How does this skill help with divergent transitions?

It provides specific code patterns for non-centered parameterization and advice on adjusting sampling parameters like target_accept to resolve geometry-related sampling issues.

Can I use standard NumPy functions inside the model?

No, this skill reinforces the use of PyTensor math operations (pm.math) inside the model context to ensure compatibility with the underlying computational graph.

Does this skill work with PyMC 3 or 4?

While many concepts overlap, this skill is specifically optimized for PyMC 5, focusing on current best practices, PyTensor integration, and the InferenceData output format.

What diagnostic metrics does it support?

It includes patterns for checking R-hat (should be 400), and Pareto k values for LOO-CV.

PyMC 5 Fundamentals

Name: PyMC 5 Fundamentals
Author: choxos

bychoxos

データサイエンスとML

Facilitates the creation, review, and optimization of Bayesian statistical models using the PyMC 5 framework.

概要

This skill provides a comprehensive foundation for building robust Bayesian models in Python using PyMC 5. It equips developers and data scientists with the necessary patterns for model specification, distribution selection, and advanced sampling techniques like NUTS and ADVI. By integrating ArviZ diagnostic workflows, the skill helps identify and resolve common issues such as divergent transitions through non-centered parameterization, ensuring that statistical inferences are both valid and efficient. It is particularly useful for those migrating from Stan or BUGS, or for anyone implementing the Bayesian workflow described in 'Statistical Rethinking'.

主な機能

Comprehensive syntax patterns for PyMC 5 model blocks and context managers
Advanced MCMC sampling configuration and troubleshooting for divergences
Complete ArviZ diagnostic suite for R-hat, ESS, and posterior predictive checks
0 GitHub stars
Extensive reference for continuous, discrete, and multivariate distributions
Guidance on non-centered parameterization and PyTensor math operations

ユースケース

Debugging sampling performance and model convergence issues
Developing hierarchical Bayesian models with complex dependency structures
Translating research models from Stan, JAGS, or BUGS into modern Python

概要

主な機能

Comprehensive syntax patterns for PyMC 5 model blocks and context managers
Advanced MCMC sampling configuration and troubleshooting for divergences
Complete ArviZ diagnostic suite for R-hat, ESS, and posterior predictive checks
0 GitHub stars
Extensive reference for continuous, discrete, and multivariate distributions
Guidance on non-centered parameterization and PyTensor math operations

ユースケース

Debugging sampling performance and model convergence issues
Developing hierarchical Bayesian models with complex dependency structures
Translating research models from Stan, JAGS, or BUGS into modern Python