Wrapper for LASSO in R

The default use of cross-validation in the R package lars is to give one k-fold average. In this wrapper, we average over the k-fold outcomes to give a more stable cross-validation function. The two programs LASSO wrapper and modified cv.lars need to be sourced. The modification (due to Howard Bondell) is so that the fraction is defined in terms of the full fit rather than the reduced data sets. Here is a typical call:

> set.seed(987)
> lasso.dr1(x=ncaa2[,1:19],y=ncaa2$y,rep=10)->out1

K (num. of folds)= 5
Num. of reps)= 10
Fraction from CV= 0.7272727
number nonzero= 16
mse= 48.36116
Variables Selected = x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x15 x17 x18

and a second call to show that the results don't change much.

> lasso.dr1(x=ncaa2[,1:19],y=ncaa2$y,rep=10)->out2

K (num. of folds)= 5
Num. of reps)= 10
Fraction from CV= 0.7373737
number nonzero= 16
mse= 48.08779
Variables Selected = x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x15 x17 x18

However, the default lars call has rep=1. Here are 4 replications showing
the variation.

> set.seed(345)
> lasso.dr1(x=ncaa2[,1:19],y=ncaa2$y,rep=1)->out3

K (num. of folds)= 5
Num. of reps)= 1
Fraction from CV= 0.5555556
number nonzero= 11
mse= 52.12135
Variables Selected = x1 x2 x3 x4 x5 x6 x7 x9 x11 x15 x17

> lasso.dr1(x=ncaa2[,1:19],y=ncaa2$y,rep=1)->out4

K (num. of folds)= 5
Num. of reps)= 1
Fraction from CV= 0.7676768
number nonzero= 16
mse= 47.34128
Variables Selected = x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x15 x17 x18

> lasso.dr1(x=ncaa2[,1:19],y=ncaa2$y,rep=1)->out5

K (num. of folds)= 5
Num. of reps)= 1
Fraction from CV= 0.5050505
number nonzero= 11
mse= 55.45815
Variables Selected = x1 x2 x3 x4 x5 x6 x7 x9 x11 x15 x17

> lasso.dr1(x=ncaa2[,1:19],y=ncaa2$y,rep=1)->out6

K (num. of folds)= 5
Num. of reps)= 1
Fraction from CV= 0.6060606
number nonzero= 14
mse= 51.41852
Variables Selected = x1 x2 x3 x4 x5 x6 x7 x9 x10 x11 x13 x15 x17 x18

Here is an illustration of the effect of the modification to cv.lars.

Without modification on the wing data set:

> lasso.dr(x=wing[,1:37],y=wing$y,rep=10)->out

K (num. of folds)= 5
Num. of reps)= 10
Fraction from CV= 0.6666667
number nonzero= 34
mse= 1.768729e-05
Variables Selected = x1 x2 x3 x4 x5 x7 x9 x10 x11 x12 x13 x14 x15 x16 x17 x23
x27 x30 x31 x33 x34 x35 x36 x37 x38 x39 x40 x41 x42 x43 x44 x45 x46 x48

With the modification:

> lasso.dr1(x=wing[,1:37],y=wing$y,rep=10)->out

K (num. of folds)= 5
Num. of reps)= 10
Fraction from CV= 0.5050505
number nonzero= 26
mse= 2.004648e-05
Variables Selected = x1 x4 x5 x7 x10 x11 x12 x13 x14 x15 x16 x17 x23 x27 x30
x34 x35 x38 x40 x41 x42 x43 x44 x45 x46 x48