This is a python implementation of the opt-NOMC algorithm, as well as sliced Wasserstein distances and kernel approximation applications accompany the publication "Demystifying Orthogonal Monte Carlo and Beyond" by Han Lin, Haoxian Chen, Tianyi Zhang, Clement Laroche, and Krzysztof Choromanski.

NOMC relies on Python 3.7 as well as some commonly-used libraries as mentioned in requirements.txt. The specific python and libraries versions should not matter too much in the implementation of our algorithm.

opt_NOMC.py implements opt-NOMC algorithm, with parameters set in config_NOMC.json. QMC.py implements QMC algorithm, with parameters set in config_QMC.json. SWD.py implements sliced Wasserstein distance experiments with multivariate gaussian as an example, and the parameters can be set in config_SWD.json.

Besides, we also includes some already calculated and optimized NOMC and QMC samples under the folder data, which can be used directly to replicate our experiments using NOMC and QMC algirhms in the paper.

If you think this project is helpful, please feel free to give a ⭐️ and cite our paper:

@article{lin2020demystifying,
  title={Demystifying orthogonal Monte Carlo and beyond},
  author={Lin, Han and Chen, Haoxian and Choromanski, Krzysztof M and Zhang, Tianyi and Laroche, Clement},
  journal={Advances in Neural Information Processing Systems},
  volume={33},
  pages={8030--8041},
  year={2020}
}