tests/testthat/old/test-CATEestimator-XhRF_autotune_hyperband.R

In soerenkuenzel/causalToolbox: Toolbox for Causal Inference with emphasize on Heterogeneous Treatment Effect Estimator

library(testthat)
test_that("Tests X_RF_autotune_hyperband", {
  context("X-RF autotune hyperband")
  set.seed(1423614230)

  feat <- iris[, -1]
  tr <- rbinom(nrow(iris), 1, .5)
  yobs <- iris[, 1]
  sample.fraction <- .75
  num_iter <- 3 ^ 2
  eta <- 3
  verbose <- TRUE
  seed <- 24750371
  nthread <- 1

  xl <- X_RF_autotune_hyperband(
    feat = feat,
    tr = tr,
    yobs = yobs,
    sample.fraction = sample.fraction,
    num_iter = num_iter,
    eta = eta,
    verbose = FALSE,
    seed = seed,
    nthread = nthread
  )
  


  expect_equal(EstimateCate(xl, feat)[1],
               -0.07525655,
               tolerance = 1e-3)


  #### real example ####
  set.seed(432)
  cate_problem <-
    simulate_causal_experiment(
      ntrain = 400,
      ntest = 10000,
      dim = 20,
      alpha = .1,
      feat_distribution = "normal",
      setup = "RespSparseTau1strong",
      testseed = 543,
      trainseed = 234
    )
  
  xl_tuned <- X_RF_autotune_hyperband(
    feat = cate_problem$feat_tr,
    yobs = cate_problem$Yobs_tr,
    tr = cate_problem$W_tr,
    num_iter = 3 ^ 2,
    verbose = FALSE
  )
  
  expect_equal(mean((
    EstimateCate(xl_tuned, cate_problem$feat_te) - cate_problem$tau_te
  ) ^ 2),
  337.1134,
  tolerance = 1)

  # ----------------------------------------------------------------------------
  set.seed(21)
  feat <- iris[, -1]
  tr <- rbinom(nrow(iris), 1, .5)
  yobs <- iris[, 1]
  num_iter <- 2 ^ 3
  eta <- 2
  verbose <- TRUE
  seed <- 24750371
  nthread <- 1
  
  xl_at <- X_RF_autotune_hyperband(
    feat = feat,
    tr = tr,
    yobs = yobs,
    num_iter = num_iter,
    eta = eta,
    verbose = FALSE,
    nthread = 1
  )
  
  
  CIs <- CateCI(
    theObject = xl_at,
    feature_new = feat,
    B = 5,
    verbose = FALSE
  )

  #theObject = xl_at; feature_new = feat; B = 5; verbose = FALSE
  
  expect_equal(as.numeric(CIs[1,]),
               c(-0.222455421, -0.454358265, 0.009447424),
               tolerance = 1e-4)

})