[mlpack] A Friendly Hello + Contribution Idea

[Ryan Curtin]

On Mon, Mar 25, 2019 at 11:00:31AM +0000, Patrick Cando wrote:
> Hi Ryan,
> 
> Cheers for coming back to me!
> 
> I agree on the GSoC; given my own deadline constraint, I probably won't be
> able to work on anything on the list.
> 
> Given that the NEAT project is proving to be quite popular, I could perhaps
> implement Novelty Search (which I believe is not part of GSoC - I also
> don't recall any emails coming in about this thus far) and add that to the
> library? It combines well with my area of interest and supervisor
> expertise. Once we get going, there's a lot we can do in that area adding
> HyperNEAT, Differential Evolution etc. In fact, I was looking to perhaps do
> a standalone implementation of Novelty Search, just to work on implementing
> an algorithm; but it would be brilliant to add that to something already
> active, as opposed to it being isolated project on Github.
> 
> Otherwise, I'm open to other areas that you think could be useful /
> interesting for the library.

Hey Patrick,

Sorry for the slightly slow response.  I don't know much about Novelty
Search, but I do think perhaps it would be nice to include in ensmallen.
Basically the keys there are just that we get the API right, but we
already have a pretty established API for unconstrained
non-differentiable functions ("arbitrary functions").  Perhaps Novelty
Search fits there?  If so, or if we want it to be able to be
constrained, we may need to rethink the constrained optimization API a
little bit.  Possibly Marcus may have more comments, but definitely this
would be nice to include.

For merging it in, basically we just need (a) a good implementation
that's hopefully as flexible as a user might want and matches the
current API (or whatever we choose for the API); (b) good documentation
(but that's pretty easy if we follow the existing recipes for ensmallen
documentation); (c) good tests.  And that's about it, although it would
be nice to run some simulations and see how novelty search compares to
other algorithms for the same type of problem.

Just to be clear, is your intention that we help do some in-detail
mentoring for you during the summer?  Or moreso that you work on the
code, open a PR, we give it a review, and everything is good to go?
Personally I don't know that I have the time to take on a full mentor
position for this project but I can make time for PR reviews, etc.

Thanks!

Ryan

-- 
Ryan Curtin    | "Hey, tell me the truth... are we still in the
ryan at ratml.org | game?" - The Chinese Waiter