MN-PCA

MN-PCA (matrix normal principal component analysis) is a powerful and intuitive PCA method through modeling the graphical noise by the matrix normal distribution. MN-PCA obtains a low-rank representation of data and the structure of the correlated noise simultaneously.

• illustrative_example.m is an illustractive example to show the projections of PCA and MN-PCA on the synthetic dataset.
• ./MN-PCA-MRL contains the main function of the maximizing regularized likelihood algorithm of MnPCA.m
• ./MN-PCA-w2 contains the matlab wrapper function MnPCAq1_wrapper.m for the minizing Wasserstein distance algorithm. The algorithm is written with PyTorch (MnPCAq1.py).

Citation

@article{Zhang2019,
  title={Matrix normal PCA for interpretable dimension reduction and graphical noise modeling},
  author={Zhang, Chihao and Gai, Kuo and Zhang, Shihua},
  journal={arXiv preprint arXiv:1911.10796},
  year={2019}
}

Tutorial

The tutorial is available at here