A technical demonstration showing why causal structure discovery alone is insufficient for making profitable decisions, and how PyMC bridges the gap to actionable causal inference.
This project responds to the ADIA Lab Causal Discovery Challenge results, which showed that supervised learning on labeled simulations achieves ~77% accuracy in causal structure discovery vs ~40% for classical methods.
The key insight: Discovering that A causes B doesn't tell you how A causes B. The functional form matters enormously for predictions and decisions.
PyMC Labs