/mlmod

Machine learning package for data-driven modeling and simulation of particle systems, materials, and complex fluids (interfaces with LAMMPS MD package)

Primary LanguageC++OtherNOASSERTION

Documentation | Examples | Paper

MLMOD: Machine Learning Methods for Data-Driven Modeling in LAMMPS

Now available with Jupyter notebooks and Python scripts for readily setting up models and simulations.

MLMOD is a Python/C++ package for utilizing machine learning methods and data-driven modeling for simulations in LAMMPS. The package provides methods for time-step integrators for dynamics and interactions using general ML model classes, including Neural Networks, Kernel Regression, and others. Models can be trained and exported from PyTorch or from other machine learning frameworks. Note, this is an early prototype alpha release with future implementations and features to come. Please provide feedback or information on bugs using the forms below.

LAMMPS is an optimized molecular dynamics package in C/C++ providing many interaction potentials and analysis tools for modeling and simulation. Interaction methods include particle-mesh electrostatics, common coarse-grained potentials, many-body interactions, and others.

Quick Start

To try automatically to install the package for Python, you can use the script quick_install.py.

python ./quick_install.py

To install directly the pre-compiled packages for Python, download one of the following

Install using

pip install -U (substitute-filename-or-url-here).whl

To test the package installed run

python -c "from mlmod_lammps.tests import t1; t1.test()"

For example models and simulations, see the notebooks and scripts in the examples folder.

Documentation pages can be found here.

Pre-compiled binaries are currently for (Debian 9+/Ubuntu and Centos 7+, Python 3.6+).

If you installed previously this package, please be sure to update to the latest version using pip install -U (substitute-filename-or-url-here).whl

Video giving a brief overview of MLMOD can be found here.

Other ways to install the package For running prototype models and simulations on a desktop, such as Windows and MacOS, we recommend using Docker container. For example, install Docker Desktop, or docker for linux, and then load a standard ubuntu container by using in the terminal docker run -it ubuntu:20.04 /bin/bash You can then use apt update; apt install python3-pip, and can then pip install and run the simulation package as above. Note use command python3 in place of python when calling scripts. Pre-installed packages in anaconda also in docker run -it atzberg/ubuntu_20_04_anaconda1 /bin/bash Use conda activate mlmod-lammps May need to update packages to the latest version.

For more information on other ways to install or compile the package, please see the documentation pages.

Please cite the paper below for this package:

MLMOD Package: Machine Learning Methods for Data-Driven Modeling in LAMMPS", P.J. Atzberger, Journal of Open Source Software, 8(89), 5620, (2023) [paper link].

@article{mlmod_atzberger,
  author    = {Paul J. Atzberger},
  journal   = {Journal of Open Source Software}, 
  title     = {MLMOD: Machine Learning Methods for Data-Driven Modeling in LAMMPS},
  year      = {2023},  
  publisher = {The Open Journal},
  volume    = {8},
  number    = {89},
  pages     = {5620},
  note      = {http://atzberger.org},
  doi       = {10.21105/joss.05620},
  url       = {https://doi.org/10.21105/joss.05620}
}

Tutorials / Talks:

  • Video giving a brief overview of MLMOD can be found here.

  • Documentation on using the package can be found here.

  • Examples can be found here.

Mailing List for Future Updates and Releases

  • Please join the mailing list for future updates and releases here.

Bugs or Issues

  • If you encounter any bugs or issues please let us know by providing information here.

Please submit usage and citation information

  • If you use this package or related methods, please let us know by submitting information here. This helps us with reporting and with further development of the package. Thanks.

Acknowledgements We gratefully acknowledge support from NSF Grant DMS-2306101, NSF Grant DMS-1616353, and DOE Grant ASCR PHILMS DE-SC0019246.

Additional Information
http://atzberger.org/

Documentation | Examples | Paper | Atzberger Homepage