FeS-PCA-Code

Repository containing a Python script that can be imported to evaluate FeS-PCA[1] and its variants. The unfederated counterparts are also available for comparison.

Files:

FeS_PCA.py: Script containing FeSK class for applying FeS-PCA and its variants to distributed data.

SPCA.py: Script containing SPCA class for applying standard SPCA and its variants to centralized data.

visualize_toy_datasets.py: Script which applies either FeS-PCA or SPCA to toy datasets for recreating Fig. 6 from [1] and Fig. 1-a, 1-d, 4-a, 4-d, 5-a, 5-d, from [2] (Note recreation of figures from [2] is not exact since precise parameters for kernel function and dataset generation were not given).

• First command line argument must be integer x in [0-5]: x mod 3 is dataset index in ['xor', 'rings', 'iris'] and ⌊x/2⌋ is method index in ['FeS-PCA/SPCA, FeSK-PCA/KSPCA].
• Choose between applying federated or unfederated algorithms by setting variable FEDERATED on line 15 to True or Flase, respectively.

data_utils.py: Script for generating toy datasets.

Bibliography:

[1] W. Briguglio, W. A. Yousef, I. Traore, and M. Mamun, “Federated Supervised principal component analysis”, SUBMITTED to IEEE Transactions on Information Forensics and Security

[2] E. Barshan, A. Ghodsi, Z. Azimifar, and M. Z. Jahromi, “Supervised principal component analysis: Visualization, classification and regression on subspaces and submanifolds”, Pattern Recognition, vol. 44, no. 7, pp. 1357–1371, 2011.