Maintainer: Laurent Callot ([email protected])

Date: 01/05/2015

This repository contains the replication material for the simulations and the empirical application in Thresholded Lasso: Variable Selection in High Dimensions. All the computations are carried using R and the knitr package for easy replication. The material is divided between 4 folders:

• application: contains a knitr file, application.Rnw, to be compiled to replicate the results of the application.
• code: contains 3 files. tlas.R contains the code for the thresholded Lasso and the simulations. The other two files contain functions to print the result to a tex table.
• data: BIS_data.Rda contains a 3 dimensional array with the data used in the application.
• mc: contains the files for the simulations. 7 files called xp_XXX.R contain the setup and call to ttlas for each of the 5 experiments. xp_all.R contains the global settings for the simulations and runs the experiments. It is the file that should be ran to compute the simulations, it takes around 3 hours to complete computations on 16 cores for 1000 replications. mc.Rnw should be compiled to generate the tables in the paper from the saved simulations statistics located in the mc/save/ folder. plot_psize.R generates figure 1, it is also based on 1000 replications making it a bit heavy to compute.