{"id":"W3134161095","doi":"10.1145/3408877.3432430","title":"STOP THE (AUTOGRADER) INSANITY: Regression Penalties to Deter Autograder Overreliance","year":2021,"lang":"en","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":47,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Insanity; Regression; Regression analysis; Computer science; Psychology; Computer security; Operations research; Actuarial science; Engineering; Business; Machine learning; Psychoanalysis; Psychiatry","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":[0.01584321,0.001292522,0.0008370848,0.002010089,0.00140093,0.002444351,0.002423053,0.001808892,0.01338189],"category_scores_gemma":[0.1665377,0.000491612,0.0004865104,0.001036829,0.001315598,0.002516731,0.002805802,0.004379221,0.005285121],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008992155,"about_ca_system_score_gemma":0.002411264,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001544048,"about_ca_topic_score_gemma":0.003180047,"domain_scores_codex":[0.9858001,0.006669041,0.001196094,0.001744239,0.003551192,0.001039237],"domain_scores_gemma":[0.8453097,0.07814267,0.02137795,0.02353277,0.02312017,0.008516656],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002590324,0.003161961,0.09805162,0.0005103287,0.0001660902,0.0006123448,0.002873137,0.02625203,0.02013715,0.02535477,0.1015328,0.7187575],"study_design_scores_gemma":[0.001717341,0.008349509,0.1583293,0.001033973,0.0003274297,0.002469749,0.002698167,0.5219407,0.08566684,0.0684364,0.1480489,0.0009817829],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.613357,0.0004251982,0.3044307,0.007392015,0.001492642,0.001008763,0.001159642,0.03530635,0.03542764],"genre_scores_gemma":[0.8819241,0.00007042685,0.104128,0.001038874,0.0001711842,0.0004634185,0.0004862918,0.001755702,0.009962068],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01584321,"threshold_uncertainty_score":0.08378786,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02189657914387818,"score_gpt":0.2754168992692508,"score_spread":0.2535203201253727,"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."}}