Open in new window / Try shogun cloud
--- Log opened Wed Jan 25 00:00:03 2012
--- Day changed Wed Jan 25 2012
@sonney2ktrees00:00
@sonney2knot everything but everything needed :)00:00
@sonney2kblackburn, I totally forgot to ask - did your exam go well?00:01
blackburnsonney2k: I consider shogun as 'core', but didn't formulated what is it to be core00:01
@sonney2kthe optical illusion one I mean :D00:01
blackburnyeah, got 4/500:01
@sonney2kperfect00:01
blackburnI am happy to get 4 for this shit hah00:01
blackburnwait a min I would show you presentation00:02
@sonney2kblackburn, that is a nice page http://scikit-learn.org/stable/auto_examples/index.html00:02
blackburnhttp://dl.dropbox.com/u/10139213/share/05-%D0%B8%D1%82%D0%B5%D1%80%D0%B0%D1%86%D0%B8%D0%BE%D0%BD%D0%BD%D1%8B%D0%B9.pdf00:02
blackburnyes, I like these guys00:02
blackburnthey rock00:02
@sonney2ksth like that for shogun would be really great for gsoc and fun for a student00:02
blackburnbut they dimreduction sucks haha00:03
blackburntheir00:03
@sonney2kblackburn, heh - nice :)00:03
blackburnsonney2k: how can one memorize this?00:03
@sonney2kby looking it up00:04
blackburnsonney2k: 28 topics totally, it is one of them00:04
@sonney2kif we manage to get in to say 2-3 years of gsoc we really have a wide base of all the basic features one usually needs in shogun00:05
@sonney2keven now I am using it for a lot of things00:06
blackburnsonney2k: about sklearn dimreduction00:06
@sonney2kI have to sleep now!00:06
blackburnI know a way (2 lines fix) how to speed up LLE00:07
blackburnbut didn't tell it to them yet00:07
@sonney2kblackburn, well their SVMs certainly are no match to shogun stuff00:07
blackburndo you think it is fair? :)00:07
@sonney2keither00:07
@sonney2kas you want00:07
@sonney2kdid they fix anything in shogun?00:07
blackburnnot really :)00:07
@sonney2khttp://jmlr.csail.mit.edu/papers/volume12/pedregosa11a/pedregosa11a.pdf00:08
@sonney2khave you read?00:08
@sonney2ktable 1 lists some experiments for very tiny data sets00:08
blackburnsonney2k: glanced over, is there anything interesting?00:08
@sonney2kshowing that shogun is slower00:08
blackburnsonney2k: well did you see my measures?00:09
@sonney2kbut that is all in the 1 second range00:09
@sonney2knot sure how they can mark 5.2 seconds vs. 5.63 seconds and then 5.2 secs as the clear winner00:09
blackburn19.43 vs 473.3400:10
blackburnseconds00:10
blackburnfor 10000 3d vectors00:10
@sonney2kwhy is our PCA slower?00:10
blackburnno idea00:10
blackburnwell it is not really optimal00:11
blackburnsonney2k: what did they measure?00:12
@sonney2kwhatever - I don't even look at data sets of size ~5000 nowadays00:13
@sonney2knothing below 100k examples :)00:13
blackburnsonney2k: hah that's funny I care too much with measures00:13
blackburntrying to make them fair..00:13
@sonney2kI guess hardly any toolkit except shogun works for that00:13
@sonney2kI wonder why reviewers didn't criticize this table00:14
@sonney2kbut hey00:14
blackburnsonney2k: okay I would BLAS the PCA just now00:14
@sonney2kI really need to sleep now00:14
blackburnor no00:14
blackburnwould better do the same00:14
@sonney2kit is blas'd IIRC?00:14
@sonney2kthe same?00:14
blackburnsonney2k: sub-mean ops could be blased too00:14
blackburnsonney2k: yeah, would better sleep too00:14
blackburnsonney2k: I'm going to submit my paper in 2 days00:15
@sonney2kanything that fits in memory is reasonably fast00:16
@sonney2kdoesn't matter if it takes 1.5 or 1.0 seconds :)00:16
@sonney2kwell I am really off now00:16
@sonney2kcu00:16
blackburnsonney2k: well scikits multiplies two 10000 * 10000 matrices when embedding 10000 vectors00:17
blackburn(we don't) :D00:17
blackburnokay see you00:17
@sonney2khttp://packages.python.org/milk/benchmarks.html00:17
blackburnso it matters sometimes00:17
@sonney2klook at this milk  :)00:17
* sonney2k is typing while asleep00:17
blackburnmilk is pretty fast and for kmeans and SVM learning it is the fastest system.00:17
blackburn:D00:17
blackburn:D00:17
blackburn:D00:18
blackburnLOL00:18
blackburnomg lol00:18
blackburnsleep well :)00:19
-!- blackburn [~qdrgsm@31.28.63.56] has quit [Quit: Leaving.]00:54
-!- Ram108 [~amma@14.96.38.9] has joined #shogun07:22
-!- Ram108 [~amma@14.96.38.9] has quit [Ping timeout: 240 seconds]08:38
-!- blackburn [~qdrgsm@31.28.63.56] has joined #shogun08:57
-!- jekintrivedi [~jekin@27.4.211.159] has joined #shogun09:01
-!- jekintrivedi [~jekin@27.4.211.159] has left #shogun []09:02
-!- jekintrivedi [~jekin@27.4.211.159] has joined #shogun09:03
-!- jekintrivedi [~jekin@27.4.211.159] has quit [Read error: Connection reset by peer]09:32
-!- blackburn [~qdrgsm@31.28.63.56] has quit [Quit: Leaving.]09:59
-!- blackburn [~qdrgsm@31.28.63.56] has joined #shogun10:00
blackburnsurprising truth, sckikits is faster on toy data ;(15:40
-!- Ram108 [~amma@219.64.146.186] has joined #shogun16:14
-!- Netsplit *.net <-> *.split quits: Ram10816:34
-!- Netsplit *.net <-> *.split quits: naywhayare, blackburn16:34
-!- Netsplit *.net <-> *.split quits: @sonney2k, sonne|work16:34
-!- Netsplit *.net <-> *.split quits: shogun-buildbot16:34
-!- Netsplit *.net <-> *.split quits: CIA-1016:34
-!- Netsplit over, joins: CIA-10, shogun-buildbot, naywhayare, @sonney2k, Ram108, sonne|work16:55
-!- blackburn [~qdrgsm@31.28.63.56] has joined #shogun17:06
-!- Ram108 [~amma@219.64.146.186] has quit [Ping timeout: 248 seconds]18:39
-!- Ram108 [~amma@14.96.105.129] has joined #shogun19:08
blackburnsonney2k: around?19:19
-!- cwidmer [8ca3fe9d@gateway/web/freenode/ip.140.163.254.157] has joined #shogun19:33
blackburnsonney2k: I would need your time today :)19:57
-!- blackburn [~qdrgsm@31.28.63.56] has quit [Quit: Leaving.]21:07
-!- blackburn [~qdrgsm@31.28.63.56] has joined #shogun21:11
shogun-buildbotbuild #115 of nightly_default is complete: Failure [failed test_1]  Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/11521:52
-!- Ram108 [~amma@14.96.105.129] has quit [Ping timeout: 245 seconds]23:22
--- Log closed Thu Jan 26 00:00:19 2012