MLR Paper

This repository contains all the code to replicate the experiments and real data analyses from Spector and Fithian (2022). Running these scripts requires a python environment with knockpy installed (version 1.3+): see https://github.com/amspector100/knockpy for installation.

Overview

The directory mlr_src contains extraneous functions used in the simulations. However, the core contribution of the paper (MLR statistics) is implemented and published in the knockpy package to ease installation.

The directory sims contains the code which actually runs the simulations. It also contains a all_sims.sh files which will replicate the exact simulation settings in the paper. The exception is that the data for a few plots was simulated directly in the final-plots/final_plots.ipynb notebook.

The code needed to replicate the three real data applications are in the real_data/ subdirectory.

In general, all .py and .sh files should be called from the directory in which they reside. For example, one should navigate to the real_data/ subdirectory before running python3.9 hiv_data.py to replicate the HIV data analysis.

Figure by figure

1. Figure 1: see final-plots/final_plots.ipynb.
2. Figure 2: generated using sims/all_sims.sh using LINEAR_FX_ARGS and LINEAR_MX_ARGS.
3. Figure 3: generated using sims/all_sims.sh using VP_ARGS.
4. Figure 4: generated using sims/all_sims.sh using SPARSE_ARGS.
5. Figure 5: see final-plots/final_plots.ipynb.
6. Figure 6: generated using sims/all_sims.sh using NONLIN_ARGS.
7. Figure 7: generated using sims/all_sims.sh using LOGISTIC_ARGS.
8. Figure 8: generated from real_data/hiv_data.py.
9. Figure 9: generated from real_data/fund_rep.py.
10. Figure 10: generated from real_data/nodewise_knock.py.
11. Figure 11: generated using sims/all_sims.sh using NONLIN_ARGS.
12. Figures 12-15: generated using real_data/hiv_data.py and final-plots/final_plots.ipynb.