SCINE Parrot is a wrapper for machine learning potentials. It is an interface to

• lifelong machine learning potentials (lMLP)
• ANI (TorchANI)
• M3GNet (MatGL)
• MACE (MACE)

Each method is represented by its own Calculator and the entire wrapper constitutes a SCINE module that can be loaded dynamically at runtime. For more information on these concepts see the Scine::Core repository.

The full list of models includes

• Model("lmlp", "<generalization_setting_file>", "")
• Model("ani", "ani2x", "")
• Model("ani", "ani1ccx", "")
• Model("ani", "ani1x", "")
• Model("m3gnet", "m3gnet-mp-2021.2.8-pes", "")
• Model("m3gnet", "m3gnet-mp-2021.2.8-direct-pes", "")
• Model("mace", "mace-mp_large", "")
• Model("mace", "mace-mp_medium", "")
• Model("mace", "mace-mp_small", "")
• Model("mace", "mace-off_large", "")
• Model("mace", "mace-off_medium", "")
• Model("mace", "mace-off_small", "")

Parrot is distributed under the BSD 3-clause "New" or "Revised" License. For more license and copyright information, see the file LICENSE.txt in the repository.

The key requirements for Parrot are the Python packages scine_utilities and scine_xtb_wrapper. These packages are available from PyPI and can be installed using pip. However, these packages can also be compiled by hand. For the latter case, please visit the repositories of each of the packages and follow their guidelines.

Parrot can be installed using pip (pip3) once the repository has been cloned:

git clone <parrot-repo>
pip install ./parrot

A non super user can install the package using a virtual environment, or the --user flag.

The documentation can be found online, or it can be built using:

cd parrot
make -C docs html

It is then available at:

<browser name> docs/build/html/index.html

In order to build the documentation, you need a few extra Python packages which are not installed automatically together with Parrot. In order to install them, run

cd parrot
pip install -r requirements-dev.txt

When publishing results obtained with SCINE Parrot, please cite the corresponding release as archived on Zenodo (please use the DOI of the respective release). In addition, when publishing results obtained with other SCINE modules we kindly ask you to cite the appropriate references.

This wrapper should also not be mistaken for the respective actual machine learning potential code it wraps. For the latter code and its citations, we refer you to the respective publication/webpage.

In case you should encounter problems or bugs, please write a short message to [email protected].