Building Search Methods with Self-Confidence in a Constraint Programming Library

In the late 1990s, Constrained Programming (CP) promised to separate the declaration of a problem from the process to solve it. This work attempts to serve this direction, by implementing and presenting a modular way to define search methods that seek solutions to arbitrary Constraint Satisfaction Problems (CSPs). The user just declares their CSP, and it can be solved using a portfolio of search methods already in place. Apart from the pluggable search methods framework for any CSP, we also introduce pluggable heuristics for our search methods. We found an efficient stochastic heuristics' paradigm that smoothly combines randomness with normal heuristics. We consider a factor of disobedience to normal heuristics, and we fine-tune it each time, according to our estimation of normal heuristics' reliability (confidence). We prove mathematically that while the disobedience factor decreases, the stochastic heuristics approximate deterministic normal heuristics. Our algebraic evidence is supported by empirical evaluations on real life problems: A new search method, namely PoPS, that exploits this heuristics' paradigm, can outperform regular well-known constructive search methods.