Active development now lives at YuZh98/combinatorial-regression.
Please head there for the latest code, examples, and documentation. This repository is kept only as an archive of the original prototype accompanying the paper "Statistical Modeling of Combinatorial Response Data."
This repository contains the original R and C++ prototype for a Bayesian framework that models combinatorial response data — integer-valued arrays subject to structural constraints, arising in settings such as surveys with skip logic, event propagation on networks, and observed matchings in ecological systems.
The approach treats the observed response as a deterministic transformation of a continuous latent variable, defined as the solution to an Integer Linear Program (ILP), and uses a Gibbs sampler with a custom hit-and-run step for posterior inference. It generalizes the classical probit data augmentation approach to the combinatorial setting.
For the maintained implementation, please use combinatorial-regression.