alpha and imbalance.penalty don't matter

parameters_dont_matter.R

library(grf)

p = 3
n = 2000
sigma = 0.1

X = matrix(2 * runif(n * p) - 1, n, p)
W = rbinom(n, 1, 0.1)
TAU = (X[,1] > 0)
Y = TAU * (W - 1/2) + sigma * rnorm(n)

W.forest = regression_forest(X, W, num.trees = 500, seed=1234)
W.hat = predict(W.forest)$predictions

Y.forest = regression_forest(X, Y, num.trees = 500, seed=1234)
Y.hat = predict(Y.forest)$predictions

Y.resid = Y - Y.hat
W.resid = W - W.hat

print(mean(Y.resid, na.rm=TRUE))

for (i in seq(100)) {

  alpha = runif(1, min = 0.01, max=0.24)
  imbalance.penalty= runif(1, min = 0.01, max=1000)
  cf <- causal_forest(X, Y, W, Y.hat = Y.hat, W.hat = W.hat,
                      num.trees = 200, tune.parameters = FALSE,
                      sample.fraction=0.5,
                      alpha=alpha,
                      imbalance.penalty = imbalance.penalty,
                      min.node.size = 1,
                      stabilize.splits = TRUE,
                      seed = 12345)
  deb_error <- predict(cf)$debiased.error
  avg_deb_error <- mean(deb_error, na.rm=TRUE)

  print("ALPHA: ")
  print(alpha)

  print("IMBALANCE PENALTY: ")
  print(imbalance.penalty)

  print("AVG DEBIASED ERROR:")
  print(avg_deb_error)
  cat("\n")

}