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

Analytical prior moments are deterministic and constant. Empirical moments shift across batches, causing training/validation distribution shifts, especially in models with variable-size data.

What is the correct pipeline order for a BayesFlow adapter?

The recommended order is: 1. Type conversion, 2. Cross-key transforms, 3. Per-key transforms, 4. Standardization, 5. Concatenation into schema slots, and 6. Dropping unused keys.

How does this skill handle dimension mismatches during concatenation?

It provides patterns for using expand_dims and broadcasting specs to ensure parameters (e.g., shape B, I) can be correctly concatenated with other data (e.g., shape B, I, K).

Can I use this skill for Hyperparameter Optimization (HPO)?

Yes, it includes patterns for the declarative AdapterSpec pattern from bayesflow-hpo, which is specifically designed for HPO-compatible adapters.

BayesFlow Adapter

Name: BayesFlow Adapter
Author: matthiaskloft

bymatthiaskloft

0•

데이터 과학 및 ML

Streamlines the creation and debugging of data preprocessing pipelines for BayesFlow simulation-based inference models.

The BayesFlow Adapter skill provides specialized guidance for building robust data pipelines that map simulator outputs to inference and summary networks. It helps developers implement the 'transform-before-standardize' workflow, utilize analytical prior moments for stable training, and handle complex cross-key reparameterizations. By enforcing correct schema slotting and providing patterns for broadcasting and custom transforms, it prevents common pitfalls like distribution shift and dimension mismatches in machine learning workflows.

주요 기능

01Schema slot mapping for inference, summary, and condition variables

02Standardization using analytical prior moments to prevent train/validation shift

03Support for custom elementwise and cross-key reparameterization transforms

04Declarative AdapterSpec pattern for HPO-compatible pipeline design

05Automated debugging for pipeline ordering and broadcasting errors

060 GitHub stars

사용 사례

01Implementing custom parameter transformations for Bayesian neural networks

02Optimizing data pipelines for high-performance hyperparameter optimization (HPO)

03Configuring data preprocessing for simulation-based inference (SBI) tasks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add matthiaskloft/claude-skills bayesflow-adapter

For use in Claude.ai and ChatGPT

주요 기능

01Schema slot mapping for inference, summary, and condition variables

02Standardization using analytical prior moments to prevent train/validation shift

03Support for custom elementwise and cross-key reparameterization transforms

04Declarative AdapterSpec pattern for HPO-compatible pipeline design

05Automated debugging for pipeline ordering and broadcasting errors

060 GitHub stars

사용 사례

01Implementing custom parameter transformations for Bayesian neural networks

02Optimizing data pipelines for high-performance hyperparameter optimization (HPO)

03Configuring data preprocessing for simulation-based inference (SBI) tasks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add matthiaskloft/claude-skills bayesflow-adapter

For use in Claude.ai and ChatGPT