Scoreplane: Model Eval

来自 App Store 来源

Scoreplane is a private model-evaluation laboratory for data scientists, ML engineers, risk teams, students, researchers, and technical reviewers who need to inspect labeled scores away from a notebook or dashboard. Import truth, score, optional record ID, subgroup, and sample weight from CSV. Move the operating threshold and see the confusion matrix, precision, recall, specificity, F1, MCC, accuracy, balanced accuracy, Brier score, and entered decision cost update together. Inspect ROC, precision-recall, and calibration views plus the same-threshold subgroup screen. KEY FEATURES - Labeled-score CSV import with quoted fields - Weighted confusion matrix and operating metrics - Interactive threshold slider - Explicit false-positive, false-negative, true-positive, and true-negative costs - Minimum-recorded-cost threshold screen - ROC and precision-recall curves with AUC - Reliability bins, Brier score, and expected calibration error - Same-threshold subgroup precision, recall, FPR, and base-rate comparison - Local PDF and JSON evaluation export - Complete 320-row holdout sample - 340 original offline model-evaluation plates - No account, ads, analytics, tracking, IAP, or subscription Scoreplane evaluates imported labels and scores under recorded assumptions; it does not certify fairness, safety, legality, causality, or production fitness. Review sampling, label quality, shift, subgroup context, uncertainty, operating costs, and deployment controls with accountable domain experts.

答案里的每个数字都来自本页上方展示的数据，可以逐条核对。