remin: Residual Minimizer - Physics Informed Neural Networks

remin is a Python module that provides a framework for developing Physics Informed Neural Networks (PINNs) using PyTorch. This module integrates PyTorch for model creation and training, Latin-Hypercube sampling for geometry creation, and also includes a small, user-friendly module for creating geometries.

Key Features

• Implementation of Physics Informed Neural Networks (PINNs) using PyTorch.
• Support for creating geometries using Latin-Hypercube sampling.
• Intuitive module for defining and creating complex geometries.
• Command-line interface (CLI) for simultaneous training of multiple models.

Installation

Prerequisites

Before using remin, you will need to install the following packages:

• PyTorch (version 2.0.0 or later)
• NumPy
• pyDOE (version 0.3.8 or later)

You can install all the requirements using pip, the Python package installer. To install PyTorch, run the following command or follow the instructions on their website to install with GPU support:

pip install torch>=2.0.0

To install NumPy, run the following command:

pip install numpy

To install pyDOE, run the following command:

pip install pyDOE>=0.3.8

Alternatively, you can install these packages using conda or any other package manager of your choice.

Once you have installed these packages, remin can be installed using pip(PyPi):

pip install remin

Usage

To use remin in your Python project, import the necessary modules as follows:

import remin.geometry as rd
import remin.func as rf
import remin.solver.residual_loss as rl
from remin.solver import Solver, make_trainer
from remin.residual import Residual, make_loader

Will be extended in future.

Contributing

If you want to contribute to remin, feel free to submit a pull request or open an issue on GitHub.

License

This project is licensed under the MIT License.

Contact

For any questions, suggestions, or feedback, please feel free to contact the maintainer at [salih.tasdelen@metu.edu.tr].

Thank you for using remin!