{"id":"W4384009705","doi":"10.1109/msr59073.2023.00075","title":"PyMigBench: A Benchmark for Python Library Migration","year":2023,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Python (programming language); Computer science; Java; Popularity; World Wide Web; Programming language; Software 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":[0.0001852405,0.00006490034,0.00006148464,0.0001940977,0.00005367805,0.0001624411,0.0005029882,0.00003745736,0.00004352019],"category_scores_gemma":[0.0002668364,0.00005842444,0.00004292328,0.0009202306,0.000008617319,0.0006368408,0.000190114,0.00005362965,0.0002004334],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001070223,"about_ca_system_score_gemma":0.00005120502,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000704812,"about_ca_topic_score_gemma":0.000002471051,"domain_scores_codex":[0.9992191,0.00001181928,0.00009440662,0.0002322693,0.0001884124,0.0002539564],"domain_scores_gemma":[0.9988182,0.0007444082,0.0000111428,0.0003344213,0.00002443066,0.00006745097],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000008043792,0.000039614,0.01582299,0.0001023366,0.00002234585,0.00002254023,0.0009401489,0.001175886,0.002544226,0.09816979,0.8074382,0.07371388],"study_design_scores_gemma":[0.0003879823,0.0001896455,0.07792543,0.00002614934,0.000002094367,0.000006013676,0.00002836689,0.7313603,0.01641519,0.0140304,0.1592943,0.0003342107],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06380295,0.00004677238,0.9239672,0.005903632,0.0005237389,0.0003703602,0.000004020458,0.003884793,0.001496507],"genre_scores_gemma":[0.675246,0.00005040797,0.3004911,0.0004258896,0.000342117,0.0003537801,0.00008343987,0.00005274618,0.02295446],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7301844,"threshold_uncertainty_score":0.2576233,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02376487517013013,"score_gpt":0.2675182896619241,"score_spread":0.243753414491794,"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."}}