PhysicsInformedOptions
Options & Configuration · AiDotNet.Models.Options
Base configuration options for physics-informed neural network models.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → PhysicsInformedOptions
Known derived types: DeepOperatorNetworkOptions, DeepRitzMethodOptions, DomainDecompositionPINNOptions, FourierNeuralOperatorOptions, GraphNeuralOperatorOptions, HamiltonianNeuralNetworkOptions, LagrangianNeuralNetworkOptions, MultiFidelityPINNOptions, MultiScalePINNOptions, PhysicsInformedNeuralNetworkOptions, UniversalDifferentialEquationsOptions, VariationalPINNOptions
For Beginners
Physics-informed models learn from data AND from the physical laws the data must obey. This is the common ancestor of their options classes.
How It Works
It declares no properties of its own. Twelve options classes derive from it — PINNs, neural operators, Hamiltonian and Lagrangian networks, the Deep Ritz method — and they share no hyperparameter universally. The PDE and boundary loss weights, for instance, belong to the PINN family and would be meaningless on FourierNeuralOperatorOptions. Hoisting them here would put a property on nine classes that have no use for it, and requiring one would make those classes throw at their own defaults.
What it does contribute is its parent: deriving from
ModelHyperparameterOptions gives every physics-informed options class
MaxGradNorm and the Require validation helpers.
Constructors
| Constructor | Summary |
|---|---|
PhysicsInformedOptions() |