PCA based on Multivariate MM-estimators with Fast and Robust Bootstrap



We provide R code and MATLAB code to compute robust PCA estimates based on multivariate MM-estimators of location and shape and to perform inference based on the fast and robust bootstrap. The code is well documented with instructions on its use. Questions can be sent to Gert Willems.

R code

Main function to compute robust PCA based on an MM-estimate of shape with inference based on the fast and robust bootstrap: MMPCA.R
Necessary additional function that computes the multivariate MM-estimates of location and shape: multiMM.R
Necessary additional function that computes S-estimates of location and scatter to find the initial S-scale for the MM-estimator: fastSmulti_location.R
Necessary additional function that executes the fast and robust bootstrap for MM-estimates: multiMMRobBoot_location.R
Necessary additional function that computes empirical influences for MM estimates as needed to construct BCa bootstrap intervals: einfsMM.R

MATLAB code

Main function to compute robust PCA based on an MM-estimate of shape with inference based on the fast and robust bootstrap: MMPCA.m
Necessary additional function that executes the fast and robust bootstrap for MM-estimates: multiMMRobBoot.m
Necessary additional function that computes the multivariate MM-estimates of location and shape as well as the initial S-scale obtained from S-estimates of location and scatter: multiMM.m

Reference

Salibian-Barrera, M., Van Aelst, S., and Willems, G. (2005), " PCA based on Multivariate MM-estimators with Fast and Robust Bootstrap, " Journal of the American Statistical Association , 101, 1198 - 1211. Paper