Automatic ODE specification inside of the model macro

[bvdmitri]

@ismailsenoz often works with ODE-based models in RxInfer and suggested an interesting idea: natively supporting ODE-like statements in the @model macro, so that users wouldn’t need to manually discretize state transition functions. I think this could be quite elegant and useful.

In principle, we could automatically handle discretization using libraries like SeeToDee.jl. One idea is to introduce a nested macro, e.g.:

@model function my_model()
    # ...
    @ode x_next = f(x, u)
    # ...
end

However, I’m unsure whether this is the right direction — macros don’t always compose nicely, and this could complicate implementation internally.

An alternative might be introducing a special symbol for such expressions, like:

@model function my_model()
    x_next ≛ f(x, u)dt
end

(or something similar).

@ismailsenoz — could you write down a few examples of how you’d ideally like to use this, based on your current ODE workflows? What variables should it involve? How to handle dt? Do we want to automatically support logdt, etc? More examples - better. Then we can brainstorm implementation strategies together with @wouterwln, potentially for GraphPPL and RxInfer.