GST-UNet is a neural framework that combines a U-Net-based architecture with iterative G-computation to estimate causal effects in spatiotemporal settings, addressing interference, temporal carryover, and time-varying confounders for Conditional Average Potential Outcome (CAPO) estimation.
Use the following commands to replicate the figures from the "GST-UNet: A Neural Framework for Spatiotemporal Causal Inference with Time-Varying Confounding" paper:
- For Table 1:
cd train && ./submit_jobs.sh
cd ../data/simulated_data && python summarize.py
This will produce the summary.csv file in data/simulated_data/linear.
- For Figure 3:
cd train && python wildfire_experiment.py
- For bootstrap confidence intervals:
cd train && ./submit_wildfire_jobs.sh