How does this skill help with sampling issues like divergences?

It provides specific guidance on increasing target acceptance rates and implementing non-centered parameterization to resolve NUTS divergences in complex models.

Can I compare different Bayesian models using this skill?

Yes, it includes standard workflows for model comparison using LOO (Leave-One-Out cross-validation) and WAIC (Watanabe-Akaike Information Criterion).

Is Variational Inference supported as an alternative to MCMC?

Yes, the skill includes patterns for using ADVI (Automatic Differentiation Variational Inference) for faster, approximate inference on large datasets.

What version of PyMC does this skill support?

This skill is optimized for PyMC version 5.x and later, utilizing the most modern API and integration with ArviZ for inference data.

Does it support hierarchical or multilevel modeling?

Absolutely. It contains templates and best practices for building hierarchical models that account for group-level and observation-level variations.

PyMC Bayesian Modeling

Name: PyMC Bayesian Modeling
Author: jimmc414

byjimmc414

•

324

Data Science & ML

Builds and validates advanced Bayesian probabilistic models using PyMC 5.x for scientific discovery and statistical inference.

About

This skill equips Claude with the specialized knowledge to perform sophisticated Bayesian inference and probabilistic programming. It provides a structured framework for the entire Bayesian workflow, from defining weakly informative priors and building hierarchical models to performing MCMC sampling with NUTS. It emphasizes best practices such as non-centered parameterization to avoid divergences and utilizes ArviZ for rigorous diagnostics like R-hat and Effective Sample Size (ESS). Whether you are performing uncertainty quantification or comparing models via LOO/WAIC, this skill ensures robust and scientifically sound results.

Key Features

Advanced MCMC sampling using the No-U-Turn Sampler (NUTS) algorithm
Hierarchical and multilevel modeling with non-centered parameterization
324 GitHub stars
Fast Variational Inference (ADVI) for rapid model exploration
Model comparison and selection using LOO and WAIC metrics via ArviZ
Automated model diagnostics for R-hat, ESS, and sampling divergences

Use Cases

Performing robust predictive modeling with principled handling of missing data
Quantifying uncertainty in complex scientific data and discovery workflows
Analyzing grouped or multilevel datasets using hierarchical Bayesian structures

About

Key Features

Advanced MCMC sampling using the No-U-Turn Sampler (NUTS) algorithm
Hierarchical and multilevel modeling with non-centered parameterization
324 GitHub stars
Fast Variational Inference (ADVI) for rapid model exploration
Model comparison and selection using LOO and WAIC metrics via ArviZ
Automated model diagnostics for R-hat, ESS, and sampling divergences

Use Cases

Performing robust predictive modeling with principled handling of missing data
Quantifying uncertainty in complex scientific data and discovery workflows
Analyzing grouped or multilevel datasets using hierarchical Bayesian structures