10:26 < bugs_> Hello, I am new to this and want to contribute to a few projects. Can someone please help me?

12:06 < shardulparab97> Hi, I am Shardul Parab, a final year student of BITS Pilani (Hyderabad), India. I am highly interested in contributing to MLPack via GSoC and have tried to make some PRs recently.

12:07 < shardulparab97> A few days ago, @atulim had suggested the topic of Capsule Networks for GSoC, I would like to ask whether the corresponding topic would be pursued for this year's GSoC.

12:07 < shardulparab97> Thank you! :)

13:02 < zoq> bugs_: Hello, here to help.

13:02 < zoq> shardulpara: Hello there, that could be an option, yes.

18:19 < jenkins-mlpack2> Project docker mlpack nightly build build #207: ABORTED in 2 days 13 hr: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/207/

18:20 < rcurtin> looks like a test hung, not sure why

18:20 < ayesdie> rcurtin: hey

18:22 < ayesdie> Here's a quick upadte: I've updated the SparseSVMFunction to work as a multiclass classifier and added tests for that. I'm a bit unsure how we might implement Kernel SVM.

18:22 < ayesdie> update*

18:23 < rcurtin> ayesdie: yeah, I think kernel SVM is not possible since it needs to be optimized in the dual space

18:23 < rcurtin> and our implementation works in the primal space

18:23 < rcurtin> but that's ok, we can just go with linear SVM for now

18:25 < rcurtin> I saw your updates via email too, so I need to review the PR again. I'm hoping I'll have a chance to do that later today or tomorrow

18:26 < ayesdie> Also, can we give the user, the option to choose any optimizer? I've not tested this but you mentioned we have to use sparse differentiable separable Optimizer.(HOGWILD)

18:27 < ayesdie> It's fine, I'll work on the tests of SparseSVM and it's binding in another branch, you can review this any time :).

18:29 < ayesdie> I also wanted to bring this to your attention that the list of GSOC ideas has "Parallel stochastic optimization methods" but no details about it down there.

18:30 < rcurtin> oh, hm, let me figure out what happened there. maybe the project was removed but the contents weren't updated

18:30 < rcurtin> and yeah, for SparseSVM, I *think* that we should be able to use more than just Hogwild, but other optimizers may not be able to take advantage of the sparsity of the gradient

18:30 < rcurtin> so long as Gradient() is templatized so that the gradient type can be either arma::mat or arma::sp_mat, it should work fine

18:31 < rcurtin> I guess we should change the name from SparseSVM to LinearSVM, by the way

18:31 < rcurtin> since it doesn't necessarily require sparse data anymore :)

18:33 < rcurtin> ok, checked on the GSoC ideas page... the 'parallel stochastic optimization methods' project was removed because Shikhar Bhardwaj did the project in 2017

18:33 < rcurtin> now of course there do exist more parallel optimization methods, so if you had a cool idea we could do it, but we don't have anything specifically written up on the ideas list :)

18:39 < ayesdie> I will try to test with the optimizers other than Hogwild. As of now, I just ran the SVMFunction on some dataset(to check if there was anything wrong with my implementation) without using any Ensmallen optimizer and the results were good. (I've placed that in a separate repository).

18:40 < rcurtin> sounds good :)

18:41 < ayesdie> Thanks for telling this rn, I'll update SparseSVM to LinearSVM before I proceed any further.

18:41 < rcurtin> sure, no problem

18:41 < rcurtin> SparseSVM -> LinearSVM is an easy change anyway :)

18:42 < ayesdie> Yes, will update the folder and filenames.

18:42 < ayesdie> Or can we just typedef it?

18:44 < rcurtin> we should update the folder and filenames; there's no reverse compatibility to care about here

18:45 < ayesdie> gotcha :)

22:05 < jenkins-mlpack2> Project docker mlpack nightly build build #208: UNSTABLE in 3 hr 45 min: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/208/