xlasso: X-learner implemented via glmnet (lasso)

In xnie/rlearner: R-learner for Heterogeneous Treatment Effect Estimation

Description

X-learner as proposed by Kunzel, Sekhon, Bickel, and Yu (2017), implemented via glmnet (lasso)

Usage

xlasso(
  x,
  w,
  y,
  alpha = 1,
  k_folds_mu1 = NULL,
  k_folds_mu0 = NULL,
  k_folds_p = NULL,
  lambda_t = NULL,
  lambda_x = NULL,
  lambda_w = NULL,
  lambda_choice = c("lambda.min", "lambda.1se"),
  mu1_hat = NULL,
  mu0_hat = NULL,
  p_hat = NULL
)

Arguments

x	the input features
w	the treatment variable (0 or 1)
y	the observed response (real valued)
alpha	tuning parameter for the elastic net
k_folds_mu1	number of folds for learning E[Y|X,W=1]
k_folds_mu0	number of folds for learning E[Y|X,W=0]
k_folds_p	number of folds for learning E[W|X]
lambda_t	user-supplied lambda sequence for cross validation in learning E[y|x,w=0] and E[y|x,w=1]
lambda_x	user-supplied lambda sequence for cross validation in learning E[d1|x] and E[d0|x] where d1 = y1 - E[y|x,w=0] and d0 = y0 - E[y|x,w=1]
lambda_w	user-supplied lambda sequence for cross validation in learning E[w|x]
lambda_choice	how to cross-validate; choose from "lambda.min" or "lambda.1se"
mu1_hat	pre-computed estimates on E[Y|X,W=1] corresponding to the input x. xlasso will compute it internally if not provided.
mu0_hat	pre-computed estimates on E[Y|X,W=0] corresponding to the input x. xlasso will compute it internally if not provided.
p_hat	pre-computed estimates on E[W|X] corresponding to the input x. xlasso will compute it internally if not provided.

Examples

## Not run: 
n = 100; p = 10

x = matrix(rnorm(n*p), n, p)
w = rbinom(n, 1, 0.5)
y = pmax(x[,1], 0) * w + x[,2] + pmin(x[,3], 0) + rnorm(n)

xlasso_fit = xlasso(x, w, y)
xlasso_est = predict(xlasso_fit, x)

## End(Not run)

xnie/rlearner documentation built on April 11, 2021, 12:49 a.m.