Schrödinger-Bridge Drift Estimation Experiments

This repository accompanies the paper on direct nonparametric estimation of Schrödinger-bridge drifts from a closed-form conditional-ratio representation.

It contains:

• synthetic model definitions,
• deterministic truth computation,
• kernel drift estimators,
• rate / CLT / adaptivity / stress-test drivers,
• YAML configurations,
• saved artifacts used for the paper figures and tables.

Repository layout

• src/sbdrift/ — core implementation
• configs/ — experiment configurations
• scripts/ — experiment drivers and summary scripts
• figures/ — paper-facing figures
• results/ — saved experiment outputs and summaries

Canonical paper-facing artifacts

The final paper primarily uses:

• results/processed/adapt_final/
• results/processed/clt_runs/
• results/processed/stress/latest/
• figures/

See results/README.md for a guide to saved outputs.

Reproducing the main experiments

Typical entry points are:

python scripts/00_preflight.py --config configs/gg_1d.yaml
python scripts/01_rate.py --config configs/gg_1d.yaml ...
python scripts/02_clt.py --config configs/gg_1d.yaml ...
python scripts/03_stress_summary_raw_only.py

Run-specific parameters, seeds, and saved outputs are documented in the processed summaries under results/processed/.

Notes

• All reported experiments were CPU-only.
• The repository contains both final reported artifacts and some exploratory outputs.
• The final paper-facing summaries are documented in results/README.md.

Citation

If you use this repository, please cite the associated paper.