{"id":"W4281381325","doi":"10.5281/zenodo.4833117","title":"ML Reproducibility Challenge 2020","year":2021,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Instituto de Ciencias del Mar y Limnología, Universidad Nacional Autónoma de México; Universiteit van Amsterdam; Institute for Catastrophic Loss Reduction","keywords":"Reproducibility; Computer science; Chromatography; Chemistry","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01646982,0.001529525,0.001620277,0.002854052,0.002174563,0.008664981,0.002381699,0.008112215,0.04240554],"category_scores_gemma":[0.07702047,0.0007440601,0.001356697,0.000845651,0.002875353,0.004085316,0.002940174,0.01160359,0.02732753],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001831407,"about_ca_system_score_gemma":0.003628244,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001600333,"about_ca_topic_score_gemma":0.00239179,"domain_scores_codex":[0.9901152,0.002414804,0.0006861931,0.001400643,0.004759326,0.000623761],"domain_scores_gemma":[0.925202,0.0362075,0.002542379,0.004616903,0.02536521,0.006066109],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003464151,0.000005481731,0.00002277738,0.00006775715,0.000009431903,0.00004479645,0.000009387921,0.00002916774,0.00007484743,0.001042555,0.9899291,0.008729937],"study_design_scores_gemma":[0.00005416053,0.00003370676,0.000170593,0.0002912365,0.00002803438,0.0001579599,0.00002247448,0.0003936737,0.0003221196,0.006924566,0.9915801,0.00002142635],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"editorial","genre_gemma":"dataset","genre_scores_codex":[0.0002178039,0.004938369,0.004022426,0.1656555,0.8115634,0.00005777931,0.000717123,0.001038218,0.0117892],"genre_scores_gemma":[0.003946949,0.002384247,0.003173655,0.07131596,0.8767262,0.00008717826,0.0008518409,0.0008969287,0.04061715],"genre_candidate":"dataset","genre_consensus":null,"teacher_disagreement_score":0.9835302,"threshold_uncertainty_score":0.1418607,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03422465548572512,"score_gpt":0.2892926889993795,"score_spread":0.2550680335136544,"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."}}