{"id":"W2572009727","doi":"","title":"Bias in algorithm portfolio performance evaluation","year":2016,"lang":"en","type":"article","venue":"International Joint Conference on Artificial Intelligence","topic":"Constraint Satisfaction and Optimization","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Solver; Computer science; Portfolio; Measure (data warehouse); Algorithm; Data mining","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06784707,0.001858478,0.002403643,0.003977875,0.001079749,0.005443977,0.002630943,0.002265092,0.002583049],"category_scores_gemma":[0.3010881,0.0007302806,0.001276846,0.005400465,0.002488333,0.005606425,0.003376229,0.002448966,0.001256475],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002971931,"about_ca_system_score_gemma":0.002662829,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001895353,"about_ca_topic_score_gemma":0.001716649,"domain_scores_codex":[0.883476,0.06706894,0.008451598,0.01062402,0.0276237,0.002755785],"domain_scores_gemma":[0.719359,0.2084341,0.01463038,0.03575306,0.02008022,0.001743204],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003906575,0.0007021968,0.1187666,0.001593141,0.002453315,0.0002500455,0.0009942913,0.3880457,0.00645961,0.05244769,0.02382886,0.400552],"study_design_scores_gemma":[0.0005294529,0.002002229,0.02895848,0.0005539659,0.0005207946,0.0006445718,0.0006067149,0.8359569,0.02444022,0.08420143,0.02134246,0.0002427939],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5174592,0.01261408,0.4264037,0.00433194,0.001185572,0.0009747125,0.003434411,0.006657309,0.02693914],"genre_scores_gemma":[0.9121903,0.0006467567,0.08055582,0.0008285005,0.0002593557,0.0004627547,0.002774927,0.001038312,0.001243263],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9321529,"threshold_uncertainty_score":0.3588139,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1751941516819778,"score_gpt":0.3382325530106571,"score_spread":0.1630384013286794,"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."}}