--- Log opened Sat Apr 14 00:00:17 2012 | ||
--- Day changed Sat Apr 14 2012 | ||
wiking | :P | 00:00 |
---|---|---|
wiking | i'm gonna test this one | 00:00 |
blackburn | wiking: I still did not get how can that be you have no features for test data | 00:00 |
blackburn | what then you have at all | 00:00 |
wiking | well i do have | 00:01 |
wiking | let's say i have 2 different features about an image | 00:01 |
wiking | 1 feature is based on the image | 00:01 |
wiking | 2 feature is based on the text around the image | 00:01 |
blackburn | text around the image? | 00:01 |
wiking | most of the cases i have 1+2 | 00:01 |
wiking | yes since the text is embedded into a text | 00:02 |
wiking | and the text is for sure not completely unrelated to the image... just in very few cases it is | 00:02 |
wiking | anyhow | 00:02 |
wiking | most the time i have 1+2 features | 00:02 |
wiking | but sometimes | 00:02 |
wiking | i only have feature 1 | 00:02 |
wiking | as i don't have the 'context' of the iamge | 00:02 |
wiking | image | 00:02 |
wiking | just the image itself | 00:03 |
blackburn | I don't think it is DA however | 00:03 |
blackburn | wiking: DA is all about different distributions of test and train data | 00:09 |
wiking | well according the slide u've sent | 00:11 |
wiking | i've got that differently | 00:11 |
wiking | http://www.eatdifferent.com/tips/sugarydrinks#drink=coke suuuuugar | 00:14 |
blackburn | wiking: which slide? | 00:16 |
wiking | http://fml.tuebingen.mpg.de/raetsch/lectures/talk-mtcompbio.pdf | 00:17 |
blackburn | wiking: yes, there is underlying assumption that target probability space is slightly different | 00:18 |
blackburn | wiking: I have seen one fantastic paper | 00:19 |
blackburn | they learned svm recognizing pedestrians on *game* | 00:19 |
blackburn | and used domain adaptation to construct real system | 00:19 |
wiking | :PP | 00:19 |
blackburn | that is the case | 00:19 |
blackburn | in your case I am unsure how can it help | 00:20 |
blackburn | I mean if you have no features it is a latent model :) | 00:20 |
wiking | well my space is slighhtly different | 00:20 |
wiking | :P | 00:20 |
blackburn | so latent DA? nice | 00:20 |
blackburn | :D | 00:20 |
wiking | nono | 00:20 |
wiking | i mean u didn't get it then | 00:20 |
wiking | so you get the 2 different features right? | 00:20 |
wiking | so i use that and train that SVM | 00:21 |
blackburn | on train data? | 00:21 |
wiking | a multiclass svm | 00:21 |
wiking | yes | 00:21 |
wiking | train data has both features | 00:21 |
wiking | features from image and features from text | 00:21 |
blackburn | and test has only text often? | 00:21 |
wiking | i mix them in a way | 00:21 |
wiking | no | 00:21 |
wiking | test has soometimes | 00:22 |
wiking | only image | 00:22 |
wiking | features | 00:22 |
blackburn | aha | 00:22 |
blackburn | right | 00:22 |
wiking | so that's where i see DA could do the trick for me | 00:22 |
blackburn | how? | 00:22 |
wiking | well i train a classifier with the both of the features | 00:23 |
wiking | and then use that | 00:23 |
wiking | in DA where i only have 1 type of feat | 00:24 |
wiking | and yes i can make it both having train/test split for 1 type of features | 00:25 |
blackburn | you didn't convince me :D | 00:26 |
blackburn | wiking: I still think that you should rather do some kernel mean matching or so | 00:29 |
wiking | seems java as a language itself is going to be copyrighted :O | 00:32 |
blackburn | what do you mean? | 00:33 |
wiking | "With the Oracle v. Google trial date set for next Monday, the Judge has asked Google and Oracle to take a position on whether a programming language is copyrightable." | 00:33 |
blackburn | oh i hate oracle | 00:34 |
blackburn | their databases are powerful crap heh | 00:35 |
blackburn | and their intentions are totally evil | 00:35 |
blackburn | cu | 01:11 |
-!- blackburn [~qdrgsm@109.226.90.219] has quit [Quit: Leaving.] | 01:12 | |
-!- harshit_ [~harshit@182.68.181.141] has quit [Ping timeout: 250 seconds] | 01:23 | |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 01:30 | |
-!- harshit_ [~harshit@182.68.181.141] has quit [Remote host closed the connection] | 02:05 | |
-!- vikram360 [~vikram360@117.192.184.89] has quit [Read error: Connection reset by peer] | 05:26 | |
-!- vikram360 [~vikram360@117.192.184.89] has joined #shogun | 05:26 | |
-!- gsomix [~gsomix@109.169.156.156] has joined #shogun | 08:56 | |
-!- gsomix [~gsomix@109.169.156.156] has quit [Client Quit] | 08:57 | |
-!- gsomix [~gsomix@109.169.156.156] has joined #shogun | 08:57 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 10:58 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection] | 11:26 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 11:26 | |
wiking | yeeeey there's a domain transfer gsoc application! | 11:42 |
-!- xiangwang [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has joined #shogun | 12:17 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 264 seconds] | 12:25 | |
-!- blackburn1 [~qdrgsm@109.226.90.219] has joined #shogun | 12:25 | |
CIA-64 | shogun: Sergey Lisitsyn master * r56eb7ca / (14 files in 6 dirs): Merge branch 'lars' of git://github.com/pluskid/shogun (+16 more commits...) - http://git.io/MFvn_Q | 12:29 |
gsomix | blackburn1, yo | 12:30 |
blackburn1 | hey | 12:30 |
wiking | blackburn1: hey... have u seen the gsoc email about domain transfer? | 12:33 |
blackburn1 | wiking: domain transfer? wtf ;) | 12:34 |
wiking | ok domain adaptation | 12:35 |
wiking | ;) | 12:35 |
blackburn1 | some mohamed's mail you mean? | 12:35 |
wiking | ye | 12:35 |
blackburn1 | well yes should I worry? :) | 12:35 |
shogun-buildbot | build #714 of libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/714 blamelist: pluskid@gmail.com | 12:35 |
wiking | worry why? | 12:35 |
wiking | it's just great that somebody wants to do it as gsoc | 12:35 |
blackburn1 | huh! there are some more | 12:36 |
blackburn1 | worry because he kinda competes with me | 12:36 |
wiking | competes? | 12:36 |
wiking | why, i thought that this is a team work | 12:36 |
wiking | :))) | 12:36 |
blackburn1 | for slot | 12:37 |
wiking | ah u want to do gsoc as well this year? | 12:37 |
wiking | i didn't know | 12:37 |
blackburn1 | surprise :D | 12:37 |
wiking | :)))) | 12:37 |
wiking | ahaaaaaa | 12:37 |
wiking | i thought you are mentoring this year | 12:37 |
blackburn1 | no, I am not an expert in anything :) | 12:38 |
blackburn1 | and undergrad student still | 12:38 |
wiking | hahahahahah well in the code of shogun you are | 12:38 |
shogun-buildbot | build #715 of libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/libshogun/builds/715 | 12:39 |
wiking | oooh damn i dont understand pgm ppm formaaat | 12:39 |
-!- blackburn1 [~qdrgsm@109.226.90.219] has quit [Ping timeout: 246 seconds] | 12:42 | |
shogun-buildbot | build #652 of r_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/652 blamelist: pluskid@gmail.com | 12:43 |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has joined #shogun | 12:45 | |
shogun-buildbot | build #674 of cmdline_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cmdline_static/builds/674 blamelist: pluskid@gmail.com | 12:47 |
wiking | nyiha | 12:47 |
wiking | something got broken | 12:47 |
shogun-buildbot | build #653 of octave_static is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/653 blamelist: pluskid@gmail.com | 12:47 |
shogun-buildbot | build #640 of python_static is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_static/builds/640 blamelist: pluskid@gmail.com | 12:48 |
shogun-buildbot | build #653 of r_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/r_static/builds/653 | 12:53 |
-!- blackburn1 [~qdrgsm@109.226.90.219] has joined #shogun | 12:53 | |
blackburn1 | wiking: what is the problem with pgm? | 12:54 |
wiking | blackburn1: nada... found out what was the problem... in the header i've got some comments but managed to remove them | 12:54 |
wiking | but we have a problem with the new merge | 12:54 |
blackburn1 | yes but it may be related to order of commits | 12:55 |
blackburn1 | let us wait a little | 12:55 |
wiking | ah | 12:55 |
blackburn1 | wiking: do you read pgm manually? | 12:56 |
wiking | noup | 12:56 |
wiking | fscanf failed | 12:56 |
wiking | ;) | 12:56 |
wiking | for reading the header | 12:56 |
blackburn1 | wiking: what do you use then? | 12:56 |
wiking | i thought it's always 4 integers are coming in a given order | 12:56 |
wiking | so i did an fscanf on the header of the pgm | 12:56 |
wiking | but then some images that i used to convert with imagemagic | 12:57 |
wiking | had some comments in them | 12:57 |
wiking | from exif or i have no idea from | 12:57 |
wiking | where | 12:57 |
shogun-buildbot | build #675 of cmdline_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/cmdline_static/builds/675 | 12:57 |
wiking | anyhow imagemagick left those comments in the header | 12:57 |
blackburn1 | I mean python's imload have no problem with comments | 12:57 |
wiking | hehehe i have my own utility ... it's a pretty lame feature extractor | 12:58 |
wiking | but still i want to see how it works | 12:58 |
wiking | mmm | 12:58 |
wiking | i saw new io classes | 12:58 |
wiking | well anyhow i have to get back somehow to be able to feed gmm with my custom created gmms... so i can start clustering | 13:00 |
wiking | damn my shell script is stupid should do something better about it | 13:01 |
CIA-64 | shogun: Victor Sadkov master * rda846bd / (2 files): Refactored to satisfy code formatting standards - http://git.io/BUCoIw | 13:01 |
CIA-64 | shogun: Sergey Lisitsyn master * r416606c / (2 files): Merge branch 'refactoring' of git://github.com/Victor-Sadkov/shogun - http://git.io/t-ve5Q | 13:01 |
shogun-buildbot | build #654 of octave_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/654 | 13:02 |
wiking | lol i've just read about new way of stealing a car: they put on your parking car of the rear window in the middle a paper, so when you sit in your car and you start the engine and start to get out from the parking slot you'll realize that there's a car at the rear window, you get out (of course leave everything in the car and the engine running) to remove the paper and at this moment somebody will jump into your car and run away with your car and all yo | 13:07 |
wiking | belongings in it ;) | 13:07 |
wiking | smart :) | 13:07 |
shogun-buildbot | build #641 of python_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_static/builds/641 | 13:07 |
blackburn1 | wiking: heh | 13:07 |
wiking | blackburn1: it's cool no? :) | 13:08 |
blackburn1 | wiking: I didn't get what exactly they put? | 13:09 |
wiking | anything | 13:09 |
wiking | a piece of paper in the middle of the rear window | 13:09 |
wiking | just that you get distracted | 13:09 |
blackburn1 | I see | 13:09 |
wiking | so that you'd get out from the car to remove it | 13:09 |
shogun-buildbot | build #500 of octave_modular is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/500 blamelist: pluskid@gmail.com | 13:11 |
shogun-buildbot | build #488 of csharp_modular is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/488 blamelist: pluskid@gmail.com | 13:12 |
shogun-buildbot | build #477 of lua_modular is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/lua_modular/builds/477 blamelist: pluskid@gmail.com | 13:18 |
-!- xiangwang [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has quit [Quit: Leaving] | 13:20 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 13:24 | |
shogun-buildbot | build #475 of python_modular is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/python_modular/builds/475 blamelist: pluskid@gmail.com, blackburn91@gmail.com | 13:35 |
blackburn1 | uh that takes a while | 13:37 |
-!- karlnapf [~heiko@host86-185-15-106.range86-185.btcentralplus.com] has joined #shogun | 13:37 | |
blackburn1 | n4nd0: could you please remind me what I promised to do? :D | 13:38 |
n4nd0 | blackburn1: hmmmm | 13:40 |
n4nd0 | I think I have it on the tip of my tongue | 13:41 |
n4nd0 | blackburn1: no way, I don't remember :( | 13:42 |
gsomix | n4nd0, hi. how are you? | 13:45 |
n4nd0 | gsomix: hey! I am fine, what about you? | 13:46 |
gsomix | n4nd0, I'm lazy, hehe. what about covertree? | 13:46 |
n4nd0 | gsomix: it's going good :) | 13:47 |
n4nd0 | I hope I will have something ready to push during this weekend | 13:47 |
gsomix | cool! | 13:47 |
n4nd0 | gsomix: what about you, what are you working on? | 13:49 |
gsomix | n4nd0, reference counting for SGVector. but I had some problems with time over the past days. | 13:51 |
n4nd0 | gsomix: aham, good luck with it anyway ;) | 13:52 |
blackburn1 | n4nd0: ah consider merging your ovo thing | 13:53 |
n4nd0 | blackburn1: yeah, but first I should test if it now it works fine | 13:54 |
n4nd0 | blackburn1: don't you think so? | 13:54 |
blackburn1 | n4nd0: why couldn't it work? | 13:54 |
blackburn1 | karlnapf: hey! | 13:54 |
karlnapf | blackburn1, hi :) | 13:55 |
karlnapf | heard there are 9 slots :D | 13:55 |
n4nd0 | blackburn1: who knows | 13:55 |
blackburn1 | karlnapf: yeap | 13:55 |
blackburn1 | karlnapf: could you please add subsetting to labels as well? ;) | 13:55 |
karlnapf | pretty cool | 13:55 |
karlnapf | blackburn1, yes its on my todo list | 13:55 |
karlnapf | sorry for being away last week, too busy | 13:56 |
blackburn1 | ok nice thanks | 13:56 |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 13:58 | |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Client Quit] | 13:59 | |
blackburn1 | n4nd0: I guess that is I promised to do | 14:00 |
blackburn1 | karlnapf: what is your next weeks workload? | 14:01 |
karlnapf | blackburn1, I just handed in my last coursework, so I will have a bit more time. However, there are still the exams in May, which are really important and therefore currently have priority | 14:03 |
blackburn1 | karlnapf: I stumbled upon maximum mean discrepancy embedding and will ask you for some advice later :) | 14:04 |
karlnapf | but I think I will be able to do some shogun stuff in between | 14:04 |
karlnapf | really? where? | 14:04 |
karlnapf | This is really cool stuff (-) | 14:04 |
karlnapf | 8-) | 14:04 |
blackburn1 | karlnapf: MMD or MMDE? | 14:04 |
blackburn1 | I assume you are MMD expert | 14:04 |
blackburn1 | MMDE is some kind of multitask stuff | 14:05 |
karlnapf | maximum mean discrepancy embedding? | 14:05 |
karlnapf | never heard it, what happens there | 14:05 |
blackburn1 | karlnapf: LLE(MVU) + MMD | 14:05 |
karlnapf | LLE(MVU)? | 14:05 |
blackburn1 | Locally Linear Embedding | 14:06 |
blackburn1 | okay idea is to use MMD as distance if I get it right | 14:06 |
blackburn1 | and construct embedding | 14:06 |
karlnapf | MMD is a distance between distributions | 14:06 |
blackburn1 | as latent space for united features space for domain | 14:06 |
blackburn1 | karlnapf: can't it be empirical? | 14:07 |
karlnapf | yes, of course, you dont know the distributions | 14:07 |
karlnapf | ok, sounds cool, do you have a paper? | 14:07 |
blackburn1 | https://home.cse.ust.hk/~jamesk/papers/aaai08.pdf | 14:07 |
karlnapf | looks interesting | 14:09 |
karlnapf | cool I am always looking for applciations of MMD | 14:09 |
karlnapf | the stuff I am doing will be kind of a large scale online two sample test, so a bit more statistics than ML | 14:10 |
blackburn1 | I see | 14:10 |
blackburn1 | so how can you apply it? | 14:10 |
karlnapf | what do you mean? | 14:11 |
karlnapf | like which applications? | 14:11 |
blackburn1 | karlnapf: yes | 14:12 |
karlnapf | the current MMD test ist in quadratic time. However, imagine you have infinite data and want the best test possible for a given amount of time | 14:12 |
karlnapf | then you do a kind of online-like kernel selection on a stream of data | 14:12 |
blackburn1 | aha | 14:12 |
karlnapf | either you optimise kernel until test says distributions are different | 14:13 |
blackburn1 | but which distributions do you compare? | 14:13 |
karlnapf | or until you can say, probably they are equal | 14:13 |
karlnapf | data comes from two distributions | 14:13 |
karlnapf | goal is a two sample test | 14:13 |
karlnapf | like: | 14:13 |
karlnapf | two sources of microarray data | 14:13 |
karlnapf | test is cancer type is equal or not | 14:14 |
karlnapf | Ill tell you more about it later :) | 14:14 |
karlnapf | currently in the early stages | 14:14 |
karlnapf | oh and thats my master project | 14:14 |
karlnapf | the gsoc stuff is a bot different | 14:14 |
karlnapf | but anyway, I will study a bit now :) see you around! | 14:15 |
blackburn1 | ok looks clearer now | 14:15 |
blackburn1 | thanks for explanation :) | 14:15 |
blackburn1 | see you | 14:15 |
shogun-buildbot | build #501 of octave_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_modular/builds/501 | 14:16 |
blackburn1 | wiking: ^ all ok :) | 14:16 |
shogun-buildbot | build #489 of csharp_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/csharp_modular/builds/489 | 14:23 |
wiking | \o/ | 14:26 |
shogun-buildbot | build #478 of lua_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/lua_modular/builds/478 | 14:29 |
shogun-buildbot | build #470 of ruby_modular is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/ruby_modular/builds/470 blamelist: sadkov993@gmail.com | 14:35 |
shogun-buildbot | build #471 of ruby_modular is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/ruby_modular/builds/471 | 15:27 |
-!- vikram360 [~vikram360@117.192.184.89] has quit [Ping timeout: 265 seconds] | 15:31 | |
-!- vikram360 [~vikram360@117.192.163.46] has joined #shogun | 15:32 | |
-!- blackburn1 [~qdrgsm@109.226.90.219] has quit [Ping timeout: 276 seconds] | 15:59 | |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 16:27 | |
-!- harshit_ [~harshit@182.68.181.141] has quit [Ping timeout: 265 seconds] | 16:35 | |
-!- pluskid [~pluskid@1.204.101.113] has joined #shogun | 16:36 | |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 16:43 | |
gsomix | sonney2k, hey. | 16:57 |
-!- harshit_ [~harshit@182.68.181.141] has quit [Ping timeout: 250 seconds] | 17:04 | |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has quit [Quit: ( www.nnscript.com :: NoNameScript 4.22 :: www.esnation.com )] | 17:17 | |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 17:17 | |
-!- pluskid [~pluskid@1.204.101.113] has quit [Ping timeout: 252 seconds] | 17:32 | |
-!- pluskid [~pluskid@173.254.214.60] has joined #shogun | 17:33 | |
-!- wiking_ [~wiking@78-23-189-112.access.telenet.be] has joined #shogun | 17:44 | |
-!- wiking_ [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] | 17:44 | |
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun | 17:44 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 244 seconds] | 17:48 | |
-!- wiking_ is now known as wiking | 17:48 | |
-!- pluskid [~pluskid@173.254.214.60] has quit [Quit: Leaving] | 17:58 | |
-!- karlnapf [~heiko@host86-185-15-106.range86-185.btcentralplus.com] has left #shogun [] | 18:06 | |
-!- harshit_ [~harshit@182.68.181.141] has quit [Quit: Leaving] | 18:33 | |
-!- vikram360 [~vikram360@117.192.163.46] has quit [Ping timeout: 276 seconds] | 19:28 | |
-!- vikram360 [~vikram360@117.192.173.199] has joined #shogun | 19:28 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection] | 19:29 | |
-!- wiking [~wiking@vpna216.ugent.be] has joined #shogun | 19:29 | |
-!- wiking [~wiking@vpna216.ugent.be] has quit [Changing host] | 19:29 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 19:29 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 19:39 | |
-!- blackburn [~qdrgsm@109.226.90.219] has joined #shogun | 19:49 | |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 19:49 | |
@sonney2k | gsomix, hey - how is it going? | 19:50 |
gsomix | sonney2k, moin. I added code to ref count in the SGVector. and I did SGVector -> SGVector& for arguments to some functions and methods. | 19:56 |
gsomix | but I have a question. can you look at code preprocessors? | 19:57 |
gsomix | methods `apply_to_feature_vector' | 19:58 |
@sonney2k | yes | 20:01 |
@sonney2k | and? | 20:01 |
gsomix | sonney2k, this method should return a reference to new SGVector on heap with SG_REF, isn't it? | 20:03 |
gsomix | or should leave as is? | 20:03 |
@sonney2k | gsomix, uh you are right | 20:07 |
@sonney2k | I am not sure if this reference (on the heap) is still valid after the function returns | 20:07 |
blackburn | hey there | 20:08 |
gsomix | blackburn, yo | 20:09 |
blackburn | sonney2k: I have to ask you to review feature learning code | 20:09 |
blackburn | would be pretty unfair :) | 20:09 |
blackburn | sonney2k: and btw I merged lars | 20:10 |
wiking | i have only a mean vector and the variance vector of a gaussian... how do i create a CGaussian from that? :) | 20:11 |
-!- Marty28 [~Marty@158.181.78.199] has joined #shogun | 20:11 | |
blackburn | wiking: what is the problem? | 20:11 |
wiking | well the covariance is a matrix... | 20:11 |
wiking | CGaussian(SGVector<float64_t> mean, SGMatrix<float64_t> cov, ECovType cov_type=FULL); | 20:12 |
blackburn | use DIAG or so | 20:12 |
blackburn | or patch it :D | 20:12 |
wiking | you mean that in case of a DIAG matrix where the diag elements are actually the variance that should be fine for creating a gaussian? | 20:12 |
blackburn | wiking: yes some kind of | 20:13 |
blackburn | 3 0 0 | 20:13 |
blackburn | 0 1 0 | 20:13 |
blackburn | 0 0 3 | 20:13 |
blackburn | would work :) | 20:13 |
wiking | and 3,1,3 would be the variance? | 20:13 |
blackburn | yes | 20:13 |
wiking | mmm is there a function to create a diag matrix where with given diagonal elements, i.e. from a vector? | 20:13 |
gsomix | sonney2k, why not? or do you mean that problem is that someone should to free returned SGVector? | 20:14 |
blackburn | wiking: where? | 20:14 |
wiking | or simply create an eye matrix | 20:14 |
wiking | mmm in shogun | 20:14 |
blackburn | what is the inteface? | 20:14 |
@sonney2k | gsomix, because the on the heap allocated object might be freed already... | 20:14 |
wiking | i'm asking whether there is one :) | 20:15 |
blackburn | wiking: there is one in cmath (create_diagonal_matrix) however if you use python it would be easier to use numpy for that | 20:15 |
wiking | so like diag (v) in matlab | 20:15 |
gsomix | sonney2k, so what is strategy to convert method's returns (SGVector)? references only in the right places? | 20:22 |
gsomix | we can not return references to local variables or references to allocated SGVector in methods. should we return copy? | 20:24 |
gsomix | sonney2k, ok. I figured out how it should work. :) | 20:49 |
shogun-buildbot | build #209 of nightly_none is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_none/builds/209 | 21:01 |
blackburn | uh how bad bad bad | 21:04 |
-!- vikram360 [~vikram360@117.192.173.199] has quit [Read error: Connection reset by peer] | 21:06 | |
-!- vikram360 [~vikram360@117.192.173.199] has joined #shogun | 21:06 | |
@sonney2k | gsomix, I actually don't know - so how do you do it? | 21:15 |
CIA-64 | shogun: Sergey Lisitsyn master * r277ba68 / (src/shogun/regression/LARS.cpp src/shogun/regression/LARS.h): Proper constructor for #438 - http://git.io/EH0RWA | 21:16 |
gsomix | sonney2k, we return copy if we have new instance. simple way. | 21:17 |
* gsomix afk | 21:17 | |
blackburn | damn how I missed that :) | 21:18 |
blackburn | sonney2k: are you ok with transfer/domain_adaptation & transfer/multitask structure? | 21:22 |
@sonney2k | gsomix, ok | 21:27 |
@sonney2k | blackburn, too tired now | 21:27 |
@sonney2k | cu | 21:27 |
blackburn | sonney2k: see ya | 21:27 |
-!- harshit_ [~harshit@182.68.181.141] has quit [Quit: Leaving] | 21:31 | |
-!- vikram360 [~vikram360@117.192.173.199] has quit [Read error: Connection reset by peer] | 21:32 | |
wiking | mmm guys | 21:46 |
gsomix | mmm? | 21:46 |
wiking | ever since i've upgraded to the latest shogun | 21:47 |
wiking | i'm having problems with combined kernel | 21:47 |
blackburn | wiking: and what is the problem? | 21:47 |
wiking | [ERROR] Index out of Range: idx_a=0/0 idx_b=0/0 | 21:47 |
-!- Marty28 [~Marty@158.181.78.199] has quit [Quit: ChatZilla 0.9.88.2 [Firefox 11.0/20120310010446]] | 21:48 | |
blackburn | wiking: ok then just fix it as I assume you want to be a shogun developer ;) | 21:48 |
wiking | hahahahah lol | 21:49 |
wiking | i wonder where has been the change | 21:49 |
blackburn | so something changed very recently? | 21:49 |
wiking | yeps | 21:49 |
-!- vikram360 [~vikram360@117.192.173.199] has joined #shogun | 21:49 | |
blackburn | then my suggestion is still here - just try to bisect it ;) | 21:50 |
blackburn | how long has it been outdated? | 21:50 |
wiking | http://snipt.org/uhXi2#expand | 21:50 |
wiking | so this is my code | 21:50 |
wiking | the getKernel function is just giving me the kernel i want | 21:51 |
wiking | ahahahahhahahah | 21:51 |
blackburn | I bet it is empty | 21:51 |
wiking | i have 10 bucks that it's my fault | 21:51 |
wiking | i've commented out the kernel.init ... part | 21:52 |
blackburn | nice | 21:52 |
wiking | hehe now it works of course | 21:52 |
wiking | jesus | 21:52 |
wiking | lets see if i'm getting the same error | 21:52 |
wiking | i mean | 21:53 |
wiking | error rate :) | 21:53 |
wiking | as the homo kernel mapper | 21:53 |
wiking | :> | 21:53 |
wiking | mmm | 21:54 |
wiking | lol | 21:54 |
wiking | even better ;) | 21:54 |
-!- genix [~gsomix@178.45.80.181] has joined #shogun | 21:54 | |
wiking | blackburn: got a sec? | 21:56 |
blackburn | yeap | 21:56 |
wiking | i've just realized that u could actually do a combined kernel like thingy | 21:56 |
wiking | with homo kernel mapping :) | 21:56 |
blackburn | some kind of CHI2 + intersection? | 21:57 |
wiking | just do the mapping for the given features | 21:57 |
wiking | yes | 21:57 |
wiking | so you take the features | 21:57 |
wiking | and then do the homo mapping | 21:57 |
wiking | the desired kernel for the given feature | 21:57 |
-!- gsomix [~gsomix@109.169.156.156] has quit [Ping timeout: 245 seconds] | 21:57 | |
wiking | and then just concatenate those features | 21:57 |
blackburn | okay actually I think this should be done with some features union helper | 21:58 |
wiking | hehe yeah | 21:58 |
wiking | well i have my own function for it at the moment | 21:58 |
blackburn | I'll implement one in a hr | 21:58 |
wiking | but i guess what we would like to have is in shogun | 21:58 |
wiking | but there's something like this no? | 21:58 |
wiking | sonney2k: was talking about | 21:59 |
wiking | the CombinedDotFeatures | 21:59 |
blackburn | hmm actually yes | 21:59 |
blackburn | yes | 21:59 |
blackburn | it makes more sense even | 21:59 |
blackburn | no need to do that | 21:59 |
-!- genix is now known as gsomix | 21:59 | |
wiking | i'll give it a shot now | 21:59 |
wiking | see how it'll react | 22:00 |
-!- karlnapf [~heiko@host86-181-156-211.range86-181.btcentralplus.com] has joined #shogun | 22:11 | |
CIA-64 | shogun: Heiko Strathmann master * r4711307 / src/shogun/classifier/mkl/MKL.cpp : | 22:22 |
CIA-64 | shogun: -added initialisation of mkl_block_norm to std value | 22:22 |
CIA-64 | shogun: -added comment about strange set_sub_kernel_weights | 22:22 |
CIA-64 | shogun: -fixed a mem leak regarding this comment - http://git.io/HhlwNQ | 22:22 |
CIA-64 | shogun: Heiko Strathmann master * r2e21eb9 / src/shogun/kernel/CustomKernel.cpp : added debug messages - http://git.io/rVvBFg | 22:22 |
CIA-64 | shogun: Heiko Strathmann master * rbf5e9ea / src/shogun/machine/KernelMachine.cpp : added debug messages - http://git.io/NlQr-w | 22:22 |
CIA-64 | shogun: Heiko Strathmann master * r24bab87 / src/shogun/classifier/mkl/MKL.cpp : SGVector instead of plain array - http://git.io/4BkXRQ | 22:22 |
CIA-64 | shogun: Heiko Strathmann master * r2f00592 / (3 files in 3 dirs): | 22:22 |
CIA-64 | shogun: Merge pull request #437 from karlnapf/master | 22:22 |
CIA-64 | shogun: memory bugfix, initialization, comments and debug messages - http://git.io/sOp24A | 22:22 |
CIA-64 | shogun: Heiko Strathmann master * rd716ae0 / examples/undocumented/libshogun/modelselection_grid_search_mkl.cpp : | 22:27 |
CIA-64 | shogun: added example for mkl modelselection (currently not working, but should | 22:27 |
CIA-64 | shogun: in the future, once bugs are fixed) - http://git.io/oUJlSA | 22:27 |
CIA-64 | shogun: Heiko Strathmann master * rc12e96c / examples/undocumented/libshogun/modelselection_grid_search_mkl.cpp : | 22:27 |
CIA-64 | shogun: Merge pull request #453 from karlnapf/master | 22:27 |
CIA-64 | shogun: added mkl modelselection example - http://git.io/-Ob1_g | 22:27 |
wiking | karlnapf: the model selection example works? | 22:37 |
karlnapf | wiking, no, not yet | 22:37 |
karlnapf | segfault | 22:37 |
karlnapf | some strange stuff | 22:37 |
blackburn | long way :) | 22:38 |
karlnapf | Luckily I could tell this guy from cambridge how to use an alternative | 22:38 |
karlnapf | (not using SVMLight) | 22:39 |
blackburn | which guy? | 22:39 |
wiking | karlnapf: i'm just trying to experiment with mkl | 22:39 |
karlnapf | he wrote on the mailing list a couple of times about mkl stuff | 22:39 |
wiking | but i'm always getting worse results than simply doing a combinedfeatures/kernel | 22:39 |
karlnapf | wiking, what are you trying to do? | 22:39 |
blackburn | ah | 22:39 |
wiking | + gmnpsvm | 22:40 |
karlnapf | I mean for simple examples, MKL should be a bit worse than standard stuff since when optimising, there are a lot of approximations happening. | 22:40 |
karlnapf | but I am not an MKL expert, soeren is | 22:40 |
blackburn | soeren hates it | 22:40 |
wiking | hahahahah | 22:40 |
karlnapf | really? | 22:40 |
karlnapf | lol | 22:41 |
wiking | yeah soeren dislikes it ;) | 22:41 |
karlnapf | why did he write so many papers then ;) | 22:41 |
wiking | good question D: | 22:41 |
blackburn | just for fun may be :D | 22:41 |
karlnapf | wiking, is the result much worse? | 22:41 |
wiking | karlnapf: well we are talking about 7-8% accuracy | 22:41 |
wiking | so 80 vs 87-88 | 22:42 |
wiking | $ | 22:42 |
wiking | % | 22:42 |
karlnapf | for binary classification? | 22:42 |
karlnapf | oh | 22:42 |
karlnapf | mmh thats a lot | 22:42 |
wiking | multi class sorry | 22:42 |
wiking | yeah i thought it should be actually better | 22:42 |
wiking | but it's not :( | 22:43 |
karlnapf | mmh, dont know, there are so many possible sources for that | 22:43 |
karlnapf | why? | 22:43 |
karlnapf | MKL is usually not better than just trying combinations by hand | 22:43 |
karlnapf | only faster | 22:43 |
wiking | mmm not faster either ;) | 22:43 |
karlnapf | hehe ;) | 22:44 |
karlnapf | Well to be honest, I dont know, havent digged into that stuff. | 22:44 |
blackburn | karlnapf: ah one more issue | 22:44 |
karlnapf | I am currently more interested to get rid of memory bugs in MKL with SVMLight :) | 22:44 |
karlnapf | this is ugly | 22:44 |
blackburn | I'd like to enable cross validation for dim reduction things :) | 22:44 |
karlnapf | people want to use model-selection and it doesnt work | 22:44 |
blackburn | I'll ask you later | 22:45 |
karlnapf | blackburn, yes nice | 22:45 |
karlnapf | I guess you would want to modify/add stuff on the ModelSelection class | 22:45 |
karlnapf | Currently, a CMachine interface is needed | 22:45 |
-!- vikram360 [~vikram360@117.192.173.199] has quit [Read error: Connection reset by peer] | 22:47 | |
blackburn | karlnapf: yeah however that would be really painful change | 22:50 |
karlnapf | blackburn, I think it would be better to extend the modelselection class | 22:50 |
karlnapf | however, you would need some shared interface between all classes that do want to do some parameter selection | 22:51 |
blackburn | karlnapf: oh I feel too much pain thinking about it :D | 22:53 |
wiking | what's with svm/ScatterSVM.h ? | 22:53 |
blackburn | wiking: LOLZ | 22:53 |
karlnapf | lol ;) | 22:53 |
wiking | what's that thing? :))) | 22:53 |
wiking | with being experimental :D | 22:53 |
blackburn | you're really going to try EACH feature of shogun | 22:53 |
wiking | i'm just checking the svm machines | 22:54 |
wiking | and i've just stumbled upon that stuff :)) | 22:54 |
blackburn | wiking: okay actually it is experimental and definitely not better than larank/gmnp | 22:54 |
wiking | ahhaha | 22:54 |
blackburn | wiking: I really believe you should address your issues not with trying different machines ;) | 22:55 |
wiking | :D | 22:56 |
wiking | wtf? | 22:56 |
blackburn | wiking: I mean your accuracy problem could be solved by analysis of results | 22:56 |
wiking | once again | 22:56 |
wiking | wtf? | 22:56 |
karlnapf | blackburn, I am thinking of moving the subset logic (stack etc) from Features/Labels to Subset class. However, that would break all objects that have a subset from older shogun version (migration can do nothing here) | 22:57 |
blackburn | karlnapf: do you understand my words? :) | 22:57 |
karlnapf | what do you think? | 22:57 |
blackburn | karlnapf: why? | 22:57 |
karlnapf | blackburn, because then the logic would be in one class | 22:57 |
karlnapf | instead of having it multiple times in different classes | 22:58 |
karlnapf | Before these stacks, everythig was easy | 22:58 |
-!- gsomix [~gsomix@178.45.80.181] has quit [Ping timeout: 265 seconds] | 22:58 | |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 22:58 | |
karlnapf | but now code gets more complicated and you dont want to have a similar copy of the recomputation of the subset stack in every class which uses it | 22:58 |
blackburn | I can hardly imagine that logic in stacks though | 22:58 |
karlnapf | its a pain anyway because normally this should be done via multiple inheritance, but we cannot use this, so each class has to implement wrapper classes anyway | 22:59 |
-!- vikram360 [~vikram360@117.192.173.199] has joined #shogun | 22:59 | |
karlnapf | just the methods add/remove | 22:59 |
blackburn | I think Soeren would not like breaking migrations | 22:59 |
karlnapf | the current active subset has to be recomputed in it | 22:59 |
karlnapf | mmh | 23:00 |
karlnapf | this stuff is all not migratable anyway | 23:00 |
karlnapf | i dont know how to translate the current changes ..... | 23:00 |
blackburn | actually my opinion is that users can do migrations manually :) | 23:00 |
karlnapf | in this case, i agree | 23:01 |
karlnapf | in other cases, we should help them | 23:01 |
karlnapf | like name changes /type changes/new variables etc | 23:01 |
karlnapf | sonney2k, are you around? whats your opinion? | 23:03 |
blackburn | he is sleeping I guess | 23:03 |
blackburn | or I don't know | 23:03 |
blackburn | actually he tends to appear randomly | 23:03 |
karlnapf | same here :) | 23:04 |
karlnapf | (not the sleeping thing) | 23:04 |
blackburn | karlnapf: what is your opinion on moving lib* -> shogun/lib | 23:05 |
blackburn | liblinear | 23:05 |
blackburn | libsvm | 23:05 |
blackburn | libocas | 23:05 |
karlnapf | from machine to lib | 23:06 |
karlnapf | mmm | 23:06 |
karlnapf | dont know | 23:06 |
karlnapf | why? | 23:06 |
blackburn | from classifier | 23:06 |
karlnapf | why not? | 23:06 |
karlnapf | because its an external library? | 23:06 |
blackburn | yes | 23:06 |
karlnapf | I was a bit confused about that fact actually when I tried using shogun for the first time | 23:06 |
karlnapf | on the other hand | 23:06 |
karlnapf | all classifier are at one place now | 23:07 |
karlnapf | you would loose that | 23:07 |
blackburn | karlnapf: actually not | 23:07 |
blackburn | karlnapf: multiclass is already forked | 23:07 |
blackburn | from classifier | 23:07 |
blackburn | and both are using libocas and liblinear | 23:07 |
blackburn | ok I'll do pull request and ask you and soeren for comments | 23:08 |
blackburn | one more thing I don't like is liblinear is called SVM_linear | 23:08 |
karlnapf | agreed | 23:10 |
karlnapf | (both things) | 23:10 |
karlnapf | I will rewrite the subset stuff now | 23:10 |
karlnapf | will be much easier | 23:10 |
blackburn | ok | 23:10 |
karlnapf | and I just thought of a way for the migrations | 23:10 |
blackburn | sunday surprise for soeren | 23:10 |
blackburn | karlnapf: do you like shogun/lib/external | 23:11 |
blackburn | for that? | 23:11 |
karlnapf | yes, that would actually be nice | 23:11 |
karlnapf | blackburn, do we got a list type for SGVectors? | 23:19 |
blackburn | list of sgvectors? | 23:19 |
blackburn | no idea | 23:19 |
karlnapf | mmh I will just use DynArray then | 23:19 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 23:21 | |
-!- vikram360 [~vikram360@117.192.173.199] has quit [Ping timeout: 265 seconds] | 23:27 | |
-!- vikram360 [~vikram360@117.192.163.162] has joined #shogun | 23:27 | |
--- Log closed Sun Apr 15 00:00:19 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!