Simulating PipeLand: A quantified consulting problem

Project with Lloyd Elliott

Traditional statistical problems such as density estimation, clustering, classification, and regression, do not always map directly onto real-world problems. For example, in consulting the results of a predictive model may be transformed by a company's policy or procedures to form a sophisticated loss function which is the real target of the optimization. In this project, the student will develop a simulated consulting problem for a fictitious utility company. We will simulate a vast network of pipes for the utility company, and simulate compounding pipe failures in the network over time. We will suppose that the company has a requirement to replace pipes according to a schedule, and we will transform the results of predictive models into replacement schedules and form an automated method to evaluate the predictive models using docker. This project will result in a realistic testing framework for predictive models which can be used to compare models, or for teaching purposes.