Tiresias: a demonstration of how-to queries

Abstract

In this demo, we will present Tiresias, the first how-to query engine. How-to queries represent fundamental data analysis questions of the form: "How should the input change in order to achieve the desired output". They exemplify an important Reverse Data Management problem: solving constrained optimization problems over data residing in a DBMS. Tiresias, named after the mythical oracle of Thebes, has complex under-workings, but includes a simple interface that allows users to load datasets and interactively design optimization problems by simply selecting actions, key performance indicators, and objectives. The user choices are translated into a declarative query, which is then processed by Tiresias and translated into a Mixed Integer Program: we then use an MIP solver to find a solution. The solution is then presented to the user as an interactive data instance. The user can provide feedback by rejecting certain tuples and/or values. Then, based on the user feedback, Tiresias automatically refines the how-to query and presents a new set of results.