{"id":"W2202127834","doi":"10.1109/iccd.2015.7357081","title":"Clustering-based revision debug in regression verification","year":2015,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Debugging; Computer science; Ranking (information retrieval); Cluster analysis; Overhead (engineering); Automation; Rank (graph theory); Data mining; Algorithmic program debugging; Machine learning; Reliability engineering; Programming language; Engineering","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005669473,0.00006373383,0.0000688769,0.0001616821,0.0000161077,0.00006399303,0.0004686447,0.00004149437,0.0000060236],"category_scores_gemma":[0.0006528075,0.00005137523,0.00001456697,0.0004646822,0.000008858142,0.0002346639,0.000131478,0.00009442979,0.0001004048],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001107365,"about_ca_system_score_gemma":0.00009891902,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003808711,"about_ca_topic_score_gemma":0.000009602331,"domain_scores_codex":[0.9991472,0.00004559577,0.0001213181,0.0002246258,0.0002991354,0.0001621399],"domain_scores_gemma":[0.999161,0.0001516582,0.00001865414,0.0005001461,0.00006900101,0.00009953088],"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":[0.0001368913,0.0004636046,0.1025977,0.0002525767,0.00000699575,0.0001783154,0.002228257,0.1975985,0.004540472,0.004583668,0.05040751,0.6370054],"study_design_scores_gemma":[0.0003999173,0.00005560954,0.01432796,0.00006682277,2.528e-7,0.000002524296,0.000006696928,0.9782552,0.002395694,0.00009570179,0.004293578,0.0001000369],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0223856,0.00009106451,0.9757581,0.0008256412,0.0001754526,0.0001042981,7.611353e-8,0.0002953389,0.0003644366],"genre_scores_gemma":[0.9171785,0.000002198844,0.08255564,0.00006506857,0.00001760939,0.000007250321,0.00000152014,0.000005712478,0.0001665579],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8947929,"threshold_uncertainty_score":0.2095021,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0510991814935841,"score_gpt":0.3141047332282426,"score_spread":0.2630055517346585,"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."}}