06:28 < krishna> Respected sir,

06:29 < krishna> I am a B.Tech student in Computer Science and Engineering from India

06:30 < krishna> Previously I have been looking into mlpack organisation from different source like github

06:30 < krishna> after going through it I wish to contribute to this open source organisation

06:31 < krishna> any input I can get regarding this?

07:17 < soonmok> there is a tutorial (http://www.mlpack.org/docs/mlpack-git/doxygen/tutorials.html)

07:54 < krishna> I wish to contribute by adding a differential evolution algorithm to the existing mlpack libraries. Can that be an effective addition to it?

08:53 < jenkins-mlpack2> Project docker mlpack nightly build build #196: STILL UNSTABLE in 3 hr 39 min: http://ci.mlpack.org/job/docker%20mlpack%20nightly%20build/196/

11:30 < zoq> krishna: Hello, sure, the methods should probably be part of the ensmallen repo: https://github.com/mlpack/ensmallen

22:02 < rcurtin> zoq: "on a compiler journey" :) I like it :)

22:42 < zoq> rcurtin: Still fighting with some mingw64+windows constraints ...

22:46 < rcurtin> not an easy journey I suppose :)

23:20 < zoq> somebody posted a link to ensmallen in the in R subreddit https://www.reddit.com/r/cpp/comments/aicl4h/ensmallen_a_headeronly_c_library_for_mathematical/

23:22 < rcurtin> lots of people misunderstanding why someone would use C++ in that thread

23:24 < zoq> not the first time I see the discussion about armadillo and autodiff

23:43 < ShikharJ> Hmm, I've been looking at the comments, and they really seem to be interested in an autodiff solver. Might be a good project for GSoC?

23:44 < rcurtin> I feel like it would be a huge amount of maintenance. but, one thing that might be useful is seeing if ensmallen could use some existing C++ autodiff package

23:47 < zoq> Agreed, getting autodiff right is a huge effort, not sure GSoC is a good fit for such a project.

23:47 < ShikharJ> That could be a good option too.

23:47 < rcurtin> it's probably enough work to get some autodiff package working with ensmallen and then write up tutorials and documentation for it, I suspect :)

23:48 < rcurtin> one of the important things will be benchmarking, since often autodiff implementations can be slower

23:50 < zoq> right, I think the paper has some compelling numbers in that direction

23:52 < ShikharJ> Some people also seem to be concerned with the non-use of Eigen and the compile times for ensmallen. How big of an issue is that?

23:52 < ShikharJ> I didn't see compile times reduce by much when we shifted ensmallen to a separate repository, so I think it might be safe to assume that it doesn't lead to a huge compile time overhead?

23:52 < rcurtin> if you're driven to using C++ for your application, it's likely that compile time doesn't really matter all that much

23:53 < rcurtin> certainly compiling some program that uses ensmallen will increase the compile time, but not by too much more than what Armadillo already does

23:53 < rcurtin> I don't understand everyone's focus on Eigen instead of Armadillo either. sure, it's what TensorFlow used... but the syntax for Armadillo is so much nicer

23:53 < rcurtin> and the fact that it wraps any LAPACK/BLAS is, in my opinion, a better thing than hand-implemented BLAS like Eigen did

23:54 < rcurtin> but... I realize that I have a biased perspective :)

23:54 < ShikharJ> Hmm, I have to agree that Armadillo's source code has got to be the most beautiful piece of code I have seen in a long time.