ProjectCausal / discoveryNeurIPS 2025★ Featured
Do-PFN
PFN Studio reproduction of Do-PFN (Robertson et al., NeurIPS 2025).
✓PFN Studio · reproduction of Robertson, Reuter, Guo et al. 2025Apache-2.0
Use when you have historical records with a lever you control (a price, dose, setpoint, treatment) plus an outcome, and you want to estimate the effect of changing that lever before committing. Ideal for what-if and intervention questions, even when hidden factors muddy the raw correlations.
Tensors loaded
83 / 83
upstream .cpkt
Bar-dist NLL
1.51 → 0.03
12 continued-pretrain steps
Params
7.3M
TabPFN-v2 backbone