Alias variables about componentarrays.jl HOT 2 OPEN

This is something I've wanted to implement for a while, and even made up a quick prototype for a few months back. It's nice for state-space modeling too because you can have a C*x + D*u output signal. Also, activation functions for NNs.

This syntax is better than the one I had, too. Although, my preference would be to have the function as the first argument.

Time permitting, I'm going to try to get something working within the next few days. There are a few changes that will have to be made to get this working, but nothing that will require too much major surgery.

from componentarrays.jl.

I think the first time I read this I missed the significance of the fact that it's looking at the upper level. Most of the use cases I have for alias components are applying functions to components within the same level.

So really this is two separate issues:

1. We need to have alias components
2. Those aliases should have a way to reference the components above them

I'd like to have a way to accomplish the first without having to have a ThisLevel.

My original thoughts were something like:

x = ComponentArray(
	a = 1,
	b = 2,
	c = AliasVar(x -> x.a + x.b)
)

So AliasVar would always take in a function of one variable, which is the ComponentArray that holds the AliasVar (in this case x). Maybe a way of accomplishing both would be something like:

x = ComponentArray(
    comp1 = ComponentArray(out = 1),
    comp2 = ComponentArray(in = AliasVar(x -> fun(UpperLevel(x).comp1.out))),
)

It's a little more verbose for this case, but it handles the first case without having to introduce a ThisLevel. Also, I think this makes it more straightforward to understand for the case where you might want to go even higher level like UpperLevel(UpperLevel(x)).

from componentarrays.jl.