tests/testthat/test-deep_meta_learners.R

In DeepLearningCausal: Causal Inference with Super Learner and Deep Neural Networks

testthat::skip_on_cran()

test_that("keras_pattc", {
  #python_ready()
  message(paste0("Python tensorflow: ", reticulate::py_module_available('tensorflow')))
  message(paste0("Python keras: ", reticulate::py_module_available('keras')))
  
  skip_if_not(reticulate::py_available(), "Python not available") 
  # Try keras3, install if missing
  if (!requireNamespace("keras3", quietly = TRUE)) {
    message("Installing keras3...")
    install.packages("keras3", repos = "https://cloud.r-project.org")
  }
  
  # Try tensorflow, install if missing
  if (!requireNamespace("tensorflow", quietly = TRUE)) {
    message("Installing tensorflow...")
    install.packages("tensorflow", repos = "https://cloud.r-project.org")
  }
  
  set.seed(1234)
  reticulate::py_set_seed(1234)
  deeppattc <- pattc_deeplearning(response.formula = support_war ~ age + female +
                            income + education +  employed + married +
                            hindu + job_loss,
                          exp.data = exp_data_full,
                          pop.data = pop_data_full,
                          treat.var = "strong_leader",
                          compl.var = "compliance",
                          compl.algorithm = "adam",
                          response.algorithm = "adam",
                          compl.hidden.layer = c(8,4,2),
                          response.hidden.layer = c(8,4,2),
                          compl.hidden_activation = "relu",
                          response.hidden_activation = "relu",
                          response.output_activation = "sigmoid",
                          response.output_units = 1,
                          response.loss = "binary_crossentropy",
                          response.metrics = "accuracy",
                          ID = NULL,
                          weights = NULL,
                          cluster = NULL,
                          compl.epoch = 50,
                          response.epoch = 100,
                          compl.validation_split = 0.2,
                          response.validation_split = 0.2,
                          compl.patience = 20,
                          response.patience = 20,
                          compl.dropout_rate = 0.1,
                          response.dropout_rate = c(0.1, 0.1, 0.1),
                          verbose = 0,
                          batch_size = 32,
                          nboot = 1000)
  expect_s3_class(deeppattc, "pattc_deeplearning")
  expect_equal(nrow(deeppattc$complier_prediction), nrow(exp_data_full))
  
})


test_that("ensemble-meta", {
  set.seed(1234)
  library(SuperLearner)
  slearner_en <- metalearner_ensemble(cov.formula = support_war ~ age + female
                                      + education + income + employed + 
                                      job_loss + hindu + political_ideology,
                                  data = exp_data_full, 
                                  treat.var = "strong_leader",
                                  meta.learner.type = "S.Learner", 
                                  SL.learners = c("SL.xgboost", "SL.glm"), 
                                  family = binomial(),
                                  conformal=TRUE,
                                  alpha=0.1,
                                  calib_frac=0.5,
                                  seed=1234)
  expect_s3_class(slearner_en, "metalearner_ensemble")
})

test_that("ensemble-pattc", {
  set.seed(1234)
  library(SuperLearner)
  pattc_en <- pattc_ensemble(response.formula = support_war ~ age + female
                            + income + education +  employed + married +
                              hindu + job_loss,
                          exp.data = exp_data_full,
                          pop.data = pop_data_full,
                          treat.var = "strong_leader",
                          compl.var = "compliance",
                          response.family = binomial(),
                          compl.SL.learners = c("SL.glm"),
                          response.SL.learners = c( "SL.glm"),
                          binary.preds = FALSE,
                          nboot = 1000)
  expect_s3_class(pattc_en, "pattc_ensemble")
})  

test_that("neuralnet-pattc", {
  set.seed(1234)
  pattc_nn <- pattc_neural(response.formula = support_war ~ age + female
                            + income + education +  employed + married +
                              hindu + job_loss,
                           exp.data = exp_data, pop.data = pop_data,
                           treat.var = "strong_leader", compl.var = "compliance",
                           compl.hidden.layer = c(2, 2),
                           response.hidden.layer = c(2, 2),
                           compl.stepmax = 1e+09, response.stepmax = 1e+09, 
                           nboot = 1000)
  expect_s3_class(pattc_nn, "pattc_neural")
})

test_that("neuralnet-xl", {
  set.seed(1234)
  xlearner_nn <- metalearner_neural(cov.formula = support_war ~ age + female
                                    + income + education +  employed + married +
                                      hindu + job_loss,
                                    data = exp_data, 
                                    treat.var = "strong_leader",
                                    meta.learner.type = "X.Learner", 
                                    stepmax = 1e+9,
                                    algorithm = "rprop+", 
                                    hidden.layer = c(2,2),
                                    act.fct = "logistic",
                                    err.fct = "sse",
                                    linear.output = FALSE, 
                                    binary.preds = FALSE)
  expect_s3_class(xlearner_nn, "metalearner_neural")
})

test_that("neuralnet-rl", {
  set.seed(1234)
  rlearner_nn <- metalearner_neural(cov.formula = support_war ~ age + female
                                    + income + education +  employed + married +
                                      hindu + job_loss,
                                    data = exp_data, 
                                    treat.var = "strong_leader",
                                    meta.learner.type = "R.Learner", 
                                    stepmax = 1e+9,
                                    algorithm = "rprop+", 
                                    hidden.layer = c(2,2),
                                    act.fct = "logistic",
                                    err.fct = "sse",
                                    linear.output = FALSE, 
                                    binary.preds = FALSE)
  expect_s3_class(rlearner_nn, "metalearner_neural")
})