17:50 < kirizaki> Hi

18:23 < kirizaki> i have some issue with Eclipse CDT, when i include libraries of mlpack/core.hpp my PC processors are going crazy, i mean, ~100% of usage

18:27 < kirizaki> i want to test it via "nano" & compile in terminal, but i think it is some Eclipse issue

18:53 < rcurtin> kirizaki: unfortunately I don't know enough about Eclipse to help

18:53 < rcurtin> do you think maybe it is trying to build an autocompletion dictionary or something for all of the mlpack classes?

18:53 < kirizaki> i already found answer, its some old eclipse bug with indexer

18:54 < rcurtin> ah, yeah, indexer, I was sort of close with my idea :)

18:54 < kirizaki> xD

19:05 < kirizaki> yes, for sure its indexer, cause after disabling it its working normally and there are some solutions for this but no time for this, gotta learn MLPACK :)

20:13 < travis-ci> mlpack/mlpack#335 (master - e09d6a4 : ryan): The build passed.

20:13 < travis-ci> Change view : https://github.com/mlpack/mlpack/compare/1eb721c663e6...e09d6a4869ed

20:13 < travis-ci> Build details : https://travis-ci.org/mlpack/mlpack/builds/89284382

21:43 < klyons> Hi all, I'm interested in contributing to mlpack

21:43 < klyons> in particular if there is still useful work to do along the lines of the ANN project mentioned in the list of summer of code projects for this past year

21:44 < rcurtin> klyons: hello there! lots of interest in neural networks :) zoq would be the person to talk to, though

21:44 < rcurtin> I can help you get started with contributing; there might be some simple bugs that you could fix to get an idea of how the library is laid out

21:45 < klyons> that's usually a better way to get moving?

21:45 < rcurtin> it depends on your preferences, really

21:45 < klyons> I've never contributed to a project this large before, I'm a physics guy up to this point

21:46 < rcurtin> ah, okay

21:46 < klyons> But I'm very interested in the algorithms in particular

21:46 < klyons> Well physics and applied math

21:46 < rcurtin> it'd probably be worth taking a look at some of the code that's in there, just to get an idea of the coding style, commenting conventions, and so forth

21:46 < rcurtin> here is a filter on the github issues for "easy" issues, but based on what I'm piecing together of your background some of these might be too easy and uninteresting: https://github.com/mlpack/mlpack/issues?q=is%3Aopen+is%3Aissue+label%3A%22D%3A+easy%22

21:47 < rcurtin> the project itself isn't that big; for the most part zoq, stereomatchingkiss (sometimes seen in here), and I are the main contributors at the moment

21:48 < rcurtin> some amount of coordination is necessary between us, of course, but it's not like firefox or any of the massive projects with huge barriers to entry and so forth

21:48 < rcurtin> because we're a small project, though, it's important that all code is documented well and tested well, to reduce time spent fixing bugs or figuring out code during later maintenance

21:49 < klyons> Are there other areas related to specific algorithms that are in particular need of help compared to say ANNs?

21:50 < klyons> I am generally interested in ML algos and am probably going to make a career shift in that direction soon

21:51 < klyons> ANNs are just one that I've found particularly interesting, but I'd prefer be useful than not

21:53 < klyons> understood by the way on the testing and documenting

21:53 < klyons> I've spent much of the past year learning how to not write code like a physicist

21:53 < rcurtin> sorry, I stepped out for a second...

21:53 < klyons> no worries at all

21:54 < rcurtin> I'm trying to think of algorithms that need help... there are a couple bugs open under the "impossible" difficulty if you really want to get your hands dirty and dig in

21:54 < rcurtin> but I would definitely not recommend that. some of those I am avoiding for my own sanity, although I'll have to tackle a few eventually

21:55 < rcurtin> one other idea might be to pick an algorithm that you don't see in mlpack, and implement it

21:55 < rcurtin> I can help you with it and merge it in; the basic requirements are that it's tested, documented, and fast, by way of comparison to a competing implementation

21:56 < rcurtin> I'm trying to think of some good examples of algorithms that wouldn't be too hard to implement that would be useful...

21:57 < klyons> yeah any advice for a good starting project would be great

21:57 < kirizaki_> Sorry if i disturb, but few days ago I interested in MLPACK with idea of ANN too ;)

21:58 < rcurtin> kirizaki_: no problem :) like I said, zoq is probably the person to talk to; he's been leading that effort

21:58 < rcurtin> klyons: do you think you might find, say, kernel regression or support vector machines interesting?

21:59 < rcurtin> I don't know what techniques you are familiar with, but those are both useful techniques that mlpack doesn't have anything for

21:59 < rcurtin> SVMs will be a little difficult, because personally I don't see much point in implementing SVMs if we can't beat libsvm's runtimes, and my understanding is that libsvm is pretty fast

21:59 < klyons> yeah i have a basic familiarity with both, and can always check references to bulk up on them

21:59 < klyons> yeah that is what i've read also

22:00 < klyons> I know that is one of their design goals as well, and they have been doing it for a long time

22:00 < rcurtin> yeah; there have been some attempts at improvements to the SMO algorithm; maybe one of those could outperform libsvm, but I've personally never had a chance to look into it

22:00 < rcurtin> I think that with a theoretically fast algorithm and a good implementation, it'd be possible to beat libsvm, but it might be a good amount of work

22:01 < klyons> I'm alright with a project being a good amount of work

22:01 < rcurtin> at the same time, something like "mlpack has really fast SVMs" would be a great selling point for the library and the implementation would probably get a decent amount of attention

22:01 < rcurtin> anyway, unfortunately, I have to head out now, but I idle in here all the time so if you leave messages I'll get back to you. or, you can also use the mailing list:

22:01 < rcurtin> http://lists.cc.gatech.edu/mailman/listinfo/mlpack/

22:01 < rcurtin> or you can open github issues for discussion too. a million ways to collaborate :)

22:01 < klyons> I would be happy to spend time with SVMs

22:02 < rcurtin> anyway, I'll be back later. I'll see if I can dig up some references on fast SVM training algorithms that are worth thinking about, but I can't promise anything. either way, stock SMO is probably the best way to go at first

22:02 < rcurtin> good to meet you, by the way :)

22:02 < klyons> Likewise, and thanks for chatting with me about this!