We implement the original PCA as a baseline, the probabilistic PCA [2] and Bayesian PCA [1]. We then also try to extend the bayesian formulation to the kernel PCA based also on the work on the Probabilistic Kernel PCA [3].

This project was done as part of a course on Bayesian Learning, taught by Remi Bardenet as part of the Master Data Science at the University of Lille.

A notebook is available with the conducted experiments (here).

• NumPy
• SciPy
• Matplotlib

To install simply clone the project :

git clone https://github.com/MaxenceGiraud/BayesianPCA
cd BayesianPCA

import bayespca as bpca

## Bayesian PCA
b = bpca.BayesianPCA()
b.fit_transform(X)

## Hinton diagram
bpca.utils.hinton(b.W.T)

• Implement original PCA using SVD and eigen decomposition
• Probabilistic PCA (via eigendecomposition)
• Implement EM algo to compute Probabilistic PCA
• Bayesian PCA using EM
• Kernel PCA
• Probabilistic Kernel PCA (EM + via eigendecomposition)
• Bayesian Kernel PCA

[1] Bishop, C. M. (1999). Bayesian PCA. MIT Press.
[2] Tipping, M. E . and C. M. Bishop (1997). Probabilistic principal component analysis. Journal of the Royal Statistical Society, B.
[3] Z. Zhang, G. Wang, D.-Y. Yeung, and J. Kwok. Probabilistic kernel principal component analysis. 2004.