Figure 2.
The rank-ordered influence function with a feature-metric set to the mean for lasso, ridge regression, and PCA-lariat methods. Note that βi delineates the ith parameter of the lasso/ridge regression and λ is the Lagrange multiplier associated with the lasso/ridge regularization where λ = 0 is the un-regularized regression. As λ increases the regularization is increased, decreasing the magnitude of all βi for i > 1 while increasing the influence of β0 [113].