{"id":"W2594091530","doi":"","title":"Automatic prediction of the severity of bugs using stack traces","year":2016,"lang":"en","type":"article","venue":"PolyPublie (École Polytechnique de Montréal)","topic":"Software Engineering Research","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Computer science; Software bug; Stack (abstract data type); Crash; Eclipse; Data mining; Function (biology); Process (computing); Software regression; Machine learning; Programming language; Software; Software quality; Software development","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":[0.0007472509,0.0001575904,0.0002384374,0.0002422942,0.0000835336,0.00004166708,0.001229192,0.0001302681,0.00001289725],"category_scores_gemma":[0.0008140416,0.0001031092,0.0001257773,0.0007672814,0.0001297405,0.0004857621,0.0004196113,0.0001672017,0.000001732676],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002781721,"about_ca_system_score_gemma":0.000269223,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0013447,"about_ca_topic_score_gemma":0.00007122675,"domain_scores_codex":[0.9982274,0.0001308932,0.000413448,0.0002679541,0.0005798248,0.0003804422],"domain_scores_gemma":[0.9978894,0.000461345,0.0002200931,0.001163869,0.0001607779,0.0001045158],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002541324,0.0003593794,0.5681316,0.000322734,0.0001125229,0.00001078007,0.000952255,0.003551627,0.2611088,0.01321296,0.0008506958,0.1513613],"study_design_scores_gemma":[0.0002559106,0.00007245321,0.4522316,0.0002015373,0.00001010511,0.00004676285,0.00001084666,0.3834456,0.1623592,0.001177636,0.00006635083,0.0001219982],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4860696,0.0001042954,0.5127277,0.0004916455,0.0000887487,0.0002252935,0.00002204599,0.000261808,0.000008860342],"genre_scores_gemma":[0.9079053,0.00001840431,0.09187758,0.00003775114,0.00002658441,0.0000410842,3.454504e-7,0.00001839012,0.00007455057],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4218358,"threshold_uncertainty_score":0.4204673,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01559150229944111,"score_gpt":0.2384178648429278,"score_spread":0.2228263625434867,"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."}}