Skip to content

Latest commit

 

History

History
54 lines (35 loc) · 1.37 KB

README.md

File metadata and controls

54 lines (35 loc) · 1.37 KB
Raw

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.