{"id":"W3081900038","doi":"","title":"Unit Test Effort Prioritization Using Combined Datasets and Deep Learning: A Cross-Systems Validation.","year":2020,"lang":"en","type":"article","venue":"Software Engineering and Knowledge Engineering","topic":"Software Testing and Debugging Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Université du Québec","funders":"","keywords":"Prioritization; Computer science; Test (biology); Artificial intelligence; Unit (ring theory); Unit testing; Cross-validation; Machine learning; Data mining; Engineering; Software; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01258254,0.001545958,0.0008807543,0.002647356,0.0006033232,0.001354971,0.002353307,0.001486449,0.001727285],"category_scores_gemma":[0.04617232,0.0004900642,0.00129762,0.001604388,0.0005546102,0.001670705,0.002093928,0.001599962,0.001076789],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001134538,"about_ca_system_score_gemma":0.001365189,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007712861,"about_ca_topic_score_gemma":0.014867,"domain_scores_codex":[0.9931266,0.003425104,0.0005483595,0.001795786,0.0008539257,0.0002500909],"domain_scores_gemma":[0.9573489,0.0238032,0.002439579,0.00836331,0.007361762,0.0006833549],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0129682,0.00850689,0.3311955,0.00197693,0.008837014,0.0004552033,0.001102454,0.208423,0.02212977,0.001959535,0.03930125,0.3631443],"study_design_scores_gemma":[0.001206787,0.003749305,0.2145876,0.0003535977,0.001775308,0.0003831077,0.0005286932,0.741353,0.02392486,0.003426196,0.008532676,0.0001788695],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9416491,0.001177641,0.04287559,0.000176635,0.0002145792,0.000356513,0.009109995,0.002878525,0.00156128],"genre_scores_gemma":[0.9374619,0.0001126993,0.04198625,0.00009896321,0.00003551196,0.0003376145,0.01867867,0.0003499654,0.0009382972],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01258254,"threshold_uncertainty_score":0.06654364,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02080212164416674,"score_gpt":0.2542011728348439,"score_spread":0.2333990511906772,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}