Optimizes the creation and debugging of BayesFlow data preprocessing pipelines for simulation-based inference.
The BayesFlow Adapter skill provides specialized guidance for building robust data pipelines that bridge simulator outputs with neural inference networks. It assists in mapping variables into specific schema slots—inference, summary, and conditions—while enforcing critical best practices like analytical standardization and correct transformation sequencing. By following these patterns, users can avoid common pitfalls such as distribution shifts caused by empirical batch statistics or incorrect tensor broadcasting, ensuring high-performance simulation-based inference workflows.
Key Features
010 GitHub stars
02Schema mapping for inference, summary, and condition variables
03Declarative AdapterSpec patterns for HPO-compatible pipelines
04Support for custom elementwise and cross-key reparameterization
05Standardization using analytical prior moments to prevent distribution shift
06Enforced sequencing of data transforms and standardization steps
Use Cases
01Building preprocessing pipelines for complex scientific simulators
02Implementing custom reparameterization logic for neural inference
03Debugging data shape and transformation ordering in BayesFlow
