functionality similar to `@named` for `SymbolicNeuralNetwork`

[TorkelE]

When using SymbolicNeuralNetwork to create a neural network to insert into a UDE, it feels natural to want the output to have the same symbolic names as the variables in which you store them (as basically is the case ofr all symbolic variables across the MTK system)

nn, theta = SymbolicNeuralNetwork(; chain, nn_name = :nn, nn_p_name = theta)

Right now this have to be done manually. However, something similar to @named used in MTk would allow

@named nn, theta = SymbolicNeuralNetwork(; chain)

with nn_name = :nn and nn_p_name = theta infered automatically. Right now, @named is used for declaring Systems, and while a check checking if it is applied to a SymbolicNeuralNetwork could be possible, it might also not make sense to overload it with other functionality.

Not sure what the best solution is. One could do a @SymbolicNeuralNetwork macro. But in this case it probably makes sense to drop the function call and directly do

@SymbolicNeuralNetwork nn, theta = chain

which works well for base, however, it might make designating additional arguments more awkward.

[SebastianM-C]

Using @named would not be appropriate here for a variety of reasons, one being that SymbolicNeuralNetwork is specifically not a System like NeuralNetworkBlock.

What SymbolicNeuralNetwork does is essentially 2 @parameters calls. This is a bit awkward since you need to special case something for this particular application only, so probably the only way to do it is via @SymbolicNeuralNetwork.

I would argue that @SymbolicNeuralNetwork theta nn = chain is closer to what we do since, the nn wraps the chain.