Generalizable Problems

[ndem0]

I started creating a gallery/zoo of different problems directly in the PINA source code (https://github.com/mathLab/PINA/blob/0.2/pina/problem/zoo/poisson_2d_square.py). Of course, it's too difficult (and probably no sense) to have a complete generalization, but I would like an easy way to change the domain (or the size of the domain), the forcing term, the boundary conditions, ...

The two options @FilippoOlivo and I thought are:

• changing the problem definition, by adding the init method in order to take input variables. The older problems should still run (they have an implicit init by default) but the new ones can use runtime variables (e.g. PoissonSquare(x_range=[0, 1], y_range=[4, 6], force_term=...))
• adding a factory to build at runtime the classes we need (here is an example https://refactoring.guru/design-patterns/factory-method/python/example). In this way, we can change everything at runtime, but it's an additional and complex layer in the source code.

What do you think? Which option is better in your opinion?

[GiovanniCanali]

In my opinion, as long as there are only a few input variables, an __init__ method makes the problem definition intuitive.

[ndem0]

Maybe once we have the 0.2 almost completed, we can try with a init problem in the zoo.
We can of course keep the factor class for future additions

[dario-coscia]

See the AdvectionProblem as example

PINA/pina/problem/zoo/advection.py

Line 81 in e105318

def __init__(self, c=1.0):