Fig. 4

Histograms and kernel density estimates (i.e., the smooth lines) demonstrating the shrinkage of regression coefficients induced by using a Laplace L(0.5) prior instead of a Normal N(0,1) prior. The data are from one classification fold. Note that a large number of slopes in the Laplace plot are compressed towards zero (indicated by the higher central spike in the left subplot than the right one)