{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004842651,0.002635716,0.0008859851,0.003765825,0.001539639,0.001669617,0.005074928,0.001117472,0.00429021],"category_scores_gemma":[0.02197083,0.001243848,0.001348627,0.007513813,0.001333237,0.004254508,0.003345111,0.002856333,0.003935419],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001794296,"about_ca_system_score_gemma":0.003823352,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0117907,"about_ca_topic_score_gemma":0.01129778,"domain_scores_codex":[0.9922529,0.001324997,0.001044541,0.001010019,0.003427672,0.0009398003],"domain_scores_gemma":[0.9885689,0.003415799,0.000919243,0.003192439,0.003049457,0.0008542696],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.003247207,0.00159896,0.04656728,0.006356047,0.0008606656,0.001533365,0.001937583,0.0801359,0.03230998,0.01516904,0.5376056,0.2726785],"study_design_scores_gemma":[0.001000652,0.001365845,0.06802088,0.000806823,0.0003990477,0.002316575,0.001333774,0.3420981,0.1204321,0.01977755,0.4418794,0.000569255],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"software","genre_gemma":"dataset","genre_scores_codex":[0.3732916,0.005901421,0.09651632,0.001987598,0.001733699,0.001411813,0.07642987,0.3892748,0.05345289],"genre_scores_gemma":[0.433985,0.002549308,0.2189085,0.0009597179,0.0001762787,0.00177419,0.2606713,0.06728411,0.01369157],"genre_candidate":"dataset","genre_consensus":null,"teacher_disagreement_score":0.0117907,"threshold_uncertainty_score":0.02561069,"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."}}