DScribe: Library of descriptors for machine learning in materials science
Lauri Himanen
Abstract
This talk introduces DScribe which is a Python package for transforming atomic structures into fixed-size numerical fingerprints. These fingerprints are often called "descriptors" and they can be used in various tasks, including machine learning, visualization, similarity analysis, etc. DScribe provides user-friendly, off-the-shelf descriptor implementations and is freely available under the open-source Apache License 2.0. This seminar will include several hands-on examples on how these descriptors can be used for different tasks, including property prediction through a regression model, data clustering through unsupervised learning, and visualization of complex chemical environments. The package currently contains implementations for Coulomb matrix, Ewald sum matrix, sine matrix, Smooth Overlap of Atomic Positions (SOAP), Many-body Tensor Representation (MBTR), Local Many-body Tensor Representation (LMBTR), Atom-centered Symmetry Function (ACSF) and the Valle-Oganov descriptor.