Open in new window / Try shogun cloud
--- Log opened Wed Apr 04 00:00:19 2012
@sonney2knite everyone00:04
n4nd0sonney2k: good night00:05
-!- harshit_ [~harshit@] has joined #shogun00:30
harshit_n4nd0: hey r u around ?00:32
n4nd0harshit_: yeah, tell me00:40
-!- harshit_ [~harshit@] has quit [Ping timeout: 252 seconds]00:45
-!- harshit_ [~harshit@] has joined #shogun00:46
harshit_n4nd0: Cant figure out what to do with "Additional info" box in gsoc proposal.00:47
harshit_Where should I upload ? my own domain or some public doc hosting site00:48
harshit_and what to upload : whole proposal( schedule + details ) or only schedule00:48
n4nd0maybe a personal webpage or files you want to them to see00:49
n4nd0I think you have to upload the whole proposal00:49
harshit_okay thanks..!00:50
harshit_btw are you done with proposal ?00:50
n4nd0are you too?00:54
n4nd0btw, I saw that you NewtonSVM has been merged today00:55
n4nd0that's cool :)00:55
harshit_n4nd0: yeah got my first patch to be merged :)01:06
harshit_feels great01:06
harshit_and "are you too ?"01:07
harshit_what did you mean ?01:07
-!- blackburn [~qdrgsm@] has joined #shogun01:07
-!- nickon [] has quit [Quit: ( :: NoNameScript 4.22 :: )]01:08
n4nd0yea I meant if you are too done with the proposal01:15
n4nd0blackburn: how did it go the work for your project?01:15
blackburnn4nd0: oh I have been fucking with m$ word for two hours trying to set up relative links01:16
blackburnwho the fuck suggested to use m$ shit..01:16
n4nd0but why were you using it?01:16
n4nd0las time I had to do it I almos commited suicide, I couldn't manage to make anything work01:17
blackburnit is needed for this god damned journal01:17
blackburnyeah I was near that minute before01:17
n4nd0oh yeah it's true, you said something about that01:17
-!- Juggy [~Razor@] has left #shogun []01:26
harshit_bye every one, have a good night :)01:39
n4nd0good night01:39
harshit_and n4nd0 : I just submitted my proposal. Need some comments by soeren  desperately.01:40
n4nd0you may ask him for his opinion01:41
-!- harshit_ [~harshit@] has quit [Remote host closed the connection]01:41
-!- av3ngr [av3ngr@nat/redhat/x-jxgralxksizjasyb] has joined #shogun01:54
CIA-64shogun: iglesias master * ra985a56 / src/shogun/converter/StochasticProximityEmbedding.h : + SPE documentation -
CIA-64shogun: Sergey Lisitsyn master * r07e97f4 / (9 files in 6 dirs): Merge branch 'dr-spe' of git:// -
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]02:24
-!- n4nd0 [] has quit [Quit: leaving]02:26
shogun-buildbotbuild #448 of csharp_modular is complete: Failure [failed compile]  Build details are at  blamelist: fernando.iglesiasg@gmail.com02:59
shogun-buildbotbuild #437 of python_modular is complete: Failure [failed test_1]  Build details are at  blamelist: fernando.iglesiasg@gmail.com04:17
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]04:26
shogun-buildbotbuild #450 of csharp_modular is complete: Success [build successful]  Build details are at
-!- pythonroar [d2198538@gateway/web/freenode/ip.] has joined #shogun04:59
shogun-buildbotbuild #438 of python_modular is complete: Success [build successful]  Build details are at
-!- Afan [] has quit [Read error: Connection reset by peer]05:19
-!- pythonroar [d2198538@gateway/web/freenode/ip.] has quit [Quit: Page closed]06:24
-!- sidsniper [~piggy@] has quit [Ping timeout: 252 seconds]06:37
-!- Afan [] has joined #shogun06:37
-!- sidsniper [~piggy@] has joined #shogun06:37
-!- pythonroar [d2198538@gateway/web/freenode/ip.] has joined #shogun08:06
-!- wiking [] has joined #shogun08:08
-!- wiking [] has quit [Changing host]08:08
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun08:08
-!- n4nd0 [] has joined #shogun08:08
-!- pythonroar_ [d2198538@gateway/web/freenode/ip.] has joined #shogun08:29
-!- pythonroar [d2198538@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]08:31
n4nd0sonne|work: any idea why the same error occurred again
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]08:38
-!- uricamic [9320543b@gateway/web/freenode/ip.] has joined #shogun08:44
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has joined #shogun08:44
blackburnuricamic: we are waiting for your contributions ;)08:48
uricamicblackburn: working on it ;-)08:52
@sonney2kuricamic, what are you working on?08:52
uricamicI want to implement bundle methods for structured output learning08:53
uricamicI have some matlab implementations08:53
blackburnsonney2k: I guess you know this guy? ;)08:54
@sonney2kblackburn, :P08:56
@sonney2kuricamic, what I meant is what you are working on as a patch08:56
uricamicI will start with some basic implementation of BMRM08:57
@sonney2kuricamic, can you do that in a few days already?08:58
uricamicI hope so08:58
@sonney2kuricamic, since deadline is on apr 6 - show us some progress via github pull request. just mark them with [Work in Progress]09:04
@sonney2khave to leave the train09:06
uricamicok, I will do that tommorow, if its ok09:06
@sonney2kand enjoy a shower in the rain :D09:06
@sonney2kuricamic, yes of course09:06
blackburnuricamic: you could also consider some small patches btw09:06
uricamicblacburn: any ideas what it should be?09:07
blackburnuricamic: I can come up with something if you need it09:07
uricamicblackburn: ok, I would appreciate it09:09
blackburnuricamic: it seems you are a kind of CV expert?09:09
uricamicI wouldn't say expert, just know some stuff :)09:10
blackburnuricamic: what about parzen window classifier?09:14
uricamicblackburn: well, I can do that09:15
blackburnuricamic: are you familiar with shogun structure/code/etc?09:16
uricamicblackburn: I started to explore the code and to get familiar with it09:19
blackburnI see09:19
blackburnthen it could be a good introduction09:20
blackburnhowever if you notice any other classifier that is not in shogun feel free to add it09:20
uricamicok, I will look at it soon09:23
-!- harshit_ [~harshit@] has joined #shogun09:26
harshit_dam n4nd0, when do you sleep man ?09:27
n4nd0harshit_: haha during "part" of the night09:27
blackburnn4nd0: aren't you sleeping?? wtf09:28
harshit_blackburn: do you know from which MATLAB/C implementation, incomplete LBP features is implemented09:28
n4nd0blackburn: I have been around here for a while now man09:28
blackburnharshit_: they are implemented in libocas09:28
blackburnn4nd0: yeah.. now I see09:28
n4nd0blackburn: actually, I have to go soon to pick up my second laundry :P09:28
harshit_I guess that "part"  has only 2 hrs right ?09:29
blackburnn4nd0: you have to rest man, we don't want you to die :D09:29
n4nd0blackburn: no worries :)09:29
harshit_blackburn : thanks,i'll have a look at it09:30
-!- harshit_ [~harshit@] has quit [Ping timeout: 260 seconds]09:43
-!- harshit_ [~harshit@] has joined #shogun09:47
-!- wiking [] has joined #shogun10:10
-!- wiking [] has quit [Changing host]10:10
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:10
-!- av3ngr [av3ngr@nat/redhat/x-jxgralxksizjasyb] has quit [Quit: That's all folks!]10:16
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]10:20
sonne|workn4nd0: I am not sure no10:24
n4nd0sonne|work: ok10:24
sonne|workmy guess is that something was already scheduled before and because the bot was busy it didn't update to a newer version10:28
blackburnsonne|work: awesome stuff10:39
sonne|workblackburn: license?10:40
blackburnsonne|work: matlab code anyway10:41
sonne|workwell mex stuff10:42
blackburnsonne|work: can't see any license at all10:42
blackburnsonne|work: I'm going to port these things10:44
sonne|workbut make sure that if you want to use their code to get a license statement10:44
blackburnsonne|work: group / tree / overlap lasso is a kind of MT so would be nice to have these10:45
blackburnsonne|work: I'd rather contact authors10:45
sonne|workwell it is kind of MKL too :)10:46
sonne|workyeah of course contact authors to get that statement10:46
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun10:48
-!- blackburn [5bdfb203@gateway/web/freenode/ip.] has quit [Quit: Page closed]10:53
wikinganybody tried to compile shogun with clang?11:02
sonne|workwiking: sure - the debian package is clang compiled11:03
wikingsince now i've just switched to it and i have a lot of warnings and errors while compiling... so i was just wondering if anybody is doing fixing so i could join11:03
sonne|workit needs a lot less memory on clang11:03
sonne|workerrors too?11:03
sonne|workthese are new then :D11:03
sonne|workfixes and reports welcome11:03
wiking[latent src] $ clang --version                                                                                                                       [wiking@welitron:~/shogun/src]11:03
wikingApple clang version 3.1 (tags/Apple/clang-318.0.58) (based on LLVM 3.1svn)11:03
wikingTarget: x86_64-apple-darwin11.3.011:03
wikingThread model: posix11:03
sonne|workshould all be trivial though...11:03
wikingyeah they are quite trivial11:04
wikingi was just wondering if anybody started fixing11:04
wikingbut then i'll do a new branch and start the cleanup11:04
sonne|workdon't do warnings for now11:05
sonne|workbut only errors11:05
sonne|workhowever, clang has much better warnings...11:05
wikingbut it's amazing how faster it compiles for me11:06
sonne|workyes and it needs just 1GB of mem not 3.7 :D11:06
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]11:13
-!- sidsniper [~piggy@] has quit [Read error: Connection reset by peer]11:17
-!- sidsniper [~piggy@] has joined #shogun11:17
-!- sidsniper [~piggy@] has quit [Ping timeout: 244 seconds]11:23
-!- sidsniper [~piggy@] has joined #shogun11:24
-!- sidsniper [~piggy@] has quit [Ping timeout: 244 seconds]11:28
-!- sidsniper [~piggy@] has joined #shogun11:28
-!- uricamic [9320543b@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]11:32
-!- sidsniper [~piggy@] has quit [Ping timeout: 250 seconds]11:33
-!- sidsniper [~piggy@] has joined #shogun11:33
-!- sidsniper [~piggy@] has quit [Ping timeout: 244 seconds]11:37
-!- sidsniper [~piggy@] has joined #shogun11:41
-!- sidsniper [~piggy@] has quit [Read error: Connection reset by peer]11:43
-!- sidsniper [~piggy@] has joined #shogun11:44
-!- sidsniper [~piggy@] has quit [Read error: Connection reset by peer]11:52
-!- sidsniper [~piggy@] has joined #shogun11:52
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun11:52
-!- wiking [~wiking@huwico/staff/wiking] has quit [Client Quit]11:54
-!- sidsniper [~piggy@] has quit [Read error: Connection reset by peer]11:55
-!- sidsniper [~piggy@] has joined #shogun11:55
-!- sidsniper [~piggy@] has quit [Ping timeout: 260 seconds]12:01
-!- sidsniper [~piggy@] has joined #shogun12:02
-!- sidsniper [~piggy@] has quit [Ping timeout: 272 seconds]12:07
-!- sidsniper [~piggy@] has joined #shogun12:08
-!- Afan [] has quit [Remote host closed the connection]12:14
-!- pythonroar_ [d2198538@gateway/web/freenode/ip.] has quit [Quit: Page closed]12:18
-!- sidsniper [~piggy@] has quit [Read error: Connection reset by peer]12:22
-!- sidsniper [~piggy@] has joined #shogun12:22
-!- sidsniper [~piggy@] has quit [Read error: Connection reset by peer]12:29
-!- sidsniper [~piggy@] has joined #shogun12:30
-!- sidsniper [~piggy@] has quit [Read error: Connection reset by peer]12:45
-!- sidsniper [~piggy@] has joined #shogun12:45
-!- sidsniper [~piggy@] has quit []12:59
-!- flxb [] has quit [Quit: flxb]13:01
-!- harshit_ [~harshit@] has quit [Remote host closed the connection]13:16
-!- blackburn [~qdrgsm@] has joined #shogun13:59
-!- harshit_ [~harshit@] has joined #shogun14:00
harshit_blackburn: libocas is not able to compile, gives an error.! what could be the problem ?14:01
blackburnI need to know the error to answer ;)14:01
harshit_error: undefined reference to 'log'14:01
harshit_in 'svmocas.c ' and 'libocas.c14:02
blackburntry to add cmath include or so14:02
blackburnput #include<math.h> and try again14:03
harshit_Its already included !14:03
blackburnnice bug then14:04
blackburnsonne|work: is there any reason to use clang?14:06
harshit_btw to run mex files, do we need to have some library or so14:06
-!- flxb [] has joined #shogun14:08
blackburnharshit_: matlab14:10
harshit_blackburn :will octave do the job ?14:10
blackburnshould work but I am unsure14:11
harshit_okay thanks14:11
-!- gsomix [~gsomix@] has joined #shogun14:22
sonne|workblackburn: clang will be (or is?) the default compiler on osx14:26
sonne|workharshit_: link with -lm14:27
harshit_sonne|work: I have seen the Makefile, in there -lm is already linked while compiling14:29
sonne|workharshit_: wrong order then try to put -lm at the end14:30
sonne|work(or beginning)14:30
harshit_same error ! have a look at what my shell says :
sonne|workgcc -msse -O3 -fPIC -fstrict-aliasing -fomit-frame-pointer -Wall -pthread -lm -O svmocas svmocas.c lib_svmlight_format.c sparse_mat.c ocas_helper.c libocas.c libqp_splx.c -lm14:35
sonne|worktry this ^14:36
harshit_sonne|work : it worked :)14:42
harshit_thanks,btw what was the problem ?14:42
sonne|workorder in which libraries are specified matters - objects that need a symbol need the library -l'd past the object14:44
harshit_yeah ! right ..14:46
harshit_sonne|work: btw I am beginning work on LBP features, hope I can complete it till next week !14:47
-!- n4nd0 [] has left #shogun []14:47
-!- harshit_ [~harshit@] has quit [Quit: Leaving]14:51
-!- emrecelikten [~Anubis@] has joined #shogun15:07
emreceliktenHello, everyone15:07
emreceliktenWould anybody be interested in implementing Gaussian HMMs in Shogun as a GSoC project?15:08
emreceliktenI mean, HMMs with multi-variate Gaussians as observations15:10
sonne|workemrecelikten: I would love to have them but we are lacking manpower aka mentors that could do this15:12
sonne|worknevertheless - if you would work on this we certainly accept  your patch15:15
emreceliktenOkay, thanks. I needed them personally when I tried Shogun some time ago, so I might try it out15:19
-!- n4nd0 [] has joined #shogun15:32
blackburnn4nd0: gsomix is in search of new ideas for his proposal, feel free to suggest something regarding interfaces/system/etc16:06
n4nd0blackburn: ok, I think of it16:07
n4nd0blackburn: what project is he applying for so I can read the idea on the webpage?16:08
blackburnn4nd0: various usability improvements16:08
blackburnn4nd0: just think how can we improve usability16:08
sonne|workblackburn, gsomix - I could also imagine to get the testsuite to run for python/python_modular :D16:09
blackburnsonne|work: not feasible :D16:09
sonne|workyeah tough - but my impression is that the list on the website is long already?!16:10
n4nd0blackburn: chef support16:10
sonne|worksgvector -> refcounts16:10
blackburnsonne|work: yes, we need chef support16:10
sonne|workASSERT -> SG_ERROR16:10
sonne|workoctave sparse modular typemaps16:10
blackburnperformance (time,memory) measures16:11
sonne|worktypemaps for crossvalidation16:11
sonne|workdirector classes16:11
sonne|workthat is a lot already16:11
blackburnsonne|work: we both (n4ndo and I) think we should be the first Chef library (
blackburnI believe it would be very useful16:12
blackburnimagine you put two svms into the mixing bowl16:12
blackburnand stir it for two minutes16:12
blackburngsomix: ping16:13
-!- PhilTillet [9d9f2503@gateway/web/freenode/ip.] has joined #shogun16:13
PhilTilletHello everybody16:13
n4nd0with 5 ml GaussianKernel16:13
sonne|workgreat work16:13
n4nd0and 100g SimpleFeatures16:13
PhilTilletsonne|work: integrating OpenCL to shogun is non-trivial from a design point of view. :D16:14
n4nd0blackburn: what about SGNDArray's serialization issue?16:15
blackburnlarge scale means there are more than 5 kg of features16:15
PhilTilletblackburn: kg? :D16:15
PhilTilletlike kilogram16:15
sonne|workn4nd0: that is too tough16:16
n4nd0PhilTillet: we plan to work on chef support
sonne|workn4nd0: should be sth you with help of heiko could attempt16:16
n4nd0sonne|work: ok16:16
PhilTilletn4nd0: what about ?16:17
blackburnn4nd0: I think better solution is to rewrite shogun with chef16:17
PhilTilletmaybe we could write an IP over Avian Carriers protocol in Chef16:18
blackburnPhilTillet: that should be our baseline for distributed stuff16:18
n4nd0this part of the article is pretty sad16:18
n4nd0On 28 April 2001, IPoAC was actually implemented by the Bergen Linux user group.[4] They sent nine packets over a distance of approximately five kilometers (three miles), each carried by an individual pigeon and containing one ping (ICMP Echo Request), and received four responses.16:18
blackburnn4nd0: ok let you reimplement shogun in chef16:19
n4nd0blackburn: I'm with it16:20
blackburnI'll check your code at 18 UTC16:20
blackburnI hope it will have been done by this moment16:20
blackburnn4nd0: later you will collaborate with PhilTillet on pigeon communicated distributed chef-based shogun16:21
blackburnthat's the plan16:21
PhilTilletgreat idea16:21
PhilTilletcan I submit a GSoC proposal about that,16:21
blackburnyes exactly16:22
blackburnother ideas became outdated16:22
n4nd0do we have a mentor for that?16:22
blackburnsonne|work: will you mentor it?16:22
blackburnn4nd0: I heard elton john wanted to mentor this16:22
-!- harshit_ [~harshit@] has joined #shogun16:22
blackburnwith a couple of songs16:23
n4nd0wow pigeon's RFC is awesome16:25
blackburnn4nd0: we should contact chef's author16:27
blackburnand tell him we want to collaborate16:27
blackburnoh damn I imagine that, it would look like all shogun developers get high16:30
blackburnor overvodked16:30
blackburnn4nd0: there was a english tutorial based on vodka word16:31
blackburnI have been vodking since I was 316:31
blackburnhe vodked yesterday16:31
blackburnI will be vodking tomorrow16:31
blackburnall tenses can be described with 'to vodka' verb16:32
blackburnokay time to rest a little hah16:33
blackburnthat is becoming dangerous16:34
-!- harshit_ [~harshit@] has quit [Ping timeout: 245 seconds]16:40
PhilTilletthis irc chan is going crazy o_o16:41
emreceliktenI'm seriously considering applying to Shogun for GSoC now ;D16:42
emreceliktenAre there any applications for C5.0? I'm sort of indecisive about a project16:43
n4nd0emrecelikten: I cannot tell you how many people have applied to that project with certainty, but there have some mails in the mailing list of people interested in that project16:48
n4nd0blackburn: one question about covertree16:49
emreceliktenn4nd0: Okay, thanks!16:50
n4nd0blackburn: why is it that we implement the covertree point structure in every class that makes use of covertree?16:50
n4nd0emrecelikten: you're welcome16:50
sonne|workgsomix: ask our vodking he will hand you some16:51
-!- harshit_ [~harshit@] has joined #shogun16:52
PhilTilletsonne|work: is it ok is the patch is not perfect from a design point of view? Making OpenCL working with current shogun interface in a sweet design is kind of a headache :p16:54
sonne|workissue a pull request - show us what you have and get feedback. normally it takes a few iterations until a patch is OK16:57
-!- sonne|work [~sonnenbu@] has left #shogun []17:19
-!- PhilTillet [9d9f2503@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]17:32
-!- n4nd0 [] has quit [Ping timeout: 245 seconds]17:34
-!- PhilTillet [9d9f2503@gateway/web/freenode/ip.] has joined #shogun17:35
PhilTilletsonney2k: is it okay if the first version of this patch is this week end? I'm having the worst week of my life because of my dumb school :D17:35
PhilTilletmy school forces me to use 1 week to create one dumb fictionnal corporation. I am forced to be with over-stressed managers who keep harassing me to ask me how to use powerpoint because "i'm an engineer"...17:38
@sonney2kblackburn, hmmhh I just compiled shogun with clang without problems...17:47
-!- PhilTillet [9d9f2503@gateway/web/freenode/ip.] has quit [Quit: Page closed]17:59
-!- harshit_ [~harshit@] has quit [Quit: Leaving]18:04
-!- n4nd0 [] has joined #shogun18:07
emreceliktenIs it okay to apply to multiple projects under Shogun?18:09
n4nd0emrecelikten: yeah, no problem18:14
n4nd0emrecelikten: GSoC general rules just apply here18:14
-!- PhilTillet [] has joined #shogun18:14
n4nd0emrecelikten: 20 proposals max and you can only be selected for one project at most18:14
emreceliktenn4nd0: Okay, thanks18:15
-!- harshit_ [~harshit@] has joined #shogun19:19
@sonney2kn4nd0, emrecelikten but it usually doesn't make sense to apply to more than 2-3 things better ask here who is applying for what19:36
@sonney2kto see if there is still a chance...19:36
-!- flxb [] has quit [Quit: flxb]19:48
emreceliktenI'm thinking abou applying to C5.0 and the one that incorporates SO-SVMs and HM-SVMs19:50
@sonney2kblackburn, I finally think I understand why development branches are a good idea19:51
@sonney2kone can mess around like hell in such branch19:51
@sonney2kand then merge cherry-pick / polished stuff19:52
@sonney2kbut it only makes sense if one collaborates on a certain feature19:52
emreceliktenSo that makes only 2 projects, which is fine I guess?19:52
-!- PhilTillet [] has quit [Ping timeout: 240 seconds]19:56
@sonney2kgsomix, btw you could add to your proposal clang/llvm support19:56
@sonney2kgsomix, i.e. fix errors and warnings to get shogun to work with clang19:57
-!- Johannes_ [55e5c1c3@gateway/web/freenode/ip.] has joined #shogun19:57
Johannes_Hello, silly question perhaps but what's an reasonable build time for shogun? Been an unreasonable amount of time and figure it's died..19:59
@sonney2kJohannes_, what hardware / how much memory?20:03
Johannes_it's an mac air 1.4dualcore with 2gb20:04
@sonney2kJohannes_, yeah well it will require between 2-3GB to compile w/ gcc20:05
Johannes_haha it just completed, thanks for the response :=20:05
@sonney2kso use clang to reduce the memory footprint...20:06
Johannes_Just got a bit worried after 15mi20:06
n4nd0Johannes_: it is a good idea to use the flag --disable-optimizations to reduce compile time20:06
@sonney2kJohannes_, or just compile libshogun - that doesn't need a lot20:06
@sonney2k./configure --interfaces=20:07
@sonney2k(will compile libshogun only)20:07
Johannes_ok cool. good to know in general20:07
Johannes_but non-existing problem solved20:07
@sonney2kbut if you compile for *all* interfaces it can take >15min!20:08
n4nd0sonney2k: I've checked before how to use CoverTree for KNN20:08
Johannes_I think perhaps I underestimated the size of it all.20:08
n4nd0sonney2k: just one thought, my plan this far is to add a boolena member, e.g. m_use_covertree20:09
n4nd0sonney2k: and just add in the methods that are already implemented something like if ( m_use_covertree )20:09
n4nd0sonney2k: get the k neighbors using the cover tree20:09
n4nd0sonney2k: else, do the stuff it is now implemented (with quick sort and so)20:09
n4nd0sonney2k: do you think it is ok like this or something would be better?20:10
n4nd0something else, I wanted to say20:10
@sonney2kn4nd0, perfectly ok - that is exactly how I would do it too20:22
n4nd0sonney2k: :) let's get to it then20:22
CIA-64shogun: Soeren Sonnenburg master * r1aa40a1 / (6 files in 4 dirs): fix a couple of clang related warnings (which infact were errors) -
* gsomix have returned from the dance floor.20:25
gsomixsonney2k, ok.20:29
blackburnsonney2k: not sure I got your idea about branches20:53
blackburnactually I have never used cherry-pick20:54
blackburnn4nd0: here?21:17
n4nd0blackburn: yeah21:17
n4nd0blackburn: tell me21:17
blackburnn4nd0: about covertree?21:17
blackburnmain reason is they are not generalized21:17
blackburni.e. while isomap and lle are similar21:18
blackburnklle has different type of distance21:18
blackburnand so on21:18
n4nd0but if it is just because of the distance, can it be parametrized with CDistance?21:18
blackburnin next iterations of 'libedrt' I am going to get it more generalized though21:18
blackburnyes but would require to change all the code21:18
blackburnI'm on the way21:19
n4nd0apart from this that just seems to be a better design21:20
blackburnn4nd0: however I'd suggest to add one new covertree point structure21:20
n4nd0do you think there is more room for improvement in libedrt?21:20
n4nd0blackburn: one new covertree point structure in KNN you mean right?21:20
blackburnn4nd0: yes21:21
n4nd0blackburn: sure, I am working on that21:21
blackburnn4nd0: the reason is simple21:21
n4nd0blackburn: something similar to SPE and the others21:21
blackburnwhile lle/isomap/spe stuff will be general21:21
blackburnI can not generalize KNN and libedrt21:21
-!- Johannes_ [55e5c1c3@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]21:22
blackburnlibedrt makes use of user_data with provided distance function21:22
blackburnbut in KNN it would be enough to have pointer to distance instance21:22
n4nd0I see21:22
n4nd0sonney2k: how is the number of proposals so far by the way?21:27
blackburnlet me guess21:28
n4nd0I say 2021:29
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]21:40
-!- wiking [] has joined #shogun21:40
-!- wiking [] has quit [Changing host]21:40
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun21:40
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has joined #shogun21:47
shogun-buildbotbuild #199 of nightly_none is complete: Success [build successful]  Build details are at
gsomixshogun-buildbot, good work. Please talk to me.21:50
blackburnforever alone21:51
gsomixonly vodka would be with me21:52
n4nd0blackburn: check there, line 19621:54
n4nd0blackburn: I think that can be done with j from i+121:54
blackburnn4nd0: yes21:55
blackburnI assume distances are symmetric21:56
n4nd0blackburn: for KNN is the same then21:56
n4nd0we have distance(features, features)21:57
blackburnn4nd0: hmm check for euclidean distance is still there22:01
blackburnI believe it is not true22:01
blackburnwhat is the difference between euclidean and any other distance?22:01
n4nd0blackburn: you believe that it is not true that they are symmetric?22:02
blackburnn4nd0: in the spe22:02
blackburnyou allow only euclidean distance22:02
blackburnwhy so?22:03
n4nd0in the toolbox they do it with Euclidean22:04
n4nd0I believed it must be that done22:04
n4nd0but it might be that another one works22:04
blackburnn4nd0: ok let us consider dataset X ;)22:04
blackburnand some distance function22:04
blackburnlets also consider distance matrix of this distance function between feature vectors of X22:05
blackburndoes there exist a dataset T with euclidean distances equal to ones of X?22:06
n4nd0yes, I think so22:06
n4nd0if you apply the same transformation (a rotation or displacement) to all the points in the dataset22:07
n4nd0I believe the Eculidean distance won't change22:07
blackburnso no matter what distance do you use22:07
n4nd0eculidean ha!22:07
blackburnsome other one distance is just an explicit transformation to T22:07
blackburnimplicit sorry22:07
n4nd0I don't know that property22:08
blackburnn4nd0: I'm pretty sure with it22:08
-!- harshit_ [~harshit@] has quit [Quit: Leaving]22:09
n4nd0blackburn: I believe you, but didn't know about it22:09
blackburnn4nd0: kernel trick does the same22:09
blackburnwhile distance and kernels have some relations I'd say itis the same22:10
blackburnn4nd0: btw do you know where kernel trick came from?22:10
n4nd0blackburn: the idea?22:11
n4nd0not really22:11
n4nd0already before in another place?22:11
blackburnaizerman, theoretical foundations of the potential functions for pattern recognition22:12
blackburnIIR the name C22:12
blackburnIIRC :)22:12
n4nd0I am checking22:13
n4nd0good memory22:13
-!- n4nd0 [] has quit [Quit: Lost terminal]22:14
-!- blackburn_ [5bde8018@gateway/web/freenode/ip.] has joined #shogun22:15
blackburn_lost connection22:15
-!- n4nd0 [] has joined #shogun22:15
blackburn_wtf :D22:15
blackburn_n4nd0: did you lost connection too22:15
n4nd0sometimes my shell dies22:15
blackburn_do you use shell irc client?22:16
n4nd0you don't?22:16
blackburn_currently I am under the hood of firewall of access point so have to use web client22:16
n4nd0oh! I thought because once I said something about it irssi22:16
blackburn_but I use pidgin22:16
-!- blackburn [5bde8018@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]22:17
n4nd0irssi is the name of the client I use22:17
blackburn_I see22:17
n4nd0it's pretty cool22:17
n4nd0I recommend you to try it22:17
n4nd0highly customizable22:17
n4nd0mine is very colorful, wanna take a look to a screenshot?22:17
blackburn_how does it handle highlights?22:18
n4nd0you can choose how to do it22:18
blackburn_I like gui popups when somebody wants me :)22:18
n4nd0now I have it configured to show me then on red and with a beep22:18
blackburn_so do you have some terminal window for that?22:18
blackburn_I do all the terminal stuff using guake22:18
n4nd0I have it in a tab22:19
blackburn_n4nd0: tab of?22:20
blackburn_btw have you ever used guake?22:20
n4nd0of the terminal, I have a tab for irssi22:20
blackburn_ah I see22:20
n4nd0it's the first time I hear of it22:20
blackburn_remember quake console22:20
blackburn_it is the same22:20
blackburn_pretty hardcore :D22:21
n4nd0I will try guake, looks cool22:21
n4nd0does it allow to divide the shell?22:21
n4nd0I use terminator because I think that that is awesome22:21
blackburn_n4nd0: split?22:21
blackburn_there are tabs but no splitting22:22
n4nd0look at this22:22
blackburn_n4nd0: I like the opacity thing22:22
blackburn_n4nd0: i.e. I press F2 and it emerges22:22
blackburn_but I can see other window (irc or anything else)22:23
blackburn_so I can see two things without any pain :D22:23
blackburn_I am afraid it is not supported22:23
blackburn_however can be emulated I think22:24
blackburn_screen? or something like that22:24
n4nd0no idea22:24
n4nd0but I really like that option22:24
n4nd0it comes handy lot of times22:24
blackburn_your matlab seems to crash often22:24
n4nd0yes ...22:24
n4nd0it is a kind of problem with gnome I think22:25
blackburn_n4nd0: TmForever, really? ;)22:25
n4nd0it happens when I have it in another workspace22:25
n4nd0I change constantly to others22:25
n4nd0I don't even remember what that is22:25
n4nd0what is it22:25
blackburn_trackmania :D22:26
n4nd0ah ok22:26
n4nd0I tried it :)22:26
n4nd0kind of fun22:26
naywhayaren4nd0: this lurker doesn't know how you can read that purple text on a black background22:26
n4nd0naywhayare: I like purple :)22:27
blackburn_naywhayare: I admit it seems it is the first your message there ;)22:27
naywhayarewell, I don't have anything useful to add to the conversation but if we're talking about colors and backgrounds and terminals I become useful :)22:28
n4nd0naywhayare: you like that stuff too?22:29
naywhayarethe contrast on my monitor just must not be good enough to have the purple stand out enough22:29
n4nd0I like personalization a lot22:29
naywhayarehaha... surprisingly yes22:29
naywhayarealthough personalization for me means stripping out all the crap I don't care about.  across two monitors I'm looking at 4 terminals and nothing else, using fluxbox because it stays out of my way22:30
* emrecelikten writes a note to self that he should visit this channel more22:30
naywhayareI think most people would refer to me as "boring"22:30
n4nd0fluxbox, that's really hardcore22:30
emreceliktenI used fluxbox too, once22:30
emreceliktenFor a year or something22:31
naywhayareI don't need much out of a window manager when all I do is open terminals and a browser, so it's a good fit for me22:31
n4nd0I didn't enjoy it that much22:31
n4nd0I didn't like the look and feel22:31
blackburn_n4nd0: what browser do you prefer? ;)22:31
n4nd0like the frames for the apps22:31
n4nd0blackburn_: lynx :P22:32
emreceliktenIt had the feeling that you feel when you visit the empty flat that you are planning to rent22:32
blackburn_oh sh22:32
naywhayareemrecelikten: I think that's the best description I've heard of fluxbox... I'll remember that one :)22:32
blackburn_n4nd0: are you for real?22:32
n4nd0blackburn_: hahaha just kidding22:32
n4nd0blackburn_: I like to use it sometimes for fun22:33
naywhayareI actually experimented with using lynx only for browsing once (so I could live my entire life inside my screen session) but the lack of images quickly made things difficult22:33
emreceliktennaywhayare: Haha, okay :)22:33
blackburn_well some guys *really* use it :D22:33
n4nd0blackburn_: but I normally go with chrome22:33
blackburn_that's quite normal then :D22:33
n4nd0blackburn_: oh didn't know about that?22:33
n4nd0what about you?22:33
n4nd0I wouldn't mind trying sth new ;)22:34
@sonney2ka long way from machine learning to colors and taste :D22:35
emreceliktenIsn't that sort of the ultimate goal?22:35
n4nd0sonney2k: haha, actually true22:35
n4nd0let's finish with some philosophy, vim or emacs? :P22:36
gsomixcat "blabla" > my_super_code.cpp22:37
n4nd0gsomix: :O22:37
gsomixecho, i mean22:38
blackburn_n4nd0: chrome just like you :)22:38
blackburn_as for vim or emacs22:38
naywhayare+1, there is no other way to edit22:38
blackburn_configuring emacs whole day long is rather way of sexual activity than editing22:39
n4nd0I have actually never tried emacs seriously22:40
n4nd0maybe I should just to make a fair comparison22:40
emreceliktenI use emacs for LaTeX editing and stuff, but I'm annoyed with it22:40
emreceliktenMight try vim one day22:40
* sonney2k started with emacs but lost a finger to it22:40
emreceliktenPoor pinky22:41
emreceliktenWell since I'm used to "crouching" with it, my pinky and emacs get along just fine22:41
naywhayareah, emacs... alt+shift+meta+footpedal+S for save22:43
gsomixStallman looks at you with disapproval.22:48
blackburn_and keeps eating some foot stuff22:50
blackburn_ah bad connection22:54
blackburn_wiking: hey, there?22:54
blackburn_sonney2k: do you think scikits don't mind to become python_modular submodule?22:57
blackburn_n4nd0: what's up with chef support?22:58
blackburn_had you completed it by 18 UTC?22:59
n4nd0blackburn_: eeeh mmm ... I contacted the author and he didn't agree22:59
wikingblackburn_: yes23:00
@sonney2kblackburn_, if we have Director* support everything will be enslaved by shogun ;-)23:00
blackburn_wiking: any progress on api? ;)23:00
wikingblackburn_: yep just want to commit23:00
blackburn_sonney2k: yes but we shall offer it to them23:00
@sonney2kblackburn_, I am just trying gcc-4.723:00
wikingand then i finish my proposal for gsoc23:00
@sonney2kthe memory footprint seems to be much better23:00
@sonney2kas in I was able to compile w/ octave requiring less than 2GB23:01
blackburn_sonney2k: that's nice23:01
@sonney2kvodking - better drink and sleep23:01
blackburn_sonney2k: no need to sleep23:02
blackburn_all one should do is to vodka23:02
blackburn_I really recommend you to change your lifestyle23:02
blackburn_once you get addicted to vodka things will go smoother23:02
blackburn_wiking: didn't you send it already?23:03
blackburn_n4nd0: convince him or bribe!23:04
n4nd0blackburn_: we'll what I can do23:04
blackburn_guys lets invest gsoc money to bribe chef author23:05
blackburn_we really need this chef support23:05
blackburn_I can't imagine how had I lived before I tried to mix 2 kg of potato and 5 kg of string features23:06
wikingblackburn_: well it's just a very shitty draft23:06
blackburn_wiking: of proposal?23:06
blackburn_wiking: you seems to be unhappy with something :)23:07
wikingblackburn_: yeah the gsoc application23:07
wikingblackburn_: too many balls and trying to joggle with them23:07
wikinghave 2 conf papers to finish23:07
wikingas well23:07
blackburn_hey will you manage to find time for gsocing?23:07
wikingblackburn_: june july and august is for that23:08
wikingi'm on 'holidays' for july-august23:08
blackburn_what about may and june?23:08
wikingmeaning i can just work on gsoc23:08
wikingjune is clear now for me as well23:09
blackburn_do not forget to mention that in the proposal23:09
wikingi only have one deadline for 1st of june23:09
wikingbut it coincides with the gsoc23:09
wikingi mean i'll use the stuff in the paper that i'm doing in gsoc23:09
wikingi guess it's not a problem...23:09
blackburn_that sounds ideal23:10
blackburn_it is exactly that shogun made for :)23:10
wikingyesterday i had a much better idea for the api23:10
wikingso it'll be more c++ alike23:11
blackburn_that's not a problem we are not tied to C23:11
wikingi'm just wondering how swig gonna handle that23:11
blackburn_wiking: what kind of pitfalls there?23:12
wikingso if i have a unary function then what swig is doing with it23:12
@sonney2kwiking, I just compiled with clang - no issues!23:12
wikingsonney2k: which version?23:12
wikingi'm doing 3.123:12
wikingi wonder what could be the diff between those two23:13
@sonney2kis that released even?23:13
blackburn_wiking: what kind of unary function?23:13
wikingblackburn_: are you familiar with the STL functions in <algorithm> ?23:13
blackburn_not really familiar23:13
wikingok here's an example23:14
@sonney2kbah why does anyone want to use stl?23:14
blackburn_do you want to make use of algorithms stuff?23:14
wikinglike for_each23:14
wikingthere you define an unary function for what to do with the container's each element23:14
wikingsonney2k: no it's just the idea from STL23:14
wikingi won't use it23:14
blackburn_okay, continue ;)23:15
wikingso for the argmax function23:15
blackburn_do you need to have this function available with swig?23:15
blackburn_I got you, some residual function would be here or so23:15
wikingi'd do the same kind of idea... that either you define your own unary function for the argmax of the latent variable23:15
blackburn_or objective23:16
wikingor it uses the predefined/bruteforce version of it23:16
wikingso now the question is23:16
wikinghow does swig handle this?23:16
blackburn_hmm so you want to implement this function using target interface language?23:16
wikingso that you can define your own function in python for example23:16
blackburn_yes there is a way23:17
blackburn_gsomix is applying for idea with director classes23:17
blackburn_so probably you will collaborate there a little23:17
blackburn_wiking: however performance would be worse23:18
blackburn_wiking: are you going to implement at least one C++ some domain specific latent svm?23:18
@sonney2kwiking, don't rely on it23:18
blackburn_like with sliding hog windows?23:19
@sonney2kwiking, it can very well be that this stuff is broken in swig23:19
blackburn_sonney2k: we probably should contact swig guys for that?23:21
wikingblackburn_: yeah i'll create an example...23:21
-!- flxb [] has joined #shogun23:21
@sonney2kand then fix it ourselves yes23:21
wikingand of course the brute force version will be there as well23:21
@sonney2kbut none of us is a swig guru23:21
wikingbut that's brute force, thus it might take much more time to do the optimization than with a specialized function...23:21
@sonney2kblackburn_, which version of gcc do you have on your machine?23:22
blackburn_sonney2k: 4.6.123:22
@sonney2kblackburn_, how much memory does it take to compile octave_modular w/ default settings for you?23:23
@sonney2kn4nd0, which gcc version do you have?23:23
n4nd0sonney2k: gcc (Ubuntu/Linaro 4.6.1-9ubuntu3) 4.6.123:23
blackburn_sonney2k: I'll measure23:23
@sonney2kanyone using some older gcc?23:23
blackburn_sonney2k: optmizations? w/ wo?23:24
blackburn_ah default23:24
@sonney2k./configure --interfaces=octave_modular23:24
blackburn_yes I know next steps23:24
blackburn_I have tried that a few times before23:24
gsomixsonney2k, 4.5.223:25
@sonney2kblackburn_, for me gcc4.7 / 4.6 / clang require 'just' 1.5-1.7GB of memory23:26
@sonney2kso I think something got improved23:26
* sonney2k install gcc-4.423:27
@sonney2kgsomix, if you could compile octave modular too and measure memory requirements - would be nice23:28
@sonney2kblackburn_, actually gcc-4.4 required 1.7-1.8GB23:33
@sonney2kso I guess the real change is that --datatypes configure flag23:34
@sonney2kwhere we support only a few datatypes now by default23:34
@sonney2k(in typemaps)23:34
blackburn_sonney2k: yes probably23:36
@sonney2kI now compiled with ./configure --cc=gcc-4.7 --cxx=g++-4.7 --datatypes=char,bool,uint8_t,uint16_t,uint32_t,uint64_t,int8_t,int16_t,int32_t,int64_t,float32_t,float64_t,floatmax_t --prefix=/usr/local/stow/shogun-1.2.0 --ruby=ruby1.9.1 --interfaces=octave_modular23:37
@sonney2kand this still requires just 1.8GB23:37
@sonney2kso what did we change?23:38
blackburn_sonney2k: ignores?23:38
@sonney2kignores and moving things from .h -> .cpp ?23:38
@sonney2ksuch big impact?23:38
@sonney2kI don't know23:38
blackburn_there are a lot classes in ignores23:38
blackburn_sonney2k: remove ignores and check ;)23:39
@sonney2kthis means though we should have some experimental release for debian then23:39
blackburn_sonney2k: 1.4g23:39
blackburn_sonney2k: so what is # of proposals?23:41
@sonney2kblackburn_, do you think we can get current git into a sane state23:43
blackburn_sonney2k: what kind of sanity you mena?23:43
@sonney2kactually I think it is in much better state than before gsoc stars :)23:43
@sonney2kblackburn_, testsuite / applications to be working again23:43
blackburn_are you still dreaming about release?23:43
blackburn_I need to check testsuite23:44
@sonney2kwe need one - or would you want to work with the shogun 1.1 debian clang compiled package?23:44
blackburn_no idea23:44
blackburn_sonney2k: okay let me compare to 1.123:45
blackburn_old testsuite now acts better than on 1.123:45
blackburn_some issues were fixed23:45
@sonney2kcross validation broken23:45
blackburn_sonney2k: what caused it?23:46
@sonney2kocas broken23:46
blackburn_sonney2k: ocas is ok23:47
emreceliktenJust submitted my hasty GSoC application23:47
@sonney2kno I mean these where all fixed in master23:47
blackburn_sonney2k: btw could you please update ocas serialized output?23:48
@sonney2kand QDA/newtonsvm/python3 stuff is not really essential for now but it should even compile w/ python3 now23:48
@sonney2kblackburn_, I updated already many of the failed tests but when attempting to push the data everything was lost23:48
@sonney2kno idea why23:48
@sonney2kgsomix, what is the python3 state btw - and the state of your proposal?23:49
@sonney2kblackburn_, anyway what I am saying is I would rather want to release something next week than to work with shogun 1.123:50
blackburn_sonney2k: agree23:51
gsomixsonney2k, i need to check type and typemaps. it seems, everything else works23:51
blackburn_sonney2k: ah yes! I'll check newton svm on my data23:51
@sonney2kblackburn_, yes please :)23:52
@sonney2kgsomix, ok then finish your proposal and then python3 stuff .. then we put this into the release too!23:52
gsomixsonney2k, and i still have not corrected examples.23:52
@sonney2kgsomix, didn't you run 2to3?23:52
@sonney2kgsomix, what happened to the patch you send a week ago or so where the examples were corrected?23:53
blackburn_sonney2k: I have asked you before but why we have -O9 flag?23:53
gsomixsonney2k, there were errors in examples.23:54
@sonney2kgsomix, in how many?23:54
-!- n4nd0 [] has quit [Quit: Lost terminal]23:54
@sonney2kand which?23:54
-!- n4nd0 [] has joined #shogun23:54
gsomixsonney2k, 2to3 works correctly. but some examples requires manual correction.23:55
gsomixsonney2k, not many. I can fix examples tomorrow23:55
-!- n4nd0_ [] has joined #shogun23:57
-!- n4nd0_ [] has quit [Client Quit]23:57
* sonney2k compiles python3 interface23:58
@sonney2kit did compile...23:59
--- Log closed Thu Apr 05 00:00:19 2012