This is a Python package to automatically build the reference set for the training of Neural Network Potentials (NNPs), and eventually other machine-learned potentials, in an automated, data-driven fashion. For that purpose, a large set of reference configurations sampled in a physically meaningful way (typically with molecular dynamics) is filtered and the most important points for the representation of the Potential Energy Surface (PES) are identified. This is done by using a set of NNPs, called a committee, for error estimates of individual configurations. By iteratively adding the points with the largest error in the energy/force prediction, the reference set is progressively extended and optimized.
Keywords:
- Active learning
- Query by committee
- Ensemble averaging
- Committee machines
- Neural Network Potentials
More information can be found in the following references:
- C. Schran, F. L. Thiemann, P. Rowe, E. A. Müller, O. Marsalek, A. Michaelides,
"Machine learning potentials for complex aqueous systems made simple",
PNAS 118, e2110077118 (2021), 10.1073/pnas.2110077118 - C. Schran, K. Brezina, O. Marsalek,
"Committee neural network potentials control generalization errors and enable active learning",
J. Chem. Phys. 153, 104105 (2020), 10.1063/5.0016004
For now, just clone the repository and source the env.sh file.
Dependencies:
- NumPy
- SciPy
- Matplotlib
- MDTraj (for scoring)
- cp2k-input-tools (for CP2K calculations)