Unit tests

[AniekMarkus]

Checks to add:

• Different input styles for trainExplore (these tests were originally there, but after updates no longer correctly testing this behavior)
  • Option 1: fully pre-specified settings file works
    • Input: settings_path (to use input settings file)
    • Optional: train_data (test with and without)  
  • Option 2: pre-specified settings file with input data works
    • Input: settings_path + other parameter (to change input settings file)
    • Optional: train_data (test with and without)  
  • Option 3: only input parameters (no settings file) works
    • Input: train_data (mandatory) + output_path (mandatory) + other parameter (to change package settings file)  
• Model -> trainExplore
  • Function should return a String
  • Number of features in model should be correct with input (min StartRulelength / max EndRulelength)
  • Model (performance) should be the same for Parallel (and BranchBound) TRUE / FALSE
  • Final model performance should satisfy contraints (as set by Accuracy, BalancedAccuracy, Specificity etc.)
    • Unless best performing models is worse than constraint -> Model is NULL
  • Given dataset -> Model is not NA / NULL etc. + Model remains unchanged for same settings etc.  
• Class always 0 (or 1) -> Model always predict 0 (or 1)
• One variable perfectly predicting outcome -> Should be selected in the model   -> Run these tests with various settings e.g. all possibilities for MAXIMIZE parameter  
• Prediction -> predictExplore Check output dimensions Expected output -> either 0 or 1 Repeated -> should give same result  
• Curve -> modelsCurveExplore + rocCurveExpore ModelsCurve returns vector of strings (where each element is a model) ROC should be between 0 and 1  
• Use adapted version expect_correct_fitPlp / expect_correct_prediction from PLP? https://github.com/OHDSI/PatientLevelPrediction/blob/main/tests/testthat/helper-expectations.R  
• Check other tests Cyclops e.g. around Rcpp (?) -> https://github.com/OHDSI/Cyclops/tree/main/tests/testthat

[AniekMarkus]

• Number of rules generated Run trainExplore for dataset binary_10.arff, check in binary_10.results file if "Total Count Cutoff Sets" is identical to # expected.

[AniekMarkus]

Idem for dataset mix_4.arff:

[AniekMarkus]

- Verifying if the generated rules are correct / the same as before

from mix_4.results ->

2 = AND 3 = OR 4 = Candidate model: '198124209' = "FALSE" AND '316139209' = "FALSE" OR '316139210' = "FALSE" Candidate model: '198124209' = "FALSE" AND '316139209' = "FALSE" OR '316139210' = "TRUE" Candidate model: '198124209' = "FALSE" AND '316139209' = "TRUE" OR '316139210' = "FALSE" Candidate model: '198124209' = "FALSE" AND '316139209' = "TRUE" OR '316139210' = "TRUE" Candidate model: '198124209' = "TRUE" AND '316139209' = "FALSE" OR '316139210' = "FALSE" Candidate model: '198124209' = "TRUE" AND '316139209' = "FALSE" OR '316139210' = "TRUE" Candidate model: '198124209' = "TRUE" AND '316139209' = "TRUE" OR '316139210' = "FALSE" Candidate model: '198124209' = "TRUE" AND '316139209' = "TRUE" OR '316139210' = "TRUE"