{"id":"W7125948450","doi":"10.1109/ase63991.2025.00077","title":"An Empirical Study of Python Library Migration Using Large Language Models","year":2025,"lang":"","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Python (programming language); Code (set theory); Benchmark (surveying); Empirical research; Unit testing","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":[0.01450466,0.0008777505,0.0004760058,0.001607055,0.0008499905,0.001579869,0.001863682,0.000965164,0.0008251935],"category_scores_gemma":[0.1129067,0.0006112765,0.0007183683,0.002609683,0.00158505,0.004491808,0.001645666,0.002078027,0.0004583188],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001846883,"about_ca_system_score_gemma":0.001496058,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01117052,"about_ca_topic_score_gemma":0.01342896,"domain_scores_codex":[0.988479,0.006101674,0.0009523752,0.001348348,0.002654436,0.000464188],"domain_scores_gemma":[0.8809332,0.0855123,0.01063594,0.01258177,0.008475722,0.001861076],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001715213,0.003508136,0.7882912,0.001240944,0.0004950311,0.001136499,0.007974189,0.0699876,0.006160842,0.003561197,0.01356124,0.1023681],"study_design_scores_gemma":[0.0003796978,0.00296838,0.3632089,0.0003368372,0.0003443228,0.001461183,0.00777377,0.5864141,0.01096531,0.004490262,0.02144583,0.0002113618],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9944881,0.0001835511,0.002644382,0.0002447932,0.00001957344,0.00008920739,0.0006637276,0.0006770481,0.0009896363],"genre_scores_gemma":[0.9866081,0.0001324105,0.009088534,0.0001151003,0.0000106308,0.0001469282,0.003147353,0.0002478223,0.0005031187],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01450466,"threshold_uncertainty_score":0.07670885,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03883258264504589,"score_gpt":0.3605659661900668,"score_spread":0.3217333835450208,"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."}}