Table 4. glmnet timings for fitting a lasso problem in various settings^†.

Setting	Correlation	Time (s) without strong rule	Time (s) with strong rule
Gaussian	0	0.99 (0.02)	1.04 (0.02)
	0.4	2.87 (0.08)	1.29 (0.01)
Binomial	0	3.04 (0.11)	1.24 (0.01)
	0.4	3.25 (0.12)	1.23 (0.02)
Cox	0	178.74 (5.97)	7.90 (0.13)
	0.4	120.32 (3.61)	8.09 (0.19)
Poisson	0	142.10 (6.67)	4.19 (0.17)
	0.4	74.20 (3.10)	1.74 (0.07)

^†There are p=20000 predictors and N =200 observations. Values shown are the mean and standard error of the mean over 20 simulations. For the Gaussian model the data were generated as standard Gaussian with pairwise correlation 0 or 0.4, and the first 20 regression coefficients equalled to 20, 19,…,1 (the rest being 0). Gaussian noise was added to the linear predictor so that the signal-to-noise ratio was about 3.0. For the logistic model, the outcome variable y was generated as above, and then transformed to {sgn(y) + 1}/2. For the survival model, the survival time was taken to be the outcome y from the Gaussian model above and all observations were considered to be uncensored.