170 likes | 479 Views
ILOG Solver Directions. Laurent Perron ILOG SA. Outline. Constraint Programming, a powerful technology The CP suite in ILOG CP faces new challenges Recent Technical Advances. Constraint Programming, a Powerful Technology. I hope this will be demonstrated by the workshop
E N D
ILOG Solver Directions Laurent Perron ILOG SA
Outline • Constraint Programming, a powerful technology • The CP suite in ILOG • CP faces new challenges • Recent Technical Advances
Constraint Programming, a Powerful Technology • I hope this will be demonstrated by the workshop • CP is a robust technology with a past of successful applications • Some famous academics success in the OR community (10 teams problem closed easily) • A large collection of deployed industrial applications based on Constraint Programming
CP Strengths • A high level modeling language allows for a rich and accurate representation of the model • Domain expertise is captured through a rich search language • Special techniques (decomposition, LNS, repair…) allows the tackling of large and difficult problems • “A clever tool for clever people”
CP weaknesses • A high level modeling language allows for a rich and accurate representation of the model • Domain expertise is captured through a rich search language • Special techniques (decomposition, LNS, repair…) allows the tackling of large and difficult problems • You need clever people to use this clever tool
ILOG CP Suite • ILOG Solver • Generic CP System • ILOG Scheduler • Detailed Scheduling Specialization • ILOG Dispatcher • Vehicule Routing and Disptaching Specialization • ILOG Configurator • Configuration System
ILOG CP Market • ILOG CP Market is defined by • ILOG Consultants • Big ISV • Specialized Solution Vendors • Technical Consulting Companies • Big Companies with Dedicated R&D • This is not a huge market • We want to enlarge our market
CP Meets New Challenges • To reach our goals, CP should be improved • At the evaluation phase • Easier to use tools • Rapid results • Low coding effort • At the implementation phase • Low technology profile • At the maintenance phase • The application should improve with time
New Rules for Tools Evaluation • People are not very technical • Maybe one week of training • They may have a limited OR background • The know basic OR rules about modeling • They may not be Computer Science experts • The cp system should be easily integrated/documented • Maybe not keen with compilers and library
News Rules for Evaluation (2) • Evaluation Phase is Limited in time • IT is made against other competing techniques • And sometimes against internal tools • We need to achieve something soon • Even if the problem is over-constrained • Even if the data are dirty • Even if the model is naive
New Rules for Application Development • The IT guy is not the OR Expert • The code will not evolve • The time devoted to Optimization is usually limited • Users will need guidance • Why this doesn’t work? • Users expect performance to improve with time • Without code evolution
Usability • To be effective, a CP solution consists of • A good model • A clever search part • We would like to remove the need for search part • At least for small problems, typically the one encountered in the evaluation phase
Robustness • Different logically equivalent formulations can lead to different runtime performances • Expand a global constraint into smaller subparts • This imply that getting a good model is an art • We do not believe this cannot be • The CP Solver should detect these cases and reformulate the model • Users do not know about different level of propagation
Evolution of the Code • In the past, to use a new technology, a new constraint or a new search construct was implemented • Users had to rewrite their code in order to benefit from it • And new users had to learn more each time
Interactivity • Users want Explanation • And useful explanations • Users want Solutions • Even with over-constrained problems • Along with explanations of why some constraints are not part of the solution
ILOG Solver Directions • Default Search • Model Reinforcement • Constraint Aggregation • Various work on robustness • Better constraints (without filtering levels) • No pathological cases • No slow propagation • Explanations and Solver Anyway
ILOG CP Directions • There is a new suite of optimization tools in ILOG • ILOG Plant Power Ops for production planning and scheduling • ILOG Transport Power Ops for routing and dispatching • ILOG Fab Power Ops for semi conductor industry