Open in new window / Try shogun cloud
--- Log opened Sun Jun 10 00:00:41 2012
n4nd0blackburn: ?00:01
blackburn:D00:01
blackburnneed matlab to test slep :)00:01
n4nd0blackburn: aham00:10
n4nd0do you need a copy of matlab?00:10
blackburnno I have one00:11
n4nd0all right then :)00:11
blackburnand I dooo nooot have crack for that00:11
blackburn:D00:11
n4nd0ooh I see00:11
blackburnsonney2k: still there?00:12
blackburnI am searching for a conference where I can present shogun :D00:12
n4nd0nice00:13
n4nd0in Russia, or you don't mind where?00:13
blackburnbetter far away :)00:14
n4nd0blackburn: just let me know if you want some help with it00:14
n4nd0I would love to participate in any way00:14
blackburnsure00:14
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]01:53
-!- emrecelikten [~Anubis@176.40.238.218] has quit [Quit: Leaving.]02:21
n4nd0good night guys02:34
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving]02:36
-!- blackburn [~blackburn@31.28.59.65] has quit [Ping timeout: 260 seconds]03:00
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun03:09
CIA-9shogun: Sergey Lisitsyn master * rbc2cf3e / (6 files in 3 dirs): Added sparse inverse covariance estimator - http://git.io/TYOcFQ04:32
-!- blackburn [~blackburn@31.28.59.65] has quit [Ping timeout: 265 seconds]05:21
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun07:26
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun07:54
-!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: No route to host]07:54
-!- wiking_ is now known as wiking07:54
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]08:10
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun08:15
-!- blackburn [~blackburn@31.28.59.65] has joined #shogun10:38
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]10:47
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun10:54
-!- blackburn [~blackburn@31.28.59.65] has quit [Ping timeout: 245 seconds]11:25
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun12:02
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 240 seconds]12:16
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun12:34
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host]12:34
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun12:34
-!- pluskid [~pluskid@111.120.62.134] has joined #shogun13:36
-!- emrecelikten [~Anubis@176.40.238.218] has joined #shogun13:54
-!- gsomix [~gsomix@80.234.31.247] has joined #shogun14:06
gsomixhi all14:06
emreceliktengsomix: Hello gsomix14:06
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 260 seconds]14:24
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 252 seconds]14:38
-!- blackburn [~blackburn@109.226.76.136] has joined #shogun14:40
-!- pluskid [~pluskid@111.120.62.134] has quit [Quit: Leaving]14:41
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun14:44
-!- blackburn [~blackburn@109.226.76.136] has left #shogun []15:50
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 246 seconds]15:56
-!- puffin444 [472e31fb@gateway/web/freenode/ip.71.46.49.251] has joined #shogun16:37
-!- puffin444 [472e31fb@gateway/web/freenode/ip.71.46.49.251] has quit [Quit: Page closed]16:58
-!- puffin444 [472e31fb@gateway/web/freenode/ip.71.46.49.251] has joined #shogun16:58
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun17:01
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host]17:01
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun17:01
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun17:24
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 244 seconds]18:27
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun18:28
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host]18:28
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun18:28
-!- wiking_ [~wiking@68.68.31.85.customer.bluemilenetworks.com] has joined #shogun19:06
-!- wiking_ [~wiking@68.68.31.85.customer.bluemilenetworks.com] has quit [Changing host]19:06
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun19:06
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 252 seconds]19:08
-!- wiking_ is now known as wiking19:08
-!- blackburn [d55784cd@gateway/web/freenode/ip.213.87.132.205] has joined #shogun19:37
gsomixsonney2k, around?20:34
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 260 seconds]20:35
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun20:36
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host]20:36
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun20:36
@sonney2kgsomix, yes - whats up?20:39
gsomixsonney2k, machines20:39
gsomixcan you look at director machines?20:39
-!- blackburn [d55784cd@gateway/web/freenode/ip.213.87.132.205] has quit [Ping timeout: 245 seconds]20:45
-!- blackburn [~blackburn@188.168.3.9] has joined #shogun20:49
@sonney2kgsomix, looks fine - now we only need some example20:57
gsomixsonney2k, have any ideas?20:58
blackburnI see one problem here21:05
blackburntrain_machine is a) protected b) have default parameter21:05
@sonney2kgsomix, you could do parzen windows with gaussian kernel (equation 55 in   https://www.cs.utah.edu/~suyash/Dissertation_html/node11.html )21:07
@sonney2kfor the directorkernelmachine21:07
blackburnhaha parzen finally21:07
blackburnor with Epanechnikov kernel!21:08
@sonney2ksounds like a disease21:08
blackburnsonney2k: no typical kernel in atypical kernel mean :)21:09
blackburnstatistic kernel I mean21:10
blackburnis there anyone to describe me what is advantage having sparse inverse covariance estimator?21:11
@sonney2kgsomix, and for linearmachine - you could implement a linear perceptron algorithm is here (page 2 on top https://www.cs.cmu.edu/~avrim/ML09/lect0126.pdf)21:11
gsomixok21:12
@sonney2kgsomix, but for the very first test just do something as simple as using alpha=-1 for negatively labeled examples and +1 otherwise for the directorkernelmachine21:13
gsomixsonney2k, ok. what about dotfeatures?21:15
@sonney2kand some random w vector in linear machines (b=0)21:18
@sonney2kgsomix, for directordotfeatures - best example would be to just implement all the required operations for some numpy double matrix21:19
-!- blackburn [~blackburn@188.168.3.9] has quit [Ping timeout: 244 seconds]21:19
@sonney2kgsomix, then you can run the code once using your director features and once using shogun's densefeatures21:19
@sonney2kand result should be the same21:19
@sonney2kgsomix, I started a directordotfeatures example - but it is not totally there yet21:20
@sonney2kgsomix, it might need some new operators in SGVector ( overload * / - + etc)21:21
@sonney2k+= / -= ...21:21
@sonney2kgsomix, got disconnected21:32
@sonney2knot sure if you've seen this21:32
@sonney2k<sonney2k> gsomix, then you can run the code once using your director features and once using shogun's densefeatures21:32
@sonney2k<sonney2k> and result should be the same21:32
@sonney2k<sonney2k> gsomix, I started a directordotfeatures example - but it is not totally there yet21:32
@sonney2k<sonney2k> gsomix, it might need some new operators in SGVector ( overload * / - + etc)21:32
@sonney2k<sonney2k> += / -= ...21:32
gsomixsonney2k, aha21:33
@sonney2kgsomix, first line was gsomix, for directordotfeatures - best example would be to just implement all the required operations for some numpy double matrix21:33
@sonney2kgsomix, does that make sense?21:33
gsomixsonney2k, yep. I think I'll start with dotfeatures.21:34
@sonney2kgsomix, can you fix the return 0 -> return false in your pull request21:42
@sonney2kI would then merge these21:42
gsomixah ^__^''21:42
gsomixsonney2k, done21:45
CIA-9shogun: Evgeniy Andreev master * rb01c399 / src/shogun/machine/DirectorKernelMachine.h : added DirectorKernelMachine - http://git.io/KAkBrA21:49
CIA-9shogun: Evgeniy Andreev master * r0df5d4d / src/shogun/machine/DirectorLinearMachine.h : added DirectorLinearMachine - http://git.io/s7NjUQ21:49
CIA-9shogun: Evgeniy Andreev master * r1914e86 / (3 files in 2 dirs): new classes added to modular - http://git.io/-os58w21:49
CIA-9shogun: Evgeniy Andreev master * r60a7f70 / (2 files): fixed returns in train_function - http://git.io/19v1oQ21:49
CIA-9shogun: Soeren Sonnenburg master * r51c3b36 / (5 files in 2 dirs): Merge pull request #578 from gsomix/director_classes - http://git.io/SKuf0w21:49
@sonney2kgsomix, btw please use git commit --amend next time21:51
@sonney2kthen push --force21:52
gsomixsonney2k, mmm? why?21:52
@sonney2kgsomix, to avoid the useless fix21:58
gsomixa, ok21:58
@sonney2kso one just has one OK commit21:58
n4nd0sonney2k: hey! can I double check with you an issue about the cutting plane algorithm we are using for SO??22:03
@sonney2kn4nd0, ask22:08
n4nd0sonney2k: http://www.cs.cornell.edu/people/tj/publications/finley_joachims_08a.pdf take a look here to second page22:08
n4nd0the algorithm, line 1022:08
n4nd0and the gist here https://gist.github.com/263448722:08
n4nd0line 4122:09
n4nd0in the paper it seems that they solve the QP each time they add a new constraint22:09
n4nd0inside the for loop22:09
n4nd0while in the gist it's outside the for loop22:09
n4nd0I wonder if both are equivalent22:10
n4nd0I am checking in some HMSVM code that Nico sent me and it seems to be done outside the for loop as well, like in the gist22:12
n4nd0this may be the right solution then22:12
@sonney2kn4nd0, both will converge to the same solution22:17
@sonney2kn4nd0, only difference is speed22:17
n4nd0ok22:17
@sonney2kn4nd0, and that really depends on the optimizer speed / argmax speed22:17
@sonney2kn4nd0, but it is clear - if you have zero costs for doing the optimization inside the loop then do it :)22:18
@sonney2kn4nd0, and actually with cplex at least ( I guess mosek too) one can use the dual simplex method and 'hotstarts'22:19
n4nd0n4nd0: I am still debugging, I think I have a bug somewhere since the weight vector learned is all zeros22:19
n4nd0and I think that must be an error even if the dataset is so simply as the one I am using22:19
n4nd0oh I am back to black ... talking to myself ^ :D22:20
@sonney2kn4nd0, well the trivial solution might mean that your example is too hard :)22:43
n4nd0sonney2k: :) I think there's no way for that to happen, I have 17 training examples of two classes, data distributed as Gaussians22:44
@sonney2kn4nd0, why so many?22:51
@sonney2kwhy not 2 examples - one in each class?22:51
n4nd0sonney2k: ok, I am going to try with that22:52
n4nd0it seems funny though22:53
@sonney2kn4nd0, rule #1 try with the simplest possible case first22:55
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking]23:01
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun23:03
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host]23:03
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun23:03
-!- gsomix [~gsomix@80.234.31.247] has quit [Ping timeout: 240 seconds]23:59
--- Log closed Mon Jun 11 00:00:41 2012