DomainDecompositionTrainingHistory<T>

Models & Types ยท AiDotNet.PhysicsInformed

Training history for domain decomposition PINN training.

Inheritance: TrainingHistory<T> โ†’ DomainDecompositionTrainingHistory<T>

Implements: IDomainDecompositionTrainingHistory<T>

How It Works

For Beginners: This class tracks the training progress of a domain decomposition PINN. It extends the base TrainingHistory with additional metrics specific to domain decomposition:

  • SubdomainLosses: Per-subdomain PDE and data losses
  • InterfaceLosses: Continuity errors at subdomain boundaries
  • PhysicsLosses: Total PDE residual across all subdomains

Key Observations During Training:

  1. Subdomain losses should decrease independently
  2. Interface losses ensure solution continuity
  3. If one subdomain loss is much higher, it may need more capacity
  4. Interface losses are critical for global solution quality

Constructors

Constructor Summary
DomainDecompositionTrainingHistory(int subdomainCount) Initializes a new instance with the specified number of subdomains.

Properties

Property Summary
List<List<T>> SubdomainLosses { get; } Gets the losses per subdomain per epoch.
List<T> InterfaceLosses { get; } Gets the interface continuity losses per epoch.
List<T> PhysicsLosses { get; } Gets the PDE residual losses per epoch.
int SubdomainCount { get; } Gets the number of subdomains.

Methods

Method Summary
void AddEpoch(T totalLoss, List<T> subdomainLosses, T interfaceLoss, T physicsLoss) Records metrics for a training epoch.