PINNeAPPle - PINNFactory

A lightweight framework for building Physics-Informed Neural Networks (PINNs) with symbolic PDE definitions using SymPy and automatic differentiation in PyTorch.

It provides:

• Flexible neural architectures (NeuralNetwork class).
• Wrapper for inverse parameter estimation (PINN).
• Factory for PDE-driven loss generation from symbolic equations (PINNFactory).

---

Installation

Clone the repository and install the requirements:

git clone https://github.com/barrosyan/pinnfactory.git
cd pinnfactory
pip install -r requirements.txt

Requirements

• torch
• numpy
• matplotlib
• sympy

---

Quick Start

1. Define your neural network

from pinn_generator import NeuralNetwork, PINN

# Example: 1 input, 1 output, 3 hidden layers with 20 neurons each
net = NeuralNetwork(num_inputs=1, num_outputs=1, num_layers=3, num_neurons=20)
pinn = PINN(net)

2. Define PDEs and conditions symbolically

from pinn_generator import PINNFactory

# PDE: u_xx + u = 0  (example)
pde_residuals = ["Derivative(u(x), (x,2)) + u(x)"]

# Boundary conditions: u(0) = 0, u(pi) = 0
conditions = [
    {"equation": "u(0)"},
    {"equation": "u(pi)"}
]

factory = PINNFactory(
    pde_residuals=pde_residuals,
    conditions=conditions,
    independent_vars=["x"],
    dependent_vars=["u"]
)

loss_fn = factory.generate_loss_function()

3. Training loop

import torch

optimizer = torch.optim.Adam(pinn.parameters(), lr=1e-3)

# Example training batch
x = torch.linspace(0, 3.14, 100).view(-1, 1).requires_grad_(True)

for epoch in range(1000):
    optimizer.zero_grad()
    loss, loss_components = loss_fn(pinn, {"collocation": (x,)})
    loss.backward()
    optimizer.step()
    if epoch % 100 == 0:
        print(f"Epoch {epoch} - Total Loss: {loss.item():.6f}")

---

Roadmap

• [ ] Add support for system of PDEs.
• [ ] Implement collocation sampling utilities.
• [ ] Add GPU support for large-scale PDE solving.
• [ ] Integrate with visualization tools for PINN training.
• [ ] Add more NN Architectures.

---

License

Apache License 2.0.
You may use, modify, and distribute this project under the terms of the Apache License 2.0.