00:54 < karthikabinav> Hi, is there an API for doing a SVM? Or do you have an convex program solver and I need to invoke the SVM convex program with that code ? I went through the API documentation and code and couldnt find either one.

01:04 < stephentu> karthikabinav: there's currently no api for doing SVM directly

01:04 < stephentu> you could run SGD on the hinge loss function with a regularizer

01:05 < stephentu> i'm hoping to get a general convex ERM module in at some point in the future

01:21 < karthikabinav> ok thanks ! I have been dabbling around with this code trying to see if I can make some contribution. This is when I figured there didnt seem to be a direct API for SVM. I was wondering if working on an implementation of it is a concrete place to start off?

01:21 < karthikabinav> I havent still thought through the details. But do you see any immediate challenges ?

09:51 < DivanshuJain> hi i am a new developer and would like to contribute

09:53 < DivanshuJain> i have experience in C++ ... can i get some guidance on how i should start ?

10:31 < zoq> DivanshuJai: Hello, the best way to get started is to download mlpack and compile it from source, then use it for some simple machine learning tasks.

10:31 < zoq> The tutorials might prove helpful: http://www.mlpack.org/tutorial.html

10:32 < zoq> Once you've got a basic feel for mlpack programs and source, you can take a look at the list of open tickets you might find something interesting: https://github.com/mlpack/mlpack/issues.

10:33 < zoq> Most of them are marked with a difficulty, so that might help you figure out some issues that you can handle. We are always interested in new algorithms so if you interested in some special field I think we can figure something out.

18:57 < stephentu> naywhayare: i started looking at this MVU stuff

18:57 < stephentu> i think i want to try to get our sdp solvers to work on it

18:58 < stephentu> do you have any code to generate test manifolds?

18:58 < stephentu> or like old test cases for mvu

19:24 < naywhayare> I had a swiss roll csv

19:24 < naywhayare> let me see if I can find it real quick

19:25 < naywhayare> http://ratml.org/misc/

19:25 < naywhayare> I put a couple in there

19:26 < naywhayare> 2_class_swiss_roll*.csv

19:26 < naywhayare> swiss_roll*.csv

19:26 < stephentu> thanks man

19:26 < naywhayare> if I remember right, if MVU works, then you should be able to unroll the swiss roll and then use a linear classifier (maybe Perceptron) to achieve perfect accuracy

19:26 < naywhayare> LinearRegression could work too

19:27 < stephentu> i think you sparked an interest in me

19:27 < naywhayare> I'll have a response about the philosophy of error handling in mlpack later today; I want to do some thinking first

19:27 < stephentu> for manifold learning

19:27 < naywhayare> :)

19:27 < stephentu> it seems really cool

19:27 < naywhayare> in my view manifold learning is still wide open; there aren't many scalable solutions

19:27 < stephentu> and if i can tie it with sdps

19:27 < stephentu> even better

19:27 < naywhayare> I haven't visited it in a few years

19:27 < naywhayare> anyway, I have to run to the store... I'll be back later

19:27 < stephentu> ya ben doesnt relaly believe in manifold learning but thats another story

19:27 < stephentu> alright peace