pymc-modeling

Modern Bayesian modeling with PyMC v5+. Key defaults: nutpie sampler (2-5x faster), non-centered parameterization for hierarchical models, HSGP over exact GPs, coords/dims for readable InferenceData, and save-early workflow to prevent data loss from late crashes.

Modeling strategy: Build models iteratively — start simple, check prior predictions, fit and diagnose, check posterior predictions, expand one piece at a time. See references/workflow.md for the full workflow.

Notebook preference: Use marimo for interactive modeling unless the project already uses Jupyter.

使用 PyMC v5+ 進行貝葉斯統計建模。在建立機率模型、指定先驗、執行 MCMC 推理、診斷收斂或比較模型時使用。涵蓋 PyMC、ArviZ、pymc-bart、pymc-extras、nutpie 和 JAX/NumPyro 後端。觸發器涉及的任務包括：貝葉斯推理、後驗採樣、分層/多級模型、GLM、時間序列、高斯過程、BART、混合模型、先驗/後驗預測檢查、MCMC 診斷、LOO-CV、WAIC、模型比較或使用 do/observe 的因果推理。 來源：pymc-labs/python-analytics-skills。