GMP-Featurizer: A parallelized Python package for efficiently computing the Gaussian Multipole features of atomic systems
share
GMP-Featurizer is a lightweight, accurate, efficient, and scalable software package for calculating the Gaussian Multipole (GMP) features <cit.> for a variety of atomic systems with elements across the periodic table. Starting from the GMP feature computation module from AmpTorch <cit.>, the capability of GMP-Featurizer has since been greatly improved, including its accuracy and efficiency, as well as the ability to parallelize on different cores, even machines. Moreover, this python package only has very few dependencies that are all standard python libraries, plus cffi for C++ code interfacing and Ray <cit.> for parallelization, making it lightweight and robust. A set of unit tests are designed to ensure the reliability of its outputs. A set of extensive examples and tutorials, as well as two sets of pseudopotential files (needed for specifying the GMP feature set), are also included in this package for its users. Overall, this package is designed to serve as a standard implementation for chemical and material scientists who are interested in developing models based on GMP features. The source code for this package is freely available to the public under the Apache 2.0 license.
READ FULL TEXT