Does this skill support both Stan and JAGS?

Yes, it includes specific implementation patterns and diagnostics for both Stan (via cmdstanr) and JAGS (via R2jags).

How does it handle model comparison?

It provides code snippets for LOO-CV (Leave-One-Out Cross-Validation), WAIC, and DIC, allowing for rigorous comparison between different model structures.

What convergence metrics does the skill prioritize?

It focuses on Rhat (aiming for 400), and monitoring for divergent transitions.

Can it help resolve 'divergent transitions' in Stan?

Yes, it provides troubleshooting steps such as increasing adapt_delta, using non-centered parameterization, or reparameterizing using Cholesky factors.

Bayesian Model Diagnostics

Name: Bayesian Model Diagnostics
Author: choxos

bychoxos

0•

Data Science & ML

Performs comprehensive MCMC diagnostic checks and posterior predictive assessments for Bayesian models implemented in Stan or JAGS.

This skill empowers Claude to rigorously evaluate the convergence and fit of Bayesian models, ensuring statistical validity in complex modeling tasks. It provides standardized patterns for checking key MCMC metrics like Rhat, Effective Sample Size (ESS), and HMC divergences, while also offering advanced visualization techniques such as rank histograms and pairs plots. Whether working with Stan (via cmdstanr) or JAGS, the skill assists in identifying geometry issues, improving parameterization, and validating model predictions through posterior predictive checks (PPC) and information criteria comparisons like LOO-CV, WAIC, and DIC.

Key Features

01Implementation of visual diagnostics including trace plots, rank histograms, and pairs plots

02Model comparison and selection using LOO-CV, WAIC, and DIC metrics

03Detection and resolution strategies for HMC divergent transitions and treedepth issues

040 GitHub stars

05Posterior predictive checks (PPC) for validating model-to-data fit with bayesplot

06Automated MCMC convergence assessment including Rhat and Effective Sample Size (ESS)

Use Cases

01Debugging non-convergent Stan models exhibiting low Effective Sample Size or high Rhat values

02Visualizing posterior distributions and parameter correlations to detect sampling bottlenecks

03Comparing multiple Bayesian model specifications to select the best predictive fit

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add choxos/biostatagent model-diagnostics

For use in Claude.ai and ChatGPT

Key Features

01Implementation of visual diagnostics including trace plots, rank histograms, and pairs plots

02Model comparison and selection using LOO-CV, WAIC, and DIC metrics

03Detection and resolution strategies for HMC divergent transitions and treedepth issues

040 GitHub stars

05Posterior predictive checks (PPC) for validating model-to-data fit with bayesplot

06Automated MCMC convergence assessment including Rhat and Effective Sample Size (ESS)

Use Cases

01Debugging non-convergent Stan models exhibiting low Effective Sample Size or high Rhat values

02Visualizing posterior distributions and parameter correlations to detect sampling bottlenecks

03Comparing multiple Bayesian model specifications to select the best predictive fit