mathLab/PINA

📦 Highlights

• A detailed overview of the updates introduced in this version is provided in the following section.

📦 Solver Architecture: Mixin-Based Design

• AutoregressiveMixin — enables autoregressive sequential prediction
• ConditionAggregatorMixin — aggregates loss contributions across conditions
• EnsembleMixin — supports ensemble of models
• GradientEnhancedMixin — adds gradient-enhanced training
• ManualOptimizationMixin — provides fine-grained control over optimization steps
• MultiModelMixin — enables multi-model configurations
• PhysicsInformedMixin — injects physics-informed loss computation
• ResidualBasedAttentionMixin — implements residual-based attention mechanisms
• + 12 more

♻️ Conditions Refactoring: Decoupled Evaluation

• DataCondition — for data-fitting constraints
• DomainEquationCondition — for PDE constraints over domains
• InputEquationCondition — for PDE constraints from input data
• InputTargetCondition — for input-output supervised pairs
• TimeSeriesCondition — for autoregressive time-series data
• GraphTimeSeriesCondition — for graph-based time-series autoregression

📦 Weighting Module

• A dedicated weighting module has been introduced, separating loss weighting strategies from the loss computation itself:
• NoWeighting — uniform weighting
• LinearWeighting — learnable linear weights
• ScalarWeighting — fixed scalar weighting
• SelfAdaptiveWeighting — gradient-norm-based adaptive weighting
• NTKWeighting — Neural Tangent Kernel-based weighting

📦 Loss Module

• SinkhornLoss — Sinkhorn divergence-based loss

✨ New Models

• Kolmogorov-Arnold Networks (KAN) - PINA v0.3 introduces support for Kolmogorov-Arnold Networks, a novel neural network architecture based on the Kolmogorov-Arnold representation theorem. The `KolmogorovArnoldNetwork` class provides a stack of `KANBlock` modules, each applying learnable spline-based transformations to the input, optionally combined with a base activation function. The implementation includes fully vectorized spline basis functions with analytical derivatives, making KANs suitable for physics-informed applications where gradient computation is essential.

📦 Equation Zoo

• BurgersEquation — 1D Burgers' equation
• AdvectionEquation — advection equation
• AcousticWaveEquation — acoustic wave equation
• AllenCahnEquation — Allen-Cahn equation
• DiffusionReactionEquation — diffusion-reaction equation
• HelmholtzEquation — Helmholtz equation
• PoissonEquation — Poisson equation
• FixedValue, FixedGradient, FixedLaplacian, FixedFlux — boundary condition equations

📦 Problem Collection

• Version 0.3 significantly expands the problem zoo with modular implementations of common physical problems:
• BurgersProblem — 1D Burgers' equation problem
• AdvectionProblem — advection operator learning problem
• AcousticWaveProblem — acoustic wave propagation problem
• AllenCahnProblem — Allen-Cahn equation problem
• DiffusionReactionProblem — diffusion-reaction problem
• HelmholtzProblem — Helmholtz equation problem
• PoissonProblem — Poisson equation problem
• + 1 more

📦 Data Module Enhancements

• The `DataModule` (formerly `PinaDataModule`) has been further refined with improved internal structure:
• Creator — handles dataset creation from conditions
• Aggregator — aggregates batched data from multiple conditions
• ConditionSubset — manages per-condition data subsets
• DataManager interface with TensorDataManager and GraphDataManager implementations
• BatchManager — coordinates batching strategies
• SingleBatchDataLoader — optimized single-batch data loading

📦 Internal Backward Compatibility

• Version 0.3 maintains backward compatibility with version 0.2 through deprecation warnings. The renamed modules and moved classes provide a smooth migration path for existing codebases.

📦 Highlightns

• We are thrilled to announce version 0.2 of the PINA library, a major release
• that brings numerous enhancements and new features while maintaining compatibility with previous versions.
• A comprehensive description of the updates introduced in this version is provided in the following section.

📦 Data management: Introduction of PINADataModule

• Automatic dataset splitting: By simply specifying the desired fractions for training, validation, and testing in the Trainer, PINA automatically handles dataset splitting and the creation of corresponding DataLoaders.
• Seamless integration with graph data: the redesigned pipeline natively supports different data types (e.g., ``torch.Tensor`` and ``torch_geometric.data.Data``) without requiring any modifications to the user's code. This is achieved through the definition of separate custom dataset classes, making the entire data-loading process fully transparent.
• Modularity and scalability: compared to the previous release, we have significantly improved the performance of the data-loading process by implementing an efficient data transfer strategy to the GPU. This is particularly beneficial for Physics-Informed problems where the dataset fits entirely in GPU memory. Additionally, we have introduced support for more advanced features of the PyTorch DataLoader, such as memory pinning, further optimizing data handling and performance.

📦 Problem collection

• Version 0.2 introduces a new module that includes a collection of common physical problems. This module serves two key purposes:
• 1. It provides modular implementations of widely-used problems, allowing users to easily import these problem classes and adjust parameters as needed.
• 2. It offers a valuable resource for users looking to create new, more complex problems in PINA style.
• In future releases, we will extend this module with additional problems.

✨ New Contributors

• We thank all the new contributors!
• [@FilippoOlivo](https://github.com/FilippoOlivo) made their first contribution in #344
• [@MatteB03](https://github.com/MatteB03) made their first contribution in #463