In order to assimilate FourierLayer to the yet to implement DeepONet, it would be nice to have a Fourier Neural Operator (FNO) constructor that creates the entire architecture.
Something like:
model = FNO(in, out, latentspace, layercount, grid, modes, activation)
Instead of having to make the entire chain yourself:
This doesn't add functionality but helps unify the API of the package. FourierLayer should still stay accessible via API as is since it offers a lot more control per layer than constructing the entire architecture with all the same global arguments.
In order to assimilate
FourierLayer
to the yet to implement DeepONet, it would be nice to have a Fourier Neural Operator (FNO) constructor that creates the entire architecture.Something like:
Instead of having to make the entire chain yourself:
This doesn't add functionality but helps unify the API of the package.
FourierLayer
should still stay accessible via API as is since it offers a lot more control per layer than constructing the entire architecture with all the same global arguments.