Validates neural posterior estimators using simulation-based calibration (SBC) and advanced diagnostic metrics.
The bayesflow-validation skill provides specialized tools for assessing the quality and reliability of BayesFlow models through Simulation-Based Calibration (SBC). It automates the generation of factorial condition grids, manages cached validation datasets, and calculates critical calibration metrics like C2ST and coverage error. By integrating threshold-based quality gates and diagnostic dashboards, this skill ensures that neural posterior estimators are well-calibrated and accurate across the entire design space, helping researchers and data scientists avoid common pitfalls in simulation-based inference such as validating only on training conditions or ignoring finite-sample calibration floors.
Key Features
01Factorial condition grid generation for design space testing
02Threshold-based quality gates for automated training loops
03Pre-computed validation dataset caching and reuse
04Diagnostic dashboard plotting for rank histograms and coverage
