Make DeepONet and FourierNeuralOperator into structs

[KirillZubov]

deeponet = NeuralOperators.DeepONet(
    Chain(
        Dense(1 => 10, Lux.tanh_fast), Dense(10 => 10, Lux.tanh_fast), Dense(10 => 10)),
    Chain(Dense(1 => 10, Lux.tanh_fast), Dense(10 => 10, Lux.tanh_fast),
        Dense(10 => 10, Lux.tanh_fast)))
julia> deeponet.name ==nothing 
true

it guess it should be something like deeponet.name == "DeepONet"

[KirillZubov]

@avik-pal is there another easy way to find out the type of Lux.NeuralOperator? is it DeepONet or FourierNeuralOperator and etc?

[avik-pal]

No don't dispatch on name, name is used for pretty printing. Use CompactLuxLayer{:DeepONet}.

[KirillZubov]

@avik-pal ok, how identify type of FourierNeuralOperator?

[avik-pal]

FNO just returns a Chain. We could modify it to return a layer that can be dispatched on.

[avik-pal]

@ayushinav can you get this done?

[ayushinav]

We can probably make a CompactLayer for FNOs too and other architectures in future, otherwise the structs would only contain the chain of the operator. Making a struct for different parameters of each operator sort of defeats the purpose of how useful CompactLayers are.
For now, I'm inclined towards creating a struct with only the chain as its parameter.