Robust covariance and precision matrix estimators. Based on the review of P.-L. Loh and X. L. Tan. (2018)

To install:


There are in total 4 robust covariance and 3 correlation estimation implemented, namely:

P.-L. Loh and X. L. Tan. (2018) then used these robust estimates in Graphical Lasso (package glasso) or Quadratic Approximation (package QUIC) to obtain sparse solutions to precision matrix

With glasso, a function robglasso stand for robust graphical LASSO is implemented. It has build in cross validation described in P.-L. Loh and X. L. Tan. (2018), for instance, to use the method with cross validation:

robglasso(data=matrix(rnorm(100),20,5), covest = cov,CV=TRUE)

Where data should be a matrix and covest should be a function that estimate the covariance e.g. anyone mentioned above. The result list contains everything from glasso output with the optimal tuning parameter found by cross validation. One can also decide fold by setting fold in robglasso. For more details see ?robglasso.