Source code and data for the paper "Necessary and Sufficient Conditions for Optimal Decision Trees using Dynamic Programming"

STreeD is a framework for optimal binary decision trees with separable optimization tasks. A separable optimization task is a task that can be optimized separately for the left and right subtree. The current STreeD Framework implements a broad set of such optimization tasks, from group fairness constraints to prescriptive policy generation. For an explanation of each application, see below. For details on what tasks are separable and how the algorithm works, see our paper.