use Armadillo RNG support

[rcurtin]

There are a couple instances where mlpack's Random() or RandInt() functions are used:

./parallel_sgd/parallel_sgd_impl.hpp:          mlpack::math::randGen);
./sa/sa_impl.hpp:  const double unif = 2.0 * math::Random() - 1.0;
./sa/sa_impl.hpp:  const double xi = math::Random();
./scd/descent_policies/random_descent.hpp:    return mlpack::math::RandInt(function.NumFeatures());

I think this shows up a lot in the tests too. Instead, we can use Armadillo's RNG here; so mlpack::math::randGen can (probably) be replaced by arma::arma_rng_cxx11_instance or something similar, and we can use that to generate random numbers.

However there is also the Armadillo configuration parameter ARMA_USE_EXTERN_CXX11_RNG; but, looking through it quickly, I don't see anything different for us to do whether or not someone using ensmallen has written #define ARMA_USE_EXTERN_CXX11_RNG.

[conradsnicta]

Armadillo has an internal class called arma_rng_cxx11, but it's not part of the public API.

The ARMA_USE_EXTERN_CXX11_RNG define is set by the armadillo cmake installer to indicate that the arma::arma_rng_cxx11_instance is available in the armadillo wrapper library. The define not meant to be used outside of that specific use case. Furthermore, it's only valid on Linux systems with gcc 4.8.3+. It currently doesn't work with macOS or windoze. (it should be trivial to make it work with macOS, now that Apple has proper support for thethread_local storage class in C++11).

Perhaps it would be sufficient to simply use the public randi(), randu(), randn() and randg() functions ?

In armadillo version 8.400+ all of the above functions were extended to output single scalars.
randg() was added in version 4.650. randi() was added in version 3.930.
randu() and randn() have been available "since forever".

If armadillo 8.400 is "too recent", it's still possible to use the above functions to generate scalars. Simply use as_scalar() on a random vector that has only one dimension.

As a sidenote, armadillo 8.400 is available in Ubuntu 18.04.

[zoq]

In case of:

std::shuffle(visitationOrder.begin(), visitationOrder.end(), mlpack::math::randGen);

it should be reasonable to use:

visitationOrder = arma::shuffel(visitationOrder);

any thoughts?

[zoq]

For the other cases I agree, it would be simple to use: arma::as_scalar(randi(...)).

[rcurtin]

Agreed on both counts---I wonder why we didn't just use arma::shuffle(visitationOrder) originally?

[zoq]

I guess the benefit is that std::shuffel works inplace; should be neglectable here.

[conradsnicta]

arma:shuffle() should detect aliasing and do an inplace shuffle:
https://gitlab.com/conradsnicta/armadillo-code/blob/9.200.x/include/armadillo_bits/op_shuffle_meat.hpp#L62
https://gitlab.com/conradsnicta/armadillo-code/blob/9.200.x/include/armadillo_bits/op_shuffle_meat.hpp#L129

[zoq]

Thanks for the clarification, should have checked first. 05a8743 implements the modifications.

[rcurtin]

I think this one is fully handled now---I see no use of any random object or anything other than Armadillo's existing public random functions.