PINNs Playground ✨

Project Highlights ✨

1. Automatically generate training history upon completion 📊
2. Easy to extend with minimal code, enabling rapid verification ✅

Done and Todo ✅🔥

Methods ✔️

• Inverse Problem
• Residual-based adaptive refinement
• Self-adaptive activation function
• Self-adaptive loss weights
• Causal sampling
• RAR-D
• More methods/algorithms

PDE examples 📝

• Burgers
• NavierStocks2D
• NavierStocks3D

Others 🛠️

• Add checkpoint
• MLP model
• Add log file
• fastapi handle request
• Visualization
• More Models

Usage 🚀

1. Install dependencies 🛠️

   ... install pytorch
   >>> pip install fastapi tqdm rich matplotlib numpy pandas imageio

2. Write your PDE in pdes, and modify config.common.py's function pde_fn(cls): to return your PDE

3. Create a train script, import PINN and config, and run PINN(config, model).train(). Check train_burgers.py for an example.

4. For inverse problems, just add params_init in the config, and modify pde_fn to accept params as input. Check train_ns_inverse.py for an example.

Add modules 🧩

This framework is designed to be modular, so you can easily add your own modules. We provide a Callback class, which can be used to add your own callback functions.