README.md

In wwrechard/screening:

screening

An R package for efficient variable screening for linear and generalized linear models.

Description

This package implements four popular variable screening algorithms, the sure independence screening1 (SIS), high-dimensional ordinary least squares projection2 (HOLP), rank-robust correlation screening3 (RRCS) and the Forward regression4 (FR) for efficient screening operation over large p small n models. SIS, HOLP and FR are implemented to work with both linear models and generalized linear models while RRCS works only for linear models. We refer the reader to the references and the R-help docs for detailed information.

Dependency

For linear model screening, the package does not rely on any external packages. For generalized linear models, HOLP relies on the package glmnet to compute the ridge-regression estimator. This functional dependency can be replaced by any user function or package that achieves the same functionality (i.e., ridge regression for generalized linear models).

Installation

To install the package in a painless way, you will need to first install devtools package and then execute the following lines in your favorate R IDE

> library(devtools)
> install_github('wwrechard/screening')
> library(screening)

The package is now ready for use. Simply call the helper by typing

> ?screening

will provide further information on the usage of the function. If there is any issue installing the devtools package, you may instead downlowd the R files from directory /R directly. The file screening.r contains all working functions required for using the algorithm.

Quick start

A simple working example. Assume we observe 200 samples with 500 dimension, stored in x and y. Following code illustrates screening using a linear model

> library(screening)
> output = screening(x, y, method = 'holp', num.select = 50)
> print (output$screen)

output$screen contains the indexes of the screened variables, which, in this case, is a 50-dimensional vector. If you opt to use extended BIC for determining the screening set, then simply replace the second line in the above code with

> output = screening(x, y, method = 'holp', ebic = TRUE)

The package contains other two testing functions linearModelTest and logisticTest. The two functions provide integrated tests on the four screening methods. The full functionality includes simulating synthetic datasets and apply screening methods and report the result. Their codes also serve as good examples of using screening. For more detailed information, please refer to their R-help docs.

Author

Xiangyu (Samuel) Wang and Chenlei Leng. We appreciate your feedback under issues of this repository.

References

1. Fan, Jianqing, and Jinchi Lv. "Sure independence screening for ultrahigh dimensional feature space." Journal of the Royal Statistical Society: Series B (Statistical Methodology) 70.5 (2008): 849-911.

2. Wang, Xiangyu, and Chenlei Leng. "High-dimensional ordinary least-squares projection for screening variables." Journal of the Royal Statistical Society: Series B (Statistical Methodology) (To appear)

3. Li, Gaorong, et al. "Robust rank correlation based screening." The Annals of Statistics 40.3 (2012): 1846-1877.

4. Wang, Hansheng. "Forward regression for ultra-high dimensional variable screening." Journal of the American Statistical Association 104.488 (2009): 1512-1524.

wwrechard/screening documentation built on May 4, 2019, 12:04 p.m.