{"id":"W4322771373","doi":"10.1007/s43069-022-00190-4","title":"Model Development and Solver Demonstrations Using Randomized Test Problems","year":2023,"lang":"en","type":"article","venue":"Operations Research Forum","topic":"Optimization and Packing Problems","field":"Engineering","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"Wilfrid Laurier University","funders":"","keywords":"Computer science; Solver; Python (programming language); Visualization; MATLAB; Scalability; Software engineering; Analytics; Development (topology); Programming language; Data science; Artificial intelligence; Database","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":[0.001003894,0.0001012613,0.0001501625,0.0003568324,0.000734506,0.0002709437,0.0000865809,0.00006506628,0.00004239336],"category_scores_gemma":[0.0002195081,0.00009370107,0.0000264155,0.0006888615,0.0001151543,0.0002621025,0.00006159127,0.0001996924,0.00009300229],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005695079,"about_ca_system_score_gemma":0.0001652367,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001864905,"about_ca_topic_score_gemma":0.0003069763,"domain_scores_codex":[0.9988695,0.00005863706,0.0002610897,0.0001573924,0.0002654031,0.0003880079],"domain_scores_gemma":[0.9994016,0.0001777892,0.000005864347,0.0001417795,0.0001710706,0.0001018998],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001257049,0.00001274995,0.00005895982,0.00002293342,0.00001856003,6.461009e-7,0.0007204403,0.9915266,0.002238651,0.003495472,0.001248306,0.0006441008],"study_design_scores_gemma":[0.003693727,0.000005593996,0.000008486954,0.00003193663,0.000003797081,0.000002567268,0.0001563649,0.9936251,0.0007604942,0.0003092513,0.001289813,0.0001128585],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.151821,0.0004814744,0.8107154,0.003167972,0.0002689074,0.003853603,0.00007021274,0.001588766,0.02803272],"genre_scores_gemma":[0.9284442,0.0003183252,0.06655895,0.00002668933,0.00002338788,0.0003750821,0.0001067106,0.00004483019,0.004101814],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7766232,"threshold_uncertainty_score":0.5649301,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1298099136693669,"score_gpt":0.3470276789235193,"score_spread":0.2172177652541523,"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."}}