StarfishDB: A Query Execution Engine for Relational Probabilistic Programming
Ouael Ben Amara, Sami Hadouaj, Niccolò Meneghetti · Proceedings of the ACM on Management of Data · 2024
We introduce StarfishDB, a query execution engine optimized for relational probabilistic programming. Our engine adopts the model of Gamma Probabilistic Databases, representing probabilistic programs as a collection of relational constraints, imposed against a generative stochastic process. We extend the model with the support for recursion, factorization and the ability to leverage just-in-time compilation techniques to speed up inference. We test our engine against a state-of-the-art sampler for Latent Dirichlet Allocation.