Open in new window / Try shogun cloud
--- Log opened Fri Mar 30 00:00:19 2012
blackburnbtw should labels created in apply be SG_REFed?00:01
@sonney2kthese are returned as new objects and so are not used internally00:02
@sonney2kso no00:02
blackburnaha I see00:02
shogun-buildbotbuild #408 of java_modular is complete: Failure [failed git]  Build details are at  blamelist: blackburn91@gmail.com00:03
@sonney2kblackburn, you killed the build :)00:05
blackburnsonney2k: lol00:05
@sonney2kdammed github issue00:06
@sonney2kblackburn, btw we should all start to use git amend00:06
blackburnfor what?00:06
@sonney2kblackburn, to not have stupid/broken git commits00:07
@sonney2kamend enables merging multiple commits into one big one00:07
@sonney2kso if sth was forgotten in one commit (missing file, other error)00:08
@sonney2kbuildbot doesn't need to build it00:08
@sonney2kand fail once00:08
@sonney2kand then succeed00:08
@sonney2kand we don't have a crap commit :)00:08
blackburnwe could also consider bigger commits00:12
blackburnactually I try to do all my commits non-breaking anything00:12
@sonney2kblackburn, bah philipps' opencl test is slower on my GPU00:51
@sonney2kby a lot ...00:51
blackburnsonney2k: than CPU?00:51
@sonney2klike 10 times00:51
@sonney2kon his it is like 5 times faster00:51
blackburnsonney2k: see you00:59
-!- blackburn [~qdrgsm@] has quit [Quit: Leaving.]01:00
-!- vikram360 [~vikram360@] has quit [Read error: Connection reset by peer]01:00
-!- vikram360 [~vikram360@] has joined #shogun01:00
-!- flxb_ [] has joined #shogun01:29
-!- flxb [] has quit [Ping timeout: 246 seconds]01:30
-!- flxb_ is now known as flxb01:31
-!- flxb_ [] has joined #shogun01:32
-!- flxb [] has quit [Ping timeout: 246 seconds]01:36
-!- flxb_ is now known as flxb01:36
-!- vikram360 [~vikram360@] has quit [Ping timeout: 246 seconds]03:11
-!- romi__ [] has quit [Ping timeout: 246 seconds]03:17
-!- romi_ [~mizobe@] has joined #shogun03:45
-!- harshit_ [~harshit@] has joined #shogun04:39
-!- harshit_ [~harshit@] has quit [Ping timeout: 246 seconds]05:37
-!- romi_ [~mizobe@] has quit [Ping timeout: 246 seconds]05:37
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]05:39
-!- vikram360 [~vikram360@] has joined #shogun05:56
-!- Miggy [~piggy@] has quit [Ping timeout: 248 seconds]06:43
-!- av3ngr [av3ngr@nat/redhat/x-lgbneiugylvjahzy] has joined #shogun06:48
-!- n4nd0 [] has joined #shogun07:34
-!- novice [b4953264@gateway/web/freenode/ip.] has joined #shogun08:00
-!- vikram360 [~vikram360@] has quit [Ping timeout: 265 seconds]08:01
-!- vikram360 [~vikram360@] has joined #shogun08:08
noviceHi, can someone help me with github installation ?08:15
noviceI am trying to set it up in a virtual machine running ubuntu08:16
noviceadded the ssh key, but im unable to test it08:16
n4nd0hi novice08:17
novicehello :)08:17
n4nd0so github is not something you really "install", you need to create an account08:18
noviceyup created account08:18
n4nd0for what you have just said, I understand you have already done this step08:18
n4nd0so now you should be able to use git08:18
n4nd0git is something you do install08:19
noviceinstalled git via sudo apt-get08:19
n4nd0ok cool08:19
noviceI am trying to follow the steps mentioned in the home page after we create an account for the first time08:19
n4nd0send me the link please08:20
noviceit says to genreate ssh key and add it08:20
novicek jus a second08:20
n4nd0so have you already generated the keys?08:20
novicein the 5th step whihc says test it out08:20
novicei get an error which says domain name not registered08:21
n4nd0what command are you executing?08:21
novicesry the exact error message is ssh: Could not resolve hostname08:21
n4nd0ssh -T git@github.com08:21
noviceyup ssh -T git@github.com08:21
n4nd0mmm I gind that weird08:22
n4nd0can you try to do: ping github.com08:22
noviceI am in my college which uses NAT, u think its anythin to do with the firewall of my college?08:22
-!- Netsplit *.net <-> *.split quits: flxb, av3ngr08:23
n4nd0well yes it might be they filtered ssh connections, port 22 basically08:23
noviceping : unkown host github.com08:24
n4nd0but that is something they normally do in incoming connections08:24
noviceping : unkown host, this si the message i get08:24
n4nd0can you browse
n4nd0eeeh that is actually weird08:25
novicei am able to see the issues of shogun and all08:25
n4nd0the shell tells you that it doesn't know where that is08:25
-!- av3ngr [av3ngr@nat/redhat/session] has joined #shogun08:25
n4nd0but the browser can08:25
noviceim using virtual box08:25
noviceand trying to set ii up in virtual box08:25
n4nd0might be something related08:25
av3ngrhey novice what network set up did you use in virtual box?08:26
n4nd0try to ping
novicei tried pinging from my windows cmd, it says similar thing08:26
novice@av3ngr : i selected teh default option in it, nat08:26
-!- n4nd0 [] has quit [Quit: Lost terminal]08:27
av3ngrnovice, does ssh work for you normally?08:27
-!- n4nd0 [] has joined #shogun08:27
novicethis is my first time using ssh :|08:27
n4nd0novice: what do you with ping to the IP address08:27
n4nd0what result do you get*08:28
av3ngrn4nd0, it might be because 22 is blocked by the college.08:28
novicelinux the terminal doenst say anyhting, from windoes host i get 100% loss08:28
noviceRequest timed out08:28
n4nd0av3ngr: could be, but now we are just trying to ping08:28
n4nd0av3ngr: it is not normal that you can browse a page and the ping doesn't work ...08:29
av3ngryea just realized that - this is odd08:29
n4nd0novice: do you browse from windows?08:29
av3ngrunless college is blocking ping requests as well08:29
novicenaa browsing from virtual machine now08:29
av3ngr(which is annoying - but often done by college admins_08:29
n4nd0I have never seen that, but as you say, may be possible08:31
novicelemme just check from my friends room, brb08:31
av3ngr << might help08:32
-!- vikram360 [~vikram360@] has quit [Ping timeout: 246 seconds]08:33
novicerequest timed out for him as well08:34
n4nd0av3ngr: yeah! those are good references, probably novice can fix the problem like that08:34
-!- vikram360 [~vikram360@] has joined #shogun08:34
novice@av3ngr : checking the links, ty :)08:34
av3ngrno worries08:34
av3ngralso you could just email the IT admins  to enable ssh for github :)08:36
-!- Netsplit over, joins: flxb08:41
-!- PSmitAalto [82e9b263@gateway/web/freenode/session] has joined #shogun08:41
-!- blackburn [~qdrgsm@] has joined #shogun08:42
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun08:52
-!- novice [b4953264@gateway/web/freenode/ip.] has quit [Ping timeout: 245 seconds]08:55
-!- novice [b4953264@gateway/web/freenode/session] has joined #shogun08:58
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]09:13
-!- vikram360 [~vikram360@] has quit [Read error: Connection reset by peer]09:16
-!- vikram360 [~vikram360@] has joined #shogun09:17
-!- novice [b4953264@gateway/web/freenode/session] has quit [Quit: Page closed]09:22
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun09:25
-!- av3ngr [av3ngr@nat/redhat/session] has quit [Changing host]09:57
-!- av3ngr [av3ngr@nat/redhat/x-awjigafhwklslbst] has joined #shogun09:57
-!- PSmitAalto [82e9b263@gateway/web/freenode/session] has quit [Changing host]09:57
-!- PSmitAalto [82e9b263@gateway/web/freenode/ip.] has joined #shogun09:57
-!- av3ngr [av3ngr@nat/redhat/x-awjigafhwklslbst] has quit [Quit: That's all folks!]10:04
-!- ada [] has joined #shogun10:08
-!- Miggy [~piggy@] has joined #shogun10:26
n4nd0blackburn: I think it really sucks for students in Iran and others not to be able to participate :S10:27
blackburnpretty unfair probably yes10:29
blackburnn4nd0: do you know anyone from iran wanting to participate?10:30
blackburnmailing list10:31
-!- jckrz [] has joined #shogun10:43
blackburnsonney2k: I give up, you was right10:49
blackburnn4nd0: damn this Soeren is wise10:50
n4nd0blackburn: what is it now? :)10:51
blackburnwhat do you mean?10:51
n4nd0you said that he was right10:51
n4nd0what did you discuss?10:52
n4nd0something about STL or another thing?10:52
blackburnthere is no need for this libedrt10:52
blackburnI spent a lot of time for useless shit10:52
n4nd0aha, why do you think so?10:52
blackburnI got stucked with it10:52
n4nd0yeah man ... I checked your pull request, there is lot of stuff there, can't we get something profitable from it?10:52
blackburnno need to rewrite things again10:52
blackburnwell it could be done in an easier way10:53
blackburnoh fuck it I'll do this later10:54
blackburnn4nd0: didn't you send a pull request with kernel multiclass stuff?10:55
n4nd0I am on it right now10:55
blackburnaha I just get hallucinated probably :D10:57
n4nd0blackburn: check this a moment please10:58
blackburnoh shit10:58
n4nd0blackburn: is there a way to make presentingPolygon use the method of myGayPolygon??10:58
n4nd0it would be useful for the kernel stuff10:58
n4nd0I want to have set_subset and remove_subset in Kernel and rewrite them in CustomKernel10:59
blackburnit should work well already10:59
-!- jckrz [] has quit [Quit: Ex-Chat]10:59
n4nd0but the trick is that KernelMulticlassMachine would call the one in CustomKernel when it is one10:59
blackburndoesn't it?10:59
n4nd0no gay polygon on the screen11:00
n4nd0just polygons11:00
blackburntry to add constructors11:00
blackburnand GayPolygon(): Polygon()11:00
blackburnoh damn man that's funny11:01
blackburngay polygon11:01
n4nd0the same11:01
blackburnwhat if you use pointers?11:02
n4nd0that's a good idea actually11:02
n4nd0blackburn: you're wise too :)11:03
blackburnso pointers make things work?11:04
blackburnthat's interesting11:04
n4nd0I can say a Gay polygon now11:04
blackburnprobably your gaypolygon was casted to polygon somehow11:04
n4nd0yeah, something like that11:04
blackburnactually for methods like that11:05
n4nd0I have studied all this stuff at school with java so I know that it can be done11:05
blackburnyou would rather use const Polygon& poly11:05
n4nd0but sometimes I screw when I say ok, how do I do this in C++11:05
blackburnbut not Polygon poly11:05
blackburntry with const Polygon&11:05
n4nd0let me see11:05
n4nd0works as well with references11:06
blackburnone more puzzle for interview LOL11:06
blackburndid you get what is going on?11:07
n4nd0I guess that when the function is called11:07
n4nd0since the argument was not a reference11:08
n4nd0but the type directly11:08
n4nd0it was up-casting it11:08
blackburnno, it copies11:08
blackburnactually it should copy this instance11:09
n4nd0then whatever I give to the function11:09
blackburnif it is not a const11:09
n4nd0if it is a Polygon11:09
blackburnone more idea11:09
n4nd0it will be a Polygon11:09
blackburnwhat if you add some field to gay polygon11:09
blackburnwill it avoid gayness still?11:09
blackburnthey are almost equal in definition I mean11:10
n4nd0I would guess that yes11:10
n4nd0but let's check11:10
blackburnbut actually good lesson there - use const &11:10
n4nd0still not gay if function gets Polygon11:11
blackburnin fact I didn't get how can you use it for kernels?11:12
n4nd0Kernel.h defines set_subset and remove_subset11:13
n4nd0or CKernel better11:13
n4nd0and CCustomKernel overrides it11:13
blackburnyeah right11:14
Miggyhi blackburn11:16
Miggydo u know if anyone is working on "Built generic multiclass learning framework" project?11:16
blackburnnot really11:17
Miggyi mean has anyone given any proposal for that project?11:18
Miggyi m very much interested in doing that project11:19
blackburnoh I don't know11:19
Miggyblackburn: when does sonney2k come...he is the mentor and i would like to discuss abt the project11:21
n4nd0well guys, gtg now11:22
n4nd0blackburn: I will push the multiclass kernel stuff a bit later11:22
blackburnn4nd0: see you11:23
Miggyn4nd0: u working on multiclass learning framework?11:23
n4nd0Miggy: oh sorry, I didn't read11:24
blackburnMiggy: just some patch11:24
blackburnMiggy: yeah try to catch him this evening11:24
n4nd0Miggy: so I am planning to apply for it, but it is not my first option, my first is SO11:24
n4nd0bye people!11:25
-!- n4nd0 [] has quit [Quit: leaving]11:25
Miggyblackburn: really finding it hard to do patch...suggest me something please11:26
Miggyany small algo or lib functions will do11:27
blackburnsorry have to go now11:28
blackburnwill be back in a hour or so11:28
Miggyok but please come with something good :)11:29
-!- flxb [] has quit [Quit: flxb]11:35
-!- flxb [~cronor@] has joined #shogun12:25
-!- flxb [~cronor@] has quit [Quit: flxb]13:00
-!- flxb [] has joined #shogun13:06
blackburnMiggy: ok still here/13:37
Miggyyes blackburn14:02
blackburnMiggy: anything you would prefer to do?14:12
-!- n4nd0 [] has joined #shogun14:18
Miggyyes smthing which is not too long..dont want to start with big code14:27
blackburnone patch could be to make use of covertree in knn classifier14:31
blackburnsimpler one is to implement some graphical regression or classification example14:33
blackburnwith some popular dataset14:33
-!- vikram360 [~vikram360@] has quit [Ping timeout: 250 seconds]14:39
-!- PhilTillet [] has joined #shogun14:51
-!- harshit_ [~harshit@] has joined #shogun14:51
harshit_hey PhilTillet14:54
harshit_and hi n4nd014:54
-!- pluskid [~chatzilla@] has joined #shogun14:54
n4nd0harshit_: hey man14:54
harshit_whats going on .. what are you working on, these days ?14:55
n4nd0harshit_: so I pushed yesterday some multiclass stuff, one-vs-one for generic14:56
n4nd0harshit_: now I am preparing an example with multiclass kernel using that stuff14:57
n4nd0harshit_: what about you?14:57
harshit_ohh great ..14:57
harshit_actually i didnt get time, last 2 days were really hectic for me.. now next 2 days are free14:58
harshit_so planning to do something big14:58
harshit_hey i had a problem14:58
harshit_in my data folder in shogun directory14:58
-!- vikram360 [~vikram360@] has joined #shogun14:58
harshit_i dont have any files14:58
harshit_i dont knw why its like that, coz of that many octave examples are not running14:59
n4nd0data is in another repo14:59
n4nd0check it on github14:59
n4nd0go to shogun user and check its repos14:59
blackburngit submodule init --update14:59
n4nd0there is another one called shogun-data14:59
n4nd0ok ... that's probably a better option :)14:59
n4nd0blackburn: thank you! I had the other repo in another place and soft links all around the place :O15:00
-!- vikram360 [~vikram360@] has quit [Ping timeout: 260 seconds]15:03
-!- vikram360 [~vikram360@] has joined #shogun15:03
-!- pluskid [~chatzilla@] has quit [Ping timeout: 260 seconds]15:13
harshit_dam, this data directory is really big !!15:19
blackburnyes it is15:22
-!- PhilTillet [] has quit [Ping timeout: 244 seconds]15:35
-!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: Connection reset by peer]15:43
-!- ada [] has quit [Quit: Leaving.]15:45
n4nd0see you guys!16:03
-!- n4nd0 [] has quit [Quit: leaving]16:04
harshit_blackburn : do you knw any multiclass svm classifier that doesnt have its example in octave_modular ?16:08
blackburnharshit_: IIRC examples of octave_modular are pretty outdated16:14
harshit_so i have created an example where 20NewsGroup dataset is being classified (Actually there is a lil bit of work left ) but i am using cgmnpsvm, whose example exists beforehand16:18
harshit_and i am not using any context, its just BOW representation of 20NewsGroup that i am using16:19
blackburnharshit_: consider using multiclassliblinear then16:21
harshit_okay, thanks blackburn16:22
blackburnharshit_: gmnp is kernelized one, no need to use it there16:22
harshit_but kernelized features might give better result . no ?16:23
harshit_never mind multiclassliblinear example also doesnt exist so i think doing that would be fruitful ..16:25
blackburnharshit_: yes if you use good kernel for that16:32
blackburnhowever even linear would work16:32
harshit_blackburn : also i was thinking to include some nlp stuff in shogun like porter's algorithm ,or any other stemming algo16:34
blackburnfeel free!16:35
harshit_do you think shogun requires it16:35
harshit_:) gr16:35
-!- gsomix [~gsomix@] has joined #shogun16:37
blackburnharshit_: not really but it shows how can you use it, it is rather application16:43
-!- ada [] has joined #shogun16:48
-!- jckrz [] has joined #shogun16:55
harshit_blackburn: So I think i should ask sonney2k before doing it .16:55
-!- ada [] has quit [Quit: Leaving.]17:06
-!- harshit_ [~harshit@] has quit [Ping timeout: 252 seconds]17:44
-!- PhilTillet [] has joined #shogun18:47
-!- flxb [] has quit [Quit: flxb]18:48
-!- harshit_ [~harshit@] has joined #shogun19:17
-!- PhilTillet_ [] has joined #shogun19:36
-!- PhilTillet [] has quit [Read error: Connection reset by peer]19:36
PhilTillet_Hi again19:43
PhilTillet_I hate jet lag.19:47
PhilTillet_I arrived at 2pm, slept more or less until now19:48
PhilTillet_Now I am working on my GSoC proposal.19:49
PhilTillet_I pretty much got some results with SVM on the GPU19:54
blackburnPhilTillet_: actually sonney2k said he has got worse results than you20:04
PhilTillet_I know, i saw his mail, the results are weird but I have seen on the internet some complaints about OpenCL on the 9600GT M20:05
PhilTillet_so I tested it on a dedicated benchmark platform and got even better results20:05
PhilTillet_that is a factor 820:05
PhilTillet_(i7 950 vs GTX 470)20:06
blackburnI see20:12
PhilTillet_I'll try to test it on more hardware though, and for more problem sizes20:14
PhilTillet_Essentially, the results should be about the same then for Matrix-Matrix product20:17
PhilTillet_as I am using somewhat the same layout of parallelization20:17
PhilTillet_as for* XD (still tired :( )20:19
-!- prabhakarr [0e8b5206@gateway/web/freenode/ip.] has joined #shogun20:33
prabhakarrAm trying to install shogun for using the "converter" tools. Everything else seems to be working except "converter". Any help is appreciated !20:34
blackburnprabhakarr: what is the exact problem?20:38
prabhakarr"from shogun.Converter import Isomap" results in "ImportError: cannot import name Isomap"; i got the recent version from git, compiled "python_modular"20:42
prabhakarrThis is from the example code given: "../examples/documented/python_modular/"20:44
CIA-64shogun: Soeren Sonnenburg master * r25f5321 / (5 files):20:46
CIA-64shogun: Merge pull request #403 from pluskid/gsoc-pullrequest-120:46
CIA-64shogun: Make the testsuite data generation script run again. -
blackburnprabhakarr: did you install lapack?20:46
-!- flxb [] has joined #shogun20:47
prabhakarrblackburn: thanks for the pointers: this is my first time in any IRC. lapack is installed20:49
blackburnprabhakarr: was it detected by shogun configure?20:49
prabhakarrblackburn: thats a good catch: ./configure gives: "Checking for Vanilla lapack and blas support ... no (missing some features: LDA, liblinear, ...)" ;20:55
blackburnprabhakarr: what is the os you use?20:55
prabhakarrblackburn: ubuntu 11.04 x86_64. installed lapack. now ./configure  recognises and gives "yes". now retrying the install20:56
blackburnprabhakarr: try to install liblapack-dev20:57
prabhakarrblackburn: thats exactly what I did. [realised as soon as you pointed out the lapack configure "no", that i was missing the dev package]. install in progress, fingers crossed :)20:58
-!- n4nd0 [] has joined #shogun21:14
-!- romi_ [~mizobe@] has joined #shogun21:18
prabhakarrblackburn: with your pointer for lapack, it seems to be working. Thanks a ton.21:24
blackburnno problem21:24
-!- prabhakarr [0e8b5206@gateway/web/freenode/ip.] has left #shogun []21:25
blackburnsonney2k: hey21:30
@sonney2kblackburn, hey21:30
blackburnsonney2k: what is # of proposals? ;)21:31
@sonney2kI am swamped in stuff21:31
@sonney2klet me check21:31
blackburnand how are you?21:31
blackburnoh that's bad21:31
blackburnsonney2k: will be 12 in a min21:31
@sonney2kthis extrapolates to 30 or so :)21:32
@sonney2kI wouldn't mind to see a lot less this year21:32
blackburnsonney2k: 1221:33
@sonney2kblackburn, what do you think of ?21:34
blackburnsonney2k: I have probably no opinion on thta21:35
-!- PhilTillet_ [] has quit [Remote host closed the connection]21:35
@sonney2kwe should be on LF only anyways21:35
blackburnare we on CRLF right now?21:35
@sonney2kor do we have any developer doing any work on win crap?21:35
@sonney2klinux is just LF21:35
@sonney2kosx too21:35
blackburnI know no one21:35
@sonney2kso there shouldn't be any danger21:36
-!- PhilTillet [] has joined #shogun21:37
@sonney2kblackburn, reviewing is a pain21:37
@sonney2kmany changes and a lot of them just whitspace21:37
blackburnsonney2k: whoaa21:37
blackburnsonney2k: does evan want to participate in gsoc?21:37
@sonney2kbut the whitespace fixes seem rightful21:38
@sonney2kfixes lots of trailing spaces21:38
blackburnyeah trailing spaces is shit21:38
blackburnsonney2k: any option for vim for that?21:38
PhilTilletsonney2k, can the source file i sent to you be considered as the "small patch"? :D21:39
-!- gsomix [~gsomix@] has quit [Ping timeout: 276 seconds]21:39
@sonney2kblackburn, btw also seems nice work21:40
@sonney2kand a neat example fetching data from uci :)21:40
blackburnyeah I thought I suggested it21:41
@sonney2kat least we never had clustering metrics and I have never heard of munkres before but like the idea of using it to resolve label ambiguity21:41
blackburnsonney2k: what is idea he applying for?21:42
n4nd0Munkres ... I used that for my sokoban solver, nice algorithm21:42
n4nd0blackburn: he is applying for SO :S21:42
blackburnI feel pain right in my ass thinking about it21:44
@sonney2kPhilTillet, you need to get into shogun too - the isolated code is very drafty (which is ok at this stage) but it would be a big big plus to have the kernel machine opencl parallelization within shogun21:48
@sonney2kPhilTillet, ahh btw the code still does not do the 1000 support vectors / 100000 test examples case21:49
@sonney2kwhich is the more common case21:49
@sonney2kPhilTillet, I know you are very busy with conferences and stuff but there is still 1 week...21:50
PhilTilletsonney2k, I am done with the conference actually, back in France :) What do you mean with "does not do" ? infinite runtime?21:51
@sonney2kPhilTillet, well you currently use a case where you have 1000 SVs and 1000 test examples21:52
PhilTilletand tested it with 10000test example too21:52
PhilTilletbut the execution time was very long :p21:52
PhilTilleti mean i can test it with 100 000 but then I have to take a coffee break lol21:52
PhilTilletand I don't like coffee, so let's say a water break21:53
-!- gsomix [~gsomix@] has joined #shogun21:54
PhilTilletok the test case for 100000 test example is launched21:54
@sonney2kPhilTillet, well 100 times longer21:55
PhilTilletI'll run it on that Benchmarking machine, should be faster :p21:55
@sonney2kyeah just some 30 seconds or so21:56
@sonney2kPhilTillet, btw I am not happy with the numerical precision though21:56
@sonney2kI don't see any reason why results should deviate by more than 1e-621:56
@sonney2k1e-3 is a lot IMO21:57
PhilTilletwell, it's 1e-3 over 1000example21:57
PhilTilletso 1e-6 per example21:57
PhilTilletthe norm2 of the difference is still zero from the tests I did21:58
@sonney2knorm2 is unfair I really mean absdiff22:03
PhilTilletAnd okay i'll try to integrate the OpenCL work into CKernelMachine22:03
@sonney2kPhilTillet, shouldn't your benchmark machine be able to do double btw?22:03
PhilTilletit can22:04
PhilTilletbut not on my mobile GPU22:04
@sonney2kthen you could directly compare for the 1000x1000 problem22:04
PhilTilletthe problem is that the GPU has to be double compatible22:04
PhilTilletwhich is not always the case :p22:04
blackburnanyone wanting to port some matlab code?22:05
@sonney2kyeah yeah but it is important to check if there is a bug (which is what I suspsect)22:05
@sonney2kblackburn, ???22:05
blackburnsonney2k: what is ??? for ? ;)22:05
@sonney2kmatlab code? what are you doing?22:06
blackburnsonney2k: easy task - port some dim reduction method22:06
blackburnanyway, n4nd0 will care about stochastic proximity embedding next days22:07
n4nd0blackburn: let's see how it goes with it ;)22:08
PhilTilletsonney2k, the SVM is still training with 100 000 examples :p22:13
PhilTilletthe training seems quite computationally expensive :p22:47
blackburnPhilTillet: what is d and N?22:47
PhilTilletN = 100 000, d=100022:48
blackburnand kernel?22:48
blackburnoh that can take a while :)22:48
@sonney2kPhilTillet, training? why training?22:48
@sonney2kor do you mean apply?22:48
PhilTilletwell the SVM is trained before apply22:48
PhilTilletdoes not need to be?22:48
@sonney2kwhy that?22:48
blackburnPhilTillet:  you probably can set alphas manually..22:49
@sonney2kjust set alphas to 122:49
@sonney2k(or random numbers again)22:49
@sonney2kbut yes training is very expensive...22:50
PhilTilleti'm sure training can also be done on GPU :P22:50
PhilTilletbut not in 1 week :D22:50
blackburnthat's something complex22:50
blackburntake a look on libsvm source22:50
PhilTilletSo I have to set the ids of the support vectors manually too?22:53
blackburnPhilTillet: yeah just create KernelMachine using custom alphas and indices22:53
PhilTilletokay XD22:54
shogun-buildbotbuild #405 of python_modular is complete: Failure [failed test_1]  Build details are at  blamelist: pluskid@gmail.com22:55
n4nd0I got a graphical example for multiclass kernel machines23:06
n4nd0what do you think guys?23:06
shogun-buildbotbuild #401 of ruby_modular is complete: Failure [failed compile]  Build details are at  blamelist: pluskid@gmail.com23:06
n4nd0it is a least curious :P23:07
blackburnn4nd0: what is the svm?23:07
n4nd0my gf asks why I am doing flowers23:07
blackburntell her because of gay polygon23:08
n4nd0LibSVM with GaussianKernel23:08
n4nd0one vs rest23:08
n4nd0I am going to try with o.vs.o23:08
n4nd0but I think it will be the same23:08
blackburnyeah a little difference23:09
n4nd0oh shit, seg fault with one vs. one, got to debug23:10
PhilTilletsonney2k, so I should add a clean apply_opencl function to CKernelMachine ?23:10
PhilTilletand if possible make it templated so that both apply_opencl<float> and apply_opencl<double> work?23:11
CIA-64shogun: Evan Shelhamer master * rb763e52 / (337 files in 23 dirs): Trailing whitespace cleanup -
CIA-64shogun: Soeren Sonnenburg master * r17ea108 / (337 files in 23 dirs):23:18
CIA-64shogun: Merge pull request #404 from shelhamer/whitespace-cleanup23:18
CIA-64shogun: Trailing whitespace cleanup -
@sonney2kPhilTillet, well first I would be very interested in seeing the results (much lower error) and comparison to 1000 SVs - 100k test23:26
@sonney2kthen a apply_gpu  makes sense where the old apply code is put into apply_cpu23:28
-!- Vuvu [~Vivan_Ric@] has joined #shogun23:28
@sonney2kand then one can set a flag such that apply() uses the cpu or gpu function23:29
-!- flxb [] has quit [Read error: Connection reset by peer]23:29
-!- flxb [] has joined #shogun23:30
shogun-buildbotbuild #410 of java_modular is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #406 of python_modular is complete: Success [build successful]  Build details are at
shogun-buildbotbuild #402 of ruby_modular is complete: Success [build successful]  Build details are at
-!- harshit_ [~harshit@] has quit [Quit: Leaving]23:57
-!- jckrz [] has quit [Quit: Ex-Chat]23:58
--- Log closed Sat Mar 31 00:00:19 2012