{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004438395,0.0001678765,0.0001501457,0.00007028775,0.0004970777,0.0006441185,0.0008530669,0.00006191833,0.00005404046],"category_scores_gemma":[0.0001811339,0.0001026984,0.0001015988,0.0006385607,0.00004384549,0.000280115,0.0005207965,0.000456184,0.0001469796],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002852712,"about_ca_system_score_gemma":0.00009666465,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002077809,"about_ca_topic_score_gemma":0.00005459144,"domain_scores_codex":[0.9982926,0.0002586972,0.0001888412,0.0004771753,0.0003919262,0.0003907949],"domain_scores_gemma":[0.9987116,0.0001688595,0.00006337842,0.0008562133,0.00007700085,0.0001229483],"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":[0.00000322565,0.00007256158,0.004169894,0.00001689472,0.00002365366,0.00005513374,0.01198283,0.0002166926,0.0004579567,0.01468266,0.008376293,0.9599422],"study_design_scores_gemma":[0.0002096811,0.00008798706,0.03155646,0.000127114,0.00000824447,0.00009023842,0.001324396,0.003944613,0.002287637,0.0003430915,0.9596837,0.0003368288],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1113045,0.001190933,0.836033,0.04665679,0.001303621,0.0002313602,5.664439e-7,0.001280156,0.001999143],"genre_scores_gemma":[0.8488111,0.00001452513,0.1391074,0.004120021,0.0001214476,0.00002218772,0.000001086798,0.00001597408,0.00778625],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9596054,"threshold_uncertainty_score":0.6211249,"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."}}