test_that("PFI works properly regression", {

  formula = "psych_well ~ gender + age + socioec_status + depression"

  hyper_nn_tune_list = list(
    learn_rate = c(-2, -1),
    hidden_units = c(3,10)
  )

  analysis_object <- preprocessing(df = sim_data, formula = formula, task = "regression")

  analysis_object <- build_model(analysis_object = analysis_object,
                             model_name = "Neural Network",
                             hyperparameters = hyper_nn_tune_list)

  analysis_object <- fine_tuning(analysis_object = analysis_object,
                             tuner = "Bayesian Optimization",
                             metrics = "rmse",
                             verbose = F)

  analysis_object <- sensitivity_analysis(analysis_object, methods = c("PFI", "SHAP", "Integrated Gradients", "Olden"))

  pfi <- analysis_object$sensitivity_analysis$PFI
  shap <- analysis_object$sensitivity_analysis$SHAP
  int_grad <- analysis_object$sensitivity_analysis$IntegratedGradients
  olden <- analysis_object$sensitivity_analysis$Olden

  expect_equal(pfi$Importance[[1]], 17.831, tolerance = 1e-1)
  expect_equal(shap$depression[1], 18.709, tolerance = 1e-1)
  expect_equal(int_grad$depression[1], 0.7978044, tolerance = 1e-1)
  expect_equal(olden$depression[1], -0.7335, tolerance = 1e-1)


})