Abstract
We approach the classic problem of discovering interesting itemsets in data using an infinite latent feature model that utilizes novel constructions of the Hierarchical Indian Buffet Process (HIBP). These constructions yield explicit expressions for the posterior, predictive and marginal distributions of the HIBP which we utilize to obtain likelihood and predictive distributions for our model resulting in an implementation that was unavailable in literature. We propose an extension to using a Poisson HIBP to allow for random count valued latent features and derive a predictive distribution for that model. We also provide a sequential sampling scheme for our model which allows sequential sampling of items by existing customers and new customers.