--- Log opened Wed Jul 18 00:00:17 2012 | ||
rieck | sonney2k: we can talk another day. guess i am now more often on irc | 00:00 |
---|---|---|
rieck | ;) | 00:00 |
rieck | btw droopy can also learn some ml/shogun stuff from this channel | 00:00 |
droopy | rieck: what riven said | 00:00 |
rieck | yes, but you need more knowledge in this field. i have trained you only with crappy hacker talk so far. | 00:01 |
blackburn | droopy: loss + regularization is a general framework in machine learning | 00:01 |
droopy | worst decision ever to invest on him :) | 00:01 |
rieck | he's making fun out of us :( | 00:02 |
blackburn | I think so | 00:02 |
rieck | droopy: seriously, it is better to learn something useful now! | 00:02 |
droopy | rieck: like traversing entryset of a map | 00:02 |
rieck | like that, for example! | 00:02 |
droopy | rieck: i'm having a long break again :-(!?! | 00:04 |
rieck | droopy: no, you can stay here and relax | 00:04 |
droopy | rieck: rofl | 00:04 |
rieck | woah, i am little surprised how well the new engine runs. i am sure that's only by chance today. | 00:05 |
blackburn | droopy: how is it going for you? | 00:06 |
droopy | sorry | 00:06 |
droopy | o/ has blackbeard! | 00:06 |
blackburn | blackbeard? :D | 00:07 |
rieck | basically, he knows roughly 10.000 messages from hacker chats | 00:07 |
rieck | droopy: you are quite restricted to security stuff unless you learn something new here. | 00:07 |
droopy | rieck: penny, from a practical point of view: for federal to use the hbgary name, there must be some sort of agreement or license, and with either an agreement of a license you usually have a method of leverage. i think it is very possible to put pressure on federal, *even* if it is technically independent. | 00:07 |
rieck | yes. | 00:07 |
rieck | droopy: that was far too detailed | 00:08 |
blackburn | I wonder what else he could say | 00:08 |
rieck | i myself don't know it. | 00:08 |
rieck | you can not read 10.000 messages | 00:09 |
droopy | rieck: yup | 00:09 |
blackburn | droopy: say something really detailed now :D | 00:10 |
droopy | i know the basics, but not any advanced things | 00:10 |
rieck | ;) | 00:11 |
blackburn | rieck: it is really unclear for me how does it work :) | 00:11 |
rieck | ~/droopy/parrot$ zcat *.gz | wc -l | 00:12 |
rieck | 14106 | 00:12 |
rieck | it's actually pretty simple. | 00:12 |
rieck | i have recordings of 14000 irc messages. droopy is looking for the most matching msgs to an input and returns the next msg from a different person | 00:12 |
droopy | rieck: poor article | 00:12 |
blackburn | rieck: ahh got it | 00:13 |
rieck | in many cases, the next msg is part of a conversation. thats why it seems to fit. actually it does not fit | 00:13 |
shogun-buildbot | build #166 of deb3 - modular_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/166 | 00:13 |
blackburn | nice idea | 00:13 |
rieck | for example, if you ask something general it works well (usually) | 00:13 |
rieck | droopy: what do you feel about this? | 00:13 |
droopy | rock on | 00:14 |
droopy | nothing? | 00:14 |
rieck | the engine fails if you ask something with real semantics. | 00:14 |
rieck | droopy: do you think that shogun will be bought by microsoft soon? | 00:14 |
droopy | rieck: its a conspiracy | 00:14 |
rieck | ;) | 00:14 |
rieck | that's luck | 00:15 |
blackburn | rieck: microsoft? oh no way | 00:15 |
rieck | blackburn: btw, do you know if something like kfa has been implemented in shogun yet? | 00:15 |
rieck | i have not followed the development, to be honest | 00:15 |
blackburn | sure I know - no :) | 00:15 |
blackburn | unfortunately or fortunately I know most of things about shogun now | 00:16 |
blackburn | rieck: do you need kfa for something? | 00:16 |
rieck | no. it seems like an old, forgotten idea from alex smola | 00:16 |
blackburn | sometimes I do something unrelated to get in shape | 00:17 |
rieck | i am wondering if it simply does not really work or is indeed a good idea to get a sparse variant of kpca | 00:17 |
blackburn | didn't you try? | 00:17 |
rieck | nope. testing is not so easy. i could write a simply matlab version, but there would be no real data to test. | 00:18 |
blackburn | I see | 00:18 |
rieck | in my research, usually the data is sequential (strings, etc.) and induces a high-dimensional vector space with millions of dimensions | 00:19 |
blackburn | rieck: actually I did try to embed strings as example of my 'toolbox' | 00:19 |
blackburn | that I've developed last year | 00:19 |
rieck | i am curious whether one can do an efficient dimension reduction on this type of data. | 00:19 |
blackburn | I even came up with kernel formulation of LTSA method | 00:20 |
blackburn | making possible to use string kernels | 00:20 |
blackburn | it is not very efficient though | 00:20 |
rieck | approaches like kpca, isomap, lle do not produce efficient models that can be applied with linear time complexity | 00:20 |
rieck | transition system analyzer? | 00:20 |
blackburn | rieck: http://shogun-toolbox.org/edrt/words.png that is the example | 00:21 |
rieck | interesting, but what is the run-time complexity of putting a new document into this space? | 00:21 |
blackburn | rieck: just like of search of k neighbors | 00:22 |
blackburn | so something log N | 00:22 |
rieck | whoah! | 00:22 |
rieck | that | 00:22 |
rieck | is slowww ;) | 00:22 |
blackburn | rieck: what is fast for you? | 00:22 |
rieck | so you need something like k-d trees, ball trees or cover trees to do this in high dimensions? | 00:22 |
blackburn | right | 00:23 |
blackburn | well if you have kernel you don't care about dimensionality I think | 00:23 |
rieck | i have not been very successful in getting these approaches to really work reliably fast | 00:23 |
blackburn | depends what is fast for you again | 00:23 |
blackburn | I can't see any way to do that faster than log N | 00:23 |
rieck | kernels work efficient if there are not used in a non-sparse kernel expansion. that is, you don't have to compute a kernel for a lot of points from the training data | 00:24 |
rieck | N is the dimension or the number of data points? | 00:24 |
blackburn | number of data points | 00:24 |
CIA-18 | shogun: Sergey Lisitsyn master * r53e4cff / (6 files in 3 dirs): Introduced L1 logistic regression - http://git.io/_VYbcA | 00:24 |
blackburn | well dimensionality will be here sure | 00:24 |
blackburn | hmm wait | 00:25 |
blackburn | it is not the only way | 00:25 |
blackburn | if we learn embedding function itself it would be faster for sure | 00:25 |
blackburn | but I can hardly imagine that | 00:26 |
blackburn | I have seen you have some software for that, how does it work? | 00:26 |
rieck | cover trees and friends guarantee O(m log N), where m is something like C^D | 00:26 |
rieck | and D is the dimensionality (intrinsic) | 00:26 |
rieck | all my software basically uses hash functions to map the data to a sparse and high-dimensional space. efficiency is then a result of sparsity | 00:27 |
gsomix | good nigth guys | 00:27 |
blackburn | I see | 00:27 |
blackburn | gsomix: nite | 00:28 |
blackburn | rieck:why to reduce dim anyway? | 00:28 |
rieck | i think it's one of the most interesting results from using kernel machines: (a) kernels don't care about dimensions; (b) kernel run-time does | 00:28 |
rieck | blackburn: some learning methods can deal with only < 100 dimensions. | 00:29 |
rieck | some of these methods are really cool and would be worth a try. | 00:29 |
rieck | if you apply kpca or something else it takes ages to convert your data to 100 dim. moreover, it will be impossible to do an on-line application | 00:29 |
rieck | that's why have been interested in kfa, which runs with something like O(Dd), where D is the intrinsic dimension of the orignal space and d the number of projected components | 00:30 |
rieck | kpca in contrast uses O(nDd) where n is the number of training data points | 00:30 |
blackburn | rieck: how long is matlab implementation? | 00:31 |
rieck | i have no idea. | 00:31 |
blackburn | rieck: is accelerated more interesting? | 00:32 |
rieck | as far as i recall, AKFA is just a clever implementation of KFA | 00:32 |
blackburn | I could try implement AKFA actually | 00:33 |
rieck | in short, kfa tries to perform a pca in the feature space, that is, finding the direction with highest variance. | 00:33 |
rieck | however, instead of determining this direction as a linear combination of the training data, kfa picks just a single data point that is closest to this direction | 00:34 |
rieck | things get a little tricky once you need to determine a second component, because kfa components (as pca components) need to be orthognonal. | 00:34 |
rieck | that's where afka does some smart tricks | 00:34 |
blackburn | algorithms looks pretty straightforward | 00:35 |
rieck | yep | 00:35 |
rieck | you can try to. however, keep in mind that i am running a research group on computer security and machine learning is not my main field of research anymore | 00:35 |
rieck | could be that kfa does not perform well in my applications and ... well, you did it just for the fun | 00:36 |
blackburn | I do everything for fun so I a not so worried :) | 00:36 |
rieck | ;) | 00:37 |
rieck | very good | 00:37 |
blackburn | are you doing only security now? | 00:37 |
rieck | i am doing security with machine learning | 00:37 |
rieck | the focus is security. machine learning is the tool. | 00:38 |
blackburn | I see | 00:38 |
blackburn | rieck: I love matrix stuff now with eigen3 :D | 00:39 |
rieck | eigen3? | 00:39 |
blackburn | yeah cool linalg library for C++ | 00:39 |
blackburn | well you probably know how ugly blas is | 00:40 |
rieck | yep | 00:40 |
blackburn | in eigen3 you just do x = a*y+b*z; | 00:40 |
blackburn | and it is vectorized using SSE if available | 00:40 |
rieck | it overloads all the operators? | 00:40 |
blackburn | yes | 00:41 |
rieck | okay. | 00:41 |
rieck | is there some python interface | 00:41 |
blackburn | well I think this approach is not applicable for python | 00:41 |
blackburn | it is optimizing compiler basically | 00:41 |
rieck | ok. but there could be some interface, e.g. using swig | 00:41 |
blackburn | I mean a*x+b*y+c*z is going to be optimized into one loop | 00:41 |
rieck | btw numpy does also overload operators but on a different level | 00:42 |
blackburn | yeah but you still going to lose that level of optimization | 00:42 |
rieck | i see. (compile-time vs. run-time optimization) | 00:42 |
blackburn | one great example - it doesn't compute extra matrix doing (A*B).trace() | 00:42 |
blackburn | it is not possible under python at all I am afraid | 00:43 |
rieck | yep | 00:43 |
rieck | that's a downside of interpretation | 00:43 |
droopy | understandable!?! | 00:44 |
rieck | hehe | 00:44 |
rieck | i am off. | 00:46 |
rieck | nice talking with you, blackburn | 00:46 |
blackburn | rieck: it is mutual | 00:46 |
blackburn | see you | 00:46 |
-!- bern4rd [4f940120@gateway/web/freenode/ip.79.148.1.32] has quit [Quit: Page closed] | 01:00 | |
shogun-buildbot | build #167 of deb3 - modular_interfaces is complete: Failure [failed test python_modular] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/167 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 01:21 |
CIA-18 | shogun: Sergey Lisitsyn master * r7565032 / (17 files in 3 dirs): Merged all SLEP solvers into one - http://git.io/n2KxDg | 03:11 |
-!- blackburn [~blackburn@109.226.92.17] has quit [Quit: Leaving.] | 03:16 | |
shogun-buildbot | build #32 of nightly_default is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_default/builds/32 | 03:17 |
shogun-buildbot | build #22 of nightly_all is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/nightly_all/builds/22 | 03:20 |
shogun-buildbot | build #157 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/157 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 03:20 |
-!- wiking [~wiking@huwico/staff/wiking] has quit [Ping timeout: 240 seconds] | 03:40 | |
-!- wiking [~wiking@info2k1.hu] has joined #shogun | 03:41 | |
gsomix | good morning | 07:10 |
gsomix | sonney2k, slice getting works. | 07:10 |
-!- gsomix [~gsomix@188.168.4.28] has quit [Ping timeout: 240 seconds] | 07:15 | |
droopy | tomatensuppe! | 08:03 |
-!- gsomix [~gsomix@r206-10.smr.ru] has joined #shogun | 08:26 | |
-!- droopy [~droopy@vmext164.gwdg.de] has quit [Quit: rieck] | 08:31 | |
-!- droopy [~droopy@vmext164.gwdg.de] has joined #shogun | 08:34 | |
-!- droopy [~droopy@vmext164.gwdg.de] has left #shogun [] | 08:34 | |
-!- droopy [~droopy@vmext164.gwdg.de] has joined #shogun | 08:35 | |
droopy | rieck: te whole room is hit | 08:35 |
-!- droopy [~droopy@vmext164.gwdg.de] has quit [Client Quit] | 08:36 | |
-!- droopy [~droopy@vmext164.gwdg.de] has joined #shogun | 08:42 | |
rieck | droopy: good to see you back | 08:42 |
droopy | rieck: hahaha | 08:42 |
-!- pluskid [~pluskid@111.120.62.38] has joined #shogun | 08:52 | |
-!- uricamic [~uricamic@2001:718:2:1634:e0a5:7bff:d55f:9553] has joined #shogun | 08:57 | |
-!- rieck [~rieck@paul.ml.tu-berlin.de] has left #shogun [] | 09:31 | |
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has joined #shogun | 10:22 | |
@sonney2k | gsomix, nice | 10:30 |
@sonney2k | droopy, so how is life? | 10:30 |
droopy | sonney2k: he is founder of hostgater bro | 10:31 |
gsomix | sonney2k, but I don't manage yet to how get [a:b, c:d] | 10:33 |
@sonney2k | gsomix, step by step... | 10:34 |
gsomix | aha, try to add getting item by [] | 10:35 |
gsomix | *trying | 10:35 |
@sonney2k | gsomix, in worst case we have to check how numpy / python do it themselves | 10:35 |
gsomix | sonney2k, so what I'm doing :) | 10:36 |
gsomix | numpy is really good python c/api manual | 10:36 |
@sonney2k | gsomix, hehe | 10:39 |
@sonney2k | gsomix, actually numpy had a array protocol first - it is them who proposed to define the python buffer protocol | 10:40 |
@sonney2k | that is why things are as they are | 10:40 |
@sonney2k | unfortunately these guys are astronomers so they deal with multidimensional (dense) arrays only | 10:41 |
@sonney2k | so no sparse stuff :/ | 10:41 |
@sonney2k | gsomix, so no idea about sparse and string features... | 10:41 |
yoo | all: which interface do you usually use ? | 10:43 |
yoo | I use libshogun c++ and I am wondering now howto easily plot my data (branching matplotlib ?) | 10:44 |
yoo | sonney2k: could you thx heiko for me: he solved the storing model bug. | 10:45 |
@sonney2k | yoo, I am using python modular most of the time | 10:51 |
yoo | compilation issue: make[1]: *** No rule to make target `../shogun/lib/slep/slep_logistic.h', needed by `classifier/FeatureBlockLogisticRegression.cpp.o'. Stop | 10:53 |
-!- hoijui [~hoijui@dslb-088-074-122-056.pools.arcor-ip.net] has joined #shogun | 10:54 | |
-!- rieck [~rieck@paul.ml.tu-berlin.de] has joined #shogun | 11:01 | |
@sonney2k | droopy, who the fsck is hostgater pro? | 11:04 |
droopy | guy from anonops hes cool | 11:05 |
droopy | zzz sonney2k | 11:05 |
@sonney2k | droopy, zzz indeed | 11:05 |
droopy | sonney2k: modelview matrix says "take the model coordinates and transform them by this before drawing" | 11:05 |
rieck | hey ;) | 11:05 |
rieck | chatting with dr**py | 11:06 |
@sonney2k | rieck, hey | 11:08 |
rieck | sonney2k: don't op droopy. i haven't fixed the kick thresholds, as i need them in another channel | 11:08 |
droopy | rieck! | 11:08 |
@sonney2k | so could you make all students happy? | 11:08 |
@sonney2k | droopy, you again! | 11:08 |
droopy | sonney2k! | 11:08 |
@sonney2k | !!! | 11:08 |
rieck | happy ... hmm depends | 11:08 |
rieck | we had an exam yesterday, average is 2.7 (german grades) | 11:09 |
rieck | i think many are not happy | 11:09 |
@sonney2k | rieck, wow you are tough | 11:09 |
gsomix | sonney2k, ok, getting item from DenseFeatures works now. | 11:10 |
gsomix | [a:b, c:d] next | 11:10 |
@sonney2k | pluskid, could you please git checkout github.com:shogun-toolbox/shogun-web.git and give some feedback? | 11:11 |
@sonney2k | gsomix, cool :) | 11:11 |
-!- blackburn [~blackburn@109.226.92.17] has joined #shogun | 11:26 | |
rieck | my first written exam was far too easy | 11:38 |
rieck | it's clear that the next time i moved to the other extreme | 11:39 |
CIA-18 | shogun: Heiko Strathmann master * r1e888a0 / (19 files in 5 dirs): Merge pull request #649 from puffin444/master (+8 more commits...) - http://git.io/o5YgbQ | 11:46 |
-!- heiko [~heiko@host86-177-183-33.range86-177.btcentralplus.com] has joined #shogun | 11:47 | |
heiko | sonney2k, around? | 11:48 |
gsomix | sonney2k, btw, what do you think about my_cool_array[a, b]? how it may works? maybe any ideas... | 11:48 |
shogun-buildbot | build #158 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/158 | 11:48 |
shogun-buildbot | build #159 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/159 blamelist: puffin444 <walke434@msu.edu>, Heiko Strathmann <heiko.strathmann@gmail.com> | 11:49 |
CIA-18 | shogun: Sergey Lisitsyn master * r5712c9a / (2 files): Fixed includes in logistic regression classes - http://git.io/Ds-BKg | 11:51 |
blackburn | droopy: hey | 11:53 |
droopy | so could you make all students happy? | 11:53 |
blackburn | droopy: are you happy? | 11:55 |
droopy | blackburn: back o/ | 11:55 |
-!- bern4rd [4f940120@gateway/web/freenode/ip.79.148.1.32] has joined #shogun | 12:14 | |
-!- fabs [~fabs@h1749472.stratoserver.net] has joined #shogun | 12:20 | |
shogun-buildbot | build #168 of deb3 - modular_interfaces is complete: Failure [failed test libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/168 blamelist: puffin444 <walke434@msu.edu> | 12:20 |
gsomix | sonney2k, oh no, it seems, that numpy.array works as map, where key is [a:b, c:d]. | 12:21 |
gsomix | bad :) | 12:22 |
gsomix | droopy, hi | 12:22 |
droopy | hi!?! | 12:22 |
blackburn | shogun-buildbot: you bastard | 12:23 |
shogun-buildbot | build #145 of deb2 - static_interfaces is complete: Failure [failed test libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/145 blamelist: puffin444 <walke434@msu.edu> | 12:23 |
shogun-buildbot | build #160 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/160 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 12:23 |
pluskid | sonney2k: awesome logo on the "download now" box! | 12:27 |
pluskid | could we make it look more 3D? | 12:27 |
CIA-18 | shogun: Sergey Lisitsyn master * r46b4d2c / (2 files in 2 dirs): Updated one more include - http://git.io/16Pwbw | 12:31 |
shogun-buildbot | build #161 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/161 | 12:33 |
rieck | where is the logo? could not see it | 12:34 |
shogun-buildbot | build #146 of deb2 - static_interfaces is complete: Failure [failed test libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/146 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 12:35 |
rieck | droopy: i hope you don't learn from shogun-buildbot. | 12:36 |
droopy | rieck: i recall that somebody's holidays start by the end of july hehe | 12:36 |
rieck | droopy: yours? | 12:37 |
droopy | rieck: just got online. | 12:37 |
gsomix | droopy, vodka? | 12:37 |
droopy | gsomix: excellent :) | 12:37 |
rieck | he's the best! | 12:39 |
-!- bern4rd [4f940120@gateway/web/freenode/ip.79.148.1.32] has quit [Quit: Page closed] | 12:51 | |
-!- blackburn [~blackburn@109.226.92.17] has quit [Quit: Leaving.] | 12:52 | |
shogun-buildbot | build #169 of deb3 - modular_interfaces is complete: Failure [failed test libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/169 blamelist: Sergey Lisitsyn <lisitsyn.s.o@gmail.com> | 13:07 |
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has joined #shogun | 14:13 | |
vojtech | rieck: Hi, sorry that didn't answer yesterday I just close my browser without looking to IRC tab. | 14:20 |
rieck | no problem | 14:21 |
rieck | i am in a meeting right now :( | 14:21 |
vojtech | no problem :) | 14:21 |
vojtech | I just came from one :) | 14:22 |
droopy | hoho. | 14:22 |
yoo | heiko: I have to go, lets see if we have time to fix this tonight. here is the c++ rewrited code http://pastebin.com/c7wKje0H | 15:27 |
yoo | heiko: ++ | 15:27 |
heiko | yoo, alright! | 15:28 |
-!- gsomix [~gsomix@r206-10.smr.ru] has quit [Remote host closed the connection] | 15:53 | |
-!- pluskid [~pluskid@111.120.62.38] has quit [Quit: Leaving] | 16:17 | |
-!- uricamic [~uricamic@2001:718:2:1634:e0a5:7bff:d55f:9553] has quit [Quit: Leaving.] | 16:35 | |
-!- rieck is now known as sub-zero | 17:03 | |
-!- gsomix [~gsomix@188.168.3.115] has joined #shogun | 17:04 | |
-!- sub-zero is now known as Guest78654 | 17:04 | |
-!- Guest78654 is now known as rieck | 17:07 | |
-!- hoijui [~hoijui@dslb-088-074-122-056.pools.arcor-ip.net] has quit [Quit: Leaving] | 17:17 | |
-!- bern4rd [4f940120@gateway/web/freenode/ip.79.148.1.32] has joined #shogun | 17:23 | |
-!- vojtech [9320543b@gateway/web/freenode/ip.147.32.84.59] has quit [Quit: Page closed] | 18:14 | |
-!- blackburn [~blackburn@109.226.92.17] has joined #shogun | 18:15 | |
blackburn | heiko: that's cool you improved tutorial | 18:20 |
heiko | blackburn, yeah, Ill continue this week, did more structure things today, but tomorrow I will actually write stuff | 18:21 |
blackburn | heiko: default borders are pretty wide | 18:22 |
heiko | blackburn, true, but these are based on typographic rules | 18:23 |
heiko | and they look much nicer ,are easier to read etc | 18:23 |
heiko | do you know the book elements of typographic style? | 18:23 |
blackburn | what do you mean? | 18:23 |
heiko | theres a book about typo | 18:23 |
heiko | its worth reading has lots of interesting tips in it | 18:24 |
blackburn | well I've read a lot of them actually :) | 18:24 |
blackburn | but still these borders are way too wide for me | 18:24 |
heiko | I mean we can change it back, was just my suggestion. I think the layout can be done later anyway. | 18:24 |
heiko | But I vote for large borders :D | 18:25 |
blackburn | most confusing is bottom margin | 18:25 |
blackburn | it is something like 5 cm probably | 18:25 |
heiko | well I dont insist on large margins :) | 18:26 |
blackburn | it all depends how do you treat it | 18:27 |
blackburn | nobody will print it probably | 18:27 |
heiko | thats true | 18:28 |
heiko | its also a matter of taste | 18:28 |
blackburn | and actually we need no fancy typographic if it is screen only | 18:29 |
blackburn | we would even need sans fonts for that | 18:29 |
heiko | mmh | 18:30 |
heiko | I dont agree, why typesetting it if its not for printing? Then we could write a webpage or similar | 18:31 |
blackburn | yeah I am asking this question too but can't say someone would print it :) | 18:31 |
blackburn | that's the problem | 18:31 |
heiko | I would print it | 18:32 |
heiko | like printed stuff so much more | 18:32 |
blackburn | ok then :) | 18:32 |
heiko | Also, this could be a nice long term goal | 18:32 |
heiko | printing a book | 18:32 |
heiko | sell hard-copies :) | 18:33 |
blackburn | ambitious | 18:33 |
-!- yoo [2eda6d52@gateway/web/freenode/ip.46.218.109.82] has quit [Quit: Page closed] | 18:35 | |
heiko | blackburn, I forgot how to enable lines and files on errors messages | 18:38 |
blackburn | io.enable_line_and_file | 18:39 |
blackburn | or vice versa in line and file | 18:39 |
heiko | thanks :) | 18:39 |
blackburn | heiko: includes should go to .cls file btw | 18:46 |
blackburn | usepackage I mean | 18:46 |
gsomix | good evening | 18:46 |
heiko | blackburn, didnt I do that? | 18:47 |
blackburn | you put it into .tex | 18:47 |
blackburn | well actually | 18:47 |
blackburn | we can do it in .tex | 18:47 |
blackburn | but then we rather should remove cls and just use book | 18:47 |
heiko | I though Ive put everything into cls | 18:50 |
@sonney2k | bern4rd, have you seen? I updated the webpage with some logo stuff... | 19:07 |
@sonney2k | gsomix, so slicing works just for 1 dimensional stuff now? | 19:08 |
@sonney2k | 'just' | 19:08 |
gsomix | sonney2k, as for python lists | 19:09 |
@sonney2k | same thing ok | 19:09 |
bern4rd | yes I have seen it, really cool | 19:09 |
gsomix | sonney2k, numpy's slicing a little harder for realisation. | 19:09 |
@sonney2k | blackburn, heiko, wiking koen-shogun everyone actually - if you have cool figures produced w/ shogun I would add them to the new website :) | 19:10 |
bern4rd | I am thinking what to do, but most of stuff is style stuff or check the content of the articles, links, etc. If you have any idea let me know. =) | 19:10 |
@sonney2k | gsomix, can you use python buffer protocol for that? | 19:10 |
blackburn | sonney2k: well shogun is not plotting library ;) | 19:10 |
heiko | sonney2k, sorry I currently dont :( | 19:11 |
gsomix | sonney2k, nope, it's not needed. just mapping protocol. | 19:11 |
@sonney2k | gsomix, so it is some typemap? | 19:11 |
gsomix | sonney2k, no, it's mechanism like other protocol for dictionary[key], where key is PyObject (e.g. [a:b] type for slicing) | 19:12 |
gsomix | *other protocols | 19:12 |
@sonney2k | gsomix, ahh I see | 19:13 |
@sonney2k | bern4rd, thinking... | 19:13 |
gsomix | sonney2k, I'm working now on get/set for DenseFeatures. and doing some fixes. ok? | 19:13 |
@sonney2k | bern4rd, actually automagically generated download links for http / ftp based on NEWS | 19:14 |
bern4rd | what you want to download? I mean you want to download each new, or the shogun version related with each new? I don't know if i am misunderstanding what you mean. | 19:19 |
@sonney2k | bern4rd, yes the shogun version related to each news item | 19:20 |
@sonney2k | so shogun1.0.0.tar.bz2 etc | 19:20 |
@sonney2k | please have a look at the news box at http://www.shogun-toolbox.org/ for the download links | 19:21 |
@sonney2k | bern4rd, actually that thing is called "Download Releases" on the old website | 19:22 |
bern4rd | I see | 19:23 |
bern4rd | ok I will try, I am going to read more about how to download links work | 19:24 |
@sonney2k | bern4rd, well they are just links that are based on the version number | 19:26 |
bern4rd | and the url would be the same? ftp://shogun-toolbox.org/shogun/releases/ | 19:28 |
@sonney2k | bern4rd, yes we will just use the very same http:// and ftp:// urls | 19:30 |
@sonney2k | bern4rd, the only other dynamic content is this spreadsheet: http://spreadsheets.google.com/ccc?key=0Aunb9cCVAP6NdDVBMzY1TjdPcmx4ei1EeUZNNGtKUHc&hl=en | 19:33 |
@sonney2k | this is what is under Feature matrix on the old website | 19:33 |
@sonney2k | I again have some python script to generate that | 19:33 |
@sonney2k | gsomix, ok I see | 19:34 |
bern4rd | ah ok, the huge matrix i didnt add to the django project. | 19:35 |
@sonney2k | yeah too much to do manually | 19:36 |
bern4rd | yes =), so I can suppose that the al the tar.bz2 files are going to be in the url i wrote before | 19:37 |
@sonney2k | bern4rd, yeah I guess - just copy the urls from the releases on the current page to know it | 19:39 |
bern4rd | and should I add the download links next to each new? | 19:39 |
@sonney2k | bern4rd, yes makes sense | 19:40 |
bern4rd | oks | 19:41 |
@sonney2k | bern4rd, maybe rename news to releases | 19:48 |
bern4rd | ok | 19:49 |
blackburn | sonney2k: will show you some svm hyperplane classic fig soon | 19:55 |
blackburn | I had pretty nice one in my thesis | 19:55 |
@sonney2k | blackburn, can you now look at the website? | 19:55 |
blackburn | sonney2k: no I didn't manage to run and never got back | 19:55 |
droopy | yo | 19:55 |
-!- bern4rd [4f940120@gateway/web/freenode/ip.79.148.1.32] has quit [Ping timeout: 245 seconds] | 19:56 | |
CIA-18 | shogun: Heiko Strathmann master * r01d6292 / (14 files in 5 dirs): Merge pull request #658 from karlnapf/master (+7 more commits...) - http://git.io/Ce_EfA | 19:59 |
shogun-buildbot | build #162 of deb1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/162 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 20:03 |
heiko | yippieh! | 20:03 |
heiko | oh these arent mine | 20:04 |
heiko | slep causes error | 20:04 |
shogun-buildbot | build #163 of deb1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb1%20-%20libshogun/builds/163 | 20:04 |
shogun-buildbot | build #170 of deb3 - modular_interfaces is complete: Failure [failed compile libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/170 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 20:05 |
shogun-buildbot | build #64 of bsd1 - libshogun is complete: Failure [failed compile] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/64 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 20:10 |
blackburn | relax I am working on it | 20:10 |
blackburn | sonney2k: https://dl.dropbox.com/u/10139213/shogun/hyperplane.png what about something like that? | 20:12 |
blackburn | looks like science lol | 20:12 |
shogun-buildbot | build #65 of bsd1 - libshogun is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/bsd1%20-%20libshogun/builds/65 | 20:16 |
shogun-buildbot | build #171 of deb3 - modular_interfaces is complete: Failure [failed test libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/171 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 20:36 |
shogun-buildbot | build #147 of deb2 - static_interfaces is complete: Failure [failed test libshogun] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/147 blamelist: Heiko Strathmann <heiko.strathmann@gmail.com> | 20:39 |
CIA-18 | shogun: Sergey Lisitsyn master * r53d384e / (27 files in 7 dirs): Refactoring of multitask submodule - http://git.io/_V6jZA | 21:18 |
blackburn | shogun-buildbot: lets see success now | 21:19 |
CIA-18 | shogun: Sergey Lisitsyn master * r20c8bce / examples/undocumented/libshogun/Makefile : Removed duplicated transfer learning examples targeting - http://git.io/1yDTeQ | 21:29 |
-!- puffin444 [62e3926e@gateway/web/freenode/ip.98.227.146.110] has joined #shogun | 21:50 | |
puffin444 | heiko, quick question. | 21:58 |
shogun-buildbot | build #172 of deb3 - modular_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb3%20-%20modular_interfaces/builds/172 | 22:02 |
shogun-buildbot | build #148 of deb2 - static_interfaces is complete: Success [build successful] Build details are at http://www.shogun-toolbox.org/buildbot/builders/deb2%20-%20static_interfaces/builds/148 | 22:08 |
@sonney2k | blackburn, is that a fancy figure? | 22:10 |
blackburn | sonney2k: not really but still a figure | 22:10 |
@sonney2k | blackburn, how about all your dimred / embedding stuff | 22:11 |
-!- romi_ [~mizobe@200-232-221-127.dsl.telesp.net.br] has joined #shogun | 22:12 | |
@sonney2k | or road signs? | 22:12 |
blackburn | sonney2k: what to do with road signs? | 22:12 |
@sonney2k | embed them? or just show classification rule... | 22:15 |
@sonney2k | with some signs left and right... | 22:15 |
@sonney2k | it is becoming more art then science but hey who cares | 22:16 |
@sonney2k | maybe we could run some benchmarks (speed) to show how fast certain things are too | 22:16 |
* audy is boosting | 22:20 | |
audy | http://d.pr/i/pZTF+#.png | 22:21 |
audy | How does that look? | 22:21 |
blackburn | something is wrong with 'ksvm' and 'glm' :) | 22:23 |
audy | blackburn they are tarded | 22:24 |
audy | KSVM works really well normally. I don't know what happened here | 22:24 |
blackburn | rieck: ping | 22:29 |
rieck | if auc < 0.5 ; swap labels of prediction | 22:33 |
blackburn | rieck: do you know internals of AKFA? | 22:35 |
blackburn | I got stucked a little with one step here | 22:35 |
rieck | to be honest, no. | 22:35 |
blackburn | ok, will dig into it a little later then | 22:35 |
blackburn | funny no implementation is available just anywhere | 22:35 |
audy | rieck I know. I'm messing around in Rattle so I have no idea what's going on behind the scenes. | 22:37 |
rieck | audy: i have no idea what you are doing. but auc < 0.5 is often an indication of swapped labels. | 22:37 |
rieck | audy: some classifiers like libsvm/liblinear tend to swap labels, as they pick +1 of the svm from the first data point. | 22:39 |
audy | I'm pretty sure KSVM swaps the labels b/c I've never seen less than 50% accuracy | 22:39 |
rieck | accuray? you mean auc | 22:39 |
rieck | area under roc | 22:39 |
audy | rieck I mean accuracy of prediction. If the prediction accuracy is less than 50%, then you can just swap the labels | 22:42 |
audy | and get a better performing predictor | 22:42 |
blackburn | for binary yes | 22:42 |
rieck | audy: hmm. let's get things straight. | 22:42 |
rieck | if the class distribution is 50/50, you are right. | 22:43 |
rieck | in any case, you need to look at the auc and not the accuracy | 22:43 |
rieck | i think shogun features some code to compute aucs for roc and pr curves | 22:47 |
@sonney2k | rieck, written by blackburn ... | 22:47 |
blackburn | sonney2k: is it bad or good? | 22:48 |
@sonney2k | blackburn, the best that could happen - well maintained code :) | 22:49 |
droopy | grin | 22:49 |
blackburn | sonney2k: it is tested and compared a few times works like charm so no insults :) | 22:51 |
rieck | even droopy thinks this is funny | 22:52 |
droopy | we're gonna be using that 2600 chan as a way of correlating data from anonops' past/wikis | 22:52 |
@sonney2k | droopy, cheer up blackburn! | 22:53 |
droopy | sonney2k: but we did not do anything illegal, such s ddos a company or taking private info i am trying to help you guys, | 22:53 |
@sonney2k | wiking, what happened? you are gone again? | 22:53 |
gsomix | droopy, ?ome to drink vodka. :) I've already bought. | 22:59 |
droopy | gsomix: it is wild :) | 22:59 |
@sonney2k | droopy, cheer up blackburn! | 22:59 |
@sonney2k | wiking, what happened? you are gone again? | 22:59 |
droopy | will i still have to deal with the angle problem from before? | 22:59 |
rieck | droopy: no, you have voda. no angle problem | 23:00 |
droopy | rieck: yeah i have an access log! | 23:00 |
gsomix | good night guys | 23:06 |
blackburn | droopy: do you have access log? | 23:06 |
droopy | what is kick banned? | 23:06 |
gsomix | droopy, nite | 23:06 |
droopy | evil[1]'s from germany | 23:06 |
@sonney2k | gsomix, nite and can I have sweet dreams about PRs for tomorrow? | 23:07 |
gsomix | sonney2k, what do you want in PR? :) | 23:08 |
@sonney2k | slicing for densefeatures :D | 23:08 |
@sonney2k | that's all :D | 23:08 |
gsomix | sonney2k, like numpy slicing? | 23:08 |
@sonney2k | yeah | 23:08 |
gsomix | ok, I do not want to sleep :) | 23:09 |
@sonney2k | gsomix, what? | 23:09 |
gsomix | time to work | 23:10 |
@sonney2k | gsomix, ??? sleep and do it tomorrow! | 23:10 |
gsomix | :( | 23:11 |
gsomix | okay | 23:11 |
gsomix | sonney2k, nite. it was not a funny joke, perhaps. | 23:12 |
@sonney2k | gsomix, no way! work now! | 23:13 |
gsomix | huh :) | 23:14 |
droopy | haha | 23:14 |
rieck | droopy: time for a last beer! | 23:43 |
droopy | sure | 23:43 |
rieck | !order beer for everybody | 23:43 |
* droopy sees: rieck orders a round of beer for everybody. | 23:43 | |
* droopy sees: The waitress arrives and serves a round of beer for everybody. | 23:44 | |
@sonney2k | droopy, order vodka for blackburn. | 23:49 |
droopy | they should tell the fbi so they can raid him | 23:49 |
@sonney2k | beer is just not enough for RUssians. | 23:49 |
blackburn | sonney2k: well 5 liters would be enough | 23:50 |
@sonney2k | droopy, cookies! | 23:50 |
droopy | sonney2k: i wouldn't suggest it just yet we could use more of his shit. | 23:50 |
@sonney2k | nite folks | 23:51 |
@sonney2k | droopy, sweet dreams! | 23:51 |
droopy | well it's not too far from mexico city but it's too out of way from ny | 23:51 |
blackburn | good night of logistic regression solvers | 23:51 |
blackburn | (for me) | 23:51 |
blackburn | :D | 23:51 |
--- Log closed Thu Jul 19 00:00:17 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!