reliability-diagram

Measures Jev classifier's confidence is really worth observability.

khalib is a classifier calibration algorithm based on Khiops

Calibration and uncertainty quantification for ranking systems.

Local-first calibration diagnostics for binary classifiers: reliability diagrams, uncertainty, paired model comparisons, separate-data recalibration, and offline reports.

Compare classifier confidence with accuracy using your own predictions. Browser-based calibration analysis and reports.

Measure how well a model's stated confidence matches how often it is right, with debiased calibration error and held-out recalibration checks

Study and Implementation of various neural network pruning techniques. Extending the lottery ticket hypothesis to structured pruning for accelerated training while maintaining uncertainty and accuracy. The focus is on simplifying the model's complexity without sacrificing its overall performance or leading to overfitting.

🎯 How much can your model decide on its own? Confidence threshold for a target accuracy, auto-handle rate, calibration and a held-out check, in one page

TypeScript binary forecast metrics: Brier score, log loss, ROC AUC, reliability diagrams, ECE and seeded day-block bootstrap intervals.

Is your model's confidence honest? ECE, Brier decomposition, reliability diagrams and temperature scaling. Zero dependencies.

Is your probability forecast honest? Brier, ECE and reliability, one temperature fitted on held-out days with day-clustered bootstrap CIs, and an error-budget cut on the gate's own scale. Stdlib Python.

Calibration auditing for probabilistic predictions: reliability diagrams, ECE, Brier decomposition, temperature scaling. Zero dependencies.