How does this skill handle reparameterization?

It provides templates for cross-key transforms where multiple variables interact. These are added early in the forward path but executed last during inversion to ensure parameters remain on their original scales.

Why should I standardize using prior moments instead of batch statistics?

Empirical batch statistics shift across different training iterations, especially with variable-size data. Using analytical prior moments ensures standardization remains constant and deterministic, preventing training/validation distribution shifts.

What is a BayesFlow adapter?

A BayesFlow adapter is a specialized preprocessing pipeline that transforms raw simulator outputs into the structured schema (inference, summary, and condition variables) required by neural networks for inference.

What is the correct order for an adapter pipeline?

You should always perform data transformations (like logs or one-hot encoding) before standardization. This ensures that you are standardizing the actual transformed distribution that the neural network will see.

BayesFlow Data Adapter

Name: BayesFlow Data Adapter
Author: matthiaskloft

bymatthiaskloft

0•

Data Science & ML

Architects and optimizes data preprocessing pipelines for simulation-based inference models using the BayesFlow framework.

This skill empowers Claude to build, debug, and refine BayesFlow adapters—the essential data preprocessing bridge between simulator outputs and neural inference networks. It provides specialized guidance on mapping simulator data to specific schema slots like inference variables and summary conditions, ensuring that data transformations and standardizations follow the correct mathematical order. By leveraging analytical prior moments for standardization and implementing custom reparameterization transforms, the skill helps prevent common pitfalls like distribution shift and incorrect broadcasting in complex Simulation-Based Inference (SBI) workflows.

Key Features

01Generates HPO-compatible declarative configurations using the AdapterSpec pattern.

02Maps simulator outputs into inference_variables, summary_variables, and inference_conditions.

03Implements analytical prior-moment standardization to prevent training distribution shifts.

04Supports custom elementwise and cross-key reparameterization transforms.

050 GitHub stars

06Enforces correct pipeline sequencing by performing transformations before standardization.

Use Cases

01Debugging model training issues caused by empirical batch standardization or incorrect transform ordering.

02Refactoring manual preprocessing code into standardized BayesFlow adapter objects.

03Building robust data pipelines for complex Simulation-Based Inference (SBI) tasks.

Key Features

01Generates HPO-compatible declarative configurations using the AdapterSpec pattern.

02Maps simulator outputs into inference_variables, summary_variables, and inference_conditions.

03Implements analytical prior-moment standardization to prevent training distribution shifts.

04Supports custom elementwise and cross-key reparameterization transforms.

050 GitHub stars

06Enforces correct pipeline sequencing by performing transformations before standardization.

Use Cases

01Debugging model training issues caused by empirical batch standardization or incorrect transform ordering.

02Refactoring manual preprocessing code into standardized BayesFlow adapter objects.

03Building robust data pipelines for complex Simulation-Based Inference (SBI) tasks.