sprcom stands for Spatial Regression of Communities and is a Bayesian statistical package designed to streamline the interpretation and modeling of very high dimensional binary and count-valued data. The underlying model assumes a low-dimensional latent structure via communities or clusters that leads to a parsimonious model. sprcom is unique in that it can also account for the dependence of these communities on covariates! A number of utility and plotting functions are included to help visualize your results. sprcom is a wrapper for a PyMC3 model and you can use any PyMC3 estimation method with it including Hamiltonian Monte Carlo and ADVI. In particular, it is especially well suited for GPU computing.
covariates, response, adjacency = load_data(...)
n_communities = 5
model = spatial_community_regression(covariates, response, adjacency, n_communities)
with model:
trace = pm.sample()
...You can install this package using pip: pip install sprcom. It requires several dependencies including PyMC3 and Seaborn.
We've included documentation to help you get up and running. Check out the florabank1-tutorial notebook for more details!
For questions or comments please contact Christopher Krapu at ckrapu@gmail.com.