|
We have developed full-scale engines based on both MAC and stochastic search and enhanced by dedicated algorithms required for the solutions of CSP's with the above characteristics.
MAC-based engines: Perform systematic search with backtracking through arc-consistent states. The engines have strong generic abilities to cope with conditional CSP, and a hierarchy of soft constraints, and have built in data-types which are designed to deal with huge domains and with some of the generic forms of constraints we encounter. The engines are used by Piparazzi, Genesys-Pro, and X-Gen.
Stochastic search engine: Performs stochastic search with an escape algorithm which is non-local, and strongly relies on learning of high-level parameters of the program topography. The engine is used by FPGen.
Analytic and specific engines: Used mainly by FPGen to solve problems which have an identifiable analytic solution, or where an analytic solution may lead to a good starting point for other heuristics.
Modeling: We designed a language template suitable for modeling the wide range of CSP's we encounter. For each of the applications, a specific modeling language was designed based on this template. A graphical tool supports language, and defining any CSP reduces to filling up forms in each of the specific languages.
| 
