/DoubleAE

Primary LanguagePythonMIT LicenseMIT

Online-compatible Unsupervised Non-resonant Anomaly Detection Repository

Repository containing all scripts used in the studies of Online-compatible Unsupervised Non-resonant Anomaly Detection model.

To train your own model, first Download the official dataset from zenodo and use the example code to prepare the datasets. To run the training, use:

python AE40Mhz.py [--single/--double/--supervised/--all] [--load] --out NAME

To train a single AE, the double + decorrelatied method, supervisedd, or all of them respectively. Trained model weights are also providedd in the weights folder that can be loaded using the --load flag.

The output of the script will create an NAME.h5 file in the base directory. Use this file to plot the results using the script plot.py

python plot.py --file NAME.h5

Different plot options are available in the script.

For any use of paper ideas and results, please cite

@article{PhysRevD.105.055006,
  title = {Online-compatible unsupervised nonresonant anomaly detection},
  author = {Mikuni, Vinicius and Nachman, Benjamin and Shih, David},
  journal = {Phys. Rev. D},
  volume = {105},
  issue = {5},
  pages = {055006},
  numpages = {9},
  year = {2022},
  month = {Mar},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevD.105.055006},
  url = {https://link.aps.org/doi/10.1103/PhysRevD.105.055006}
}