SuSiE-PCA¶

SuSiE PCA is a scalable Bayesian variable selection technique for sparse principal component analysis

SuSiE PCA is the abbreviation for the sum of single effects model in principal component analysis (SuSiE PCA). We develop SuSiE PCA for an efficient variable selection in PCA when dealing with high dimensional data with sparsity, and for quantifying uncertainty of contributing features for each latent component through posterior inclusion probabilities (PIPs). We implement the model with the JAX library developed by Google which enable the fast training on CPU, GPU or TPU.

Contents¶