[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M73J152C7E3WVMM2QADFFWFD":3,"skill-same:01M73J152C7E3WVMM2QADFFWFD":48},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":32,"updatedAt":33,"install":34,"resources":47},"01M73J152C7E3WVMM2QADFFWFD","pymc-bayesian-modeling",null,"Bayesian modeling with PyMC 5: priors, likelihood, NUTS\u002FADVI sampling, diagnostics (R-hat, ESS), LOO\u002FWAIC comparison, prediction. Hierarchical, logistic, GP variants; predictive checks.","bg-szy\u002FTOP-SKILLS","bg-szy","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F106576190?v=4","01M73A4E8KPV67SFXRY8XZGNBX","ShuZhiyong","01M73A4E96MQ294MHM74TZBVQD",6,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fawesome-skills\u002Fpymc-bayesian-modeling\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: \"pymc-bayesian-modeling\"\ndescription: \"Bayesian modeling with PyMC 5: priors, likelihood, NUTS\u002FADVI sampling, diagnostics (R-hat, ESS), LOO\u002FWAIC comparison, prediction. Hierarchical, logistic, GP variants; predictive checks.\"\nlicense: \"Apache-2.0\"\n---\n\n# PyMC Bayesian Modeling\n\n## Overview\n\nPyMC is a Python library for Bayesian statistical modeling and probabilistic programming. It provides an expressive syntax for defining probabilistic models and efficient inference via MCMC (NUTS) and variational methods (ADVI). This skill covers the full Bayesian modeling cycle from model specification through diagnostics, comparison, and prediction.\n\n## When to Use\n\n- Estimating parameters with full uncertainty quantification (credible intervals, not just point estimates)\n- Fitting hierarchical\u002Fmultilevel models to grouped or nested data\n- Performing prior and posterior predictive checks to validate model assumptions\n- Comparing candidate models using information criteria (LOO-CV, WAIC)\n- Building regression models (linear, logistic, Poisson) in a Bayesian framework\n- Handling missing data or measurement error as latent parameters\n- Modeling time series with autoregressive or random walk priors\n- Generating posterior predictions for new observations with uncertainty bounds\n- Use **Stan\u002FPyStan** instead for compiled, more scalable Bayesian inference on large models; use **statsmodels** for frequentist statistical tests\n\n## Prerequisites\n\n- **Python packages**: `pymc >= 5.0`, `arviz`, `numpy`, `matplotlib`\n- **Data**: NumPy arrays or pandas DataFrames with numeric columns\n- **Environment**: CPU sufficient for most models; GPU via JAX backend for large models\n\n```bash\npip install pymc arviz numpy matplotlib\n# Optional: JAX backend for GPU acceleration\npip install pymc[jax]\n```\n\n## Quick Start\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Simulate data\nnp.random.seed(42)\nX = np.random.randn(100)\ny = 2.5 + 1.3 * X + np.random.randn(100) * 0.5\n\n# Build and fit model\nwith pm.Model() as model:\n    alpha = pm.Normal(\"alpha\", mu=0, sigma=5)\n    beta = pm.Normal(\"beta\", mu=0, sigma=5)\n    sigma = pm.HalfNormal(\"sigma\", sigma=1)\n    mu = alpha + beta * X\n    y_obs = pm.Normal(\"y_obs\", mu=mu, sigma=sigma, observed=y)\n    idata = pm.sample(1000, tune=1000, chains=4, random_seed=42)\n\nprint(az.summary(idata, var_names=[\"alpha\", \"beta\", \"sigma\"]))\n# Expected: alpha ~ 2.5, beta ~ 1.3, sigma ~ 0.5\n```\n\n## Workflow\n\n### Step 1: Prepare Data\n\nStandardize continuous predictors for better sampling efficiency. Use named coordinates for readable models and ArviZ integration.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Load data\nX = np.random.randn(200, 3)  # 200 obs, 3 predictors\ny = X @ np.array([1.0, -0.5, 0.3]) + np.random.randn(200) * 0.8\n\n# Standardize predictors\nX_mean, X_std = X.mean(axis=0), X.std(axis=0)\nX_scaled = (X - X_mean) \u002F X_std\n\n# Define coordinates for named dimensions\ncoords = {\n    \"predictors\": [\"var1\", \"var2\", \"var3\"],\n    \"obs_id\": np.arange(len(y)),\n}\nprint(f\"Data shape: X={X_scaled.shape}, y={y.shape}\")\n```\n\n### Step 2: Define Model and Set Priors\n\nSpecify the model structure inside a `pm.Model()` context. Use weakly informative priors, `dims` for named dimensions, and `HalfNormal` or `Exponential` for scale parameters.\n\n```python\nwith pm.Model(coords=coords) as model:\n    # Priors — weakly informative, not flat\n    alpha = pm.Normal(\"alpha\", mu=0, sigma=1)\n    beta = pm.Normal(\"beta\", mu=0, sigma=1, dims=\"predictors\")\n    sigma = pm.HalfNormal(\"sigma\", sigma=1)\n\n    # Linear predictor\n    mu = alpha + pm.math.dot(X_scaled, beta)\n\n    # Likelihood\n    y_obs = pm.Normal(\"y_obs\", mu=mu, sigma=sigma, observed=y, dims=\"obs_id\")\n\n# Inspect model variables\nprint(model.basic_RVs)  # Lists: [alpha, beta, sigma, y_obs]\n```\n\n### Step 3: Prior Predictive Check\n\nValidate that priors produce plausible data ranges before fitting. Adjust priors if simulated data is unreasonable.\n\n```python\nwith model:\n    prior_pred = pm.sample_prior_predictive(samples=1000, random_seed=42)\n\n# Check prior-implied data range\nprior_y = prior_pred.prior_predictive[\"y_obs\"].values.flatten()\nprint(f\"Prior predictive range: [{prior_y.min():.1f}, {prior_y.max():.1f}]\")\nprint(f\"Observed data range:    [{y.min():.1f}, {y.max():.1f}]\")\n\naz.plot_ppc(prior_pred, group=\"prior\", num_pp_samples=100)\n```\n\n### Step 4: Sample Posterior (MCMC)\n\nRun NUTS sampling with multiple chains. Include `log_likelihood=True` if you plan model comparison later.\n\n```python\nwith model:\n    idata = pm.sample(\n        draws=2000,\n        tune=1000,\n        chains=4,\n        target_accept=0.9,\n        random_seed=42,\n        idata_kwargs={\"log_likelihood\": True},\n    )\n\nprint(f\"Posterior shape: {idata.posterior['beta'].shape}\")\n# Expected: (4 chains, 2000 draws, 3 predictors)\n```\n\n### Step 5: Diagnose Sampling\n\nCheck convergence before interpreting results. All three diagnostics (R-hat, ESS, divergences) must pass.\n\n```python\n# Summary with convergence diagnostics\nsummary = az.summary(idata, var_names=[\"alpha\", \"beta\", \"sigma\"])\nprint(summary[[\"mean\", \"sd\", \"hdi_3%\", \"hdi_97%\", \"r_hat\", \"ess_bulk\"]])\n\n# R-hat convergence check\nbad_rhat = summary[summary[\"r_hat\"] > 1.01]\nif len(bad_rhat) > 0:\n    print(f\"WARNING: {len(bad_rhat)} parameters with R-hat > 1.01\")\n    print(bad_rhat[[\"r_hat\"]])\n\n# Effective sample size check\nlow_ess = summary[summary[\"ess_bulk\"] \u003C 400]\nif len(low_ess) > 0:\n    print(f\"WARNING: {len(low_ess)} parameters with ESS \u003C 400\")\n\n# Divergence check\nn_div = idata.sample_stats.diverging.sum().item()\ntotal = len(idata.posterior.draw) * len(idata.posterior.chain)\nprint(f\"Divergences: {n_div}\u002F{total} ({n_div \u002F total * 100:.2f}%)\")\n\n# Visual diagnostics — trace plots and rank plots\naz.plot_trace(idata, var_names=[\"alpha\", \"beta\", \"sigma\"])\naz.plot_rank(idata, var_names=[\"alpha\", \"beta\", \"sigma\"])\n```\n\n### Step 6: Posterior Predictive Check\n\nValidate model fit by comparing simulated data from the posterior to observed data.\n\n```python\nwith model:\n    pm.sample_posterior_predictive(idata, extend_inferencedata=True, random_seed=42)\n\naz.plot_ppc(idata, num_pp_samples=100)\n# Blue = observed data, grey = posterior simulations\n# Systematic deviations indicate model misspecification\n```\n\n### Step 7: Compare Models\n\nUse LOO-CV or WAIC to compare candidate models. Lower information criterion is better.\n\n```python\n# Fit multiple models with log_likelihood=True, then compare\n# Example: compare linear vs a second model\nidatas = {\"linear\": idata}  # add more fitted models here\n\ncomparison = az.compare(idatas, ic=\"loo\")\nprint(comparison[[\"rank\", \"elpd_loo\", \"p_loo\", \"d_loo\", \"weight\"]])\n\n# Check LOO reliability via Pareto-k diagnostics\nloo_result = az.loo(idata, pointwise=True)\nhigh_k = (loo_result.pareto_k > 0.7).sum().item()\nprint(f\"Observations with Pareto-k > 0.7: {high_k}\")\n# Interpretation: Dloo \u003C 2 = similar models; Dloo > 10 = strong evidence\n\naz.plot_compare(comparison)\n```\n\n### Step 8: Generate Predictions\n\nProduce posterior predictions for new data with full uncertainty propagation.\n\n```python\nX_new = np.array([[0.5, -1.0, 0.2]])\nX_new_scaled = (X_new - X_mean) \u002F X_std\n\nwith model:\n    pm.set_data({\"X_scaled\": X_new_scaled})\n    post_pred = pm.sample_posterior_predictive(\n        idata.posterior, var_names=[\"y_obs\"], random_seed=42\n    )\n\ny_pred = post_pred.posterior_predictive[\"y_obs\"]\nprint(f\"Predicted mean: {y_pred.mean().item():.3f}\")\nprint(f\"94% HDI: {az.hdi(y_pred, hdi_prob=0.94).values}\")\n```\n\n## Key Parameters\n\n| Parameter | Default | Range \u002F Options | Effect |\n|-----------|---------|-----------------|--------|\n| `draws` | `1000` | `500`-`10000` | Number of posterior samples per chain |\n| `tune` | `1000` | `500`-`5000` | Warmup iterations (discarded); increase for complex posteriors |\n| `chains` | `4` | `2`-`8` | Number of independent chains; minimum 4 for reliable R-hat |\n| `cores` | all CPUs | `1`-`N` | Parallel chains; set equal to `chains` for full parallelism |\n| `target_accept` | `0.8` | `0.8`-`0.99` | NUTS acceptance rate; increase to reduce divergences |\n| `init` | `\"auto\"` | `\"adapt_diag\"`, `\"jitter+adapt_diag\"`, `\"advi\"` | Initialization strategy for sampler |\n| `random_seed` | `None` | any int | Seed for reproducibility |\n| `idata_kwargs` | `{}` | `{\"log_likelihood\": True}` | Store log-likelihood for LOO\u002FWAIC model comparison |\n| `method` (pm.fit) | `\"advi\"` | `\"advi\"`, `\"fullrank_advi\"`, `\"svgd\"` | Variational inference algorithm |\n| `n` (pm.fit) | `10000` | `5000`-`100000` | VI optimization iterations |\n| `samples` (prior pred) | `500` | `100`-`5000` | Prior predictive samples for validation |\n\n## Key Concepts\n\n### Prior\u002FDistribution Selection Guide\n\n| Distribution | Use When | Key Parameters |\n|-------------|----------|----------------|\n| `Normal(mu, sigma)` | Unbounded real-valued parameter (standardized data) | `mu`: center, `sigma`: spread |\n| `HalfNormal(sigma)` | Scale\u002Fstandard deviation parameter (positive) | `sigma`: spread of positive half |\n| `Exponential(lam)` | Scale parameter, alternative to HalfNormal | `lam`: rate (1\u002Fmean) |\n| `StudentT(nu, mu, sigma)` | Robust alternative to Normal (outlier-resistant) | `nu`: degrees of freedom (\u003C10 = heavier tails) |\n| `Beta(alpha, beta)` | Probability or proportion in [0,1] | `alpha=beta=2`: weakly informative |\n| `Gamma(alpha, beta)` | Positive parameter (rate, concentration) | `alpha`: shape, `beta`: rate |\n| `LogNormal(mu, sigma)` | Positive parameter with multiplicative effects | `mu`, `sigma`: of underlying Normal |\n| `LKJCorr(n, eta)` | Correlation matrix prior | `eta=1`: uniform; `eta>1`: prefer identity |\n| `Dirichlet(a)` | Probability vector (sums to 1) | `a`: concentration; uniform if all equal |\n| `Bernoulli(p \u002F logit_p)` | Binary outcome likelihood | Use `logit_p` for numerical stability |\n| `Poisson(mu)` | Count data (equidispersed) | `mu`: rate; use NegBinomial if overdispersed |\n| `NegativeBinomial(mu, alpha)` | Overdispersed count data | `alpha`: dispersion (smaller = more overdispersion) |\n\n### Diagnostic Thresholds\n\n| Metric | Threshold | Interpretation | Action if Failed |\n|--------|-----------|---------------|------------------|\n| R-hat | \u003C 1.01 | Chains converged | Run longer chains; check multimodality |\n| ESS bulk | > 400 | Sufficient independent samples | Increase `draws`; reparameterize |\n| ESS tail | > 400 | Reliable tail estimates | Increase `draws` |\n| Divergences | 0 | NUTS explored successfully | Increase `target_accept`; non-centered param. |\n| Pareto-k (LOO) | \u003C 0.7 | LOO estimate reliable | Use WAIC or k-fold CV |\n| Max tree depth | \u003C 10 | No trajectory truncation | Reparameterize or increase `max_treedepth` |\n\n### Model Variants Overview\n\n| Problem Type | Recipe | Likelihood | Key Feature |\n|-------------|--------|------------|-------------|\n| Grouped\u002Fnested data | Hierarchical Model | Normal (varies) | Non-centered parameterization, partial pooling |\n| Binary outcome | Logistic Regression | Bernoulli | `logit_p` link function |\n| Nonlinear\u002Fspatial | Gaussian Process | Normal | Kernel-based covariance, flexible shape |\n| Count data | (use Poisson in Workflow) | Poisson \u002F NegBinomial | Log link; NegBinomial for overdispersion |\n| Time series | (see references) | AR \u002F GaussianRandomWalk | Autoregressive coefficients |\n| Mixture\u002Fclustering | (see references) | Mixture \u002F NormalMixture | Component weights via Dirichlet |\n\n## Common Recipes\n\n### Recipe: Hierarchical Model\n\nWhen to use: data has natural grouping (patients within hospitals, students within schools). Non-centered parameterization avoids divergences from funnel geometry.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\nn_groups, n_per_group = 5, 30\ngroup_idx = np.repeat(np.arange(n_groups), n_per_group)\ngroup_names = [f\"group_{i}\" for i in range(n_groups)]\n\n# Simulated grouped data\ntrue_alphas = np.random.normal(3.0, 1.5, n_groups)\ny_obs = np.random.normal(true_alphas[group_idx], 0.5)\n\nwith pm.Model(coords={\"groups\": group_names}) as hierarchical_model:\n    # Hyperpriors (population level)\n    mu_alpha = pm.Normal(\"mu_alpha\", mu=0, sigma=5)\n    sigma_alpha = pm.HalfNormal(\"sigma_alpha\", sigma=2)\n\n    # Non-centered parameterization\n    alpha_offset = pm.Normal(\"alpha_offset\", mu=0, sigma=1, dims=\"groups\")\n    alpha = pm.Deterministic(\n        \"alpha\", mu_alpha + sigma_alpha * alpha_offset, dims=\"groups\"\n    )\n\n    # Observation level\n    sigma = pm.HalfNormal(\"sigma\", sigma=1)\n    y = pm.Normal(\"y\", mu=alpha[group_idx], sigma=sigma, observed=y_obs)\n\n    idata_hier = pm.sample(2000, tune=1000, target_accept=0.95, random_seed=42)\n\nprint(az.summary(idata_hier, var_names=[\"mu_alpha\", \"sigma_alpha\", \"alpha\", \"sigma\"]))\n```\n\n### Recipe: Logistic Regression\n\nWhen to use: binary outcome variable (success\u002Ffailure, disease\u002Fhealthy). Use `logit_p` for numerical stability instead of computing probabilities directly.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Simulated binary outcome\nnp.random.seed(42)\nn, p = 200, 3\nX_lr = np.random.randn(n, p)\ntrue_beta = np.array([1.0, -0.5, 0.3])\nprob = 1 \u002F (1 + np.exp(-(0.2 + X_lr @ true_beta)))\ny_binary = np.random.binomial(1, prob)\n\nwith pm.Model() as logistic_model:\n    alpha = pm.Normal(\"alpha\", mu=0, sigma=2)\n    beta = pm.Normal(\"beta\", mu=0, sigma=2, shape=p)\n    logit_p = alpha + pm.math.dot(X_lr, beta)\n    y = pm.Bernoulli(\"y\", logit_p=logit_p, observed=y_binary)\n\n    idata_logit = pm.sample(2000, tune=1000, target_accept=0.9, random_seed=42)\n\nsummary = az.summary(idata_logit, var_names=[\"alpha\", \"beta\"])\nprint(summary[[\"mean\", \"sd\", \"hdi_3%\", \"hdi_97%\"]])\n# Odds ratios\nbeta_samples = idata_logit.posterior[\"beta\"].values.reshape(-1, p)\nprint(f\"Odds ratios (median): {np.median(np.exp(beta_samples), axis=0)}\")\n```\n\n### Recipe: Gaussian Process\n\nWhen to use: modeling unknown nonlinear functions, spatial data, or smooth latent processes. Computationally expensive for n > 1000; consider sparse approximations for larger datasets.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Simulated nonlinear data\nnp.random.seed(42)\nX_gp = np.sort(np.random.uniform(0, 10, 60))[:, None]\ny_gp = np.sin(X_gp[:, 0]) + np.random.randn(60) * 0.3\n\nwith pm.Model() as gp_model:\n    # GP hyperparameters\n    ls = pm.Gamma(\"lengthscale\", alpha=2, beta=1)\n    eta = pm.HalfNormal(\"amplitude\", sigma=2)\n    sigma_noise = pm.HalfNormal(\"noise\", sigma=0.5)\n\n    # Covariance function\n    cov = eta**2 * pm.gp.cov.Matern52(1, ls=ls)\n    gp = pm.gp.Marginal(cov_func=cov)\n\n    # Marginal likelihood\n    y_ = gp.marginal_likelihood(\"y\", X=X_gp, y=y_gp, sigma=sigma_noise)\n    idata_gp = pm.sample(1000, tune=1000, target_accept=0.9, random_seed=42)\n\n# Predict on new points\nX_new_gp = np.linspace(0, 10, 100)[:, None]\nwith gp_model:\n    f_pred = gp.conditional(\"f_pred\", X_new_gp)\n    pred_samples = pm.sample_posterior_predictive(idata_gp, var_names=[\"f_pred\"])\n\nprint(f\"GP predictions shape: {pred_samples.posterior_predictive['f_pred'].shape}\")\n```\n\n## Expected Outputs\n\n- **Trace plots** (`az.plot_trace`) -- chain mixing visualization; healthy chains look like \"fuzzy caterpillars\"\n- **Posterior summary** (`az.summary`) -- mean, SD, HDI, R-hat, ESS per parameter\n- **Prior\u002Fposterior predictive plots** (`az.plot_ppc`) -- simulated vs observed data distributions\n- **Forest plots** (`az.plot_forest`) -- coefficient estimates with credible intervals\n- **Model comparison table** (`az.compare`) -- ranked models with LOO\u002FWAIC, weights, warnings\n- **Rank plots** (`az.plot_rank`) -- uniform histograms indicate good mixing\n- **Energy plot** (`az.plot_energy`) -- HMC energy transition diagnostics\n- **InferenceData file** (`idata.to_netcdf(\"results.nc\")`) -- serialized results for later analysis\n\n## Troubleshooting\n\n| Problem | Cause | Solution |\n|---------|-------|----------|\n| Divergent transitions | Posterior geometry difficult for NUTS (funnels, ridges) | Increase `target_accept=0.95`; use non-centered parameterization; add stronger priors |\n| Low ESS (\u003C 400) | High autocorrelation between samples | Increase `draws=5000`; reparameterize to reduce correlation; try QR decomposition for correlated predictors |\n| R-hat > 1.01 | Chains have not converged | Increase `tune=2000, draws=5000`; check for multimodality; initialize with ADVI |\n| Slow sampling | Complex posterior or large dataset | Use ADVI for initialization; reduce model complexity; increase `cores`; try JAX backend |\n| Pareto-k > 0.7 in LOO | Influential observations affecting LOO estimate | Use WAIC instead; investigate influential points; consider k-fold CV |\n| Hit max tree depth | Model geometry requires long trajectories | Reparameterize model; increase `max_treedepth` parameter |\n| `SamplingError` at start | Poor initialization | Try `init=\"adapt_diag\"` or `init=\"jitter+adapt_diag\"`; provide `initvals` via MAP |\n| Biased posterior (prior dominates) | Priors too strong relative to data | Weaken priors (increase sigma); check prior predictive; verify data is correctly passed |\n| `ValueError: logp = -inf` | Parameter hit impossible region | Check data for NaN\u002FInf; ensure positive params use `HalfNormal`\u002F`Gamma`; verify likelihood matches data type |\n\n## Best Practices\n\n1. **Use weakly informative priors**, not flat priors. Flat priors (Uniform over large ranges) cause sampling difficulties and are rarely appropriate.\n2. **Always run prior predictive checks** before fitting. If prior-implied data is implausible, adjust priors before sampling.\n3. **Standardize continuous predictors** to mean=0, sd=1 for better NUTS geometry and faster sampling.\n4. **Use non-centered parameterization** for hierarchical models: `offset ~ Normal(0,1); param = mu + sigma * offset` instead of `param ~ Normal(mu, sigma)`.\n5. **Run at least 4 chains** for reliable R-hat and ESS diagnostics. Never trust a single chain.\n6. **Check all three diagnostics** (R-hat \u003C 1.01, ESS > 400, 0 divergences) before interpreting posteriors.\n7. **Include `log_likelihood=True`** when fitting if you plan to compare models via LOO or WAIC.\n8. **Use `dims` instead of `shape`** for named dimensions -- integrates with ArviZ for labeled summaries and subsetting.\n9. **Report credible intervals** (HDI), not just point estimates. The posterior distribution is the result, not a single number.\n10. **Save results** with `idata.to_netcdf(\"results.nc\")` for reproducibility and later re-analysis.\n\n## Bundled Resources\n\nTwo reference files provide deeper detail for on-demand consultation:\n\n- **`references\u002Fdistributions_inference.md`** -- Distribution catalog organized by category (continuous, discrete, multivariate, mixture, time series, special\u002Fmodifiers) with full parameter signatures, support, and common uses. Sampling and inference methods (NUTS, Metropolis, Slice, CompoundStep, SMC, ADVI, fullrank ADVI, SVGD, MAP) with code examples and a method selection guide table. Reparameterization tricks (non-centered, QR decomposition) with code.\n  Covers: consolidated from original `references\u002Fdistributions.md` (320 lines) and `references\u002Fsampling_inference.md` (424 lines).\n  Relocated inline: core distribution selection guide table (Key Concepts), diagnostic threshold table (Key Concepts), basic ADVI usage (Recipe: Variational Inference in original, covered in references here).\n  Omitted: shape broadcasting examples (trivial NumPy-style `shape=5`), basic `dims` usage (covered in Workflow Step 2), prior predictive sampling (covered in Workflow Step 3).\n\n- **`references\u002Fadvanced_workflows.md`** -- Model comparison workflow (LOO, WAIC, Pareto-k reliability checks, model averaging with weighted predictions), comprehensive diagnostic report generation (trace plots, rank plots, autocorrelation, energy, ESS evolution), prior-posterior comparison patterns, data preparation best practices (standardization, centering, missing data imputation), prior selection guidelines (weakly informative vs informative with domain knowledge), named dimensions (`dims`) usage with xarray subsetting, save\u002Fload patterns (NetCDF, pickle). Mixture model pattern.\n  Covers: consolidated from original `references\u002Fworkflows.md` (526 lines).\n  Relocated inline: core diagnostic checking logic (Workflow Step 5), model comparison basics (Workflow Step 7), prior predictive check (Workflow Step 3), linear\u002Flogistic\u002Fhierarchical model code (Recipes).\n  Omitted: complete monolithic workflow template (redundant with 8-step Workflow section).\n\n### Per-reference-file disposition (all originals)\n\n**Original `references\u002F` (3 files):**\n1. `distributions.md` (320 lines) -> consolidated with `sampling_inference.md` into new `references\u002Fdistributions_inference.md`\n2. `sampling_inference.md` (424 lines) -> consolidated with `distributions.md` into new `references\u002Fdistributions_inference.md`\n3. `workflows.md` (526 lines) -> migrated as new `references\u002Fadvanced_workflows.md`\n\n**Original `scripts\u002F` (2 files):**\n1. `model_diagnostics.py` (350 lines) -> `check_diagnostics()` logic (R-hat, ESS, divergences, tree depth checks) inlined in Workflow Step 5; `create_diagnostic_report()` plot generation patterns described in Expected Outputs and referenced in `references\u002Fadvanced_workflows.md`; `compare_prior_posterior()` utility covered in `references\u002Fadvanced_workflows.md`\n2. `model_comparison.py` (387 lines) -> `compare_models()` and `check_loo_reliability()` patterns inlined in Workflow Step 7 (using `az.compare` and `az.loo` directly); `model_averaging()` pattern covered in `references\u002Fadvanced_workflows.md`; `cross_validation_comparison()` guidance covered in `references\u002Fadvanced_workflows.md`; `plot_model_comparison()` is a thin wrapper around `az.plot_compare` (inlined in Step 7)\n\n## Related Skills\n\n- **arviz (planned)** -- posterior visualization and diagnostics; PyMC returns ArviZ InferenceData objects\n- **bambi (planned)** -- formula-based interface to PyMC (R-style `y ~ x1 + x2` syntax); simpler API for standard GLMs\n- **numpyro (planned)** -- JAX-based probabilistic programming; faster NUTS via hardware acceleration\n- **statsmodels-statistical-modeling** -- frequentist regression and GLMs; use when Bayesian framework is not needed\n- **scikit-learn-machine-learning** -- ML classification\u002Fregression; use when prediction accuracy matters more than uncertainty quantification\n\n## References\n\n- [PyMC documentation](https:\u002F\u002Fwww.pymc.io\u002Fprojects\u002Fdocs\u002Fen\u002Fstable\u002F) -- official API reference and tutorials\n- [PyMC GitHub](https:\u002F\u002Fgithub.com\u002Fpymc-devs\u002Fpymc) -- source code and issue tracker\n- [ArviZ documentation](https:\u002F\u002Fpython.arviz.org\u002Fen\u002Fstable\u002F) -- posterior analysis and visualization\n- [Bayesian Modeling and Computation in Python](https:\u002F\u002Fbayesiancomputationbook.com\u002F) -- textbook using PyMC\n- [PyMC examples gallery](https:\u002F\u002Fwww.pymc.io\u002Fprojects\u002Fexamples\u002Fen\u002Flatest\u002F) -- curated notebook collection\n","# PyMC Bayesian Modeling\n\n## Overview\n\nPyMC is a Python library for Bayesian statistical modeling and probabilistic programming. It provides an expressive syntax for defining probabilistic models and efficient inference via MCMC (NUTS) and variational methods (ADVI). This skill covers the full Bayesian modeling cycle from model specification through diagnostics, comparison, and prediction.\n\n## When to Use\n\n- Estimating parameters with full uncertainty quantification (credible intervals, not just point estimates)\n- Fitting hierarchical\u002Fmultilevel models to grouped or nested data\n- Performing prior and posterior predictive checks to validate model assumptions\n- Comparing candidate models using information criteria (LOO-CV, WAIC)\n- Building regression models (linear, logistic, Poisson) in a Bayesian framework\n- Handling missing data or measurement error as latent parameters\n- Modeling time series with autoregressive or random walk priors\n- Generating posterior predictions for new observations with uncertainty bounds\n- Use **Stan\u002FPyStan** instead for compiled, more scalable Bayesian inference on large models; use **statsmodels** for frequentist statistical tests\n\n## Prerequisites\n\n- **Python packages**: `pymc >= 5.0`, `arviz`, `numpy`, `matplotlib`\n- **Data**: NumPy arrays or pandas DataFrames with numeric columns\n- **Environment**: CPU sufficient for most models; GPU via JAX backend for large models\n\n```bash\npip install pymc arviz numpy matplotlib\n# Optional: JAX backend for GPU acceleration\npip install pymc[jax]\n```\n\n## Quick Start\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Simulate data\nnp.random.seed(42)\nX = np.random.randn(100)\ny = 2.5 + 1.3 * X + np.random.randn(100) * 0.5\n\n# Build and fit model\nwith pm.Model() as model:\n    alpha = pm.Normal(\"alpha\", mu=0, sigma=5)\n    beta = pm.Normal(\"beta\", mu=0, sigma=5)\n    sigma = pm.HalfNormal(\"sigma\", sigma=1)\n    mu = alpha + beta * X\n    y_obs = pm.Normal(\"y_obs\", mu=mu, sigma=sigma, observed=y)\n    idata = pm.sample(1000, tune=1000, chains=4, random_seed=42)\n\nprint(az.summary(idata, var_names=[\"alpha\", \"beta\", \"sigma\"]))\n# Expected: alpha ~ 2.5, beta ~ 1.3, sigma ~ 0.5\n```\n\n## Workflow\n\n### Step 1: Prepare Data\n\nStandardize continuous predictors for better sampling efficiency. Use named coordinates for readable models and ArviZ integration.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Load data\nX = np.random.randn(200, 3)  # 200 obs, 3 predictors\ny = X @ np.array([1.0, -0.5, 0.3]) + np.random.randn(200) * 0.8\n\n# Standardize predictors\nX_mean, X_std = X.mean(axis=0), X.std(axis=0)\nX_scaled = (X - X_mean) \u002F X_std\n\n# Define coordinates for named dimensions\ncoords = {\n    \"predictors\": [\"var1\", \"var2\", \"var3\"],\n    \"obs_id\": np.arange(len(y)),\n}\nprint(f\"Data shape: X={X_scaled.shape}, y={y.shape}\")\n```\n\n### Step 2: Define Model and Set Priors\n\nSpecify the model structure inside a `pm.Model()` context. Use weakly informative priors, `dims` for named dimensions, and `HalfNormal` or `Exponential` for scale parameters.\n\n```python\nwith pm.Model(coords=coords) as model:\n    # Priors — weakly informative, not flat\n    alpha = pm.Normal(\"alpha\", mu=0, sigma=1)\n    beta = pm.Normal(\"beta\", mu=0, sigma=1, dims=\"predictors\")\n    sigma = pm.HalfNormal(\"sigma\", sigma=1)\n\n    # Linear predictor\n    mu = alpha + pm.math.dot(X_scaled, beta)\n\n    # Likelihood\n    y_obs = pm.Normal(\"y_obs\", mu=mu, sigma=sigma, observed=y, dims=\"obs_id\")\n\n# Inspect model variables\nprint(model.basic_RVs)  # Lists: [alpha, beta, sigma, y_obs]\n```\n\n### Step 3: Prior Predictive Check\n\nValidate that priors produce plausible data ranges before fitting. Adjust priors if simulated data is unreasonable.\n\n```python\nwith model:\n    prior_pred = pm.sample_prior_predictive(samples=1000, random_seed=42)\n\n# Check prior-implied data range\nprior_y = prior_pred.prior_predictive[\"y_obs\"].values.flatten()\nprint(f\"Prior predictive range: [{prior_y.min():.1f}, {prior_y.max():.1f}]\")\nprint(f\"Observed data range:    [{y.min():.1f}, {y.max():.1f}]\")\n\naz.plot_ppc(prior_pred, group=\"prior\", num_pp_samples=100)\n```\n\n### Step 4: Sample Posterior (MCMC)\n\nRun NUTS sampling with multiple chains. Include `log_likelihood=True` if you plan model comparison later.\n\n```python\nwith model:\n    idata = pm.sample(\n        draws=2000,\n        tune=1000,\n        chains=4,\n        target_accept=0.9,\n        random_seed=42,\n        idata_kwargs={\"log_likelihood\": True},\n    )\n\nprint(f\"Posterior shape: {idata.posterior['beta'].shape}\")\n# Expected: (4 chains, 2000 draws, 3 predictors)\n```\n\n### Step 5: Diagnose Sampling\n\nCheck convergence before interpreting results. All three diagnostics (R-hat, ESS, divergences) must pass.\n\n```python\n# Summary with convergence diagnostics\nsummary = az.summary(idata, var_names=[\"alpha\", \"beta\", \"sigma\"])\nprint(summary[[\"mean\", \"sd\", \"hdi_3%\", \"hdi_97%\", \"r_hat\", \"ess_bulk\"]])\n\n# R-hat convergence check\nbad_rhat = summary[summary[\"r_hat\"] > 1.01]\nif len(bad_rhat) > 0:\n    print(f\"WARNING: {len(bad_rhat)} parameters with R-hat > 1.01\")\n    print(bad_rhat[[\"r_hat\"]])\n\n# Effective sample size check\nlow_ess = summary[summary[\"ess_bulk\"] \u003C 400]\nif len(low_ess) > 0:\n    print(f\"WARNING: {len(low_ess)} parameters with ESS \u003C 400\")\n\n# Divergence check\nn_div = idata.sample_stats.diverging.sum().item()\ntotal = len(idata.posterior.draw) * len(idata.posterior.chain)\nprint(f\"Divergences: {n_div}\u002F{total} ({n_div \u002F total * 100:.2f}%)\")\n\n# Visual diagnostics — trace plots and rank plots\naz.plot_trace(idata, var_names=[\"alpha\", \"beta\", \"sigma\"])\naz.plot_rank(idata, var_names=[\"alpha\", \"beta\", \"sigma\"])\n```\n\n### Step 6: Posterior Predictive Check\n\nValidate model fit by comparing simulated data from the posterior to observed data.\n\n```python\nwith model:\n    pm.sample_posterior_predictive(idata, extend_inferencedata=True, random_seed=42)\n\naz.plot_ppc(idata, num_pp_samples=100)\n# Blue = observed data, grey = posterior simulations\n# Systematic deviations indicate model misspecification\n```\n\n### Step 7: Compare Models\n\nUse LOO-CV or WAIC to compare candidate models. Lower information criterion is better.\n\n```python\n# Fit multiple models with log_likelihood=True, then compare\n# Example: compare linear vs a second model\nidatas = {\"linear\": idata}  # add more fitted models here\n\ncomparison = az.compare(idatas, ic=\"loo\")\nprint(comparison[[\"rank\", \"elpd_loo\", \"p_loo\", \"d_loo\", \"weight\"]])\n\n# Check LOO reliability via Pareto-k diagnostics\nloo_result = az.loo(idata, pointwise=True)\nhigh_k = (loo_result.pareto_k > 0.7).sum().item()\nprint(f\"Observations with Pareto-k > 0.7: {high_k}\")\n# Interpretation: Dloo \u003C 2 = similar models; Dloo > 10 = strong evidence\n\naz.plot_compare(comparison)\n```\n\n### Step 8: Generate Predictions\n\nProduce posterior predictions for new data with full uncertainty propagation.\n\n```python\nX_new = np.array([[0.5, -1.0, 0.2]])\nX_new_scaled = (X_new - X_mean) \u002F X_std\n\nwith model:\n    pm.set_data({\"X_scaled\": X_new_scaled})\n    post_pred = pm.sample_posterior_predictive(\n        idata.posterior, var_names=[\"y_obs\"], random_seed=42\n    )\n\ny_pred = post_pred.posterior_predictive[\"y_obs\"]\nprint(f\"Predicted mean: {y_pred.mean().item():.3f}\")\nprint(f\"94% HDI: {az.hdi(y_pred, hdi_prob=0.94).values}\")\n```\n\n## Key Parameters\n\n| Parameter | Default | Range \u002F Options | Effect |\n|-----------|---------|-----------------|--------|\n| `draws` | `1000` | `500`-`10000` | Number of posterior samples per chain |\n| `tune` | `1000` | `500`-`5000` | Warmup iterations (discarded); increase for complex posteriors |\n| `chains` | `4` | `2`-`8` | Number of independent chains; minimum 4 for reliable R-hat |\n| `cores` | all CPUs | `1`-`N` | Parallel chains; set equal to `chains` for full parallelism |\n| `target_accept` | `0.8` | `0.8`-`0.99` | NUTS acceptance rate; increase to reduce divergences |\n| `init` | `\"auto\"` | `\"adapt_diag\"`, `\"jitter+adapt_diag\"`, `\"advi\"` | Initialization strategy for sampler |\n| `random_seed` | `None` | any int | Seed for reproducibility |\n| `idata_kwargs` | `{}` | `{\"log_likelihood\": True}` | Store log-likelihood for LOO\u002FWAIC model comparison |\n| `method` (pm.fit) | `\"advi\"` | `\"advi\"`, `\"fullrank_advi\"`, `\"svgd\"` | Variational inference algorithm |\n| `n` (pm.fit) | `10000` | `5000`-`100000` | VI optimization iterations |\n| `samples` (prior pred) | `500` | `100`-`5000` | Prior predictive samples for validation |\n\n## Key Concepts\n\n### Prior\u002FDistribution Selection Guide\n\n| Distribution | Use When | Key Parameters |\n|-------------|----------|----------------|\n| `Normal(mu, sigma)` | Unbounded real-valued parameter (standardized data) | `mu`: center, `sigma`: spread |\n| `HalfNormal(sigma)` | Scale\u002Fstandard deviation parameter (positive) | `sigma`: spread of positive half |\n| `Exponential(lam)` | Scale parameter, alternative to HalfNormal | `lam`: rate (1\u002Fmean) |\n| `StudentT(nu, mu, sigma)` | Robust alternative to Normal (outlier-resistant) | `nu`: degrees of freedom (\u003C10 = heavier tails) |\n| `Beta(alpha, beta)` | Probability or proportion in [0,1] | `alpha=beta=2`: weakly informative |\n| `Gamma(alpha, beta)` | Positive parameter (rate, concentration) | `alpha`: shape, `beta`: rate |\n| `LogNormal(mu, sigma)` | Positive parameter with multiplicative effects | `mu`, `sigma`: of underlying Normal |\n| `LKJCorr(n, eta)` | Correlation matrix prior | `eta=1`: uniform; `eta>1`: prefer identity |\n| `Dirichlet(a)` | Probability vector (sums to 1) | `a`: concentration; uniform if all equal |\n| `Bernoulli(p \u002F logit_p)` | Binary outcome likelihood | Use `logit_p` for numerical stability |\n| `Poisson(mu)` | Count data (equidispersed) | `mu`: rate; use NegBinomial if overdispersed |\n| `NegativeBinomial(mu, alpha)` | Overdispersed count data | `alpha`: dispersion (smaller = more overdispersion) |\n\n### Diagnostic Thresholds\n\n| Metric | Threshold | Interpretation | Action if Failed |\n|--------|-----------|---------------|------------------|\n| R-hat | \u003C 1.01 | Chains converged | Run longer chains; check multimodality |\n| ESS bulk | > 400 | Sufficient independent samples | Increase `draws`; reparameterize |\n| ESS tail | > 400 | Reliable tail estimates | Increase `draws` |\n| Divergences | 0 | NUTS explored successfully | Increase `target_accept`; non-centered param. |\n| Pareto-k (LOO) | \u003C 0.7 | LOO estimate reliable | Use WAIC or k-fold CV |\n| Max tree depth | \u003C 10 | No trajectory truncation | Reparameterize or increase `max_treedepth` |\n\n### Model Variants Overview\n\n| Problem Type | Recipe | Likelihood | Key Feature |\n|-------------|--------|------------|-------------|\n| Grouped\u002Fnested data | Hierarchical Model | Normal (varies) | Non-centered parameterization, partial pooling |\n| Binary outcome | Logistic Regression | Bernoulli | `logit_p` link function |\n| Nonlinear\u002Fspatial | Gaussian Process | Normal | Kernel-based covariance, flexible shape |\n| Count data | (use Poisson in Workflow) | Poisson \u002F NegBinomial | Log link; NegBinomial for overdispersion |\n| Time series | (see references) | AR \u002F GaussianRandomWalk | Autoregressive coefficients |\n| Mixture\u002Fclustering | (see references) | Mixture \u002F NormalMixture | Component weights via Dirichlet |\n\n## Common Recipes\n\n### Recipe: Hierarchical Model\n\nWhen to use: data has natural grouping (patients within hospitals, students within schools). Non-centered parameterization avoids divergences from funnel geometry.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\nn_groups, n_per_group = 5, 30\ngroup_idx = np.repeat(np.arange(n_groups), n_per_group)\ngroup_names = [f\"group_{i}\" for i in range(n_groups)]\n\n# Simulated grouped data\ntrue_alphas = np.random.normal(3.0, 1.5, n_groups)\ny_obs = np.random.normal(true_alphas[group_idx], 0.5)\n\nwith pm.Model(coords={\"groups\": group_names}) as hierarchical_model:\n    # Hyperpriors (population level)\n    mu_alpha = pm.Normal(\"mu_alpha\", mu=0, sigma=5)\n    sigma_alpha = pm.HalfNormal(\"sigma_alpha\", sigma=2)\n\n    # Non-centered parameterization\n    alpha_offset = pm.Normal(\"alpha_offset\", mu=0, sigma=1, dims=\"groups\")\n    alpha = pm.Deterministic(\n        \"alpha\", mu_alpha + sigma_alpha * alpha_offset, dims=\"groups\"\n    )\n\n    # Observation level\n    sigma = pm.HalfNormal(\"sigma\", sigma=1)\n    y = pm.Normal(\"y\", mu=alpha[group_idx], sigma=sigma, observed=y_obs)\n\n    idata_hier = pm.sample(2000, tune=1000, target_accept=0.95, random_seed=42)\n\nprint(az.summary(idata_hier, var_names=[\"mu_alpha\", \"sigma_alpha\", \"alpha\", \"sigma\"]))\n```\n\n### Recipe: Logistic Regression\n\nWhen to use: binary outcome variable (success\u002Ffailure, disease\u002Fhealthy). Use `logit_p` for numerical stability instead of computing probabilities directly.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Simulated binary outcome\nnp.random.seed(42)\nn, p = 200, 3\nX_lr = np.random.randn(n, p)\ntrue_beta = np.array([1.0, -0.5, 0.3])\nprob = 1 \u002F (1 + np.exp(-(0.2 + X_lr @ true_beta)))\ny_binary = np.random.binomial(1, prob)\n\nwith pm.Model() as logistic_model:\n    alpha = pm.Normal(\"alpha\", mu=0, sigma=2)\n    beta = pm.Normal(\"beta\", mu=0, sigma=2, shape=p)\n    logit_p = alpha + pm.math.dot(X_lr, beta)\n    y = pm.Bernoulli(\"y\", logit_p=logit_p, observed=y_binary)\n\n    idata_logit = pm.sample(2000, tune=1000, target_accept=0.9, random_seed=42)\n\nsummary = az.summary(idata_logit, var_names=[\"alpha\", \"beta\"])\nprint(summary[[\"mean\", \"sd\", \"hdi_3%\", \"hdi_97%\"]])\n# Odds ratios\nbeta_samples = idata_logit.posterior[\"beta\"].values.reshape(-1, p)\nprint(f\"Odds ratios (median): {np.median(np.exp(beta_samples), axis=0)}\")\n```\n\n### Recipe: Gaussian Process\n\nWhen to use: modeling unknown nonlinear functions, spatial data, or smooth latent processes. Computationally expensive for n > 1000; consider sparse approximations for larger datasets.\n\n```python\nimport pymc as pm\nimport arviz as az\nimport numpy as np\n\n# Simulated nonlinear data\nnp.random.seed(42)\nX_gp = np.sort(np.random.uniform(0, 10, 60))[:, None]\ny_gp = np.sin(X_gp[:, 0]) + np.random.randn(60) * 0.3\n\nwith pm.Model() as gp_model:\n    # GP hyperparameters\n    ls = pm.Gamma(\"lengthscale\", alpha=2, beta=1)\n    eta = pm.HalfNormal(\"amplitude\", sigma=2)\n    sigma_noise = pm.HalfNormal(\"noise\", sigma=0.5)\n\n    # Covariance function\n    cov = eta**2 * pm.gp.cov.Matern52(1, ls=ls)\n    gp = pm.gp.Marginal(cov_func=cov)\n\n    # Marginal likelihood\n    y_ = gp.marginal_likelihood(\"y\", X=X_gp, y=y_gp, sigma=sigma_noise)\n    idata_gp = pm.sample(1000, tune=1000, target_accept=0.9, random_seed=42)\n\n# Predict on new points\nX_new_gp = np.linspace(0, 10, 100)[:, None]\nwith gp_model:\n    f_pred = gp.conditional(\"f_pred\", X_new_gp)\n    pred_samples = pm.sample_posterior_predictive(idata_gp, var_names=[\"f_pred\"])\n\nprint(f\"GP predictions shape: {pred_samples.posterior_predictive['f_pred'].shape}\")\n```\n\n## Expected Outputs\n\n- **Trace plots** (`az.plot_trace`) -- chain mixing visualization; healthy chains look like \"fuzzy caterpillars\"\n- **Posterior summary** (`az.summary`) -- mean, SD, HDI, R-hat, ESS per parameter\n- **Prior\u002Fposterior predictive plots** (`az.plot_ppc`) -- simulated vs observed data distributions\n- **Forest plots** (`az.plot_forest`) -- coefficient estimates with credible intervals\n- **Model comparison table** (`az.compare`) -- ranked models with LOO\u002FWAIC, weights, warnings\n- **Rank plots** (`az.plot_rank`) -- uniform histograms indicate good mixing\n- **Energy plot** (`az.plot_energy`) -- HMC energy transition diagnostics\n- **InferenceData file** (`idata.to_netcdf(\"results.nc\")`) -- serialized results for later analysis\n\n## Troubleshooting\n\n| Problem | Cause | Solution |\n|---------|-------|----------|\n| Divergent transitions | Posterior geometry difficult for NUTS (funnels, ridges) | Increase `target_accept=0.95`; use non-centered parameterization; add stronger priors |\n| Low ESS (\u003C 400) | High autocorrelation between samples | Increase `draws=5000`; reparameterize to reduce correlation; try QR decomposition for correlated predictors |\n| R-hat > 1.01 | Chains have not converged | Increase `tune=2000, draws=5000`; check for multimodality; initialize with ADVI |\n| Slow sampling | Complex posterior or large dataset | Use ADVI for initialization; reduce model complexity; increase `cores`; try JAX backend |\n| Pareto-k > 0.7 in LOO | Influential observations affecting LOO estimate | Use WAIC instead; investigate influential points; consider k-fold CV |\n| Hit max tree depth | Model geometry requires long trajectories | Reparameterize model; increase `max_treedepth` parameter |\n| `SamplingError` at start | Poor initialization | Try `init=\"adapt_diag\"` or `init=\"jitter+adapt_diag\"`; provide `initvals` via MAP |\n| Biased posterior (prior dominates) | Priors too strong relative to data | Weaken priors (increase sigma); check prior predictive; verify data is correctly passed |\n| `ValueError: logp = -inf` | Parameter hit impossible region | Check data for NaN\u002FInf; ensure positive params use `HalfNormal`\u002F`Gamma`; verify likelihood matches data type |\n\n## Best Practices\n\n1. **Use weakly informative priors**, not flat priors. Flat priors (Uniform over large ranges) cause sampling difficulties and are rarely appropriate.\n2. **Always run prior predictive checks** before fitting. If prior-implied data is implausible, adjust priors before sampling.\n3. **Standardize continuous predictors** to mean=0, sd=1 for better NUTS geometry and faster sampling.\n4. **Use non-centered parameterization** for hierarchical models: `offset ~ Normal(0,1); param = mu + sigma * offset` instead of `param ~ Normal(mu, sigma)`.\n5. **Run at least 4 chains** for reliable R-hat and ESS diagnostics. Never trust a single chain.\n6. **Check all three diagnostics** (R-hat \u003C 1.01, ESS > 400, 0 divergences) before interpreting posteriors.\n7. **Include `log_likelihood=True`** when fitting if you plan to compare models via LOO or WAIC.\n8. **Use `dims` instead of `shape`** for named dimensions -- integrates with ArviZ for labeled summaries and subsetting.\n9. **Report credible intervals** (HDI), not just point estimates. The posterior distribution is the result, not a single number.\n10. **Save results** with `idata.to_netcdf(\"results.nc\")` for reproducibility and later re-analysis.\n\n## Bundled Resources\n\nTwo reference files provide deeper detail for on-demand consultation:\n\n- **`references\u002Fdistributions_inference.md`** -- Distribution catalog organized by category (continuous, discrete, multivariate, mixture, time series, special\u002Fmodifiers) with full parameter signatures, support, and common uses. Sampling and inference methods (NUTS, Metropolis, Slice, CompoundStep, SMC, ADVI, fullrank ADVI, SVGD, MAP) with code examples and a method selection guide table. Reparameterization tricks (non-centered, QR decomposition) with code.\n  Covers: consolidated from original `references\u002Fdistributions.md` (320 lines) and `references\u002Fsampling_inference.md` (424 lines).\n  Relocated inline: core distribution selection guide table (Key Concepts), diagnostic threshold table (Key Concepts), basic ADVI usage (Recipe: Variational Inference in original, covered in references here).\n  Omitted: shape broadcasting examples (trivial NumPy-style `shape=5`), basic `dims` usage (covered in Workflow Step 2), prior predictive sampling (covered in Workflow Step 3).\n\n- **`references\u002Fadvanced_workflows.md`** -- Model comparison workflow (LOO, WAIC, Pareto-k reliability checks, model averaging with weighted predictions), comprehensive diagnostic report generation (trace plots, rank plots, autocorrelation, energy, ESS evolution), prior-posterior comparison patterns, data preparation best practices (standardization, centering, missing data imputation), prior selection guidelines (weakly informative vs informative with domain knowledge), named dimensions (`dims`) usage with xarray subsetting, save\u002Fload patterns (NetCDF, pickle). Mixture model pattern.\n  Covers: consolidated from original `references\u002Fworkflows.md` (526 lines).\n  Relocated inline: core diagnostic checking logic (Workflow Step 5), model comparison basics (Workflow Step 7), prior predictive check (Workflow Step 3), linear\u002Flogistic\u002Fhierarchical model code (Recipes).\n  Omitted: complete monolithic workflow template (redundant with 8-step Workflow section).\n\n### Per-reference-file disposition (all originals)\n\n**Original `references\u002F` (3 files):**\n1. `distributions.md` (320 lines) -> consolidated with `sampling_inference.md` into new `references\u002Fdistributions_inference.md`\n2. `sampling_inference.md` (424 lines) -> consolidated with `distributions.md` into new `references\u002Fdistributions_inference.md`\n3. `workflows.md` (526 lines) -> migrated as new `references\u002Fadvanced_workflows.md`\n\n**Original `scripts\u002F` (2 files):**\n1. `model_diagnostics.py` (350 lines) -> `check_diagnostics()` logic (R-hat, ESS, divergences, tree depth checks) inlined in Workflow Step 5; `create_diagnostic_report()` plot generation patterns described in Expected Outputs and referenced in `references\u002Fadvanced_workflows.md`; `compare_prior_posterior()` utility covered in `references\u002Fadvanced_workflows.md`\n2. `model_comparison.py` (387 lines) -> `compare_models()` and `check_loo_reliability()` patterns inlined in Workflow Step 7 (using `az.compare` and `az.loo` directly); `model_averaging()` pattern covered in `references\u002Fadvanced_workflows.md`; `cross_validation_comparison()` guidance covered in `references\u002Fadvanced_workflows.md`; `plot_model_comparison()` is a thin wrapper around `az.plot_compare` (inlined in Step 7)\n\n## Related Skills\n\n- **arviz (planned)** -- posterior visualization and diagnostics; PyMC returns ArviZ InferenceData objects\n- **bambi (planned)** -- formula-based interface to PyMC (R-style `y ~ x1 + x2` syntax); simpler API for standard GLMs\n- **numpyro (planned)** -- JAX-based probabilistic programming; faster NUTS via hardware acceleration\n- **statsmodels-statistical-modeling** -- frequentist regression and GLMs; use when Bayesian framework is not needed\n- **scikit-learn-machine-learning** -- ML classification\u002Fregression; use when prediction accuracy matters more than uncertainty quantification\n\n## References\n\n- [PyMC documentation](https:\u002F\u002Fwww.pymc.io\u002Fprojects\u002Fdocs\u002Fen\u002Fstable\u002F) -- official API reference and tutorials\n- [PyMC GitHub](https:\u002F\u002Fgithub.com\u002Fpymc-devs\u002Fpymc) -- source code and issue tracker\n- [ArviZ documentation](https:\u002F\u002Fpython.arviz.org\u002Fen\u002Fstable\u002F) -- posterior analysis and visualization\n- [Bayesian Modeling and Computation in Python](https:\u002F\u002Fbayesiancomputationbook.com\u002F) -- textbook using PyMC\n- [PyMC examples gallery](https:\u002F\u002Fwww.pymc.io\u002Fprojects\u002Fexamples\u002Fen\u002Flatest\u002F) -- curated notebook collection\n",{"name":5,"license":15,"description":7},"master","63f17866001c5ca854bf2fa0b678ce277133de5b","skills\u002Fawesome-skills\u002Fpymc-bayesian-modeling",0,"https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS","https:\u002F\u002Fbg-szy.github.io\u002FTOP-SKILLS\u002F","https:\u002F\u002Fgithub.com\u002Fbg-szy","https:\u002F\u002Fwww.ecust.edu.cn\u002F","a student of East China University of Science and Technology","2026-11-15 03:38:02",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add bg-szy\u002FTOP-SKILLS --skill pymc-bayesian-modeling",{"tool":40,"command":41},"bunx","bunx skills add bg-szy\u002FTOP-SKILLS --skill pymc-bayesian-modeling",{"tool":43,"command":44},"pnpm","pnpm dlx skills add bg-szy\u002FTOP-SKILLS --skill pymc-bayesian-modeling","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fawesome-skills\u002Fpymc-bayesian-modeling\u002FSKILL.md",true,[],[49,58,65,72,79,86],{"skillNo":50,"name":51,"nameZh":6,"description":52,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":53,"licenseFileUrl":6,"sourceHtmlUrl":54,"categories":55,"keywords":56,"publishedAt":57},"01M7DXW9N826PKXK174PVW3038","serpmaven","Scrape Google search results (SERP) to JSON with SERP Maven (serpmaven) - a real Chrome browser on a residential proxy with sticky sessions. Run it as a CLI, a Python library, a local HTTP search API (`serve`), an MCP server or a Docker image; set the proxy login, workers, pages and cost tracking. Use when the user wants Google results, rank tracking, a self-hosted SERP API, or live Google search for an AI agent.","MIT","https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fmarketplace\u002Fserpmaven\u002FSKILL.md",[],[],"2026-11-15 14:19:17",{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":53,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":57},"01M7DXVHSTWDV65475QEQ915M7","revnu-landing-page-audit","Find out why a landing page isn't turning visitors into signups, demos or sales. Audits one page against what the ad, post or search that sends people there promised, then checks the first screen, proof, call to action, friction and mobile. Returns at most 7 prioritized fixes, plus what is fine and what could not be checked; rewrites specific lines and fixes the page in the repo when asked. Never promises a conversion lift. Works without a Revnu account. Use when the user says or asks: \"audit my landing page\", \"why isn't this page converting\", \"review my homepage\", \"my ads get clicks but no signups\", \"people bounce from the pricing page\", \"CRO review\", \"improve conversion rate\", \"is my hero section clear\", \"roast my landing page\", \"rewrite my headline\", \"what's wrong with this page\", \"landing page feedback\", \"make the signup page convert better\". Not for SEO or AI-search audits, page speed engineering on its own, or building a new page from scratch.","https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fmarketplace\u002Frevnu-landing-page-audit\u002FSKILL.md",[],[],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":53,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":57},"01M7DXV8X2D7B3VTD4HHYYHTGR","revnu-growth-context","Write down once who buys, what they're stuck on, what it costs, what proof you have and what you've already tried, so every growth task starts from facts instead of a questionnaire. Creates or updates .agents\u002Frevnu-growth.md from the repo, the live site and the user, and imports an existing product-marketing.md. Use when the user says \"set up our marketing context\", \"who is our ideal customer\", \"write our ICP\", \"describe the product for marketing\", \"positioning doc\", \"what's our brand voice\", \"remember this about our customers\", \"log what we tried\", \"what have we learned from marketing\", \"we changed pricing, update the context\", \"update our growth notes\", \"summarize the business for the agent\", \"import my product-marketing.md\", \"start a growth doc\", or before a first marketing, SEO, outreach or launch task when no context file exists. Works without a Revnu account.","https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fmarketplace\u002Frevnu-growth-context\u002FSKILL.md",[],[],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":53,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":57},"01M7DXV0DGJ1G0A60PAHC6J3C5","revnu-get-listed","Get onto the third-party pages AI answers and search cite when buyers ask for the best tool in your category: \"best X\" roundups, directories, review sites, GitHub awesome lists, community threads and comparison pages. Finds which ones are cited, ranks them by fit and effort, and drafts each submission, PR or pitch by the target's own rules. Use when the user says \"get us listed\", \"which directories should we submit to\", \"how do we get into best-of lists\", \"why does ChatGPT recommend competitors and not us\", \"get mentioned in AI answers\", \"submit to awesome lists\", \"get on G2\", \"where are our competitors listed\", \"pitch listicle authors\", \"find backlink opportunities\", \"launch directories\", \"get into roundups\", \"Product Hunt alternatives\", \"where should we be listed\", \"get cited by Perplexity\". Never submits, pays or posts without a yes. Works without a Revnu account.","https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fmarketplace\u002Frevnu-get-listed\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":53,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":57},"01M7DXTSPMGT9D6DSS0HP9BSTA","revnu-ai-readiness-audit","Audit whether your website gives ChatGPT, Perplexity, Claude, Gemini and Google's AI Overviews something worth quoting, and fix the pages that don't. Use when the user says or asks \"is my site ready for AI search\", \"audit my site for AI answers\", \"AEO audit\", \"GEO audit\", \"why doesn't ChatGPT quote our pricing\", \"make our pages citable by AI\", \"do I need llms.txt\", \"does schema help with ChatGPT\", \"check our structured data\", \"is our content visible without JavaScript\", \"AI SEO audit\", \"optimize our homepage for AI answers\", \"LLM SEO check\", \"why does AI get our product wrong\", \"fix our site for AI Overviews\". Checks fetchability and indexability, answers in the raw HTML, entity facts, schema (including JS-injected markup), headings, dates and llms.txt, then gives at most 10 prioritised fixes and edits the repo when asked. Works without a Revnu account.","https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fmarketplace\u002Frevnu-ai-readiness-audit\u002FSKILL.md",[],[],{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":53,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":57},"01M7DXSZHCVTRS620GC3S97JZV","proxy-troubleshooting","Diagnose residential proxy failures, with NodeMaven's gateway as the reference - 407 with correct credentials, 406, 410, 500, a connection that hangs, ERR_TUNNEL_CONNECTION_FAILED in a browser, Playwright or Puppeteer ignoring proxy credentials, a sticky session that changes IP, geo-targeting that is silently ignored, passwords with special characters, and unexpected traffic from headless Chrome. Use when a proxy connection fails, returns the wrong location, or costs more traffic than expected.","https:\u002F\u002Fgithub.com\u002Fbg-szy\u002FTOP-SKILLS\u002Fblob\u002Fmaster\u002Fskills\u002Fmarketplace\u002Fproxy-troubleshooting\u002FSKILL.md",[],[]]