[Breaking] Change weights generation to follow WeightInitializers.jl

[MartinuzziFrancesco]

In order to streamline the package and creation of reservoir computing models the creation of the weights of the matrices should follow WeightInitializers.jl

[Jay-sanjay]

Hi @MartinuzziFrancesco I guess I can take a look at this, if you can elaborate a bit what exactly needs to be done here.

[MartinuzziFrancesco]

Hi, of course! So at the moment the construction of the input layers matrices and reservoir matrices uses abstract typing and dispatch over different structs of the different architectures for the matrices and it is a bit cumbersome. I wanted to simplify it a bit and make it follow the standard in WeightInitializers.jl. This way we would also start an easier way to control the randomness in ReservoirComputing.jl. For example SparseRandReservoir would look like this:

using ReservoirComputing, Random
rng = MersenneTwister(42)

# Passing kwargs with explicit rng call
res_cl = rand_sparse(rng; sparsity=0.1, spectral_radius=1.0)
#res_cl is now a callable function that takes in the size of the reservoir
reservoir_weights = res_cl(rng, 10, 10)

# Passing kwargs with default rng call
res_cl = rand_sparse(; sparsity=0.1, spectral_radius=1.0)
#res_cl is now a callable function that takes in the size of the reservoir
reservoir_weights = res_cl(10, 10)

For details of the implementation you can refer to the source in WeightInitializers. I could sketch rand_sparse during this week if that would be of help. This needs to be done for both input layers and reservoirs.

Of course after this we need to modify the ESN constructor but that should be rather straightforward.

This issue is part of a streamlining in order to be more in line with (F)Lux as model building philosophy. This should also make contributions a little bit easier

[MartinuzziFrancesco]

@Jay-sanjay I have started a PR in #193, you can follow the template to create the remaining functions for the reservoir and input layers. I will need to think about a couple of details (and tests as well)

[Jay-sanjay]

Thanks @MartinuzziFrancesco for the template, I will start to work on it with other functions

[MartinuzziFrancesco]

Addressed in #196 and finalized in #203