Table 2. Timings for the generalized gradient procedure for solving the lasso (Gaussian case)^†.

p	Time (s) without strong rule	Time (s) with strong rule
200	10.37 (0.38)	5.50 (0.26)
500	23.21 (0.69)	7.38 (0.28)
1000	43.34 (0.85)	8.94 (0.22)
2000	88.58 (2.73)	12.02 (0.39)

^†N =100 samples are generated in each case, with all entries N(0,1) and no signal (regression coefficients are 0). A path of 100 λ-values is used, spanning the entire operating range. Values shown are the mean and standard deviation of the mean, over 20 simulations. The times are somewhat large, because the programs were written in the R language, which is much slower than C or Fortran. However the relative timings are informative.