--- Log opened Fri Apr 13 00:00:19 2012 | ||
-!- av3ngr [av3ngr@nat/redhat/x-cvkzvkolelblsktc] has joined #shogun | 01:47 | |
-!- blackburn [~qdrgsm@83.234.54.186] has quit [Ping timeout: 252 seconds] | 02:01 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 02:37 | |
-!- vikram360 [~vikram360@117.192.160.228] has quit [Ping timeout: 260 seconds] | 03:01 | |
-!- vikram360 [~vikram360@117.192.180.120] has joined #shogun | 04:45 | |
-!- vikram360 [~vikram360@117.192.180.120] has quit [Ping timeout: 252 seconds] | 04:53 | |
-!- vikram360 [~vikram360@117.192.180.208] has joined #shogun | 04:53 | |
-!- siddharth_ [~siddharth@14.139.82.6] has joined #shogun | 05:18 | |
-!- siddharth_ [~siddharth@14.139.82.6] has left #shogun [] | 05:18 | |
-!- PhilTillet [~Philippe@npasserelle10.minet.net] has quit [Ping timeout: 272 seconds] | 07:35 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 07:47 | |
-!- av3ngr [av3ngr@nat/redhat/x-cvkzvkolelblsktc] has quit [Ping timeout: 240 seconds] | 08:25 | |
-!- av3ngr [av3ngr@nat/redhat/x-xtwvnksjinqwtsqn] has joined #shogun | 08:36 | |
sonne|work | hi n4nd0 - a few warnings are on your account: http://www.shogun-toolbox.org/buildbot/builders/python_modular/builds/472/steps/compile/logs/warnings%20%2824%29 | 08:38 |
---|---|---|
n4nd0 | sonne|work: ok, I will take of them, thank you! | 08:42 |
-!- av3ngr [av3ngr@nat/redhat/x-xtwvnksjinqwtsqn] has quit [Quit: That's all folks!] | 08:43 | |
sonne|work | n4nd0: btw are you still happily covertree-ing? | 08:44 |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 08:49 | |
n4nd0 | sonne|work: yeah | 08:56 |
n4nd0 | sonne|work: right now I am stuck with some university work, but I during my shogun time I am covertree-ing :) | 08:57 |
n4nd0 | sonne|work: why? | 08:57 |
sonne|work | n4nd0: I got used to hearing break through's from you side :) | 08:57 |
n4nd0 | sonne|work: :D I am trying to control myself more now | 08:58 |
sonne|work | too bad | 08:58 |
sonne|work | :) | 08:58 |
n4nd0 | I discover yesterday though an interesting thing about JL covertree construction in my integration within shogun | 08:59 |
n4nd0 | sonne|work: I can show you now | 08:59 |
n4nd0 | sonne|work: the context is in CKNN:apply | 09:02 |
n4nd0 | sonne|work: here distance is set so distance(train_features, query) | 09:02 |
n4nd0 | sonne|work: but I need to build two cover trees before being able to do the queries | 09:03 |
n4nd0 | sonne|work: one with the train_features, and another with the queries | 09:03 |
n4nd0 | sonne|work: to build these trees I need distances, between those features | 09:03 |
n4nd0 | sonne|work: i.e to build the cover tree with train_features I need that distance is distance(train_features, train_features) | 09:03 |
n4nd0 | sonne|work: and to build the one for the queties: distance(queries, queries) | 09:04 |
n4nd0 | sonne|work: I have managed to do that using this trick http://snipt.org/uhRh3#expand | 09:04 |
n4nd0 | sonne|work: I think it is a bit ugly, but it does work! | 09:04 |
sonne|work | yeah for that one would need what you have said | 09:05 |
n4nd0 | sonne|work: the problem is that it turns out that the calls to replace_[rhs|lhs] take more time than the creation of the cover trees | 09:05 |
sonne|work | a way to compute distance for rhs,rhs stuff | 09:05 |
sonne|work | what? | 09:06 |
sonne|work | which distance? | 09:06 |
n4nd0 | sonne|work: what do you mean? | 09:06 |
sonne|work | for which distance does it take more tiem? | 09:06 |
sonne|work | time | 09:06 |
n4nd0 | I don't think it is really the distance | 09:07 |
n4nd0 | but the calls to replace_lhs and replace_rhs | 09:07 |
n4nd0 | in JL's code, both calls (for set_of_points and set_of_queries in the code) to batch_create | 09:07 |
n4nd0 | take less time than in my shogun code, using the replace methods in between | 09:08 |
n4nd0 | within my shogun code the portion of code I have pasted for you takes 0.2482(s) | 09:09 |
n4nd0 | while for JL's, the two calls to batch create take 0.053942(s) | 09:10 |
n4nd0 | oh wait | 09:10 |
n4nd0 | this may be caused as well for the upper_bound distance thing I told you about! | 09:11 |
n4nd0 | sonne|work: haha I think I did my first breakthrough of the day :D | 09:11 |
sonne|work | I don't understand a thing | 09:12 |
sonne|work | you mean upper_bound stuff is the reason why JL's thing is so fast | 09:12 |
sonne|work | but you don't use it? | 09:13 |
sonne|work | in shogun | 09:13 |
n4nd0 | I have not implemented it yet | 09:13 |
sonne|work | and that is why it is so slow? | 09:13 |
n4nd0 | I think that the reason why I see that the construction of the trees is slow in shogun | 09:13 |
n4nd0 | maybe is because of the thing of the upper_bound distance | 09:13 |
n4nd0 | do you understand? | 09:14 |
n4nd0 | I am not that good at explaining sometimes :S | 09:14 |
sonne|work | yes OK | 09:15 |
sonne|work | but it is already faster right? | 09:15 |
sonne|work | 4 times I mean? | 09:15 |
sonne|work | so you can speed it up quite a bit with this upper_bound stuff | 09:16 |
n4nd0 | exactly | 09:16 |
n4nd0 | right now it is already faster than the thing we had before | 09:16 |
n4nd0 | not as fast as standalone JL though | 09:16 |
n4nd0 | it might approach more standalanoe JL doing the upper_bound thing | 09:17 |
sonne|work | nice! | 09:18 |
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun | 09:21 | |
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] | 09:21 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 09:21 | |
sonne|work | n4nd0: now I wish you had time to try this out | 09:22 |
n4nd0 | sonne|work: the cover tree? | 09:22 |
sonne|work | yes | 09:26 |
sonne|work | the upper bound stuff w/i shogun | 09:26 |
n4nd0 | ah I will get some time to work on it today, I'm almost done with the test I wanted to do for my school project | 09:28 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Quit: wiking] | 09:39 | |
sonne|work | great | 09:42 |
sonne|work | let me know :) | 09:42 |
-!- blackburn [~qdrgsm@83.234.54.186] has joined #shogun | 09:43 | |
blackburn | n4nd0: hey there I heard some success with covertree? | 09:47 |
n4nd0 | blackburn: yeah, I managed to make it work from shogun | 09:47 |
n4nd0 | blackburn: is still work in progress though | 09:48 |
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has joined #shogun | 09:49 | |
-!- wiking [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] | 09:49 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 09:49 | |
-!- blackburn [~qdrgsm@83.234.54.186] has quit [Read error: Connection reset by peer] | 09:55 | |
-!- pluskid [~chatzilla@111.120.58.220] has joined #shogun | 09:56 | |
-!- blackburn [~qdrgsm@83.234.54.186] has joined #shogun | 09:56 | |
wiking | yo | 10:00 |
sonne|work | wiking: yo | 10:01 |
wiking | friday the 13th had it's course already on me :) | 10:01 |
sonne|work | wiking - yeah you missed out 2 lines with warning fixes http://www.shogun-toolbox.org/buildbot/builders/python_modular/builds/472/steps/compile/logs/warnings%20%2824%29 | 10:02 |
sonne|work | or was it sth else ?\ | 10:02 |
wiking | hahaha sure something else... internet got broken, warm water in the house suddenly stopped working | 10:02 |
wiking | 2? i just see one preprocessor/PNorm.h:37: | 10:04 |
wiking | pull request for fixing that error sent out | 10:07 |
CIA-64 | shogun: Viktor Gal master * r92348cf / src/shogun/preprocessor/PNorm.h : Fix for PNorm constructor doxygen comment - http://git.io/rp21jA | 10:09 |
CIA-64 | shogun: Soeren Sonnenburg master * r6c0a969 / src/shogun/preprocessor/PNorm.h : | 10:09 |
CIA-64 | shogun: Merge pull request #447 from vigsterkr/master | 10:09 |
CIA-64 | shogun: yet another fix for PNorm doxygen comment - http://git.io/BYlOsA | 10:09 |
sonne|work | wiking: this one preprocessor/PNorm.cpp:120: warning: 'norm' may be used uninitialized in this function | 10:09 |
-!- pluskid_ [~chatzilla@li400-235.members.linode.com] has joined #shogun | 10:09 | |
sonne|work | actually more critical | 10:09 |
wiking | mmm | 10:10 |
wiking | hahahahah THE FIST line and i don't see it :) | 10:10 |
-!- pluskid [~chatzilla@111.120.58.220] has quit [Ping timeout: 260 seconds] | 10:11 | |
-!- pluskid_ is now known as pluskid | 10:11 | |
wiking | s/FIST/FIRST/ | 10:11 |
wiking | actually that's a good question what it should be... afaik the best is to return 1.0 rather than 0.0 | 10:11 |
wiking | although that's not 'correct' but it won't hurt as much as if i would return 0.0 and then the caller right away uses the return value without checking to divide a number => divide by zero | 10:12 |
wiking | agree? | 10:12 |
sonne|work | wiking: huh? | 10:19 |
sonne|work | you should init it with 0 | 10:19 |
* sonne|work thinks wiking didn't have enough sleep | 10:19 | |
wiking | and if that gets returned...? | 10:19 |
sonne|work | it doesn't | 10:19 |
wiking | it might... | 10:19 |
sonne|work | no | 10:20 |
sonne|work | in all if () { } sections norm is computed | 10:20 |
sonne|work | only in the first one it is incremented | 10:20 |
sonne|work | that is why you need the init to 0 | 10:20 |
wiking | well it still can happen that 0 gets back | 10:21 |
wiking | but that's the caller's problem then... | 10:21 |
sonne|work | and an independent problem | 10:21 |
wiking | ok | 10:21 |
sonne|work | no matter what you set norm to | 10:21 |
blackburn | THE FIST | 10:22 |
wiking | :>>> | 10:22 |
wiking | pull req sent | 10:23 |
blackburn | hey guys have you heard how nice is DPRK's rocket? | 10:23 |
wiking | ? | 10:23 |
pluskid | rocket ? | 10:24 |
blackburn | http://news.yahoo.com/north-korean-missile-failed-launch-u-military-012804451.html totally broken lol | 10:24 |
blackburn | wiking: should I merge your stuff? :) | 10:24 |
wiking | blackburn: first one is already merged :) | 10:25 |
wiking | but yeah would be great | 10:25 |
wiking | ok | 10:26 |
wiking | now it's time to go back and try to speed up gmm | 10:26 |
wiking | :D | 10:27 |
-!- vikram360 [~vikram360@117.192.180.208] has quit [Ping timeout: 246 seconds] | 10:27 | |
CIA-64 | shogun: Viktor Gal master * r7f6d2d2 / src/shogun/preprocessor/PNorm.cpp : PNorm: initialise the norm to 0.0 in the beginning - http://git.io/71I6pg | 10:27 |
CIA-64 | shogun: Sergey Lisitsyn master * r1f77383 / src/shogun/preprocessor/PNorm.cpp : Merge branch 'master' of git://github.com/vigsterkr/shogun - http://git.io/VqTbVw | 10:27 |
wiking | yey | 10:28 |
blackburn | wiking: I glanced over your twitter and I admit I do not understand anything | 10:28 |
wiking | some of it is on hungarian :) | 10:28 |
blackburn | yes I meant that ;) | 10:28 |
blackburn | what is csinalok? | 10:29 |
wiking | do | 10:29 |
wiking | actually i'm doing | 10:29 |
wiking | csinalni = to do | 10:29 |
wiking | csinalok = i'm doing | 10:29 |
wiking | or i do | 10:29 |
n4nd0 | use a lot twitter guys, I am making a project to rank users and stuff! | 10:29 |
wiking | depends on the context | 10:29 |
blackburn | n4nd0: same with your twitter btw | 10:29 |
blackburn | wiking: there is no difference between I'm doing/to do in russian :D | 10:30 |
n4nd0 | blackburn: like in Swedish, no continuous forms! | 10:30 |
blackburn | however you probably know that | 10:30 |
blackburn | that is one of the reasons why we fail with speaking english properly sometimes | 10:30 |
n4nd0 | blackburn: do you have articles in Russian? | 10:32 |
blackburn | bingo | 10:32 |
n4nd0 | blackburn: like "the house" | 10:32 |
blackburn | no | 10:32 |
n4nd0 | the thing I have noticed the most in Russian speakers is that | 10:32 |
blackburn | that's why I (and other russians) mix it up as well hah | 10:32 |
blackburn | however I get used with it a little | 10:33 |
blackburn | earlier it was worse | 10:33 |
n4nd0 | I don't think you do that one, at least I didn't notice | 10:33 |
blackburn | no, I still do a lot of mistakes and crazy sentences | 10:34 |
blackburn | but year ago it was really worse hah, right sonne|work? | 10:34 |
blackburn | n4nd0: we also have cases (dative,genitive,etc) there | 10:36 |
n4nd0 | blackburn: that must be difficult | 10:36 |
blackburn | not while you are native with it :D | 10:36 |
blackburn | for example | 10:36 |
blackburn | there is a fish | 10:36 |
blackburn | and | 10:36 |
blackburn | there is no fish | 10:37 |
blackburn | would be | 10:37 |
blackburn | ah bad example | 10:37 |
blackburn | word with same cases :D | 10:37 |
blackburn | I mean word is changing in different cases | 10:37 |
n4nd0 | like | 10:37 |
n4nd0 | the fish is red | 10:37 |
n4nd0 | I like fish | 10:37 |
blackburn | ???? ??????? | 10:37 |
blackburn | ? ????? ???? | 10:37 |
blackburn | yes | 10:37 |
n4nd0 | when it is subject or DC | 10:38 |
n4nd0 | I actually know to say fish in Russian | 10:38 |
blackburn | ryyba? ;) | 10:38 |
n4nd0 | yeah | 10:38 |
n4nd0 | haha | 10:38 |
blackburn | n4nd0: aren't cases there in spanish? | 10:39 |
wiking | blackburn: ;) just saying ... usually indo-european language based people cannot grasp the idea of having only 3 tenses: past, present and future ;p | 10:39 |
n4nd0 | blackburn: no, no cases | 10:39 |
-!- uricamic [9320543b@gateway/web/freenode/ip.147.32.84.59] has joined #shogun | 10:41 | |
blackburn | wiking: isn't english indo-european? :) | 10:41 |
blackburn | ah only | 10:42 |
wiking | yes it is and it has about 8 tenses :) | 10:42 |
blackburn | yes got you | 10:42 |
blackburn | uricamic: can it be that you do not precompute kernel matrix in your gs krr? | 10:44 |
blackburn | in case it is too large | 10:44 |
blackburn | okay have to go guys | 10:44 |
blackburn | wiking: how much tenses there are in hungarian then? | 10:45 |
wiking | as said 3 | 10:45 |
wiking | :) | 10:45 |
uricamic | blackburn: yes, I can change that | 10:45 |
n4nd0 | blackburn, wiking the tenses in Spanish are not normal ... too many | 10:46 |
wiking | n4nd0: it's latin based language it's normal | 10:46 |
wiking | ;) | 10:46 |
blackburn | ok see you | 10:47 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection] | 10:47 | |
-!- blackburn [~qdrgsm@83.234.54.186] has quit [Quit: Leaving.] | 10:47 | |
-!- wiking [~wiking@vpnc224.ugent.be] has joined #shogun | 10:47 | |
-!- wiking [~wiking@vpnc224.ugent.be] has quit [Changing host] | 10:47 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 10:47 | |
n4nd0 | too many anyway :) | 10:47 |
pluskid | only 1 tense in Chinese ! :D | 10:47 |
pluskid | I guess | 10:47 |
n4nd0 | :O | 10:47 |
n4nd0 | you don't have difference between past / present / future at least? | 10:47 |
pluskid | we do not change form of a word | 10:48 |
pluskid | but add more words to indicate the tense | 10:48 |
pluskid | so there's no difference between "eat" and "eating" | 10:48 |
pluskid | but the meaning depends on the context | 10:49 |
n4nd0 | not even eat and ate? | 10:49 |
pluskid | there will be words in the context to indicate whether "eat" or "eating" | 10:49 |
pluskid | but there are multiple tenses in Japanese | 10:50 |
pluskid | though Chinese and Japanese looks very similar, they are quite different actually | 10:50 |
pluskid | btw, shogun (the japanese word) is "jiang jun" in Chinese :p | 10:51 |
pluskid | the same characters, different pronounciations | 10:51 |
pluskid | and the same meaning | 10:52 |
n4nd0 | curious | 10:52 |
sonne|work | uricamic: can you do any kind of benchmark for KRR? | 10:52 |
sonne|work | uricamic: and one more thing - could you create a *separate* pull request for the GS-KRR? | 10:53 |
uricamic | sonne|work: I plan to implement also the version which will not precompute the kernel matrix for large data. What kind of benchmark do you mean? | 10:53 |
sonne|work | uricamic: just measure if GS is faster than the pinv based one | 10:54 |
sonne|work | btw I would do it slightly differently | 10:54 |
uricamic | sonne|work: I have tried to create another pull request, but it didn't work | 10:54 |
sonne|work | gsomix, n4nd0, wiking, pluskid - please help uricamic with this :) | 10:55 |
pluskid | uricamic: are you in the mailing list? | 10:55 |
uricamic | is it possible that I can have only one opened pull request? | 10:55 |
uricamic | pluskid: yes | 10:55 |
pluskid | there was an email describing the work flow | 10:55 |
sonne|work | uricamic: I would just add a new train function to KRR | 10:55 |
shogun-buildbot | build #651 of octave_static is complete: Failure [failed test_1] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/651 blamelist: sonne@debian.org | 10:56 |
sonne|work | and the epsilon parameter | 10:56 |
wiking | !!!! sonne|work :> | 10:56 |
sonne|work | not create a new class for that - I mean the rest is the same :) | 10:56 |
pluskid | I think you have to create multiple branches to make multiple pull request | 10:56 |
sonne|work | wiking: I promise I didn't do anything! | 10:56 |
wiking | :)) | 10:56 |
wiking | why did it fail now? :) | 10:56 |
sonne|work | something wrong with perceptron | 10:57 |
uricamic | sonne|work: ok, that sounds reasonable, I will change it | 10:57 |
sonne|work | a rhyme :) | 10:57 |
sonne|work | uricamic: yeah then one can have a 'master' train function that selects some selected solver that one can specify as arg to constructor | 10:57 |
sonne|work | wiking: I don't even see any error?! | 10:58 |
uricamic | pluskid: what was the subject of that email ? | 10:58 |
n4nd0 | uricamic: yeah I agree with pluskid, you need multiple branches for multiple pull requests | 10:59 |
pluskid | uricamic: git workflow | 10:59 |
pluskid | uricamic: this link: http://article.gmane.org/gmane.comp.ai.machine-learning.shogun/2537 | 11:00 |
uricamic | pluskid: thanks | 11:00 |
pluskid | u r wel~ ^_^ | 11:00 |
* sonne|work is happy to see some interaction here :) | 11:02 | |
-!- vikram360 [~vikram360@117.192.180.208] has joined #shogun | 11:22 | |
n4nd0 | sonne|work: I got news | 11:28 |
* sonne|work listens up | 11:31 | |
sonne|work | n4nd0: what news? | 11:31 |
-!- gsomix [~gsomix@188.168.4.7] has quit [Ping timeout: 264 seconds] | 11:31 | |
n4nd0 | sonne|work: so doing this upper bounded computation | 11:32 |
n4nd0 | the time improves, but just a little bit | 11:32 |
n4nd0 | before the construction time in shogun was like 5 times the one in standalone JL | 11:33 |
* pluskid is trying some IRC command | 11:33 | |
n4nd0 | now it is sth between 4 and 5 times | 11:33 |
* n4nd0 likes this IRC command :) | 11:33 | |
* pluskid thinks it cool, too | 11:34 | |
n4nd0 | the thing is that the distance function used in standalone JL is pretty optimized | 11:35 |
n4nd0 | take a look | 11:35 |
n4nd0 | http://snipt.org/uhSe9 | 11:35 |
sonne|work | ahh ok | 11:35 |
n4nd0 | the loop is unrolled I think | 11:36 |
n4nd0 | the big issue is that it assumes that the dimension of the points is multiple of 8 | 11:36 |
sonne|work | and it is based on floats | 11:37 |
-!- pluskid [~chatzilla@li400-235.members.linode.com] has quit [Quit: ChatZilla 0.9.88.2 [Firefox 11.0/20120312181643]] | 11:37 | |
n4nd0 | does that make a big difference too? | 11:37 |
sonne|work | factor 2 at least | 11:37 |
n4nd0 | ok | 11:37 |
n4nd0 | so what do you think? | 11:37 |
sonne|work | we don't want to switch to float | 11:44 |
sonne|work | but all other speedups are welcome :) | 11:44 |
n4nd0 | so since that implementation assumes that the dimension of the features is multiple of 8 | 11:46 |
n4nd0 | (if the points given are not, there is a function that pads them with zeroes) | 11:47 |
n4nd0 | how would you handle that in shogun? | 11:47 |
xiangwang | hi, does shogun include the cutting-plane algorithms? | 11:47 |
n4nd0 | xiangwang: hey! we have OCAS :) | 11:47 |
xiangwang | in which folder of classifier ? thanks | 11:48 |
n4nd0 | classifier/svm I guess | 11:48 |
n4nd0 | yeah I see it in classifier/svm/SVMOcas | 11:49 |
n4nd0 | sonne|work: do you think it pays off to reallocate CFeatures* lhs and rhs to make them multiple of 8? | 11:50 |
n4nd0 | could it be done in another way? | 11:50 |
sonne|work | n4nd0: well one could use this algorithm for the multiple of 8 dims that fit and for the remainder the slow one | 11:58 |
n4nd0 | sonne|work: I like that solution | 11:59 |
n4nd0 | sonne|work: we document it, that is recommended to use covertree with multiples of 8 dimensions :) | 11:59 |
sonne|work | no | 12:00 |
sonne|work | I mean sth else | 12:00 |
sonne|work | one can do fast processing as long as there are blocks of 8 dims to process | 12:00 |
sonne|work | only for the remainder we use the slow method | 12:00 |
sonne|work | that is fast too so no problem | 12:00 |
n4nd0 | aham | 12:01 |
n4nd0 | ok | 12:01 |
-!- elango_ [~elango@92-244-6-194.customers.ownit.se] has joined #shogun | 12:01 | |
n4nd0 | also, this can only be done for SimpleFeatures, can't be? | 12:01 |
sonne|work | yes | 12:03 |
sonne|work | so I think this should be a function in dot/simplefeatures then | 12:03 |
sonne|work | but hey covertree first then this speedup stuff :) | 12:03 |
n4nd0 | all right | 12:04 |
n4nd0 | avout the warnings before | 12:06 |
n4nd0 | ups aBout | 12:06 |
sonne|work | yes? | 12:06 |
n4nd0 | they're related with the other covertree so when I do a pull request with the new one | 12:06 |
n4nd0 | that code disappears | 12:07 |
sonne|work | ok | 12:07 |
n4nd0 | I mean, no need to fix them before right? | 12:07 |
sonne|work | looking forward to it then | 12:07 |
shogun-buildbot | build #652 of octave_static is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/octave_static/builds/652 | 12:08 |
-!- vikram360 [~vikram360@117.192.180.208] has quit [Ping timeout: 276 seconds] | 12:15 | |
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun | 12:17 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Read error: Operation timed out] | 12:19 | |
-!- wiking_ is now known as wiking | 12:19 | |
-!- vikram360 [~vikram360@117.192.180.208] has joined #shogun | 12:28 | |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has joined #shogun | 12:43 | |
blackburn | xiangwang: MulticlassOCAS and SVMOcas are cutting-plane based (libocas) | 12:44 |
-!- elango_ [~elango@92-244-6-194.customers.ownit.se] has left #shogun ["Leaving"] | 12:47 | |
blackburn | sonne|work: hey now I understand you a little more with 10 hrs of sleeping for last 3 days :D | 12:48 |
blackburn | n4nd0: my suggestion is to add a flag/option for upper_bounded thing | 12:53 |
n4nd0 | blackburn: if SimpleFeatures are provided then do the special one? | 12:55 |
blackburn | n4nd0: why features? | 12:55 |
n4nd0 | blackburn: I think this thing can only be done for SimpleFeatures | 12:57 |
n4nd0 | http://snipt.org/uhSe9 | 12:57 |
blackburn | n4nd0: I'd it depends on distance rather than features | 12:57 |
n4nd0 | but don't you think that the function I have posted can just be done if we have SimpleFeatures? | 12:58 |
n4nd0 | we can upper bound for all ok (as long as Euclidian is used this far) | 12:58 |
n4nd0 | but to do it with that unrolled loop and the multiples of 8 | 12:59 |
blackburn | I mean some distances could provide that | 13:00 |
blackburn | and could not | 13:00 |
blackburn | sonne|work: can all these fancy string kernels/distances be upper bounded? | 13:00 |
blackburn | n4nd0: however you said it is not as fast as we expected with this upper bound? | 13:02 |
n4nd0 | sorry I was afk | 13:08 |
n4nd0 | blackburn: it is expected because it is not just the upper bound what makes JL faster but also the optimizations | 13:08 |
blackburn | n4nd0: loop unrollment? | 13:08 |
n4nd0 | blackburn: probably | 13:09 |
n4nd0 | sonne|work pointed out that the floats also, we use double | 13:09 |
blackburn | n4nd0: loop unrollment is dangerous thing to implement explicitly I think | 13:09 |
blackburn | we can't be sure it would be faster everywhere | 13:09 |
blackburn | and actually blas is unrolled so we should stick to blas I think | 13:10 |
n4nd0 | but there we cannot cut the computation if the upper bound is passed right? | 13:10 |
blackburn | we can compute batches with blas | 13:11 |
blackburn | n4nd0: DOUBLE PRECISION FUNCTION DNRM2(N,X,INCX) | 13:11 |
blackburn | could you please try that? | 13:11 |
blackburn | or SNRM2 | 13:11 |
blackburn | n4nd0: it is pretty interesting how did they implement that http://www.netlib.org/blas/dnrm2.f | 13:12 |
n4nd0 | blackburn: why don't you like just to implement that piece of code? | 13:13 |
n4nd0 | I mean the distance used in JL | 13:13 |
blackburn | n4nd0: implement where? | 13:14 |
n4nd0 | to optimize the distance computation for cover tree | 13:14 |
blackburn | check if it is faster than dnrm2 ;) | 13:14 |
n4nd0 | how can dnrm2 be faster? | 13:15 |
n4nd0 | don't you think this one is already very specific? | 13:15 |
blackburn | in which means? | 13:17 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection] | 13:18 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 13:18 | |
n4nd0 | it looks like they have optimized the distance computation so it goes very fast here | 13:18 |
blackburn | I am still think we should stick to blas ;) | 13:18 |
blackburn | it is optimized as well | 13:19 |
n4nd0 | ok | 13:19 |
blackburn | n4nd0: for example dnrm2 is more stable | 13:19 |
n4nd0 | once I start to do this part of the optimization we can test it | 13:19 |
blackburn | yes just test | 13:19 |
-!- xiangwang_ [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has joined #shogun | 13:23 | |
sonne|work | n4nd0: I think blas would be even faster | 13:26 |
n4nd0 | we should try then | 13:26 |
-!- PhilTillet [~Philippe@157.159.42.154] has joined #shogun | 13:29 | |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Ping timeout: 276 seconds] | 13:34 | |
n4nd0 | see you later guys | 13:37 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 13:37 | |
blackburn | sonne|work: don't you think we should integrate cblas there? | 13:38 |
blackburn | it is a common way I'd say | 13:39 |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 13:40 | |
sonne|work | blackburn: yes that is what I said | 13:40 |
blackburn | no | 13:40 |
blackburn | I mean to codebase | 13:40 |
blackburn | sonne|work: many libs are integrating blas into their codebase | 13:40 |
blackburn | however I see one great pitfall there - it would make packing .deb harder, right? | 13:41 |
sonne|work | no | 13:43 |
blackburn | bad way? | 13:43 |
sonne|work | I don't want atlas/blas/gotoblas or anything in our codebase | 13:43 |
blackburn | sonne|work: I am actually worried with arpack packaging - that's why I was thinking about it | 13:44 |
blackburn | however better way would be to help with arpack package maintanence | 13:45 |
sonne|work | yes | 13:45 |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Ping timeout: 265 seconds] | 13:46 | |
blackburn | sonne|work: how does it work? do maintaners have some public repo? | 13:49 |
sonne|work | there is a pts page | 13:50 |
sonne|work | for debian for each package | 13:50 |
sonne|work | but you can also do apt-get source <nameofpkg> | 13:50 |
sonne|work | then modify debian/ dir | 13:51 |
blackburn | sonne|work: what if I want to fix something? | 13:52 |
-!- PhilTillet [~Philippe@157.159.42.154] has quit [Ping timeout: 245 seconds] | 13:52 | |
blackburn | one should send patches to maintainers? | 13:52 |
sonne|work | create a patch ans submit a bugreport via reportbug <packagename> | 13:52 |
blackburn | got it | 13:52 |
blackburn | thanks | 13:52 |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 13:53 | |
blackburn | sonne|work: btw did you get used to gnome3? | 13:54 |
sonne|work | yes | 13:54 |
blackburn | I am becoming angrier with unity | 13:54 |
blackburn | day-by-day | 13:54 |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 13:56 | |
harshit_ | blackburn: hi ! | 13:57 |
blackburn | hi | 13:57 |
harshit_ | TSVM is almost complete, but there is some problem, should I create a pull request for TSVM ? | 13:58 |
blackburn | yes | 13:58 |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Ping timeout: 246 seconds] | 13:58 | |
harshit_ | also do you happen to have old svmlin library's example with you, their link is broken in web | 13:59 |
harshit_ | I sent a mail to vikas(author ) but no reply till now | 14:00 |
blackburn | no, I haven't | 14:00 |
harshit_ | I think soeren might have it, coz he is author of svmlin in shogun ? Please ask him when he comes online, I may not be able to catch him. | 14:02 |
blackburn | sonne|work: ^ | 14:02 |
harshit_ | oh he is online !! | 14:02 |
blackburn | partially | 14:02 |
blackburn | but I am afraid he just merged it | 14:03 |
harshit_ | what did he merge ?? | 14:04 |
blackburn | svmlin to shogun's codebase | 14:05 |
harshit_ | no no , I am not asking for its code, that is available. All I am asking for is its 'standard' examples,which are available in a separate archive which in not downloadable | 14:06 |
harshit_ | Want to test shogun's TSVM and svmlin TSM on that dataset before proceeding | 14:07 |
blackburn | ah no idea | 14:09 |
harshit_ | no problem, btw how to issue *(some message) on IRC ? | 14:10 |
* blackburn this one? | 14:10 | |
blackburn | / me | 14:10 |
harshit_ | yeah | 14:10 |
blackburn | without spaces | 14:10 |
harshit_ | I cant see any thing | 14:11 |
harshit_ | coming | 14:11 |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 14:11 | |
harshit_ | / checking some new command | 14:12 |
blackburn | "/me something" | 14:12 |
* harshit_ hello world | 14:12 | |
harshit_ | got it | 14:13 |
harshit_ | lol | 14:13 |
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun | 14:15 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 246 seconds] | 14:18 | |
-!- wiking_ is now known as wiking | 14:18 | |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Ping timeout: 260 seconds] | 14:24 | |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 14:32 | |
-!- pluskid [~pluskid@111.120.58.220] has joined #shogun | 14:34 | |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Ping timeout: 265 seconds] | 14:38 | |
-!- n4nd0 [~fjig@karmosin-03.csc.kth.se] has joined #shogun | 14:39 | |
pluskid | blackburn, have you looked at the LARS code? | 14:39 |
blackburn | pluskid: yes | 14:39 |
pluskid | any feedback? | 14:40 |
blackburn | ok not really I think it should be merged | 14:40 |
blackburn | I'll do that once I get to home | 14:41 |
pluskid | cool, thanks | 14:41 |
pluskid | any suggestions on what to work on next? | 14:42 |
pluskid | small ones | 14:42 |
blackburn | yes | 14:42 |
blackburn | however not so small.. | 14:42 |
pluskid | hmm, let's see what's it | 14:42 |
blackburn | pluskid: have you seen how did we organized multiclass machine? | 14:44 |
pluskid | no, haven't investigated that yet | 14:44 |
pluskid | any problems there? | 14:44 |
blackburn | yes as new multiclass machines (CS liblinear and multiclass ocas) are in multiclass | 14:45 |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 14:45 | |
sonne|work | blackburn, pluskid: actually one could do the new subset stuff from heiko for oVo ? | 14:45 |
sonne|work | OvO :D | 14:45 |
blackburn | and are inherited from multiclass machine | 14:45 |
blackburn | sonne|work: oVo looks nice ;) | 14:45 |
blackburn | sonne|work: yes it is the task too | 14:45 |
blackburn | sonne|work: my intention is to move larank and gmnp to multiclass | 14:46 |
sonne|work | yeah but first finalize what n4nd0 started no? | 14:46 |
blackburn | sonne|work: it is pretty unrelated | 14:46 |
blackburn | while that stuff was generic and solvers do not make use of it | 14:46 |
n4nd0 | what is left from the part I did? | 14:46 |
blackburn | I'll finish that on this weekend | 14:47 |
blackburn | n4nd0: ovo subsetting | 14:47 |
blackburn | pluskid: so the task could be | 14:47 |
blackburn | to move larank to multiclass | 14:47 |
* pluskid is listening | 14:47 | |
blackburn | and adjust it accordingly | 14:47 |
pluskid | ok | 14:47 |
n4nd0 | blackburn: didn't Heiko did that? I understood the thing that was left was to test if the results were now correct | 14:47 |
blackburn | n4nd0: remember there is no add_subset for labels and kernels? | 14:48 |
sonne|work | blackburn: but wouldn't he need it if he does other ECOC schemes? | 14:48 |
sonne|work | ahh labels | 14:48 |
* pluskid waiting for you guys to arrive at a conclusion | 14:48 | |
sonne|work | yes true that needs to be done | 14:48 |
sonne|work | too | 14:48 |
blackburn | pluskid: currently larank is derived from multiclasssvm | 14:49 |
pluskid | so we should make it derive from multiclass machine? | 14:49 |
blackburn | exactly | 14:49 |
blackburn | this can be done with rebasing multiclasssvm actually | 14:49 |
pluskid | ok, I'll try that | 14:49 |
pluskid | is multiclasssvm derived from multiclass machine? | 14:50 |
blackburn | that could be painful so just let me know if you don't like do such refactoring | 14:50 |
pluskid | yeah, I haven't read those code yet | 14:50 |
blackburn | no, multiclasssvm should be inherited from kernelmulticlassmachine I think | 14:50 |
blackburn | sonne|work: agree? | 14:50 |
pluskid | I'll try to look at it and ask here if I have problems :) | 14:51 |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Ping timeout: 260 seconds] | 14:51 | |
pluskid | BTW: do we have publically available online doxygen doc | 14:52 |
pluskid | for our dev code? | 14:52 |
pluskid | latest code, automatically updated doc? | 14:52 |
n4nd0 | blackburn: I don't quite remember it that much but I think I already did the thing for kernels | 14:53 |
sonne|work | blackburn: yes makes sense | 14:53 |
n4nd0 | blackburn: https://github.com/iglesias/shogun/commit/01eb524f812fff9cc9380d269351f2835bc24b62 | 14:53 |
sonne|work | pluskid: no | 14:53 |
sonne|work | but you can just run make doc | 14:53 |
pluskid | sonne|work, yes, but that makes my local repo dirty | 14:54 |
pluskid | :p | 14:54 |
pluskid | I mean git status shows some changed files in doc directory | 14:54 |
pluskid | when I compile the doc | 14:54 |
n4nd0 | blackburn: I think that what is left to do is if it is correctly working after Heiko's job | 14:55 |
pluskid | n4nd0, what you are talking about is that: what I intend to do is already done before? | 14:56 |
n4nd0 | pluskid: no no, no worries | 14:56 |
n4nd0 | pluskid: another thing | 14:56 |
pluskid | ok :) | 14:56 |
n4nd0 | pluskid: another think that appeared in the conversation ;9 | 14:56 |
blackburn | pluskid: http://dl.dropbox.com/u/10139213/shogun/refactor.jpg | 14:57 |
pluskid | got it | 14:57 |
* n4nd0 has weak fingers for parentheses today | 14:57 | |
pluskid | oops | 14:57 |
pluskid | have to connect a VPN to see files in dropbox | 14:57 |
blackburn | mad skillz painting loool | 14:57 |
n4nd0 | blackburn: haha that explanation ^ is just awesome | 14:58 |
blackburn | damn did they block dropbox? | 14:58 |
pluskid | have you guys heard about the Great Fire Wall in China? | 14:58 |
blackburn | yes | 14:58 |
n4nd0 | yeah ... | 14:58 |
pluskid | they block many valuable sites | 14:58 |
pluskid | will be back in a while after a reconnection | 14:59 |
-!- pluskid [~pluskid@111.120.58.220] has quit [Quit: Leaving] | 14:59 | |
n4nd0 | blackburn: have you checked the commit ^ ? | 14:59 |
-!- pluskid [~pluskid@173.254.214.60] has joined #shogun | 14:59 | |
blackburn | n4nd0: oh how did I miss that/ | 15:00 |
blackburn | wait | 15:00 |
blackburn | is it merged? | 15:00 |
n4nd0 | blackburn: no | 15:00 |
blackburn | what is you waiting for ;) | 15:00 |
n4nd0 | blackburn: I think we decided to wait after Heiko' | 15:01 |
blackburn | ah | 15:01 |
pluskid | what is the refactor figure mean? | 15:01 |
pluskid | is this already done or waiting to be done? | 15:01 |
blackburn | that should be done | 15:01 |
blackburn | left (as is) => right (should be) | 15:01 |
pluskid | so they all become disconnected? | 15:01 |
blackburn | no my links were overlapped by package :D | 15:02 |
blackburn | structure is obvious | 15:02 |
blackburn | the only thing - multiclasssvm should inherit from multiclass | 15:02 |
pluskid | cool figure btw :) | 15:03 |
blackburn | I think we should make use of uml for such things - much easier to grasp | 15:04 |
-!- xiangwang_ [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has quit [Remote host closed the connection] | 15:05 | |
blackburn | cu | 15:10 |
-!- blackburn [5bdfb203@gateway/web/freenode/ip.91.223.178.3] has quit [Quit: Page closed] | 15:10 | |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Remote host closed the connection] | 15:19 | |
-!- wiking [~wiking@vpnc122.ugent.be] has joined #shogun | 15:20 | |
-!- wiking [~wiking@vpnc122.ugent.be] has quit [Changing host] | 15:20 | |
-!- wiking [~wiking@huwico/staff/wiking] has joined #shogun | 15:20 | |
-!- xiangwang [~chatzilla@159.226.60.224] has joined #shogun | 15:22 | |
-!- vikram360 [~vikram360@117.192.180.208] has quit [Ping timeout: 265 seconds] | 15:28 | |
-!- harshit_ [~harshit@182.68.181.141] has quit [Quit: Leaving] | 15:28 | |
-!- xiangwang [~chatzilla@159.226.60.224] has quit [Ping timeout: 260 seconds] | 15:29 | |
-!- vikram360 [~vikram360@117.192.184.89] has joined #shogun | 15:37 | |
wiking | heheh it's great to do cuda calculations on your desktop machine... suddenly the whole gui became unresponsice :P | 15:52 |
-!- karlnapf [~heiko@dhcp-186-176.internal.eduroam.ucl.ac.uk] has joined #shogun | 15:56 | |
-!- uricamic [9320543b@gateway/web/freenode/ip.147.32.84.59] has quit [Quit: Page closed] | 16:03 | |
-!- xiangwang [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has joined #shogun | 16:13 | |
sonne|work | karlnapf: hey - the man is back :D | 16:29 |
karlnapf | sonne|work hi there :) yes busy week here | 16:30 |
karlnapf | last coursework deadline today | 16:30 |
karlnapf | family visiting ... | 16:30 |
karlnapf | exam preparation | 16:30 |
sonne|work | tough tough | 16:31 |
karlnapf | how are things? | 16:31 |
sonne|work | did you see that we got 9 slots? | 16:31 |
sonne|work | of which we will give back 1 due to not enough mentors? | 16:32 |
karlnapf | what? wow :) | 16:32 |
karlnapf | no not yet | 16:32 |
karlnapf | man | 16:32 |
karlnapf | awsome this is | 16:33 |
sonne|work | if 2/5th of them stay with us after GSoC we are in shogun heaven | 16:33 |
karlnapf | yes | 16:34 |
karlnapf | very cool | 16:34 |
karlnapf | this will probably happen | 16:34 |
karlnapf | how many proposals were there in the end? | 16:34 |
sonne|work | 48 | 16:35 |
sonne|work | from 38 students | 16:35 |
sonne|work | still a lot | 16:35 |
sonne|work | probably a dozen w/o patch though | 16:35 |
sonne|work | (reject) | 16:35 |
karlnapf | indeed, very nice congratulations for the 9 slots | 16:35 |
karlnapf | whats next? | 16:36 |
karlnapf | ranking probably | 16:37 |
karlnapf | sonne|work, talk to you later got a meeting now. I will be around in the evening and on the weekend | 16:39 |
sonne|work | cu | 16:40 |
wiking | jeeeeey | 16:41 |
wiking | finally we are on the same page with alex! | 16:41 |
wiking | \o/ | 16:41 |
-!- karlnapf [~heiko@dhcp-186-176.internal.eduroam.ucl.ac.uk] has quit [Ping timeout: 252 seconds] | 16:43 | |
wiking | now it's coding time \o/ | 16:45 |
wiking | and god damn cuda :) | 16:46 |
-!- karlnapf [~heiko@dhcp-186-176.internal.eduroam.ucl.ac.uk] has joined #shogun | 16:49 | |
-!- pluskid [~pluskid@173.254.214.60] has quit [Quit: Leaving] | 16:51 | |
wiking | mmm was there a final decision whether shogun will depend on viennaCL ? | 16:53 |
-!- n4nd0 [~fjig@karmosin-03.csc.kth.se] has quit [Quit: leaving] | 16:53 | |
sonne|work | no not yet | 16:54 |
sonne|work | optional in any case | 16:54 |
wiking | i mean that the opencl solution will be actually based on viennacl? | 16:55 |
wiking | anyhow i'll give my comments on the pull request as well | 16:56 |
wiking | send | 17:01 |
wiking | t | 17:01 |
-!- PhilTillet [~Philippe@npasserelle10.minet.net] has joined #shogun | 17:19 | |
-!- sonne|work [~sonnenbu@194.78.35.195] has left #shogun [] | 17:22 | |
-!- gsomix [~gsomix@178.45.43.67] has joined #shogun | 17:31 | |
gsomix | home, sweet home | 17:31 |
gsomix | hello | 17:31 |
-!- PhilTillet [~Philippe@npasserelle10.minet.net] has quit [Ping timeout: 245 seconds] | 17:38 | |
-!- PhilTillet [~Philippe@npasserelle10.minet.net] has joined #shogun | 18:06 | |
-!- xiangwang [~xiangwang@2001:cc0:2020:2013:f2de:f1ff:fe50:9012] has quit [Remote host closed the connection] | 18:11 | |
-!- gsomix [~gsomix@178.45.43.67] has quit [Ping timeout: 260 seconds] | 18:19 | |
-!- gsomix [~gsomix@178.45.43.67] has joined #shogun | 18:20 | |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has joined #shogun | 18:44 | |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has quit [Quit: ( www.nnscript.com :: NoNameScript 4.22 :: www.esnation.com )] | 18:53 | |
gsomix | sonney2k, moin. how are you? | 19:19 |
-!- naywhayare [~ryan@spoon.lugatgt.org] has quit [Remote host closed the connection] | 19:27 | |
-!- blackburn [~qdrgsm@109.226.90.219] has joined #shogun | 19:42 | |
wiking | blackburn: have u ever worked with DomainAdaptationSVM? | 19:48 |
blackburn | nope | 19:48 |
blackburn | planning to actually | 19:48 |
blackburn | are you going to try whole shogun? :) | 19:48 |
wiking | some stuff | 19:48 |
wiking | i'd like to test something | 19:48 |
wiking | i have a pretty well trained svm | 19:49 |
wiking | but for some images i don't have some features | 19:49 |
blackburn | at all? | 19:49 |
wiking | yep | 19:49 |
wiking | since some of the features are actually textual features | 19:49 |
wiking | that are not dependent on the image itself | 19:49 |
wiking | but where the image resided | 19:49 |
wiking | and i have some random images about which i have no other info | 19:49 |
wiking | so some features are basically missing | 19:49 |
wiking | and i was wondering if i could still | 19:50 |
wiking | classify those images | 19:50 |
blackburn | damn you have hard task :D | 19:50 |
wiking | this DomainAdaptationSVM | 19:50 |
wiking | seems to be something | 19:50 |
wiking | that is worth a shot | 19:50 |
blackburn | is it a multiclass problem you have? | 19:52 |
wiking | yes | 19:52 |
blackburn | would need to design new DA SVM then :D | 19:53 |
blackburn | how much time you have? | 19:53 |
wiking | yeah just realized from the code | 19:53 |
wiking | that actually it seems to be binary classifier | 19:53 |
blackburn | so no need to multiclass? | 19:53 |
wiking | i would | 19:53 |
wiking | but src/shogun/classifier/svm/DomainAdaptationSVM.h is binary right? | 19:54 |
wiking | mm yeah | 19:54 |
wiking | since it's derived from SVMLight | 19:54 |
wiking | damn i thought i could just use it out of box :) | 19:55 |
blackburn | yes | 19:56 |
wiking | :(( | 19:56 |
blackburn | wiking: while this DA SVM was implemented by chris I can ask him | 19:58 |
blackburn | and we can come up with linear multiclass da svm | 19:58 |
blackburn | da svm lol | 19:59 |
wiking | :> | 19:59 |
wiking | yeah would be great to extend it | 20:00 |
wiking | mmm it has a kernelized version as well... | 20:00 |
blackburn | wiking: yes while it is dual | 20:01 |
blackburn | wiking: okay dinner time now and after I'll send chris a mail | 20:02 |
wiking | cool | 20:02 |
wiking | u can cc me as well | 20:02 |
blackburn | sure | 20:02 |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 20:26 | |
blackburn | wiking: are you particularly interested in kernelized version? | 20:38 |
wiking | well that'd be great | 20:38 |
wiking | but i can do a hkm shit :))) | 20:38 |
-!- PhilTillet [~Philippe@npasserelle10.minet.net] has quit [Ping timeout: 260 seconds] | 20:41 | |
blackburn | wiking: http://fml.tuebingen.mpg.de/raetsch/lectures/talk-mtcompbio.pdf | 20:50 |
blackburn | on last pages there are some DA techniques | 20:50 |
-!- PhilTillet [~Philippe@157.159.42.154] has joined #shogun | 21:01 | |
-!- gsomix [~gsomix@178.45.43.67] has quit [Read error: Connection reset by peer] | 21:01 | |
-!- puffin444 [230bf329@gateway/web/freenode/ip.35.11.243.41] has joined #shogun | 21:01 | |
-!- gsomix [~gsomix@178.45.44.70] has joined #shogun | 21:16 | |
blackburn | wiking: sent | 21:17 |
-!- PhilTillet [~Philippe@157.159.42.154] has quit [Ping timeout: 260 seconds] | 21:23 | |
-!- naywhayare [~ryan@spoon.lugatgt.org] has joined #shogun | 21:24 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 21:46 | |
blackburn | wiking: time to take pencil! deriving CS with DA | 22:05 |
blackburn | ;) | 22:05 |
wiking | ? | 22:05 |
wiking | cs =? | 22:05 |
blackburn | crammer singer | 22:06 |
blackburn | my favourite multiclass formulation with simple dual ;) | 22:06 |
wiking | hahahah | 22:07 |
wiking | tell me tell me | 22:07 |
wiking | i make NOW :) | 22:08 |
blackburn | wiking: make what? ;) | 22:10 |
wiking | da with cs | 22:10 |
blackburn | I'm going to implement it once I formulate it | 22:10 |
blackburn | however that would be not so trivial | 22:10 |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has joined #shogun | 22:11 | |
blackburn | afaik liblinear solver do not support easy linear term setting | 22:11 |
wiking | Log-likelihood: 860587008.000000 | 22:12 |
wiking | :D | 22:12 |
wiking | gmm-ing my dataset :P | 22:12 |
blackburn | log? oh that's nice | 22:12 |
n4nd0 | good night guys! | 22:14 |
wiking | well this would have had taken like about a week to calculate | 22:14 |
blackburn | n4nd0: nite | 22:14 |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Quit: leaving] | 22:14 | |
wiking | w/o having a nvidia tesla | 22:14 |
blackburn | wiking: is the dataset so big? | 22:15 |
wiking | yep | 22:15 |
wiking | 5*10^6 vectors with each of 64 dimension | 22:15 |
wiking | it was weird though | 22:16 |
wiking | had to do a *10^3 for each vector | 22:16 |
wiking | as otherwise gmm couldn't find any cluster centers | 22:16 |
wiking | so i don't know if on the end this is going to work out :) | 22:17 |
wiking | i'm still amazed by this tesla... now it's 116W is being used now when it's running on 100% | 22:17 |
blackburn | wiking: why not kmeans? | 22:17 |
wiking | blackburn: since i'm trying to generate some fisher vectors | 22:17 |
wiking | and i'd be interested as well in the soft clustering of gmm | 22:18 |
blackburn | wiking: I see | 22:18 |
blackburn | I just recalled there is an algorithm called ?????? | 22:18 |
blackburn | :D | 22:18 |
-!- puffin444 [230bf329@gateway/web/freenode/ip.35.11.243.41] has quit [Quit: Page closed] | 22:20 | |
blackburn | wiking: idea of regularized DA svm is to | 22:21 |
blackburn | optimize 1/2 ||w-w0||^2 + C \sum_i \xi_i with same s.t. | 22:22 |
blackburn | w0 is SVM trained on source domain | 22:22 |
wiking | ah | 22:22 |
blackburn | and predictor is | 22:22 |
blackburn | <w,x>+<w0,x> | 22:22 |
blackburn | sounds reasonable but I need to think about it in MC mean | 22:23 |
-!- gsomix [~gsomix@178.45.44.70] has quit [Ping timeout: 276 seconds] | 22:32 | |
-!- karlnapf [~heiko@dhcp-186-176.internal.eduroam.ucl.ac.uk] has left #shogun [] | 22:46 | |
@sonney2k | blackburn, wiking well you can just use all the w_0 you obtained from the original training and learn new w's | 22:50 |
blackburn | sonney2k: how? | 22:51 |
@sonney2k | so it is basically the same thing like two-class - one just chooses the corresponding w_0 | 22:51 |
@sonney2k | blackburn, I assume you do true MC | 22:51 |
blackburn | CS | 22:51 |
@sonney2k | so you have one-vs-rest classification in the end | 22:51 |
blackburn | sonney2k: the problem is: | 22:52 |
blackburn | min 1/2 ||w-w0||^2 + C .. | 22:52 |
blackburn | s.t. <w,x>-... | 22:52 |
blackburn | not w-w0 | 22:52 |
blackburn | so it can't be 'virtual' w-w0 | 22:52 |
-!- harshit_ [~harshit@182.68.181.141] has quit [Read error: Connection reset by peer] | 22:52 | |
blackburn | sonney2k: got me? | 22:52 |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 22:53 | |
@sonney2k | not yet | 22:54 |
blackburn | sonney2k: I mean dual task will change in that case | 22:54 |
@sonney2k | min 0.5||w||^2 +0.5 ||w-w0||^2 + ... | 22:54 |
blackburn | sonney2k: and liblinear does not provide an easy way | 22:54 |
blackburn | wow what is it? | 22:54 |
@sonney2k | same thing like you use | 22:55 |
blackburn | but grad will change | 22:55 |
@sonney2k | by some constant | 22:56 |
@sonney2k | IIRC there is some gamma or whatever factor in front of the second term | 22:56 |
blackburn | sonney2k: but still needs some patching there right? | 22:57 |
@sonney2k | so you can scale gamma and get | 22:57 |
blackburn | sonney2k: have you seen I've got a competitor? | 22:57 |
@sonney2k | 0.5 ||w||^2 + gamma w^T w_0 + ... | 23:00 |
@sonney2k | yes you need to do some patching | 23:00 |
@sonney2k | in the same way liblinear is patched now for binary svm | 23:00 |
@sonney2k | not just one :D | 23:01 |
blackburn | sonney2k: but one is going to contribute ;) | 23:01 |
blackburn | sonney2k: do you think following is reasonable then? | 23:02 |
blackburn | 0.5 \sum_m ||w_m||^2 + gamma \sum_m wm^T w_0m + C \sum_i \xi_i | 23:02 |
blackburn | however I can hardly see what is the trade-off we want to achieve there | 23:03 |
blackburn | sonney2k: what is the role of w^T w_0? | 23:04 |
blackburn | ah stupid question it is w-w0 | 23:04 |
@sonney2k | gtg | 23:08 |
@sonney2k | cu | 23:08 |
blackburn | wiking: we should employ sonney2k more in future ;) | 23:11 |
wiking | :> | 23:19 |
-!- nickon [~noneedtok@dD5774105.access.telenet.be] has quit [Quit: ( www.nnscript.com :: NoNameScript 4.22 :: www.esnation.com )] | 23:23 | |
@sonney2k | re | 23:23 |
wiking | re | 23:23 |
@sonney2k | blackburn, and the ||w0||^2 part is const so can be left out from objective | 23:24 |
blackburn | sonney2k: yes that's clear | 23:24 |
blackburn | sonney2k: what stays unclear is how that addition changes the solution? | 23:25 |
@sonney2k | depends on value of gamma :) | 23:28 |
blackburn | I mean does the solution change if we +1 to gradient? | 23:29 |
blackburn | hmm it seems so but strange | 23:29 |
-!- wiking_ [~wiking@78-23-189-112.access.telenet.be] has joined #shogun | 23:33 | |
-!- wiking_ [~wiking@78-23-189-112.access.telenet.be] has quit [Changing host] | 23:33 | |
-!- wiking_ [~wiking@huwico/staff/wiking] has joined #shogun | 23:33 | |
blackburn | sonney2k: how many multitask proposals then? | 23:35 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 248 seconds] | 23:36 | |
-!- wiking_ is now known as wiking | 23:36 | |
wiking | have i missed something? :S | 23:37 |
wiking | this was the last message i've seen: blackburn: hmm it seems so but strange | 23:37 |
blackburn | wiking: no, the only thing is I asked how many multitask proposals we received | 23:37 |
wiking | ah there was a gsoc idea about this? | 23:41 |
blackburn | wiking: what is do you think I am applying for then? ;) | 23:43 |
wiking | mm maybe i should have tried this project :))) | 23:44 |
blackburn | wiking: will you stop changing areas some day? :D | 23:44 |
-!- harshit_ [~harshit@182.68.181.141] has quit [Read error: Connection reset by peer] | 23:44 | |
wiking | hahaha | 23:44 |
wiking | no i just need this one | 23:44 |
wiking | but | 23:45 |
wiking | i can test with 1 class | 23:45 |
wiking | to see how it would work | 23:45 |
-!- harshit_ [~harshit@182.68.181.141] has joined #shogun | 23:45 | |
wiking | that'll do out of box... | 23:45 |
blackburn | wiking: it doesn't depend on your problem already - I got interested to implement that :D | 23:47 |
wiking | :>> | 23:47 |
wiking | there's code | 23:52 |
wiking | :) | 23:52 |
--- Log closed Sat Apr 14 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!