fastmatch

Fast matching estimators for causal inference under unconfoundedness. Interpretable, fast alternative to doubleML under unconfoundedness and overlap.

Supports

• Estimand: ATT, ATE
• Estimator: vanilla KNN, bias-corrected KNN (the latter lets you use standard errors using Rai/Otsu 2017)
• Nearest Neighbours Solvers: scikit, faiss, voyager
  • the latter two are very fast and can be used with GPU support, and can also be exported to a production environment where adding new units and future nearest-neighbours queries are very fast

Uses modern vectorDBs for knn search

• faiss
• voyager

To install, run

pip install git+https://github.com/py-econometrics/fastmatch.git

Install with gpu support: pip install fastmatch[gpu]

1"""
2.. include:: ../README.md
3"""
4
5from .fastmatch import Matching