grf: generalized random forests

Assessing Treatment Effect Variation inObservational Studies: Results from a Data Challenge

A pluggable package for forest-based statistical estimation and inference. GRF currently provides non-parametric methods for least-squares regression, quantile regression, and treatment effect estimation (optionally using instrumental variables).

In addition, GRF supports 'honest' estimation (where one subset of the data is used for choosing splits, and another for populating the leaves of the tree), and confidence intervals for least-squares regression and treatment effect estimation.

This package is currently in beta, and we expect to make continual improvements to its performance and usability.

Authors

This package is written and maintained by Julie Tibshirani ([email protected]), Susan Athey, and Stefan Wager.

The repository first started as a fork of the ranger repository -- we owe a great deal of thanks to the ranger authors for their useful and free package.

Installation

The latest release of the package can be installed through CRAN:

install.packages("grf")

Any published release can also be installed from source:

install.packages("https://raw.github.com/swager/grf/master/releases/grf_0.9.3.tar.gz", repos = NULL, type = "source")

Note that to install from source, a compiler that implements C++11 is required (clang 3.3 or higher, or g++ 4.8 or higher). If installing on Windows, the RTools toolchain is also required.

Usage Examples

library(grf)

# Generate data.
n = 2000; p = 10
X = matrix(rnorm(n*p), n, p)
X.test = matrix(0, 101, p)
X.test[,1] = seq(-2, 2, length.out = 101)

# Perform treatment effect estimation.
W = rbinom(n, 1, 0.5)
Y = pmax(X[,1], 0) * W + X[,2] + pmin(X[,3], 0) + rnorm(n)
tau.forest = causal_forest(X, Y, W)
tau.hat = predict(tau.forest, X.test)
plot(X.test[,1], tau.hat$predictions, ylim = range(tau.hat$predictions, 0, 2), xlab = "x", ylab = "tau", type = "l")
lines(X.test[,1], pmax(0, X.test[,1]), col = 2, lty = 2)

# Estimate the conditional average treatment effect on the full sample (CATE).
estimate_average_effect(tau.forest, target.sample = "all")

# Estimate the conditional average treatment effect on the treated sample (CATT).
# Here, we don't expect much difference between the CATE and the CATT, since
# treatment assignment was randomized.
estimate_average_effect(tau.forest, target.sample = "treated")

# Add confidence intervals for heterogeneous treatment effects; growing more trees is now recommended.
tau.forest = causal_forest(X, Y, W, num.trees = 4000)
tau.hat = predict(tau.forest, X.test, estimate.variance = TRUE)
sigma.hat = sqrt(tau.hat$variance.estimates)
plot(X.test[,1], tau.hat$predictions, ylim = range(tau.hat$predictions + 1.96 * sigma.hat, tau.hat$predictions - 1.96 * sigma.hat, 0, 2), xlab = "x", ylab = "tau", type = "l")
lines(X.test[,1], tau.hat$predictions + 1.96 * sigma.hat, col = 1, lty = 2)
lines(X.test[,1], tau.hat$predictions - 1.96 * sigma.hat, col = 1, lty = 2)
lines(X.test[,1], pmax(0, X.test[,1]), col = 2, lty = 1)

For examples on how to use other types of forest, including those for quantile regression and causal effect estimation using instrumental variables, please see the documentation directory.

Developing

In addition to providing out-of-the-box forests for quantile regression and causal effect estimation, GRF provides a framework for creating forests tailored to new statistical tasks. If you'd like to develop using GRF, please consult the development guide.

References

Susan Athey, Julie Tibshirani and Stefan Wager. Generalized Random Forests, 2016. [arxiv]