Contents:

      Intro

      Code





Intro


This website accompanies the paper:
  • Selective inference for group-sparse linear models.
    Fan Yang, Rina Foygel Barber, Prateek Jain, & John Lafferty. arXiv:1607.08211

This paper develops post-selection confidence intervals for the effect size of a group of variables in a linear regression, which can be applied after a group-sparse model selection procedure such as forward stepwise regression, iterative hard thresholding, or the group lasso.

Here we give code in the R language providing an implementation of the selective confidence intervals and p-values for each of the three model selection methods, and reproducing the simulation results and real data results in the paper.





Code


Implementation:
  • Implementation of the selective inference methods for forward stepwise regression: FS.R
  • Implementation of the selective inference methods for iterative hard thresholding: IHT.R
  • Implementation of the selective inference methods for the group lasso: GL.R

Simulation: this script reproduces the results in the paper (note that this script runs for several days in order to run multiple trials - to reset to a lower number of trials, modify the parameters B, B_GL, B_pvals).
  • The script: simulation.R
  • A few more functions needed for the simulations: functions.R
  • The functions FS.R, IHT.R, GL.R from above are also required.

Real data experiment: this script reproduces the results in the paper using the county health data set (details on the data set: http://www.countyhealthrankings.org/).