Regularizing Lasso: a Consistent Variable Selection Method

Table 1 provides the average computational time (in minutes) for the eight methods under the simulation settings. SIS clearly requires the least computational effort, whereas RLASSO as well as Scout require much longer computational time. But all methods except RLASSO(CLIME) can be computed under a reasonable amount of time for p = 5000 and n = 100. RLASSO(CLIME) takes much longer because of inverting a matrix of 5000 dimension. However, 790.8 minutes of computation may still be acceptable. In an unreported simulation with p = 2000 and the same other settings, the average computational time for RLASSO(CLIME) is 46.7 minutes.