Skip to content

Latest commit

 

History

History
67 lines (42 loc) · 2.81 KB

README.md

File metadata and controls

67 lines (42 loc) · 2.81 KB
Raw

PhysNet DER modified

This is a new version of the PhysNet DER code for working with PES. The code computes energy, forces, charges, and dipole. The model also computes the following two modified versions of DER:

  • DER-Lipzchitz: This version uses a modified loss function that includes a term for Lipzchitz correction. It is adapted from here. The original reference can be found here.
  • DER-Multidimensional: This version uses a multidimensional prior, the Normal Inverse Wishart (NIW) distribution. It is adapted from here. The original reference can be found here.

Additionally, the code includes the basic version of DER, which is the original PhysNet DER.

To Do:

  • [] Debug the code

Using PhysNet Torch

Requirements

  • Python 3.8
  • PyTorch 1.10
  • Torch-ema 0.3
  • TensorBoardX 2.4

Using Physnet-DER

Setting up the environment

We recommend to use Miniconda for the creation of a virtual environment.

Once in miniconda, you can create a virtual environment called physnet_torch from the .yml file with the following command

conda env create --file environment.yml

To activate the virtual environment use the command:

conda activate physnet_torch

Running the code

For training the model, you can use the following command:

python run_train.py @input.inp

NOTE: You must define the type of DER that is wished to used. The options are: simple, lipz, and MD. If this is not defined, the code will not run.

Contact

This is still a work in progress, if you have questions or problems please contact:

L.I.Vazquez-Salazar, email: luisitza.vazquezsalazar@unibas.ch

Reference

  • Vazquez-Salazar, L. I.; Boittier, E. D.; Meuwly, M. Uncertainty quantification for predictions of atomistic neural networks. Chem. Sci. 2022, 13 (44), 13068-13084. DOI: 10.1039/D2SC04056E
  • Vazquez-Salazar, L.I.; Käser S.; Meuwly, M. Outlier-Detection for Reactive Machine Learned Potential Energy Surfaces. arXiv e-prints 2024, arXiv: arXiv:2402.17686.