bayespecon¶
Bayesian Spatial Econometric Models
The bayespecon package is designed to make it simpler to fit, diagnose, and interpret Bayesian spatial econometric regression models. It provides a suite of classes for building commmonly-used models using a straightforward API. Each model is implemented as a class that defines how spatial effects are represented, and the ‘main’ portion of the model specification is given using the familiar Wilkinson format via formulaic (but you can pass design matrices if you prefer).
Each model class uses PySAL graph objects to represent spatial weights, \(W\), (or sparse matrices if you prefer) providing thorough integration with the scientific Python and spatial analysis ecosystems. Estimation is handled by a custom Gibbs sampler tuned for spatial models, or passed to pymc to sample with NUTS instead.
Main Features:
Wide variety of spatial econometric models using Wilkinson formulas and PySAL
GraphobjectsMarginal (direct and indirect) effects for models with spatial terms
Fast log-determinant functions for evaluating spatial terms
Custom Gibbs sampler & PyMC/NUTS alternative
Generate synthetic datasets using a known data-generating process for each model