bigdata (version 0.1)

lasso.stars: Stability Approach to Regularization Selection for Lasso

Description

Implements the Stability Approach to Regularization Selection (StARS) for Lasso

Usage

lasso.stars(x, y, rep.num = 20, lambda = NULL, nlambda = 100, 
lambda.min.ratio = 0.001, stars.thresh = 0.1, sample.ratio = NULL, 
alpha = 1, verbose = TRUE)

Arguments

x

The n by d data matrix representing n observations in d dimensions

y

The n-dimensional response vector

rep.num

The number of subsampling for StARS. The default value is 20.

lambda

A sequence of decresing positive numbers to control regularization. Typical usage is to leave the input lambda = NULL and have the program compute its own lambda sequence based on nlambda and lambda.min.ratio. Users can also specify a sequence to override this. Use with care - it is better to supply a decreasing sequence values than a single (small) value.

nlambda

The number of regularization paramters. The default value is 100.

lambda.min.ratio

The smallest value for lambda, as a fraction of the uppperbound (MAX) of the regularization parameter which makes all estimates equal to 0. The program can automatically generate lambda as a sequence of length = nlambda starting from MAX to lambda.min.ratio*MAX in log scale. The default value is 0.001.

stars.thresh

The threshold of the variability in StARS. The default value is 0.1. The alternative value is 0.05. Only applicable when criterion = "stars"

sample.ratio

The subsampling ratio. The default value is 10*sqrt(n)/n when n>144 and 0.8 when n<=144< code="">, where n is the sample size.

alpha

The tuning parameter for the elastic-net regression. The default value is 1 (lasso).

verbose

If verbose = FALSE, tracing information printing is disabled. The default value is TRUE.

Value

An object with S3 class "stars" is returned:

path

The solution path of regression coefficients (in an d by nlambda matrix)

lambda

The regularization parameters used in Lasso

opt.index

The index of the optimal regularization parameter.

opt.beta

The optimal regression coefficients.

opt.lambda

The optimal regularization parameter.

Variability

The variability along the solution path.

Details

StARS selects the optimal regularization parameter based on the variability of the solution path. It chooses the least sparse graph among all solutions with the same variability. An alternative threshold 0.05 is chosen under the assumption that the model is correctly specified. In applications, the model is usually an approximation of the true model, 0.1 is a safer choice. The implementation is based on the popular package "glmnet".

References

1.Han Liu, Kathryn Roeder and Larry Wasserman. Stability Approach to Regularization Selection (StARS) for High Dimensional Graphical Models. Advances in Neural Information Processing Systems, 2010. 2.Jerome Friedman, Trevor Hastie and Rob Tibshirani. Regularization Paths for Generalized Linear Models via Coordinate Descent. Journal of Statistical Software, Vol.33, No.1, 2008.

See Also

bigdata-package

Examples

#generate data
x = matrix(rnorm(50*80),50,80)
beta = c(3,2,1.5,rep(0,77))
y = rnorm(50) + x%*%beta

#StARS for Lasso
z1 = lasso.stars(x,y)
summary(z1)
plot(z1)

#StARS for Lasso
z2 = lasso.stars(x,y, stars.thresh = 0.05)
summary(z2)
plot(z2)

#StARS for Lasso
z3 = lasso.stars(x,y,rep.num = 50)
summary(z3)
plot(z3)