07:30 < alsc> rsurtin: thx for the explanation!

07:33 < alsc> I could use some help in setting up the L-BFGS and mini batch as well :)... not knowing much about the details of the algorithm makes it really hard to guess. NumBasis() = 5, for example?

14:44 < nick8642_> hello

14:45 < zoq> nick8642_: Hello there.

15:02 < hodor12345678> I am above average in C++ and Python.But today when I was watching the codebase of MLpack I was totally overwhelmed.Please tell me where should I start.Please somebody guide me.

15:12 < zoq> I see, mlpack uses lots of different C++ paradigms including a lot of template metaprogramming, so I would suggest to get familiar with some techniques first, we listed a bunch of helpful resources here: http://www.mlpack.org/gsoc.html

16:01 < alsc> rcurtin: I noticed there's an ANN directory but it looks more like a placeholder. are ANN implemented in mlpack?

16:03 < alsc> otherwise maybe you could point me towards an ANN implementation that uses armadillo?

16:37 < rcurtin> alsc: I've always used default parameter for L-BFGS, to be honest, I've never played much with tuning it because there are so many parameters to the algorithm

16:38 < rcurtin> in general, choosing all of those parameters smartly would require a good understanding of what the loss surface is, which will be dependent on both the dataset and the function being optimized...

16:38 < rcurtin> so I think it would be hard to know

16:38 < rcurtin> the ANN directory is artificial neural networks; if you're looking for approximate nearest neighbor search, you can actually use the KNN class for that

16:39 < rcurtin> you can just specify an epsilon value > 0 to the constructor and the results will be approximate

16:39 < rcurtin> if you actually were looking for neural networks, there are some nice examples for now in the src/mlpack/tests/ directory, like recurrent_network_test.cpp, feedforward_network_test.cpp, etc.

16:39 < rcurtin> but the ANN code is not stable yet, so the code is found in the git master branch but not in 2.2.5