spca: Sparse Principal Component Analysis via penalised matrix decomposition

MATLAB implementation of the sparse principal component analysis method proposed in Witten et al. (2009).

Quick start

load spca_example
[coeff,score] = spca(X,0.5,'K',2);

X is a data matrix with observations in rows, the second argument is a parameter between 0 and 1 controlling coefficient vector sparsity: 0 is maximal sparsity, 1 is no sparsity. The name-value argument 'K' controls how many principal components are calculated.

Optimising the sparsity parameter

c = tunespca(X,'K',2);
[coeff,score] = spca(X,c,'K',2);

The function tunespca finds optimal values for the sparsity parameter using the missing value imputation scheme proposed in Witten et al. (2009). The parameter is optimised sequentially for each successive component, c(1)is the optimal value for the first component etc.

For more information about the arguments and behaviour, see

help spca

and the example script spca_example.m.

References

Witten, Daniela M., Robert Tibshirani, and Trevor Hastie. "A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis." Biostatistics 10.3 (2009): 515-534.