pinn-kit 0.1.5

A toolkit for Physics-Informed Neural Networks

PINN-kit

A toolkit for Physics-Informed Neural Networks (PINNs). This package provides tools and utilities for implementing and training physics-informed neural networks for solving differential equations.

Features

• Easy-to-use interface for defining physics-informed neural networks
• Support for various types of differential equations
• Flexible domain handling utilities for arbitrary input variables
• Advanced meshgrid functionality for multi-dimensional problems
• Training and evaluation tools

Installation

PINN-kit supports macOS, Linux, and Windows. You can install it using pip, which will also automatically install all required dependencies, including torch.

Create a Virtual Environment (Recommended)

It is recommended to use a virtual environment with Python 3.12 to avoid dependency conflicts.

macOS & Linux

python3.12 -m venv venv
source venv/bin/activate

Windows

Open Command Prompt or PowerShell and run:

python3.12 -m venv venv
venv\Scripts\activate

Install PINN-kit

macOS & Linux

pip install pinn-kit

Windows

Open Command Prompt or PowerShell and run:

pip install pinn-kit

Note: There is no need to install torch separately; it will be installed automatically with pinn-kit.

PINN-kit is easy to use

from pinn_kit import PINN, Domain, convert_to_meshgrid
import numpy as np

# Create a domain with flexible variable definition
domain = Domain([
    ('x', -1, 1),  # x-coordinate bounds
    ('y', -1, 1),  # y-coordinate bounds
    ('t', 0, 1)    # time bounds
])

# Sample points using various strategies
x_arr, y_arr, t_arr = domain.sample_points(
    num_samples=1000,
    sampler="lhs_classic"  # Latin Hypercube Sampling
)

# Create meshgrid from 1D arrays (new flexible interface)
x_1d = np.linspace(-1, 1, 50)
y_1d = np.linspace(-1, 1, 50)
t_1d = np.linspace(0, 1, 10)
xx, yy, tt = convert_to_meshgrid([x_1d, y_1d, t_1d])

# Initialize a PINN
pinn = PINN([3, 20, 20, 1])  # 3 inputs, 2 hidden layers, 1 output

# Define the loss terms
def loss():
    return loss_function

# Train the network
pinn.train_model([x_arr, y_arr, t_arr], loss_list, num_epochs=1000)

Documentation

For detailed documentation, please visit our documentation page.

License

This project is licensed under the MIT License - see the LICENSE file for details.