Learning Optimal Decision Trees Under Memory Constraints

Existing algorithms for learning optimal decision trees can be put into two categories: based on the use of Mixed Integer Programming (MIP) solvers and dynamic programming (DP) itemsets. While DP are fastest, their main disadvantage compared to MIP-based approaches is that amount memory these may require find an solution not bounded. Consequently, some datasets only executed machines with large amounts memory. In this paper, we propose first DP-based algorithm operates under constraints. Core contributions work include: (1) strategies freeing when too much used by algorithm; (2) effective approach recovering tree parts freed. Our experiments demonstrate a favorable trade-off between constraints run times our algorithm.