyn: Im actually about to send an reply20:42
@HeikoSwhich users could easily set up themselved20:42
@HeikoSs20:42
lisitsynpickle27: okay sure!20:42
@HeikoSsonney2k: but is very hard to maintain20:42
@HeikoSif we offer it20:42
lisitsynpickle27: I am on vacations still so I'd try to do my best to get into algorithms and help with further planning20:43
@HeikoSsonney2k: at least for grid-search20:43
@HeikoSsonney2k: gradient based things are different20:43
pickle27lisitsyn: yeah no prob, I have still got a bunch of things I want to get done before getting fully into example mode20:43
lisitsynpickle27: is everything alright with your masters thesis?20:44
pickle27lisitsyn: yeah I submitted and my defence is on Aug 1!20:44
lisitsynoh I see20:44
pickle27I meant shogun things to get done!20:44
lisitsynwhat do you mean? by aug 1?20:45
pickle27I mean there is a few more techniques I want to finish coding for shogun before I start putting all my attention to the final examples20:45
pickle27Aug 1 is when I present my thesis20:46
lisitsynahh alright20:46
-!- iglesiasg [~Fernando@s83-179-44-135.cust.tele2.se] has joined #shogun20:51
-!- mode/#shogun [+o iglesiasg] by ChanServ20:52
@iglesiasghi all20:52
-!- zxtx [~zv@cpe-75-83-151-252.socal.res.rr.com] has quit [Ping timeout: 248 seconds]20:52
@sonney2khey iglesiasg - back home?21:01
@iglesiasgsonney2k: hi! yeah, I got here yesterday at night21:01
@iglesiasgsonney2k: are you already in holiday?21:01
@HeikoSiglesiasg: hi!21:02
@iglesiasgHeikoS: thanks for setting up the doodle!21:02
@sonney2kiglesiasg, yeah watching the sun set over the sea currently21:02
@HeikoSiglesiasg: yes, good to have the meeting soon21:02
@HeikoSsonney2k: nice :D21:02
@iglesiasgsonney2k: it sounds nice21:03
@sonney2kdoing (late) nips reviews21:03
@sonney2kall rejects IMHO :/21:03
@iglesiasghaha21:04
@sonney2kHeikoS, never get yourself's paper reviewed by me21:04
@sonney2kI don't remember accepting papers21:04
@iglesiasgwe better hope for our papers not to be reviewed by you :P21:04
@HeikoSsonney2k:  maybe you have mine :)21:04
@HeikoSkameleon21:05
@iglesiasgHeikoS: did you submit sth to NIPS?21:05
@HeikoSiglesiasg: yes21:05
@HeikoSsome kernel mcmc stuff21:05
@iglesiasgHeikoS: interesting, I wish you luck with it!21:05
@sonney2kiglesiasg, well we all have seen HeikoS talks so we can be pretty sure that this gets accepted21:08
@HeikoSiglesiasg: thanks!21:08
@HeikoSsonney2k: writing is different ;)21:08
@HeikoSbut I think it will be fine21:08
@iglesiasgsonney2k: btw, I was gladly surprised with Berlin. I found it an amazing place21:14
@sonney2kiglesiasg, yeah I like it too :)21:15
@HeikoSit is amazing :) I would find it hard to work here though - so many other nice things to do :)21:15
@HeikoSgoing now, see you!21:21
@iglesiasgHeikoS, sonney2k: BTW, could any of you take a quick look at thoralf PR when you get some time?21:22
@HeikoSiglesiasg: I did, I think its ok, but I dont know enough of makefiles t omerge it, sonney2k, wiking, or lisitsyn should have a look21:23
@iglesiasgHeikoS: all right21:23
@iglesiasgwe are in the same boat about Makefiles hehe21:23
@iglesiasgbye HeikoS21:24
@sonney2kiglesiasg, yeah had a look - not in mergeable state yet21:25
@iglesiasgsonney2k: because of the crash in travis build?21:26
@iglesiasgI think that may actually be my fault, it seems to be something is CustomMahalanobisDistance21:26
@sonney2kiglesiasg, no not because of that21:29
-!- HeikoS [~heiko@f052017018.adsl.alicedsl.de] has quit [Quit: Leaving.]21:53
-!- shogun-notifier- [~irker@7nn.de] has quit [Quit: transmission timeout]22:20
-!- zxtx [~zv@rrcs-76-79-81-162.west.biz.rr.com] has joined #shogun22:31
gsomixrefactoring, refactoring, refactoring22:43
@sonney2kiglesiasg, wrt the PR I have to discuss a couple of things ...22:59
@iglesiasgsonney2k: the LMNN PR, right? Tell me about it.23:01
@iglesiasgsonney2k: after Saturday's conversation I realized there is no need to make the methods static to avoid including LMNNImpl.h in LMNN.h, just need to do this forward declaration of the class CLMNNImpl. So I was thinking of doing that and make the methods non-static23:03
-!- pickle27 [~Kevin@d67-193-243-174.home3.cgocable.net] has quit [Quit: Leaving]23:30
--- Log closed Tue Jul 16 00:00:33 2013