bayespecon.GaussianGibbsPriors¶
-
class bayespecon.GaussianGibbsPriors(beta_mu=
0.0, beta_sigma=1000000.0, sigma2_alpha=2.0, sigma2_beta=1.0, rho_lower=-0.999, rho_upper=0.999, sigma_sigma=10.0)[source]¶ Prior hyperparameters for Gaussian spatial Gibbs.
- Parameters:¶
- beta_mu : float or ndarray¶
Prior mean for β. Scalar is broadcast to all coefficients.
- beta_sigma : float or ndarray¶
Prior standard deviation for β. Scalar is broadcast.
- sigma2_alpha : float¶
Shape hyperparameter of the
InverseGamma(sigma2_alpha, sigma2_beta)prior on σ². Matches the NUTS path exactly so that posteriors — and therefore LOO/WAIC — agree between the two samplers. Conjugate with the Gaussian likelihood, so the σ² block is an exact closed-form draw (LeSage 2009 convention).- sigma2_beta : float¶
Scale (rate) hyperparameter of the InverseGamma prior on σ². Models typically resolve this to
Var(y)at construction so the prior mean is scale-aware.- rho_lower : float¶
Lower bound for ρ/λ (from spectral stability).
- rho_upper : float¶
Upper bound for ρ/λ (from spectral stability).
-
__init__(beta_mu=
0.0, beta_sigma=1000000.0, sigma2_alpha=2.0, sigma2_beta=1.0, rho_lower=-0.999, rho_upper=0.999, sigma_sigma=10.0)[source]¶
Methods
__init__([beta_mu, beta_sigma, ...])Attributes