NormalGammaPrior.shape

NormalGammaPrior.shape: float

The Gamma shape \(\lambda\) of every local variance.

Sets the marginal’s excess kurtosis, \(3 / \lambda\): one is the Laplace of the Bayesian lasso, and smaller values pile more mass at zero while fattening the tails. Below one half the variance conditional at an exact zero is improper, which _draw_scales() guards with a floor. Written theta in the label. Must be positive.