Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....they propagate values through the constraints according to the ordering (this is called executing) Unfortunately, the local propagation algorithms lost generality so they can only be applied to special sets of constraints and comparators. The local propagation algorithms DeltaBlue [13] SkyBlue [12], QuickPlan [15] DETAIL [10] Houria [8] can solve only equality constraints, e.g. linear equations over reals. The exception is the Indigo algorithm [3] for solving inequalities that combines local propagation and refining method. We borrowed the main idea behind the Indigo algorithm, i.e. the ....

Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, 1993.

....systems by constraint hierarchies. The problem is probably hidden in hierarchical constraint solvers which utilize different concepts than traditional constraint satisfaction. Additionally, most current hierarchical constraint solvers are constructed for locally better comparators only [8,9,11,12], and thus they are not suitable for inter hierarchy comparison. Also, if the constraint solvers with intrahierarchy comparison are computational hungry, the solvers supporting inter hierarchy comparison are even more demanding for computational power. Despite the current spare usage of ....

....T14. The other benefit of constraint hierarchies is the possibility to use very weak constraints to establish stability of solution that is crucial for many up to date applications of constraint technology. Currently, this feature is widely used in the construction of graphical user interfaces [11,12]. By stability of the solution we mean as small as possible difference between the original solution and the tuned solution obtained by solving the system of constraints. Constraint hierarchies and inter hierarchy comparison can be succesfully applied to traditional applications, like geometric ....

[Article contains additional citation context not shown here]

Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993

....potential locality of typical constraint networks, e.g. in graphical user interfaces. These algorithms gradually solve constraint hierarchies by selecting uniquely satisfiable constraints repeatedly. Many local propagation algorithms were developed for special purposes (DeltaBlue [18] SkyBlue [17], QuickPlan [20] DETAIL [10] Houria [9] Indigo [5] Our generalized approach to solving constraint hierarchies is also primary based 2 Another method for describing over constrained systems is PCSP (Partial Constraint Satisfaction Problems) 3 In [23] the authors also introduce ....

Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993

....systems by constraint hierarchies. The problem is probably hidden in hierarchical constraint solvers which utilize different concepts than traditional constraint satisfaction. Additionaly, most current hierarchical constraint solvers are constructed for locally better comparators only [8,9,11,12], and thus they are not suitable for inter hierarchy comparison. Also, if the constraint solvers with intra hierarchy comparison are computational hungry, the solvers supporting inter hierarchy comparison are even more demanding for computational power. 4 There exists definition of constraint ....

....The constraint network is constructed during the planning phase, while the valuation of variables is computed during the propagation phase. We will use the features of the planning phase here. The planning phase is well known from hierarchical constraint solvers DeltaBlue [12] and SkyBlue [11]. In [3] we 5 Note, that the hierarchy has no solution if we work with the domain of reals. 10 generalized the planning phase in such a way that all types of constraints and comparators, including globally better, are supported. During the planning phase, each constraint is assigned into a ....

Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993

....and suffer accordingly. Second prototype One way to reduce the impact of equation manipulation is to avoid redoing it when unnecessary. Pika regenerates the simulation equations from scratch every time the set of active MFs changes. We have reimplemented Pika (as Pika2) using SkyBlue [ Sannella, 1992 ] a hierarchical constraint manager. SkyBlue effectively maintains a causal ordering graph which can be updated incrementally as MFs activate and deactivate. Only those equations whose causal direction changes (or which form new simultaneities) must Pika2 resolve. Also, Pika2 caches solutions ....

M. Sannella. The SkyBlue Constraint Solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, December 1992.

....constraint in each step. In addition, when a variable is repeatedly updated, e.g. by user operation, it can easily evaluate only the necessary constraints to get a new solution. However, local propagation is restricted in some ways. Most local propagation algorithms (DeltaBlue [15] SkyBlue [16], QuickPlan [18] DETAIL [9] Houria [7] can solve only equality constraints, e.g. linear equations over reals. The exception is the Indigo [6] algorithm for solving inequalities that combines local propagation and refining method. The local propagation algorithms also usually use ....

Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993

....only the necessary constraints to get a new solution. This straightforward execution phase is paid off by a foregoing planning phase that choose the order of constraints to satisfy. Local propagation is also restricted in some ways. Most local propagation algorithms (DeltaBlue [20] SkyBlue [19], QuickPlan [22] DETAIL [12] Houria [10] can solve only equality constraints, e.g. linear equations over reals. The exception is the Indigo [6] algorithm for solving inequalities that combines local propagation and refining method. We borrowed the main idea behind the Indigo algorithm, i.e. ....

....it is a functional constraint cell) The first five points of the constraint network definition are obvious conditions from traditional constraint graphs extended to cover the constraint cells. Thus, the proposed constraint network is a generalization of the former concept of constraint graphs [10,19,20,22]. The new conditions 6 and 7 of the Definition 10 are the contribution of this work. They help us to keep linearity and thus effectiveness of the execution algorithm. As the execution algorithm, computing values of variables, traverses downstream the constraint network, it has to be sure that ....

Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993 25

....only the necessary constraints to get a new solution. This straightforward execution phase is paid off by a foregoing planning phase that choose the order of constraints to satisfy. 6 Local propagation is also restricted in some ways. Most local propagation algorithms (DeltaBlue [18] SkyBlue [17], QuickPlan [20] DETAIL [10] Houria [9] can solve only equality constraints, e.g. linear equations over reals. The exception is the Indigo [5] algorithm for solving inequalities that combines local propagation and refining method. We borrowed the main idea behind the Indigo algorithm, i.e. ....

....it is a functional constraint cell) The first five points of the constraint network definition are obvious conditions from traditional constraint graphs extended to cover the constraint cells. Thus, the proposed constraint network is a generalization of the former concept of constraint graphs [9,17,18,20]. The new conditions 6 and 7 of the Definition 4 are the contribution of this work. They help us to keep linearity and thus effectiveness of the execution algorithm. As the execution algorithm, computing values of variables, traverses downstream the constraint network, it has to be sure that using ....

Sannella, M., The SkyBlue Constraint Solver, Tech. Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993

....of our approach, in languages based on constraint solvers, dependencies are expressed in terms of constraints between instance variables. When the value of such a constrained variable is modified, a propagation algorithm tries to satisfy the constraint, modifying linked variables ( FBB92, MGZ92, San93] However, constraints are not expressed in terms of object interactions, so some inter object dependencies are difficult to express as constraints between instance variables. Moreover, some limitations on types of components are imposed by the constraint solver. Finally, as Wilk said in ....

....A static analysis of the link graph is important in order to detect problems due to the global control of dependencies more precisely and rapidly. The semantics of flo language is described in centaur [Bor87] and we are now working on static detection of cycles and computation of reaction chains [San93] Acknowledgement We would like to thank all the reviewers of this paper, in particular Pierre Cointe and Pierre Berlandier. ....

M. Sannella. The skyblue constraint solver. Technical Report 92-07-01, Department of Computer Science and Engineering, University of Washington, February 1993.

....(ThingLab II [44, 43] and Multi Garnet [56] has used the locally predicate better comparator. This choice has been based primarily on pragmatic rather than aesthetic or theoretical grounds: the existence of efficient incremental algorithms DeltaBlue [20] and a derivative algorithm SkyBlue [55] for finding LPB solutions. For user interface applications, we do have extensive experience in the practical use of LPB [57] It also been used by a considerable number of researchers at other institutions as well. LPB has generally proved quite satisfactory. However, for precise layout ....

Michael Sannella. The SkyBlue Local Propagation Solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, December 1992.

....re executes each time of any of its variables (or the subparts of those variables) changes identity. The execution of a constructor might trigger additional constructors to execute. This process continues until the remaining constraints are primitive constraints, which are solved by SkyBlue (Sannella 1993), the incremental local propagation solver used for Kaleidoscope 93 primitives. SkyBlue is a new algorithm that extends our earlier DeltaBlue algorithm (Freeman Benson, Maloney, and Borning 1990; Sannella, Maloney, Freeman Benson, and Borning 1993) to allow constraints with multiple outputs, and ....

Sannella, M. (1993, February). The SkyBlue Constraint Solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington.

....in the Kmachine, but the cost is small compared to the tremendous performance advantage of incremental constraint satisfaction over repeatedly solving all constraints. Further, this design allows us to exploit the incremental properties of our local propagation solver [FreemanBenson et al. 90] Sannella 93] See Section 4.1 for more details. Implementing Constraint Imperative Programming Languages: The Kaleidoscope 93 Virtual Machine page 9 of 13 since p1 and p2 contain objects of class Point. The Point object in p3 becomes the sum of p1 and p2. The constructor is not re executed after the ....

....colors, and strings. We term these constraints value constraints, since the satisfaction of these constraints finds values for objects within their domains. The primitive constraint solver we currently use in the Kaleidoscope 93 implementation is CobaltBlue. CobaltBlue is an extension of SkyBlue [Sannella 93] and can solve simultaneous equations, multiple output and nonunique constraints incrementally by local propagation. However, the K machine design is general enough to accommodate other solvers. CIP languages combine imperative constructs such as object identity and class membership with ....

[Article contains additional citation context not shown here]

Michael Sannella. The SkyBlue constraint solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993.

.... DeltaBlue was the basis of the ThingLab II interactive user interface development environment [5, 10] SkyBlue is more general successor to DeltaBlue that satisfies cycles of methods by calling external solvers and supports multi output methods (methods that set multiple output variables) [7, 8]. The Multi Garnet package [9] uses the SkyBlue solver to add support for multi way constraints and constraint hierarchies to the Garnet user interface toolkit [6] As constraint solvers have been applied to larger problems it has become clear that there is a need for constraint network debugging ....

Michael Sannella. The SkyBlue constraint solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993.

....re executes each time of any of its variables (or the subparts of those variables) changes identity. The execution of a constructor might trigger additional constructors to execute. This process continues until the remaining constraints are primitive constraints, which are solved by SkyBlue [Sannella 93] the incremental local propagation solver used for Kaleidoscope 93 primitives. SkyBlue is a new algorithm that extends our earlier DeltaBlue algorithm [Freeman Benson et al. 90, Sannella et al. 93] to allow constraints with multiple outputs, and that provides better handling of cycles. ....

Michael Sannella. The SkyBlue Constraint Solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993.

....achieve the speed of DeltaBlue in this case. SkyBlue is currently being used as the constraint solver in several applications (see Section 6) SkyBlue implementations are available from the author. Detailed information on SkyBlue, including complete pseudocode, is available in a technical report [7]. 2 Method Graphs A SkyBlue constraint is represented by one or more methods. Each method is a procedure that reads the values of a subset of the constraint s variables (the method s input variables) and calculates values for the remaining variables (the method s output variables) that satisfy ....

....graph becomes very large. Section 5 presents several techniques used in the complete algorithm that significantly improve the efficiency of SkyBlue for large constraint graphs. More detailed information on SkyBlue, including complete pseudocode for the algorithm, is available in a technical report [7]. 4.1 Adding Constraints When a new constraint is added to the constraint graph it may be possible to alter the method graph to enforce it by selecting a method for the constraint, switching the selected methods of enforced constraints with the same or stronger strength, and possibly revoking one ....

Michael Sannella. The SkyBlue Constraint Solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, February 1993.

....a suitable set of values into a compound object. SkyBlue also handles cycles of constraints in a more graceful fashion than DeltaBlue, which will make it possible to link in specialized, more powerful constraint solvers as needed (see Section 1.6. 3) Additional details of SkyBlue are given in [San92] 1.5 Multi Garnet At the beginning of the project, we set a number of goals for Multi Garnet: 1. To extend Garnet to use the SkyBlue constraint solver. 2. To change Garnet s programming style as little as possible, so that existing Garnet programs could be adapted without major changes. In ....

Michael Sannella. The SkyBlue Constraint Solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, October 1992.

....[Sannella Borning 92] has used the locally predicate better comparator. This choice has been based primarily on pragmatic rather than aesthetic or theoretical grounds: the existence of efficient incremental algorithms DeltaBlue [Freeman Benson et al. 90] and a derivative algorithm SkyBlue [Sannella 92] for finding LPB solutions. For user interface applications, we do have extensive experience in the practical use of LPB [Sannella et al. 93] It also been used by a considerable number of researchers at other institutions as well. LPB has generally proved quite satisfactory. However, for ....

Michael Sannella. The SkyBlue Constraint Solver. Technical Report 92-07-02, Department of Computer Science and Engineering, University of Washington, December 1992.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC