Guides

Task-oriented guides for common workflows. Each guide solves a specific problem and assumes you've completed the Getting Started tutorial.

---

Defining Problems

• Define a Custom Problem

  Write your own ODE callable, fields, and parameters from scratch.

• Promote a Constant to a Parameter

  Turn a fixed value into a learnable quantity with a one-line change.

• Forward vs Inverse Problems

  Understand when to use each mode and how the loss structure differs.

• PDE Forward Problems

  Set up boundary conditions, multi-dimensional domains, and PDE residuals.

Training and Data

• Use CSV Data

  Load real experimental observations instead of synthetic data.

• Tune Hyperparameters

  Adjust learning rate, architecture, loss weights, and collocation density.

• Loss Weighting Strategies

  Balance physics enforcement against data fitting.

• Lightning vs Core Training

  Choose between PyTorch Lightning and a raw training loop.

Understanding the Design

• Architecture

  How AnyPINN's layered design compares to other PINN libraries.