{"id":"W2281407436","doi":"10.1002/spe.2382","title":"DataMill: a distributed heterogeneous infrastructure forrobust experimentation","year":2015,"lang":"en","type":"article","venue":"Software Practice and Experience","topic":"Parallel Computing and Optimization Techniques","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Benchmarking; Dilemma; Computer science; Competition (biology); Empirical research; Compiler; Process (computing); Data science; Operating system","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":[0.02461815,0.002129774,0.002111505,0.002537429,0.001213106,0.004587815,0.008815679,0.001631964,0.02028852],"category_scores_gemma":[0.04333647,0.001770196,0.001698347,0.002843902,0.002565789,0.007911494,0.008558767,0.00561823,0.01060891],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001674803,"about_ca_system_score_gemma":0.003598587,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001961838,"about_ca_topic_score_gemma":0.001535482,"domain_scores_codex":[0.9746056,0.009724885,0.001968788,0.004317257,0.008043054,0.001340393],"domain_scores_gemma":[0.9521503,0.01529376,0.002110322,0.02361317,0.004622922,0.002209474],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.007374956,0.002682527,0.01248349,0.003014281,0.0009760521,0.001369943,0.002618153,0.09623821,0.05715713,0.08694759,0.3547587,0.3743789],"study_design_scores_gemma":[0.002006329,0.001017789,0.005725706,0.000438975,0.0001828332,0.0003498631,0.0004528443,0.5195968,0.06289583,0.1045916,0.3023066,0.0004348045],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"software","genre_scores_codex":[0.01937853,0.0004791552,0.6320801,0.001801316,0.0005343584,0.001713487,0.008119203,0.3196067,0.01628717],"genre_scores_gemma":[0.2437067,0.0006018791,0.6372669,0.001365367,0.000230096,0.005433973,0.03278997,0.0695895,0.009015664],"genre_candidate":"software","genre_consensus":null,"teacher_disagreement_score":0.02461815,"threshold_uncertainty_score":0.1301948,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02637679283596986,"score_gpt":0.3103458532838202,"score_spread":0.2839690604478504,"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."}}