tests/testthat/test-model_tuning.R

In MLwrap: Machine Learning Modelling for Everyone

# test_that("fine_tuning Bayesian Optimization works properly classification", {
#
#   formula = "psych_well_bin ~ gender + age + socioec_status + depression"
#
#   hyper_nn_tune_list = list(
#     learn_rate = c(-2, -1),
#     hidden_units = c(3,10)
#   )
#
#   set.seed(123)
#
#   analysis_object <- preprocessing(df = sim_data, formula = formula, task = "classification")
#
#   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",
#                              metric = "roc_auc",
#                              verbose = F)
#
#   fit <- analysis_object$tuner_fit
#
#   expect_equal(length(fit$.iter), 8)
#
#   expect_equal(fit$.predictions[[8]]$hidden_units[1], 9)
#
#   expect_equal(fit$.predictions[[8]]$activation[1], "tanh")
#
#   expect_equal(fit$.predictions[[8]]$.pred_High[1], 0.9140839, tolerance = 5e-2)
#
# })
#
# test_that("fine_tuning Grid Search CV 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)
#   )
#
#   set.seed(123)
#
#   analysis_object <- preprocessing(df = sim_data, formula = formula, task = "regression")
#
#   analysis_object <- build_model(analysis_object = analysis_object,
#                              model_name = "Neural Network",
#                              hyperparameters = list(
#                                learn_rate = c(1e-3,1e-1),
#                                hidden_units = c(3,10),
#                                activation = "sigmoid"
#                              ))
#
#   analysis_object <- fine_tuning(analysis_object = analysis_object,
#                              tuner = "Grid Search CV",
#                              metrics = "rmse",
#                              verbose = F)
#
#   fit <- analysis_object$tuner_fit
#
#   expect_equal(fit$.predictions[[1]]$hidden_units[1], 3)
#
#   expect_equal(fit$.predictions[[1]]$learn_rate[1], 1.002305, tolerance = 1e-2)
#
#   expect_equal(fit$.predictions[[1]]$.pred[1], 69.78333, tolerance = 5e-2)
#
# })
#
# test_that("Check fine_tuning wrong metric",{
#
#   formula = "psych_well ~ gender + age + socioec_status + depression"
#
#   hyper_nn_tune_list = list(
#     learn_rate = c(-2, -1),
#     hidden_units = c(3,10)
#   )
#
#   set.seed(123)
#
#   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)
#
#   expect_error(fine_tuning(analysis_object = analysis_object,
#                              tuner = "Bayesian Optimization",
#                              metrics = "roc_auc",
#                              verbose = F))
#
# })
#
# test_that("Check fine_tuning plot_results not Boolean",{
#
#   formula = "psych_well ~ gender + age + socioec_status + depression"
#
#   hyper_nn_tune_list = list(
#     learn_rate = c(-2, -1),
#     hidden_units = c(3,10)
#   )
#
#   set.seed(123)
#
#   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)
#
#   expect_error(fine_tuning(analysis_object = analysis_object,
#                            tuner = "Bayesian Optimization",
#                            metrics = "rmse",
#                            verbose = F,
#                            plot_results = "re"))
#
# })
#
# test_that("Check fine_tuning tuner typo",{
#
#   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)
#
#   expect_error(fine_tuning(analysis_object = analysis_object,
#                            tuner = "Bayesian Optimisation",
#                            metrics = "rmse",
#                            verbose = F))
#
# })
#
#