Many-body implementation of gradient-domain machine learning potentials
Project description
mbGDML
Create, use, and analyze many-body gradient-domain machine learning potentials.
About • Features • Installation • License
About
Atomistic insight is fundamental for computational predictive studies of chemical and physical processes. Machine learning potentials (or force fields) can reproduce high-level ab initio calculations without the substantial costs. Gradient-domain machine learning (GDML), a kernel-based method, directly learns the relationship between atomic coordinates and interatomic forces with only hundreds of data points. However, it uses the inverse internuclear distance descriptor which sacrifices generalized transferability to other species or number of atoms.
Many-body GDML (mbGDML), is a route for GDML size-transferable potentials by using GDML to learn n-body interactions for use in a many-body expansion approach. Every aspect of the process from preparing energy+gradient calculations, creating data sets, training GDML models, and making predictions is taken care of in this user-friendly Python package.
Disclaimer: This package is still under active development and is not ready for production.
Features
Creating mbGDML models:
- Partition structures into monomers, dimers, trimers, etc.
- Structure and data set breadcrumb trails using MD5 hashes.
- Simple GDML training interface using CPUs or GPUs.
Using mbGDML models:
- Energy and force predictions.
- Calculator interface with the Atomic Simulation Environment (ASE).
Analyzing mbGDML models:
- Avoid recalculating energies and forces by storing predictions into predict sets (npz files).
- Visually examine structural similarity and prediction accuracy using dimensionality reduction with UMAP.
- Analyze n-body contributions and predictions with heat maps.
Installation
You can install mbGDML by using pip install mbgdml
.
Or, the latest development version can be installed directly from the GitHub repository.
git clone https://github.com/keithgroup/mbGDML
cd mbGDML
pip install .
License
Distributed under the MIT License. See LICENSE
for more information.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.