bayesian-stats

What problem does it solve?

Bayesian statistical inference turns uncertain parameters into full posterior distributions by combining prior beliefs with observed data, so you can quantify uncertainty and compare models without relying on single point estimates.

Core Features & Use Cases

• Bayesian workflow end-to-end: prior selection and prior predictive checks, NUTS sampling in PyMC 5.x, and posterior predictive checks to validate fit.
• Diagnostics you can act on: convergence checks using R-hat and effective sample size (ESS), plus divergence detection to flag problematic posteriors.
• Model comparison and uncertainty-aware decisions: LOO-CV model comparison with ArviZ and hierarchical (multilevel) partial pooling; supports Bayesian A/B testing with direct probability statements.
• Use case: You’re analyzing a metric with small sample sizes and suspected group effects (e.g., multiple stores or clinics) and you need an uncertainty-aware estimate plus a decision-ready comparison between candidate models.

Quick Start

Use this Skill to fit a Bayesian hierarchical regression to your grouped dataset using PyMC 5.x with NUTS sampling, verify convergence, run posterior predictive checks, and compare competing models via LOO-CV in ArviZ.