{"id":"W2107105626","doi":"","title":"Optimal unbiased estimators for evaluating agent performance","year":2006,"lang":"en","type":"article","venue":"","topic":"Advanced Bandit Algorithms Research","field":"Decision Sciences","cited_by":23,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Estimator; Variance (accounting); Outcome (game theory); Computer science; Minimum-variance unbiased estimator; Set (abstract data type); Bias of an estimator; Mathematical optimization; Mathematics; Statistics; Mathematical economics; Economics","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":[0.00329818,0.0001350604,0.0002108966,0.0002290903,0.0003177798,0.0002257695,0.0006045959,0.0000500466,0.0008179655],"category_scores_gemma":[0.002574934,0.00009364553,0.0001042638,0.0006362209,0.000103582,0.0004263236,0.0001205039,0.0001032794,0.0004981013],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008257548,"about_ca_system_score_gemma":0.0001206145,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002529846,"about_ca_topic_score_gemma":0.00001006911,"domain_scores_codex":[0.9964482,0.00006841915,0.0005619024,0.0004956345,0.001942397,0.0004834859],"domain_scores_gemma":[0.9969292,0.00179133,0.0001305143,0.0004558685,0.0005961957,0.00009686549],"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":[0.0001462972,0.0001210349,0.006381548,0.00001371715,0.00001065361,0.000006603217,0.00007886455,0.7084318,0.001431262,0.0006362168,0.02501241,0.2577296],"study_design_scores_gemma":[0.0006882613,0.0002195987,0.007976484,0.000006259345,0.000003630245,0.000005242589,0.0001160306,0.9706879,0.008468191,0.004227338,0.00744776,0.0001532842],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4521399,0.00003182976,0.5451235,0.0001902032,0.0001575814,0.0003787384,0.000008460389,0.00005712421,0.001912623],"genre_scores_gemma":[0.66738,0.000001438119,0.3151765,0.00005092613,0.0001506345,0.00009352749,0.000009008365,0.00001729535,0.01712063],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2622561,"threshold_uncertainty_score":0.8956149,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2599298327605407,"score_gpt":0.5149763499626402,"score_spread":0.2550465172020995,"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."}}