DScribe: Library of descriptors for machine learning in materials science

Published: 2 Oct 2019 | Version 1 | DOI: 10.17632/vzrs8n8pk6.1

Description of this data

DScribe is a software package for machine learning that provides popular feature transformations (“descriptors”) for atomistic materials simulations. DScribe accelerates the application of machine learning for atomistic property prediction by providing user-friendly, off-the-shelf descriptor implementations. The package currently contains implementations for Coulomb matrix, Ewald sum matrix, sine matrix, Many-body Tensor Representation (MBTR), Atom-centered Symmetry Function (ACSF) and Smooth Overlap of Atomic Positions (SOAP). Usage of the package is illustrated for two different applications: formation energy prediction for solids and ionic charge prediction for atoms in organic molecules. The package is freely available under the open-source Apache License 2.0.

Experiment data files

This data is associated with the following publication:

DScribe: Library of descriptors for machine learning in materials science

Published in: Computer Physics Communications

Latest version

  • Version 1


    Published: 2019-10-02

    DOI: 10.17632/vzrs8n8pk6.1

    Cite this dataset

    Himanen, Lauri; Jäger, Marc O.J.; Morooka, Eiaki V.; Canova, Filippo Federici; Ranawat, Yashasvi S.; Gao, David Z.; Rinke, Patrick; Foster, Adam S. (2019), “DScribe: Library of descriptors for machine learning in materials science”, Mendeley Data, v1 http://dx.doi.org/10.17632/vzrs8n8pk6.1


Views: 15
Downloads: 6


Computational Physics


Apache-2.0 Learn more

The files associated with this dataset are licensed under a Apache License 2.0 licence.

What does this mean?

Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form.