Why is RNG discipline important in these simulators?

Using a passed-in NumPy Generator instead of global random states ensures that simulations are perfectly reproducible, which is critical for scientific validation.

Why does this skill emphasize float32?

BayesFlow and its underlying deep learning frameworks require float32 to avoid dtype mismatches and ensure computational performance during network training.

How does the skill handle variable data sizes?

It provides patterns for subclassing the Simulator class and using the @allow_batch_size decorator to ensure observation sizes are drawn per-batch for clean stacking.

Can this help with debugging batch-shape errors?

Yes, it provides specific implementation rules and decorators to ensure batch dimensions are handled correctly, preventing common shape-related crashes in training.

What is a BayesFlow simulator?

It is a modular data-generating process (DGP) consisting of a prior, a likelihood, and optional meta functions used to train neural networks for Bayesian inference.

BayesFlow Simulator Builder

Name: BayesFlow Simulator Builder
Author: matthiaskloft

bymatthiaskloft

0•

Data Science & ML

Streamlines the creation, modification, and debugging of BayesFlow 2.x simulators for simulation-based inference.

This skill provides expert guidance for implementing BayesFlow 2.x data-generating processes, including priors, likelihood functions, and meta-parameters. It enforces best practices such as strict RNG discipline for reproducibility, proper batch-shape normalization using the @allow_batch_size decorator, and float32 data typing for neural network compatibility. Whether building simulators for fixed-size data or complex variable-size observation models, this skill ensures your generative models are correctly structured for seamless integration with BayesFlow inference networks and adapters.

Key Features

010 GitHub stars

02Automated handling of variable-size data via batch-shape normalization

03Validation of output dictionaries and float32 data type consistency

04Standardized simulator creation using functional and subclassing patterns

05Enforcement of RNG discipline for reproducible simulation-based inference

06Configuration management for priors using frozen dataclasses and PriorSpec

Use Cases

01Building generative models for amortized Bayesian inference

02Debugging batch-shape errors and dtype mismatches in BayesFlow pipelines

03Implementing hierarchical models with variable observation counts

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add matthiaskloft/claude-skills bayesflow-simulator

For use in Claude.ai and ChatGPT

Key Features

010 GitHub stars

02Automated handling of variable-size data via batch-shape normalization

03Validation of output dictionaries and float32 data type consistency

04Standardized simulator creation using functional and subclassing patterns

05Enforcement of RNG discipline for reproducible simulation-based inference

06Configuration management for priors using frozen dataclasses and PriorSpec

Use Cases

01Building generative models for amortized Bayesian inference

02Debugging batch-shape errors and dtype mismatches in BayesFlow pipelines

03Implementing hierarchical models with variable observation counts

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add matthiaskloft/claude-skills bayesflow-simulator

For use in Claude.ai and ChatGPT