Author: Joseph W. Abbott, PhD Student @ Lab COSMO, EPFL
Note: work-in-progress!
A proof-of-concept framework for torch-based equivariant learning of scalar
fields and tensorial properties expanded in the angular basis. This package
provides the building blocks for end-to-end torch
-based learning and
prediction pipelines interfaced with metatensor
, a storage format for
atomistic data.
The subpackage rholearn
contains modules for loss functions, datasets,
dataloaders, models, and training, allowing gradient-based workflows with
minibatching to be built.
The subpackage rhocalc
contains infrastructure to interface the core
functionality in rholearn
with Quantum Chemistry codes. This involves routines
to generate learning targets and parse outputs into metatensor
format for
input in the ML workflow. Currently, only a simple interface (via ase
calculators) with the electronic structure code FHI-aims
is implemented, with
a focus on the generation of scalar fields expanded onto a fitted RI basis.
Some of the software modules rho_learn
conbines into a workflow are described
below:
-
rascaline
: Luthaf/rascaline. This is used to transform xyz coordinates of systems into a suitable (equivariant) structural representation. Hererascaline
calculators for generating a spherical expansion and performing CLebsch-Gordan density correlations are used to build$\lambda$ -SOAP descriptors. -
metatensor
: lab-cosmo/metatensor. This is a storage format for atomistic machine learning, allowing an efficient way to track data and associated metadata for a wide range of atomistic systems and objects.rho_learn
interfaces with subpackagesmetatensor-torch
andmetatensor-learn
to build a custom ML workflow for density learning. -
chemiscope
: lab-cosmo/chemiscope. This package is used a an interactive visualizer and property explorer for the molecular data from which the structural representations are built.
Pre-requisite: a working conda
installation. With this, follow the installation
instructions below.
git clone https://github.com/jwa7/rho_learn
cd rho_learn
conda env create --file install/environment.yaml
conda activate rho
./install/extra-pip-packages.sh
pip install .
In the case of error bash: ./install/extra-pip-packages.sh: Permission denied
you
might have to change the permission using
chmod +x ./install/extra-pip-packages.sh
before running
./install/extra-pip-packages.sh
.
TBC
-
Symmetry-Adapted Machine Learning for Tensorial Properties of Atomistic Systems, Phys. Rev. Lett. 120, 036002. DOI: 10.1103/PhysRevLett.120.036002
-
SALTED (Symmetry-Adapted Learning of Three-dimensional Electron Densities), GitHub: github.com/andreagrisafi/SALTED, Andrea Grisafi, Alan M. Lewis.
-
Transferable Machine-Learning Model of the Electron Density, ACS Cent. Sci. 2019, 5, 57−64. DOI: 10.1021/acscentsci.8b00551
-
Atom-density representations for machine learning, J. Chem. Phys. 150, 154110 (2019). DOI: 10.1063/1.5090481
-
Learning the Exciton Properties of Azo-dyes, J. Phys. Chem. Lett. 2021, 12, 25, 5957–5962. DOI: 10.1021/acs.jpclett.1c01425
-
Impact of quantum-chemical metrics on the machine learning prediction of electron density, J. Chem. Phys. 155, 024107 (2021), DOI: 10.1063/5.0055393