/AMIPPaper

Primary LanguageTeXApache License 2.0Apache-2.0

AdversarialInfluenceWorkbench

This repository contains code to reproduce our paper, An Automatic Finite-Sample Robustness Metric: When Can Dropping a Little Data Make a Big Difference? by Tamara Broderick, Ryan Giordano, and Rachael Meager.

The writing directory is writing/output/. There is data processing code in both examples and in writing/applications.

The best guide to reproducing the paper and its analyses is found in writing/output/makefile. To run it, first

  1. Set the GIT_REPO_LOC variable at the top of makefile to point to the full path of the location of the cloned AMIPPaper repository
  2. Run make all in the output directory.
  3. Follow the instructions to download the needed data.
  4. Continue to run make all and follow the instructions until the paper succesfully compiles.

To clear paper output, run make clean.

To pre-process data for individual analyses, you can run any of the subsidiary targets:

  • make sim_data
  • make cash_data
  • make mc_data
  • make mc_data

Note that the most complicated analysis, the mixture model, requires more effort to run than the R analysis.

In order to simply compile the paper without running any of the individual analyses, you can uncompress the file writing/output/applications_data.tgz, which contains the processed data we used for our original paper. The contents of the arxiv should replace the contents of the writing/output/applications_data/ directory, and satisfy the makefile target $(PP_DATA). These can also be used to sanity check your own runs against ours.

If you have problems reproducing any aspect of the pipeline, please send Ryan an email.