--- Log opened Fri Feb 24 00:00:19 2012 | ||
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Read error: Connection reset by peer] | 00:03 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 00:03 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Read error: Connection reset by peer] | 01:02 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 01:02 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 244 seconds] | 01:55 | |
-!- axitkhurana [~Akshit@180.149.49.231] has joined #shogun | 02:26 | |
-!- axitkhurana [~Akshit@180.149.49.231] has left #shogun [] | 03:37 | |
-!- axitkhurana [~akshit@180.149.49.231] has joined #shogun | 04:13 | |
-!- abhilb [75ce3638@gateway/web/freenode/ip.117.206.54.56] has joined #shogun | 04:34 | |
-!- axitkhurana [~akshit@180.149.49.231] has quit [Remote host closed the connection] | 05:16 | |
-!- blackburn [~qdrgsm@188.168.5.134] has joined #shogun | 07:21 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 07:34 | |
-!- blackburn [~qdrgsm@188.168.5.134] has quit [Ping timeout: 260 seconds] | 07:58 | |
-!- abhilb [75ce3638@gateway/web/freenode/ip.117.206.54.56] has quit [Quit: Page closed] | 09:15 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has quit [Ping timeout: 240 seconds] | 14:12 | |
-!- n4nd0 [~nando@n189-p80.kthopen.kth.se] has joined #shogun | 15:11 | |
-!- tjn4 [458618b7@gateway/web/freenode/ip.69.134.24.183] has joined #shogun | 15:42 | |
-!- tjn4 [458618b7@gateway/web/freenode/ip.69.134.24.183] has quit [Client Quit] | 15:42 | |
-!- n4nd0 [~nando@n189-p80.kthopen.kth.se] has quit [Quit: leaving] | 16:31 | |
-!- blackburn [~qdrgsm@188.122.248.245] has joined #shogun | 18:14 | |
-!- karlnapf [~heiko@host86-180-120-164.range86-180.btcentralplus.com] has joined #shogun | 19:35 | |
karlnapf | sonney2k, around? | 19:39 |
---|---|---|
karlnapf | blackburn, around? | 19:39 |
blackburn | hey | 19:40 |
blackburn | hello | 19:40 |
blackburn | yes | 19:40 |
karlnapf | hi | 19:40 |
karlnapf | how is it going? | 19:40 |
blackburn | fine, what about you? | 19:40 |
karlnapf | same here :) | 19:41 |
karlnapf | very busy | 19:41 |
blackburn | yeah, feel the same sometimes | 19:42 |
blackburn | karlnapf: yesterday we have discussed ideas with soeren | 19:42 |
karlnapf | ah nice | 19:42 |
blackburn | you probably better to check logs | 19:42 |
karlnapf | was about asking about that | 19:43 |
karlnapf | will do | 19:43 |
karlnapf | many new things? | 19:43 |
blackburn | no, no new things at all I'd say | 19:43 |
blackburn | I am going to do multitask learning, you know? | 19:43 |
karlnapf | wow, nice one :) | 19:44 |
karlnapf | I had some lectures by massi pontil about that | 19:44 |
blackburn | hmm | 19:45 |
blackburn | nice | 19:45 |
blackburn | I don't know much about it yet | 19:45 |
blackburn | :) | 19:45 |
karlnapf | me neither, just the lecture, havent studied it so well yet, but its probably a nice thing and very recent | 19:45 |
blackburn | I checked your part and it is pretty complex as well | 19:46 |
karlnapf | the mmd/HSIC? | 19:46 |
blackburn | yeah no idea what it means but probably ys | 19:47 |
karlnapf | these are really nice things in my eyes | 19:47 |
karlnapf | I hope I will do my Master project also about this | 19:47 |
karlnapf | About kernel selection for these things | 19:48 |
blackburn | master? I though you are phd.. | 19:48 |
blackburn | I'm going to do my bachelor project about road sign recognition | 19:48 |
karlnapf | no, I am a Master student, hopefully phd in autumn :) | 19:49 |
blackburn | ah last year masters? | 19:49 |
karlnapf | nice, so lots of image recognition stuff | 19:49 |
blackburn | I see | 19:49 |
karlnapf | its a one year program here | 19:49 |
blackburn | ah | 19:49 |
blackburn | I see | 19:49 |
blackburn | hmm something is wrong with svmocas | 19:50 |
blackburn | just got 16% accuracy | 19:50 |
blackburn | :D | 19:50 |
karlnapf | mmh | 19:50 |
karlnapf | thats not good | 19:50 |
blackburn | while liblinear was 96% | 19:50 |
karlnapf | I have beed fiddeling around with multiclass svm recently | 19:50 |
karlnapf | multiclassMKL also | 19:50 |
blackburn | yeah I'm doing multiclass as well | 19:50 |
karlnapf | and I think shogun desparetely needs sigmoid fitting (also for multiclass) | 19:51 |
karlnapf | oh ok. | 19:51 |
blackburn | why not to use genuine multiclass svms? | 19:51 |
karlnapf | I have a lot of changes in my local repo, will commit when I find some time | 19:51 |
karlnapf | whats that? | 19:51 |
blackburn | like gmnpsvm | 19:51 |
blackburn | no need to do sigmoid fitting | 19:51 |
karlnapf | ah | 19:51 |
karlnapf | mmh | 19:51 |
blackburn | takes a lot of time to train though | 19:51 |
karlnapf | well, but users may want to have probs for standard svm | 19:52 |
blackburn | ah | 19:52 |
karlnapf | really? Just wrote some code for it | 19:52 |
blackburn | I thought you mean different thing | 19:52 |
karlnapf | works pretty fine | 19:52 |
karlnapf | ah ok | 19:52 |
karlnapf | I also implemented it for one-against-one svms | 19:52 |
karlnapf | Results are nice | 19:52 |
blackburn | one against one? | 19:52 |
karlnapf | When we discussed about how to integrate it into shogun I will do | 19:52 |
karlnapf | yes for multiclass svm | 19:52 |
blackburn | I don't like OvO :) | 19:53 |
blackburn | had 903 svms here | 19:53 |
karlnapf | one against one currently only has this vote stuff | 19:53 |
blackburn | 43*42/2 | 19:53 |
karlnapf | oh :) | 19:53 |
karlnapf | well thats probably not feasible | 19:53 |
karlnapf | but if its used, its nice to have prob as output | 19:53 |
blackburn | sure | 19:53 |
blackburn | let me describe what I meant | 19:53 |
karlnapf | go ahead :) | 19:54 |
blackburn | when you do some OvR or OvO | 19:54 |
blackburn | scaling can vary much | 19:54 |
blackburn | and it is not fitted very well | 19:54 |
blackburn | so both vojtech and soeren told me it depends on such scaling | 19:54 |
blackburn | that's why gmnpsvm and larank works better | 19:54 |
karlnapf | ah ok | 19:55 |
karlnapf | how does it work, whats the output/input? | 19:55 |
blackburn | what is 'it'? | 19:55 |
blackburn | both gmnp and larank is OvR | 19:55 |
blackburn | but it trains like genuine multiclass svm | 19:55 |
karlnapf | mmh ok | 19:56 |
karlnapf | I have got to do some reading on this stuff | 19:56 |
blackburn | \sum_k ||w_k|| + C \sum_{i \ne k} \sum_j \xi_i^j | 19:56 |
karlnapf | I am currently facing a problem with 4 classes | 19:56 |
karlnapf | I would love to have a multi class MKL machine which output probabilities for this | 19:56 |
blackburn | I see | 19:56 |
blackburn | do you find MKL useful? | 19:57 |
karlnapf | kind of like the idea | 19:57 |
karlnapf | my problem is from bioinformatics | 19:57 |
blackburn | ah | 19:57 |
karlnapf | I have two kernels: one based on biochemical data, and one based only on sequence oinformation | 19:57 |
blackburn | I was curious if it is applicable to not bio data | 19:57 |
karlnapf | And the kernel weights would give me a kind of measure how important the features are | 19:57 |
karlnapf | MMhh, dont really know | 19:57 |
karlnapf | but I like combining kernels | 19:57 |
blackburn | there was one nice guy here in irc, he wrote HomogeneousKernelMap preprocessor | 19:58 |
blackburn | interesting stuff | 19:58 |
karlnapf | cool ,will check it out :) | 19:58 |
blackburn | maps features to linear space | 19:59 |
blackburn | where dot product is a kernel | 19:59 |
blackburn | explicitly I mean | 19:59 |
karlnapf | which dimension? | 19:59 |
blackburn | x3 | 19:59 |
blackburn | more -> better | 19:59 |
blackburn | but x3 is sufficient | 20:00 |
karlnapf | kk | 20:00 |
blackburn | I lose only 0.5% of accuracy | 20:00 |
blackburn | but linear | 20:00 |
blackburn | making me possible to use liblinear with 3m train time | 20:00 |
blackburn | instead of libsvm with 6h :) | 20:00 |
karlnapf | year | 20:00 |
karlnapf | liblinear is so fucking fast | 20:00 |
blackburn | svmocas is faster | 20:01 |
blackburn | but I am afraid of its results now | 20:01 |
karlnapf | btw I currently can reproduce a segfault using it (with modelselection) will try to fix that these days | 20:01 |
blackburn | you were pretty active last month | 20:01 |
blackburn | :) | 20:01 |
karlnapf | yes, want to integrate everythign I am using/doing here into shogun | 20:02 |
blackburn | in fact I think you should have write rights.. | 20:02 |
karlnapf | especially for model, there is lots of stuff to do | 20:02 |
karlnapf | well, I dont care much about this, also, the time I invest varies a lot | 20:02 |
blackburn | same here | 20:02 |
blackburn | I have been doing nothing about development since january I guess | 20:03 |
karlnapf | Think its ok, its opensource, youre not paid for this :) | 20:03 |
blackburn | yeah | 20:03 |
karlnapf | I think youre doing great stuff here anyway :) | 20:03 |
blackburn | but gsoc is paid so going to be active this summer | 20:03 |
blackburn | :D | 20:04 |
karlnapf | yes, ineed :) | 20:04 |
karlnapf | blackburn, I have some work to do so will leave now | 20:04 |
karlnapf | see you later | 20:04 |
blackburn | okay | 20:04 |
blackburn | see yah | 20:04 |
blackburn | ya | 20:04 |
blackburn | :) | 20:04 |
karlnapf | :) | 20:04 |
karlnapf | bye | 20:04 |
@sonney2k | karlnapf, hi - yes | 20:15 |
karlnapf | sonney2k, hi, I was about leaving now, just wanted to hear about the gsoc stuff, but talked a bit with blackburn and read yesterdays channel log | 20:16 |
blackburn | sonney2k: gmnpsvm (JS kernel) 97.14%, liblinear (OvR,JS) 96.00% | 20:17 |
@sonney2k | karlnapf, I just updated the webpage | 20:18 |
karlnapf | sonney2k, ok cool, will have a look, btw did you talk to arthur? | 20:18 |
blackburn | hmm a little different features | 20:18 |
@sonney2k | blackburn, already model selected? | 20:18 |
blackburn | well same C | 20:19 |
@sonney2k | karlnapf, no | 20:19 |
blackburn | sonney2k: the problem - I had 16% with svmocas | 20:19 |
blackburn | I am afraid it is broken | 20:19 |
@sonney2k | blackburn, hmmhh I guess we introduced a bug at some point :( | 20:19 |
-!- axitkhurana [~akshit@180.149.49.231] has joined #shogun | 20:19 | |
karlnapf | sonney2k, ok, have to go now, see you | 20:19 |
blackburn | I fixed one before | 20:20 |
@sonney2k | karlnapf, regarding arthur | 20:20 |
@sonney2k | is it OK if I put him as mentor? | 20:20 |
@sonney2k | I think it makes sense... | 20:20 |
karlnapf | sonney2k, I have asked him, but he is quite busy currently. I think he will agree and it makes sense. Will see him next week and then ask | 20:20 |
@sonney2k | karlnapf, yes ICMl deadline today | 20:21 |
karlnapf | exactly ;) | 20:21 |
karlnapf | ok, bye then | 20:22 |
@sonney2k | karlnapf, cu! | 20:22 |
blackburn | sonney2k: anyway, with gmnp I had 97.32% with HIK and 97.14% with HIK approximation with that homogay kernel map | 20:23 |
blackburn | so I'd say it is pretty nice thing | 20:23 |
@sonney2k | basically all the same | 20:24 |
blackburn | yes, so it is pretty good approximation | 20:24 |
-!- karlnapf [~heiko@host86-180-120-164.range86-180.btcentralplus.com] has quit [Quit: Leaving.] | 20:24 | |
blackburn | sonney2k: I did rejects in very simple way | 20:24 |
blackburn | just with thresholding I get 99% accuracy among non-rejected images | 20:24 |
blackburn | pretty nice for me :) | 20:25 |
-!- axitkhurana [~akshit@180.149.49.231] has quit [Ping timeout: 260 seconds] | 21:36 | |
-!- n4nd0 [~nando@s83-179-44-135.cust.tele2.se] has joined #shogun | 21:38 | |
-!- blackburn [~qdrgsm@188.122.248.245] has quit [Quit: Leaving.] | 23:08 | |
--- Log closed Sat Feb 25 00:00:19 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!