fastM
Multivariate M-estimators are usually computed using a fixed-point algorithm. As shown in Duembgen et al. (2016) a partial Newton-Raphson procedure applied to the second order Taylor expansion of the target function can make the computation considerably faster. We implement this new algorithm for the multivariate M-estimator of location and scatter using weights coming from the multivariate t-distribution (Kent et al., 1994), its symmetrized version, Tyler's shape matrix (Tyler, 1987) and Duembgen's shape matrix (Duembgen, 1998). For the symmetrized M-estimators we work with incomplete U-statistics to accelerate our procedures initially.
Duembgen, L. (1998), On Tyler's M-functional of scatter in high dimension, Annals of Institute of Statistical Mathematics, 50, 471--491.
Duembgen, L., Nordhausen, K. and Schuhmacher, H. (2016), New algorithms for M-estimation of multivariate location and scatter, Journal of Multivariate Analysis, 144, 200--217. 10.1016/j.jmva.2015.11.009
Kent, J.T., Tyler, D.E. and Vardi, Y. (1994), A curious likelihood identity for the multivariate t-distribution, Communications in Statistics, Theory and Methods, 23, 441--453.
Tyler, D.E. (1987), A distribution-free M-estimator of scatter, Annals of Statistics, 15, 234--251.