Mixing Constraint Programming, Linear Programming and Learning Methods for a scheduling engine

ILOG develops a new line of product called Constraint Programming Optimization (CPO) whose emphasis is the ease of use. That means a modeling layer based on constraint programming methods and automatic search. We present the model components that are developed for resource scheduling: optional interval variable, Simple Temporal Network (STN) of temporal constraints, 2-SAT graph of logical constraints, function of interval, cumulation and sequence. We illustrate the expressive power and ease of assemble complex model with these tools. Then we describe the main algorithmic facilities that are embedded in the automatic search: propagation algorithm, decision heuristics search tree traversal, linear relaxation, and improvement methods based on Large Neighborhood Search (LNS). We show how having these methods efficiently collaborating and insist upon learning techniques we use as meta-heuristics.