tests/testthat/test-classifiers.R

In emeyers/NeuroDecodeR: Decode Information from Neural Activity

# if one needs to regenrate the test data this function can be used
# source('generate_data_to_test_NDR.R')



load("example_ZD_train_and_test_set.Rda")
rm(training_set, test_set)



# Helper test functions ----------------------------------------------


test_reasonable_classification_accuracy <- function(cl){
  
  test_that("classification results are reasonably accurate", {
    
    # prediction_results <- get_predictions(cl, normalized_training_set, normalized_test_set)
    if (class(cl) == "cl_poisson_naive_bayes") {
      prediction_results <- get_predictions(cl, count_training_set, count_test_set)
    } else {
      prediction_results <- get_predictions(cl, normalized_training_set, normalized_test_set)
    }
    
    accuracies <- prediction_results |>
      dplyr::group_by(test_time) |>
      dplyr::summarize(mean_accuracy = mean(actual_labels == predicted_labels))
    
    expect_gt(dplyr::filter(accuracies, test_time == "stimulus")$mean_accuracy, .7)
    expect_lt(dplyr::filter(accuracies, test_time == "baseline")$mean_accuracy, .31)
    
  })
  
}



test_shuffle_results_at_chance <- function(cl){
  
  test_that("classification results on shuffled data are around chance", {
    
    #prediction_results <- get_predictions(cl, shuffled_normalized_training_set, shuffled_normalized_test_set)
    prediction_results <- get_predictions(cl, shuffled_count_training_set, shuffled_count_test_set)
    
    accuracies <- prediction_results |>
      dplyr::group_by(test_time) |>
      dplyr::summarize(mean_accuracy = mean(actual_labels == predicted_labels))
    
    expect_lt(dplyr::filter(accuracies, test_time == "stimulus")$mean_accuracy, .3)
    expect_lt(dplyr::filter(accuracies, test_time == "baseline")$mean_accuracy, .3)
    
  })
}
  
  


# test cl_max_correlation -----------------------------------------------------

test_that("the cl_max_correlation() classifier is working", {
  
  # test classifier has required methods and returns correctly formatted results
  cl <- cl_max_correlation()
  expect_null(test_valid_classifier(cl))

  test_reasonable_classification_accuracy(cl)
  test_shuffle_results_at_chance(cl)
  
  # test the classifier works when no decision values are returned
  cl <- cl_max_correlation(return_decision_values = FALSE)
  test_reasonable_classification_accuracy(cl)
  
})



test_that("the cl_max_correlation classifier constructor correctly adds objects to an ndr container", {
  
  raw_cl <- cl_max_correlation()
  expect_equal(class(raw_cl), "cl_max_correlation")
  
  fp <- fp_zscore()
  cl_combined_with_an_fp <- fp |> cl_max_correlation()
  expect_equal(class(cl_combined_with_an_fp), "ndr_container")
  
  an_ndr_container <- ndr_container()
  cl_combined_with_container <- an_ndr_container |> cl_max_correlation()
  expect_equal(class(cl_combined_with_container), "ndr_container")
  
})





# test cl_poison_naive_bayes --------------------------------------------------

test_that("the cl_poisson_naive_bayes() classifier is working", {
  
  cl <- cl_poisson_naive_bayes()
  expect_null(test_valid_classifier(cl))

  test_reasonable_classification_accuracy(cl)
  test_shuffle_results_at_chance(cl)
  
  # test the classifier works when no decision values are returned
  cl <- cl_poisson_naive_bayes(return_decision_values = FALSE)
  test_reasonable_classification_accuracy(cl)

})



test_that("the cl_poisson_naive_bayes classifier constructor correctly adds objects to an ndr container", {
  
  raw_cl <- cl_poisson_naive_bayes()
  expect_equal(class(raw_cl), "cl_poisson_naive_bayes")
  
  fp <- fp_zscore()
  cl_combined_with_an_fp <- fp |> cl_poisson_naive_bayes()
  expect_equal(class(cl_combined_with_an_fp), "ndr_container")
  
  an_ndr_container <- ndr_container()
  cl_combined_with_container <- an_ndr_container |> cl_poisson_naive_bayes()
  expect_equal(class(cl_combined_with_container), "ndr_container")
  
})




# test cl_svm -----------------------------------------------------------------

test_that("the cl_svm() claissifier is working", {
  
  cl <- cl_svm()
  expect_null(test_valid_classifier(cl))
  test_reasonable_classification_accuracy(cl)
  test_shuffle_results_at_chance(cl)

  # test the classifier works when no decision values are returned
  cl <- cl_svm(return_decision_values = FALSE)
  test_reasonable_classification_accuracy(cl)
  
  
  cl <- cl_svm(kernel = "linear")
  expect_null(test_valid_classifier(cl))
  test_reasonable_classification_accuracy(cl)
  test_shuffle_results_at_chance(cl)

})



test_that("the cl_svm classifier constructor correctly adds objects to an ndr container", {
  
  raw_cl <- cl_svm()
  expect_equal(class(raw_cl), "cl_svm")
  
  fp <- fp_zscore()
  cl_combined_with_an_fp <- fp |> cl_svm()
  expect_equal(class(cl_combined_with_an_fp), "ndr_container")
  
  an_ndr_container <- ndr_container()
  cl_combined_with_container <- an_ndr_container |> cl_svm()
  expect_equal(class(cl_combined_with_container), "ndr_container")
  
})