Code for experiments in "Post-Estimation Smoothing: a Simple Baseline for Learning with Side Information".
This repo is in progress, and will continue to be updated (details below). Stay tuned!
- Simulation experiments.
Code for the simulation results is in the folder pes/simulations.
- sim_fxns.py houses the main functions needed to run the simulations
- simulations_tls.ipynb and simulations_ols.ipynb run the simulations and produce the figures in the paper. The only difference between these notebooks is that simualtions_tls using the total least squares solve to instantiate unsmoothed predictions, whereas simulations_ols using ordinary least squares.
If you'd like to save figures (by setting the flag save_fig = True in the notebooks), make sure you have the following directory: p-es/figs_output/simulations.
-
Video experiments. (coming soon, please contact me if you're interested in using the analysis sooner!)
-
Housing experiments. (coming soon, subject to data availability, please contact me if you're interested in using the analysis sooner!)