Figure 1: A graphical illustration of the properties of three different penalty functions. The eclipses represent the contours of the objective functions (A-C). The square, round, and star shapes represent the lasso, ridge, and bridge constraint, respectively. The dots are the points where contours are “tangent” to the constraints, i.e., the regularized estimates. Note that, in lasso (A) or bridge (C), the constraint is discontinuous at zero. If the contour first touches the constraint at point zero, the corresponding parameter estimate is zero and variable selection is achieved [34].