A Python package with data-driven pipelines for dynamical systems machine learning.
The SimulAI toolkit provides easy access to state-of-the-art models and algorithms for physics-informed machine learning. Currently, it includes the following methods described in the literature:
- Physics-Informed Neural Networks (PINNs)
- Deep Operator Networks (DeepONets)
- Variational Encoder-Decoders (VED)
- Koopman AutoEncoders (experimental)
- Operator Inference (OpInf)
In addition to the methods above, many more techniques for model reduction and regularization are included in SimulAI. See documentation.
Python version requirements: [3.6, 3.10)
pip install simulai-toolkit
conda install -c conda-forge simulai-toolkit
It is strongly recommended that you have a Miniconda distribution installed. You can contribute code to SimulAI by following the steps below:
- Fork this repository
- Git clone the forked repository
- Create a conda environment
conda create -n simulai -f environment.yml
- Activate the newly created environment
conda activate simulai
cdto the root directory- Set PYTHONPATH environment variable
export PYTHONPATH=.
- Run pytest
pytest tests/
SimulAI supports multiprocessing with MPI.
In order to use it, you will need a valid MPI distribution, e.g. MPICH, OpenMPI.
conda install -c conda-forge mpich gcc
If you have problems installing gcc using the command above, we recommend you to install it using Homebrew.
Please, refer to the SimulAI API documentation before using the toolkit.
Additionally, you can refer to examples in the examples folder.
This software is licensed under Apache 2.0.
The following references in the literature.
If you find SimulAI to be useful, please consider citing it in your published work:
@software{simulai,
author = {IBM},
title = {SimulAI},
url = {https://github.ibm.com/simulai/simulai},
version = {},
date = {},
}