Physics-Informed Neural Networks - Workshop for ML4SCIENCE & ENGINEERING SUMMER SCHOOL by @inductiva

This repository contains a hands-on workshop on implementing Physics-Informed Neural Networks (PINNs) in PyTorch for the ML4SCIENCE & ENGINEERING SUMMER SCHOOL, organized by Inductiva.ai @FEUP. You will solve both a forward problem (Burgers' equation) and an inverse problem (1D Euler equations) via Jupyter notebooks.

Adapted from TUM Physics-Informed Machine Learning lab session on "Physics-Informed Machine Learning", @tummfm

Repository Structure

├── data/           # Raw data files used by the notebooks
│   ├── burgers_shock.mat
│   └── 1DEuler_data.npy
├── to_do/           # Notebooks for you to complete
│   ├── 1-forward_pinn.ipynb
│   └── 2-inverse_pinn.ipynb
├── solved/         # Completed solution notebooks
│   ├── 1-forward_pinn.ipynb
│   └── 2-inverse_pinn.ipynb
└── README.md       # This file

• data/: Contains all necessary datasets (.mat and .npy) for the exercises. These files are loaded by the notebooks.
• todo/: Contains starter notebooks with scaffolding and instructions. Your tasks are marked within code cells and markdown sections.
• solved/: Provides fully worked-out solutions, including plotting and analysis. Use these to check your work.

Prerequisites

• Python 3.11+

• Git

• Visual Studio Code with the following extensions:

  • Python (ms-python.python)
  • Jupyter (ms-toolsai.jupyter)

Optional but recommended:

• Anaconda or Miniconda for environment management.

Installation

1. Clone the repository

   git clone https://github.com/biromiro/pinn-workshop-aptwind.git
   cd pinn-workshop-aptwind

2. Create and activate a virtual environment

   Using venv:

   python -m venv venv
   # On Linux/Mac
   source venv/bin/activate
   # On Windows (PowerShell)
   .\venv\Scripts\Activate.ps1

   Or using Conda:

   conda create --name pinn-workshop
   conda activate pinn-workshop

3. Install dependencies

   Otherwise, install packages directly:

   pip install torch numpy scipy matplotlib tqdm jupyter ipywidgets widgetsnbextension

   or, with conda:

   conda install pytorch numpy scipy matplotlib tqdm jupyter ipywidgets widgetsnbextension

Running the Notebooks in VS Code

1. Open the folder

   • Launch VS Code and select File → Open Folder..., then choose this repository.

2. Select the Python interpreter

   • Press Ctrl+Shift+P, type Python: Select Interpreter, and choose the interpreter from your venv or Conda environment.

3. Open a notebook

   • Navigate to the todo/ folder and open one of the .ipynb files.

4. Run cells

   • Click the Run All button in the top toolbar to execute every cell in order. Alternatively, run cells individually using the ▶️ icons.

5. Switch to solution

   • After working through to_do/, compare your results with the solution in the corresponding notebook under solved/.

Tips

• GPU Support: The notebooks detect CUDA or Apple MPS devices automatically. Ensure your system has a supported GPU and drivers.
• Loss Monitoring: Look at training-loss plots to check convergence. You can adjust learning rates or network architectures as an exercise.

License

This workshop content is provided under the MIT License. See LICENSE for details.