17:03 < danny474> hello guys! Im new to mlpack. I've looked up a couple of tutorials and source codes, and I want to contribute. Can someone guide me on how to get started coz choosing an algorithm to work on can be confusing. Thanx.

17:43 < zoq> danny474: The best way to get started is to download and compile mlpack, and go through the tutorials; especially the parts of them that involve the C++ interface. That will help give you an idea of how the code is laid out and the code standards that are used; probably you already finished the task.

17:43 < zoq> danny474: It is mainly up to you to choose the task and algorithm that suits your interests and skills. There are a few tickets open in the bug tracker; you can take a look on the bug tracker and see if you find any of them interesting. Or you can contribute a new algorithm, we can talk about this if you like.

17:49 < danny474> thanx. Is an SVM package not available in mlpack yet?

17:52 < zoq> danny474: You are right, currently there isn't any svm code in mlpack.

17:56 < danny474> i would love to work on a new algorithm, but its difficult deciding on what. can u suggest something?

18:02 < zoq> hm, It's hard to recommend something out of the blue, because I don't know in which field you are interested. Maybe you can give us some hints? If you are interested in trees I think naywhayare would like to see new trees.

18:04 < danny474> hmm, ive mostly worked on svm and nn recently.

18:04 < danny474> broadly speaking

18:10 < zoq> I think if you are interested in svm's you can start by writing a generic svm that works with different kernels. This is just a suggestion, maybe someone the channel has any ideas? But it looks like they are all idle.

18:15 < danny474> ok thanx. ill look into it. Any idea why wasnt svm picked up yet, coz surely its very popular.

18:20 < zoq> I'm not sure, maybe there was none with an higher interest in svm's. I think naywhayare can enlighten us once he is back, I think in a couble of hours or minutes ... not sure

18:21 < zoq> *couple of hours

18:58 < naywhayare> danny474: zoq: if someone's going to write an SVM implementation, it should outperform libsvm, which might be pretty difficult

18:58 < naywhayare> if you're up for a challenge, then that might be a good one :)

19:00 < danny474> i thought of that :P will have to think of something else

19:06 < naywhayare> I think zoq's advice is right on; it's hard to recommend something. maybe best for you to take a look at what mlpack has and either aim to improve that implementation or implement something new entirely?

19:06 < naywhayare> one guy has been working on deep neural nets and wrote a sparse autoencoder and softmax regression, but it hasn't been assembled into a complete deep learning framework yet

19:12 < danny474> maybe i can try looking into something on those lines...

19:18 < danny474> stacked ae, denoising ae, convnets, rbm ,dbm ... lots can be implemented ... though deep techniques require immense infrastructural firepower at times. I start with some small steps.

19:18 < danny474> *I'll

19:21 < naywhayare> yeah; if you sketch up a plan for an API or something, or maybe a way to templatize the autoencoders, that would probably be a great start

19:21 < naywhayare> I'm more than happy to iterate on ideas until we come to a good solution

19:25 < danny474> Ok fine. I'll look into that. Do u have anything specific in mind wrt templatizing the autoencoders ?

19:26 < naywhayare> I don't know enough about autoencoders at the moment

19:26 < naywhayare> but the template parameters should minimize duplicated code and also provide a lot of flexibility

19:26 < naywhayare> but I'm not sure what types of flexibility are needed for autoencoders... things like number of neurons, etc., can be set at runtime through the constructor

19:30 < danny474> ok i look more in the present code on them ...

19:30 < danny474> *i'll

20:05 < ajinkya> do we still have the R bindings for mlpack somewhere ?

20:05 < naywhayare> I don't think we ever had R bindings, but there's the RcppMLPACK project by Qiang Kou: github.com/thirdwing/RcppMLPACK

20:05 < ajinkya> i remember Patrick Mason worked on it when I was at GT but i am not able to find it

20:06 < ajinkya> aah cool

20:06 < ajinkya> thanks!

20:06 < naywhayare> Patrick worked on some MATLAB bindings

20:06 < naywhayare> I think there's still in src/mlpack/bindings/, but I don't think they work anymore

20:06 < naywhayare> I really want to find some student this summer to do the automatic bindings project

20:06 < naywhayare> also, how's it going in California? :)

21:21 < ajinkya> i was thinking of taking up the R bindings to get started with contributions again :)

21:22 < ajinkya> but yes we should definitely make bindings a priority in next releases ... i think that will boost the overall usage of mlpack

21:23 < ajinkya> .. and yes about Cali, its been nice to me so far :) some exciting work where I am getting to play with ml libraries (finally!)

22:04 < naywhayare> ajinkya: I agree, bindings are quite important

22:04 < naywhayare> I can't find the time to do them myself though

22:23 < ajinkya> i was not suggesting that either :)

22:26 < ajinkya> i will look more into RcppMLPACK but is it enough to be a R binding for mlpack? do you have any experience with it ? wondering why is it not pushed back into mlpack yet