This is the repository for the RADNET architecture for polarization and electronic dielectric function predictions in periodic solids, and Raman intensities calculations.
Results obtained with this repo have been published as Olivier Malenfant-Thuot et al 2024 J. Phys.: Condens. Matter 36 425901.
figures/: Contains codes and scripts to generate the data and figures of the paper
radnet/: Contains the code package to install to use the RADNET architecture
scripts/: Contains the various prewritten scripts used in the paper and based on the package. See the Usage section for more details on scripts. Every script can be called with --help to get an explanation of all input parameters.
To install, simply git clone the repository and run
pip install . (or pip install -e . for local modifications)
In all cases the --help argument will give a more detailed description of every input parameter.
trainer.py:
Trains a model instance from a .h5 train dataset and a set of defined hyperparameters
inference.py:
Uses a trained model instance and a .h5 testing dataset (or the validation part of a training dataset) to evaluate the MAE and RMSE of the model
predict_raman.py:
Uses a trained model instance and a position input file (ase compatible) to predict all necessary values to the calculation of a Raman spectra.
predict_spectrum.py:
Combines the quantities computed with predict_raman.py to output Raman intensities