PyMC Bayesian Modeling

Overview

PyMC is a Python library for Bayesian modeling and probabilistic programming. Build, fit, validate, and compare Bayesian models using PyMC's modern API (version 6.x+), including hierarchical models, MCMC sampling (NUTS), variational inference, posterior predictive checks, and model comparison (LOO, WAIC).

Current Version and Setup

PyMC 6.0.1 is the current stable release as of June 2026. It requires Python 3.12+, uses PyTensor 3 as the computational graph backend, and defaults to compiled backends such as Numba. For reproducible local environments, pin the version:

bash
uv pip install "pymc[nutpie]==6.0.1"

The nutpie extra enables the faster Rust/Numba NUTS implementation. If using NumPyro or BlackJAX, install those optional sampler dependencies in the same environment and pin them in the project lockfile.

When to Use This Skill

This skill should be used when:

• Building Bayesian models (linear/logistic regression, hierarchical models, time series, etc.)
• Performing MCMC sampling or variational inference
• Conducting prior/posterior predictive checks
• Diagnosing sampling issues (divergences, convergence, ESS)
• Comparing multiple models using information criteria (LOO, WAIC)
• Implementing uncertainty quantification through Bayesian methods
• Working with hierarchical/multilevel data structures
• Handling missing data or measurement error in a principled way

Standard Bayesian Workflow

Follow this workflow for building and validating Bayesian models:

1. Data Preparation

python
import pymc as pm
import arviz as az
import numpy as np

# Load and prepare data
X = ...  # Predictors
y = ...  # Outcomes

# Standardize predictors for better sampling
X_mean = X.mean(axis=0)
X_std = X.std(axis=0)
X_scaled = (X - X_mean) / X_std

Key practices:

• Standardize continuous predictors (improves sampling efficiency)
• Center outcomes when possible
• Handle missing data explicitly (treat as parameters)
• Use named dimensions with coords for clarity

2. Model Building

python
coords = {
    'predictors': ['var1', 'var2', 'var3'],
    'obs_id': np.arange(len(y))
}

with pm.Model(coords=coords) as model:
    # Mutable data container so prediction data can be swapped later
    X_data = pm.Data('X_scaled', X_scaled, dims=('obs_id', 'predictors'))

    # Priors
    alpha = pm.Normal('alpha', mu=0, sigma=1)
    beta = pm.Normal('beta', mu=0, sigma=1, dims='predictors')
    sigma = pm.HalfNormal('sigma', sigma=1)

    # Linear predictor
    mu = alpha + pm.math.dot(X_data, beta)

    # Tie the observed variable's shape to X_data for out-of-sample prediction
    y_obs = pm.Normal('y_obs', mu=mu, sigma=sigma, observed=y, shape=X_data.shape[0], dims='obs_id')

Key practices:

• Use weakly informative priors (not flat priors)
• Use HalfNormal or Exponential for scale parameters
• Use named dimensions (dims) instead of shape when possible
• Use pm.Data() for values that will be updated for predictions

3. Prior Predictive Check

Always validate priors before fitting:

python
with model:
    prior_pred = pm.sample_prior_predictive(draws=1000, random_seed=42)

# Visualize
az.plot_ppc(prior_pred, group='prior')

Check:

• Do prior predictions span reasonable values?
• Are extreme values plausible given domain knowledge?
• If priors generate implausible data, adjust and re-check

4. Fit Model

python
with model:
    # Optional: Quick exploration with ADVI
    # approx = pm.fit(n=20000)

    # Full MCMC inference
    idata = pm.sample(
        draws=2000,
        tune=1000,
        chains=4,
        target_accept=0.9,
        random_seed=42,
        idata_kwargs={'log_likelihood': True}  # For model comparison
    )

Key parameters:

• draws=2000: Number of samples per chain
• tune=1000: Warmup samples (discarded)
• chains=4: Run 4 chains for convergence checking
• target_accept=0.9: Higher for difficult posteriors (0.95-0.99)
• Include log_likelihood=True for model comparison
• If using PyMC 6 sampler-specific kwargs, avoid deprecated nuts_sampler_kwargs; pass explicit NUTS kwargs through nuts={...} when needed

5. Check Diagnostics

Use the diagnostic script:

python
from scripts.model_diagnostics import check_diagnostics

results = check_diagnostics(idata, var_names=['alpha', 'beta', 'sigma'])

Check:

• R-hat < 1.01: Chains have converged
• ESS > 400: Sufficient effective samples
• No divergences: NUTS sampled successfully
• Trace plots: Chains should mix well (fuzzy caterpillar)

If issues arise:

• Divergences → Increase target_accept=0.95, use non-centered parameterization
• Low ESS → Sample more draws, reparameterize to reduce correlation
• High R-hat → Run longer, check for multimodality

6. Posterior Predictive Check

Validate model fit:

python
with model:
    pm.sample_posterior_predictive(idata, extend_inferencedata=True, random_seed=42)

# Visualize
az.plot_ppc(idata)

Check:

• Do posterior predictions capture observed data patterns?
• Are systematic deviations evident (model misspecification)?
• Consider alternative models if fit is poor

7. Analyze Results

python
# Summary statistics
print(az.summary(idata, var_names=['alpha', 'beta', 'sigma']))

# Posterior distributions
az.plot_posterior(idata, var_names=['alpha', 'beta', 'sigma'])

# Coefficient estimates
az.plot_forest(idata, var_names=['beta'], combined=True)

8. Make Predictions

python
X_new = ...  # New predictor values
X_new_scaled = (X_new - X_mean) / X_std

with model:
    pm.set_data({'X_scaled': X_new_scaled}, coords={'obs_id': np.arange(len(X_new_scaled))})
    post_pred = pm.sample_posterior_predictive(
        idata,
        var_names=['y_obs'],
        predictions=True,
        random_seed=42
    )

# Extract prediction intervals
y_pred_mean = post_pred.predictions['y_obs'].mean(dim=['chain', 'draw'])
y_pred_hdi = az.hdi(post_pred.predictions, var_names=['y_obs'])

Common Model Patterns

Linear Regression

For continuous outcomes with linear relationships:

python
with pm.Model() as linear_model:
    alpha = pm.Normal('alpha', mu=0, sigma=10)
    beta = pm.Normal('beta', mu=0, sigma=10, shape=n_predictors)
    sigma = pm.HalfNormal('sigma', sigma=1)

    mu = alpha + pm.math.dot(X, beta)
    y = pm.Normal('y', mu=mu, sigma=sigma, observed=y_obs)

Use template: assets/linear_regression_template.py

Logistic Regression

For binary outcomes:

python
with pm.Model() as logistic_model:
    alpha = pm.Normal('alpha', mu=0, sigma=10)
    beta = pm.Normal('beta', mu=0, sigma=10, shape=n_predictors)

    logit_p = alpha + pm.math.dot(X, beta)
    y = pm.Bernoulli('y', logit_p=logit_p, observed=y_obs)

Hierarchical Models

For grouped data (use non-centered parameterization):

python
with pm.Model(coords={'groups': group_names}) as hierarchical_model:
    # Hyperpriors
    mu_alpha = pm.Normal('mu_alpha', mu=0, sigma=10)
    sigma_alpha = pm.HalfNormal('sigma_alpha', sigma=1)

    # Group-level (non-centered)
    alpha_offset = pm.Normal('alpha_offset', mu=0, sigma=1, dims='groups')
    alpha = pm.Deterministic('alpha', mu_alpha + sigma_alpha * alpha_offset, dims='groups')

    # Observation-level
    mu = alpha[group_idx]
    sigma = pm.HalfNormal('sigma', sigma=1)
    y = pm.Normal('y', mu=mu, sigma=sigma, observed=y_obs)

Use template: assets/hierarchical_model_template.py

Critical: Always use non-centered parameterization for hierarchical models to avoid divergences.

Poisson Regression

For count data:

python
with pm.Model() as poisson_model:
    alpha = pm.Normal('alpha', mu=0, sigma=10)
    beta = pm.Normal('beta', mu=0, sigma=10, shape=n_predictors)

    log_lambda = alpha + pm.math.dot(X, beta)
    y = pm.Poisson('y', mu=pm.math.exp(log_lambda), observed=y_obs)

For overdispersed counts, use NegativeBinomial instead.

Time Series

For autoregressive processes:

python
with pm.Model() as ar_model:
    sigma = pm.HalfNormal('sigma', sigma=1)
    rho = pm.Normal('rho', mu=0, sigma=0.5, shape=ar_order)
    init_dist = pm.Normal.dist(mu=0, sigma=sigma)

    y = pm.AR('y', rho=rho, sigma=sigma, init_dist=init_dist, observed=y_obs)

Model Comparison

Comparing Models

Use LOO or WAIC for model comparison:

python
from scripts.model_comparison import compare_models, check_loo_reliability

# Fit models with log_likelihood
models = {
    'Model1': idata1,
    'Model2': idata2,
    'Model3': idata3
}

# Compare using LOO
comparison = compare_models(models, ic='loo')

# Check reliability
check_loo_reliability(models)

Interpretation:

• Δloo < 2: Models are similar, choose simpler model
• 2 < Δloo < 4: Weak evidence for better model
• 4 < Δloo < 10: Moderate evidence
• Δloo > 10: Strong evidence for better model

Check Pareto-k values:

• k < 0.7: LOO reliable
• k > 0.7: Consider WAIC or k-fold CV

Model Averaging

When models are similar, average predictions:

python
from scripts.model_comparison import model_averaging

averaged_pred, weights = model_averaging(models, var_name='y_obs')

Distribution Selection Guide

For Priors

Scale parameters (σ, τ):

• pm.HalfNormal('sigma', sigma=1) - Default choice
• pm.Exponential('sigma', lam=1) - Alternative
• pm.Gamma('sigma', alpha=2, beta=1) - More informative

Unbounded parameters:

• pm.Normal('theta', mu=0, sigma=1) - For standardized data
• pm.StudentT('theta', nu=3, mu=0, sigma=1) - Robust to outliers

Positive parameters:

• pm.LogNormal('theta', mu=0, sigma=1)
• pm.Gamma('theta', alpha=2, beta=1)

Probabilities:

• pm.Beta('p', alpha=2, beta=2) - Weakly informative
• pm.Uniform('p', lower=0, upper=1) - Non-informative (use sparingly)

Correlation matrices:

• pm.LKJCholeskyCov('chol', n=n_vars, eta=2, sd_dist=pm.HalfNormal.dist(1)) - Preferred covariance prior
• pm.LKJCorr('corr', n=n_vars, eta=2) - Correlation-only prior; eta=1 uniform, eta>1 prefers identity

For Likelihoods

Continuous outcomes:

• pm.Normal('y', mu=mu, sigma=sigma) - Default for continuous data
• pm.StudentT('y', nu=nu, mu=mu, sigma=sigma) - Robust to outliers

Count data:

• pm.Poisson('y', mu=lambda) - Equidispersed counts
• pm.NegativeBinomial('y', mu=mu, alpha=alpha) - Overdispersed counts
• pm.ZeroInflatedPoisson('y', psi=psi, mu=mu) - Excess zeros
• pm.HurdleNegativeBinomial('y', psi=psi, mu=mu, alpha=alpha) - Excess zeros plus overdispersion

Binary outcomes:

• pm.Bernoulli('y', p=p) or pm.Bernoulli('y', logit_p=logit_p)

Categorical outcomes:

• pm.Categorical('y', p=probs)

See: references/distributions.md for comprehensive distribution reference

Sampling and Inference

MCMC with NUTS

Default and recommended for most models:

python
idata = pm.sample(
    draws=2000,
    tune=1000,
    chains=4,
    target_accept=0.9,
    random_seed=42
)

Adjust when needed:

• Divergences → target_accept=0.95 or higher
• Slow sampling → Use ADVI for initialization
• Discrete parameters → Use pm.Metropolis() for discrete vars

Variational Inference

Fast approximation for exploration or initialization:

python
with model:
    approx = pm.fit(n=20000, method='advi')

    # Use for initialization
    initvals = approx.sample(return_inferencedata=False)[0]
    idata = pm.sample(initvals=initvals)

Trade-offs:

• Much faster than MCMC
• Approximate (may underestimate uncertainty)
• Good for large models or quick exploration

See: references/sampling_inference.md for detailed sampling guide

Diagnostic Scripts

Comprehensive Diagnostics

python
from scripts.model_diagnostics import create_diagnostic_report

create_diagnostic_report(
    idata,
    var_names=['alpha', 'beta', 'sigma'],
    output_dir='diagnostics/'
)

Creates:

• Trace plots
• Rank plots (mixing check)
• Autocorrelation plots
• Energy plots
• ESS evolution
• Summary statistics CSV

Quick Diagnostic Check

python
from scripts.model_diagnostics import check_diagnostics

results = check_diagnostics(idata)

Checks R-hat, ESS, divergences, and tree depth.

Common Issues and Solutions

Divergences

Symptom: idata.sample_stats.diverging.sum() > 0

Solutions:

1. Increase target_accept=0.95 or 0.99
2. Use non-centered parameterization (hierarchical models)
3. Add stronger priors to constrain parameters
4. Check for model misspecification

Low Effective Sample Size

Symptom: ESS < 400

Solutions:

1. Sample more draws: draws=5000
2. Reparameterize to reduce posterior correlation
3. Use QR decomposition for regression with correlated predictors

High R-hat

Symptom: R-hat > 1.01

Solutions:

1. Run longer chains: tune=2000, draws=5000
2. Check for multimodality
3. Improve initialization with ADVI

Slow Sampling

Solutions:

1. Use ADVI initialization
2. Reduce model complexity
3. Increase parallelization: cores=8, chains=8
4. Use variational inference if appropriate

Best Practices

Model Building

1. Always standardize predictors for better sampling
2. Use weakly informative priors (not flat)
3. Use named dimensions (dims) for clarity
4. Non-centered parameterization for hierarchical models
5. Check prior predictive before fitting

Sampling

1. Run multiple chains (at least 4) for convergence
2. Use target_accept=0.9 as baseline (higher if needed)
3. Include log_likelihood=True for model comparison
4. Set random seed for reproducibility

Validation

1. Check diagnostics before interpretation (R-hat, ESS, divergences)
2. Posterior predictive check for model validation
3. Compare multiple models when appropriate
4. Report uncertainty (HDI intervals, not just point estimates)

Workflow

1. Start simple, add complexity gradually
2. Prior predictive check → Fit → Diagnostics → Posterior predictive check
3. Iterate on model specification based on checks
4. Document assumptions and prior choices

Resources

This skill includes:

References (references/)

• distributions.md: Comprehensive catalog of PyMC distributions organized by category (continuous, discrete, multivariate, mixture, time series). Use when selecting priors or likelihoods.

• sampling_inference.md: Detailed guide to sampling algorithms (NUTS, Metropolis, SMC), variational inference (ADVI, SVGD), and handling sampling issues. Use when encountering convergence problems or choosing inference methods.

• workflows.md: Complete workflow examples and code patterns for common model types, data preparation, prior selection, and model validation. Use as a cookbook for standard Bayesian analyses.

Scripts (scripts/)

• model_diagnostics.py: Automated diagnostic checking and report generation. Functions: check_diagnostics() for quick checks, create_diagnostic_report() for comprehensive analysis with plots.

• model_comparison.py: Model comparison utilities using LOO/WAIC. Functions: compare_models(), check_loo_reliability(), model_averaging().

Templates (assets/)

• linear_regression_template.py: Complete template for Bayesian linear regression with full workflow (data prep, prior checks, fitting, diagnostics, predictions).

• hierarchical_model_template.py: Complete template for hierarchical/multilevel models with non-centered parameterization and group-level analysis.

Quick Reference

Model Building

python
with pm.Model(coords={'var': names}) as model:
    # Priors
    param = pm.Normal('param', mu=0, sigma=1, dims='var')
    # Likelihood
    y = pm.Normal('y', mu=..., sigma=..., observed=data)

Sampling

python
idata = pm.sample(draws=2000, tune=1000, chains=4, target_accept=0.9)

Diagnostics

python
from scripts.model_diagnostics import check_diagnostics
check_diagnostics(idata)

Model Comparison

python
from scripts.model_comparison import compare_models
compare_models({'m1': idata1, 'm2': idata2}, ic='loo')

Predictions

python
with model:
    pm.set_data({'X_data': X_new})
    pred = pm.sample_posterior_predictive(idata, predictions=True)

Additional Notes

• PyMC integrates with ArviZ for visualization and diagnostics; PyMC 6 / ArviZ 1 use xarray DataTree while retaining familiar groups such as .posterior and .posterior_predictive
• Use pm.model_to_graphviz(model) to visualize model structure
• Save results with idata.to_netcdf('results.nc')
• Load with az.from_netcdf('results.nc')
• For very large models, consider minibatch ADVI or data subsampling