--- Log opened Tue Jan 24 00:00:19 2012 | ||
-!- Netsplit *.net <-> *.split quits: @sonney2k | 09:19 | |
-!- Netsplit over, joins: sonney2k | 09:22 | |
-!- Ram108 [~amma@14.96.135.89] has joined #shogun | 15:30 | |
-!- Ram108 [~amma@14.96.135.89] has quit [Ping timeout: 240 seconds] | 16:55 | |
-!- blackburn [~qdrgsm@31.28.63.56] has joined #shogun | 17:02 | |
-!- Ram108 [~amma@115.117.244.206] has joined #shogun | 17:22 | |
-!- Ram108 [~amma@115.117.244.206] has quit [Ping timeout: 248 seconds] | 18:03 | |
-!- Ram108 [~amma@14.99.66.58] has joined #shogun | 18:16 | |
blackburn | sonney2k: will you sign against elsevier? :) | 20:54 |
---|---|---|
blackburn | I would protest about every non-free shit if I was a scientist hah | 20:54 |
-!- mode/#shogun [+o sonney2k] by ChanServ | 21:44 | |
-!- Ram108 [~amma@14.99.66.58] has quit [Ping timeout: 244 seconds] | 22:53 | |
@sonney2k | blackburn, I thoroughly hate all this non-open access crap | 23:16 |
blackburn | wow you are here | 23:17 |
blackburn | :) | 23:17 |
@sonney2k | Nowadays I consider it good policy to not review nor publish for / in such journals | 23:17 |
@sonney2k | blackburn, yeah but will go to bed soon | 23:17 |
blackburn | sonney2k: I did not commited any idea yet sorry | 23:18 |
@sonney2k | blackburn, did gsoc 2012 already get announced? | 23:18 |
blackburn | no no | 23:18 |
blackburn | not yet | 23:18 |
@sonney2k | I was working on the debian packages (the last days...) | 23:18 |
blackburn | for shogun? | 23:18 |
@sonney2k | uploaded like 4 revisions... | 23:18 |
@sonney2k | yeah | 23:18 |
blackburn | nice | 23:19 |
@sonney2k | weird errors / warnings with clang... | 23:19 |
@sonney2k | https://buildd.debian.org/status/package.php?p=shogun | 23:19 |
blackburn | sonney2k: is there any way to automatize it?? | 23:19 |
@sonney2k | I don't understand why it fails on powerpc. | 23:19 |
@sonney2k | blackburn, well we could have a buildbot building nightly debian packages for debian unstable | 23:19 |
blackburn | we could or we can? | 23:20 |
@sonney2k | but that will usually fail due to whatever stuff is broken in debian unstable :) | 23:20 |
@sonney2k | it is a matter of typing one command on the cmdline but still work | 23:20 |
blackburn | sonney2k: did I told you accuracy on these road signs? | 23:21 |
@sonney2k | blackburn, at some point yes but no idea where you are now | 23:21 |
blackburn | well stucked on 96% | 23:21 |
@sonney2k | did you already use virtual images? | 23:21 |
blackburn | no, not really | 23:22 |
blackburn | I don't really want to use it | 23:22 |
@sonney2k | like slight rotation / translation | 23:22 |
@sonney2k | that will improve another few percent | 23:22 |
blackburn | I'm trying to find a way to localize images a little | 23:22 |
blackburn | there is a bad background | 23:22 |
@sonney2k | blackburn, regarding GSoC | 23:22 |
blackburn | my descriptor is gradient based so I need to clean up images | 23:22 |
@sonney2k | I think we should change the old website quite a bit | 23:23 |
blackburn | yeah we def. should | 23:23 |
@sonney2k | I mean I would want to have very clear focus | 23:23 |
@sonney2k | not thousands of ideas like last time | 23:23 |
blackburn | I would want to work on two things | 23:23 |
blackburn | + 10 more dim reduction algos | 23:24 |
blackburn | or ECOC stuff | 23:24 |
blackburn | well I will have much time so I guess actually I will work on everything | 23:24 |
@sonney2k | What I could imagine is: a) structured output learning b) multiclass learning c) input / output stuff d) cross-validation e) merging of SVM solvers f) opencv integration ... | 23:25 |
@sonney2k | it sounded like chris wants to mentor domain adaptation algorithms, we might have a mentor for gaussian processes too | 23:28 |
blackburn | what the f is domain adaptation? | 23:29 |
@sonney2k | learn on data A but apply the method to dataset B (which has a slighly different statistics - so adjust for that) | 23:30 |
blackburn | well I would be interested too | 23:30 |
blackburn | hahh | 23:30 |
@sonney2k | we have stuff like this already in there | 23:31 |
@sonney2k | blackburn, this year is much easier for shogun | 23:34 |
@sonney2k | we have buildbots and a better structure | 23:34 |
blackburn | yes, much better | 23:34 |
@sonney2k | so all these things are just more or less minor extensions | 23:34 |
@sonney2k | that all integrate rather well | 23:34 |
blackburn | and at least I will help you much ;) | 23:35 |
@sonney2k | I guess our chances to get in are not so bad given that chris di bona liked my summary :) | 23:35 |
blackburn | hah yes, we had some fame | 23:36 |
@sonney2k | I wish we would find someone to do a) structured output learning b) multiclass stuff (I am particularly interested in this - the nips ecoc/tree paper sounded fast + accurate) | 23:37 |
blackburn | my gf bachelors thesis would be about ecoc hah | 23:37 |
blackburn | work/thesis/no idea how to call it properly | 23:38 |
@sonney2k | I also would wish cross validation could get some syntactic sugar and more wide use | 23:38 |
@sonney2k | I hope Vojtech will mentor again this year | 23:39 |
blackburn | what would he mentor? | 23:39 |
@sonney2k | he could really help to get all these SVM solvers in | 23:39 |
@sonney2k | he is the ueberexpert there | 23:39 |
blackburn | in SVM solvers? | 23:40 |
@sonney2k | yes - he is the author of Ocas / GMNPSVM, ... | 23:40 |
blackburn | sonney2k: I still feel unhappiness mixed with depression when looking into svm folder | 23:40 |
@sonney2k | that won't change. way to difficult/time consuming to rewrite these things | 23:41 |
blackburn | I hope I would see any progress in my life hahah | 23:41 |
CIA-10 | shogun: Soeren Sonnenburg master * r210ae0a / (4 files in 3 dirs): fix various errors/warnings occurring with clang - http://git.io/OMcgNg | 23:41 |
@sonney2k | ahh btw, some nice tutorial would be a *great* task too | 23:41 |
blackburn | I hope 40 more years would be enough | 23:41 |
@sonney2k | so applications etc | 23:42 |
blackburn | I believe student can't write nice tutorial.. | 23:42 |
@sonney2k | why not? | 23:42 |
blackburn | if it wasn't phd student = expert in machine learning | 23:42 |
@sonney2k | task could be to define some nice data set based on mentor input | 23:42 |
@sonney2k | then run all alogirhtms | 23:43 |
@sonney2k | algorithms on that | 23:43 |
@sonney2k | with nice images | 23:43 |
blackburn | well only if he was really educated in ml | 23:43 |
@sonney2k | and write text + formulas that he can find in the code / papers | 23:43 |
@sonney2k | last year we had such ML experts | 23:43 |
blackburn | who? | 23:43 |
@sonney2k | we didn't accept them | 23:44 |
blackburn | ah | 23:44 |
@sonney2k | PhD students... | 23:44 |
@sonney2k | not sure how many students we can mentor this year | 23:45 |
@sonney2k | probably around 5 again - not more | 23:45 |
@sonney2k | less is better if we find really good ones | 23:45 |
@sonney2k | of your type btw :) | 23:45 |
blackburn | hah thanks | 23:46 |
* blackburn is playing with opencv | 23:46 | |
@sonney2k | btw parallelization is also sth | 23:47 |
@sonney2k | but I guess should be done by me | 23:47 |
@sonney2k | not a student | 23:47 |
blackburn | sonney2k: no, parallelization should not be done by student | 23:47 |
blackburn | he would stuck and etc | 23:47 |
@sonney2k | yeah, btw boosting algos are still missing from shogun | 23:47 |
@sonney2k | I would love to have some of these too | 23:47 |
blackburn | and I consider doing vectorization not parallelization.. | 23:48 |
blackburn | but I still have no machine with good graphics | 23:48 |
blackburn | don't really want to buy new one for that :D | 23:48 |
@sonney2k | I am not a big fan of this GPU stuff yet | 23:50 |
@sonney2k | just have no applications where GPUs really could help | 23:50 |
blackburn | sonney2k: well my stuff would run much faster with gpus.. | 23:50 |
@sonney2k | ok I think we have a much better picture now what we could do | 23:51 |
@sonney2k | so it is easy to write something up | 23:51 |
@sonney2k | blackburn, yeah you have some quadratic/cubic algorithms in there | 23:51 |
@sonney2k | that certainly helps | 23:51 |
@sonney2k | and all nicely gpu-parallelizable | 23:51 |
blackburn | sonney2k: let multiclass stuff be done by me in feb-apr | 23:51 |
blackburn | I would need it anyway | 23:52 |
@sonney2k | multiclass stuff involvement can start with simple stuff but can be arbitrarily tough | 23:52 |
@sonney2k | so a student could as well do the harder stuff | 23:52 |
blackburn | sonney2k: well I really need it so I'll try to get it done even in feb | 23:53 |
blackburn | including NIPS things | 23:53 |
blackburn | I'm pretty scaried with it but I hope to get through hah | 23:54 |
@sonney2k | hehe | 23:54 |
blackburn | sonney2k: for me would be nice to work with chris, we could write one more article :D | 23:54 |
@sonney2k | looking forward to this code - I will probably be your first user | 23:54 |
@sonney2k | there are some ecoc libraries around already though | 23:55 |
blackburn | yes I am aware of it | 23:55 |
@sonney2k | http://mloss.org/software/view/238/ | 23:55 |
blackburn | sonney2k: I was thinking about generic ensembles and generic multiclass machines | 23:56 |
@sonney2k | that will involve quite some refactoring | 23:56 |
blackburn | it should be done really *generic* | 23:56 |
blackburn | heh | 23:56 |
@sonney2k | we have general OVO / OVR in MulticlassSVM | 23:56 |
@sonney2k | at least for classify | 23:56 |
@sonney2k | so one could probably get rid of LibSVMMultiClass or so | 23:57 |
blackburn | no, I mean something like constructing trees manually | 23:57 |
@sonney2k | for training (which just does OVO) | 23:57 |
blackburn | ecocs, etc | 23:57 |
blackburn | and not only svm | 23:57 |
@sonney2k | blackburn, the code is not just for SVM | 23:57 |
@sonney2k | it is only encapsulated in MulticlassSVM | 23:58 |
blackburn | sonney2k: one more idea! | 23:58 |
blackburn | regressors | 23:58 |
blackburn | a lot of | 23:58 |
@sonney2k | just because we had no multiclassmachine | 23:58 |
@sonney2k | true | 23:58 |
@sonney2k | linear regression is totally missing | 23:58 |
@sonney2k | and these are quite some low hanging fruits | 23:58 |
blackburn | but I believe we should be 'minimized' | 23:59 |
blackburn | we should not implement everything.. | 23:59 |
blackburn | mahalanobis distance could be gsoc-entering task hah | 23:59 |
--- Log closed Wed Jan 25 00:00:03 2012 |
Generated by irclog2html.py 2.10.0 by Marius Gedminas - find it at mg.pov.lt!