因果森林 Causal Forest

介绍

$$\hat{\theta}(x) = \argmin_{\theta} \sum_{i=1}^n K_x(X_i)\cdot \left( Y_i - \hat{q}(X_i, W_i) - \theta \cdot (T_i - \hat{f}(X_i, W_i)) \right)^2$$

算法实现

Python 包 econml 和 R 包 grf 都有实现该算法。

参考文献

Recursive Partitioning for Heterogeneous Causal Effects - arXiv

[Submitted on 5 Apr 2015 (v1), last revised 30 Dec 2015 (this version, v3)]

Estimation and Inference of Heterogeneous Treatment Effects ...

[Submitted on 14 Oct 2015 (v1), last revised 10 Jul 2017 (this version, v4)]

Generalized Random Forests - arXiv

[Submitted on 5 Oct 2016 (v1), last revised 5 Apr 2018 (this version, v4)]

Machine Learning Methods Economists Should Know About

[Submitted on 24 Mar 2019]