5.9 KiB
USING THE GECODE MODULE (#Gecode)
There are two ways to use the gecode interface from YAP. The original approach, designed by Denys Duchier, requires loading the library:
:- use_module(library(gecode)).
A second approach is closer to CLP(FD), and is described in:
- \ref Gecode_and_ClPbBFDbC
In what follows, we refer the reader to the~\cite{gecode} manual for the necessary background.
CREATING A SPACE
Space := space
CREATING VARIABLES
Unlike in Gecode, variable objects are not bound to a specific Space. Each one actually contains an index with which it is possible to access a Space-bound Gecode variable. Variables can be created using the following expressions:
IVar := intvar(Space,SPEC...) BVar := boolvar(Space) SVar := setvar(Space,SPEC...)
where SPEC... is the same as in Gecode. For creating lists of variables use the following variants:
IVars := intvars(Space,N,SPEC...) BVars := boolvars(Space,N,SPEC...) SVars := setvars(Space,N,SPEC...)
where N is the number of variables to create (just like for XXXVarArray in Gecode). Sometimes an IntSet is necessary:
ISet := intset([SPEC...])
where each SPEC is either an integer or a pair (I,J) of integers. An IntSet describes a set of ints by providing either intervals, or integers (which stand for an interval of themselves). It might be tempting to simply represent an IntSet as a list of specs, but this would be ambiguous with IntArgs which, here, are represented as lists of ints.
Space += keep(Var) Space += keep(Vars)
Variables can be marked as "kept". In this case, only such variables will be explicitly copied during search. This could bring substantial benefits in memory usage. Of course, in a solution, you can then only look at variables that have been "kept". If no variable is marked as "kept", then they are all kept. Thus marking variables as "kept" is purely an optimization.
CONSTRAINTS AND BRANCHINGS
all constraint and branching posting functions are available just like in Gecode. Wherever a XXXArgs or YYYSharedArray is expected, simply use a list. At present, there is no support for minimodel-like constraint posting. Constraints and branchings are added to a space using:
Space += CONSTRAINT
Space += BRANCHING
For example:
Space += rel(X,'IRT_EQ',Y)
arrays of variables are represented by lists of variables, and constants are represented by atoms with the same name as the Gecode constant (e.g. 'INT_VAR_SIZE_MIN').
SEARCHING FOR SOLUTIONS
SolSpace := search(Space)
This is a backtrackable predicate that enumerates all solution spaces (SolSpace). It may also take options:
SolSpace := search(Space,Options)
Options is a list whose elements maybe:
restart to select the Restart search engine threads=N to activate the parallel search engine and control the number of workers (see Gecode doc) c_d=N to set the commit distance for recomputation a_d=N to set the adaptive distance for recomputation
EXTRACTING INFO FROM A SOLUTION
An advantage of non Space-bound variables, is that you can use them both to post constraints in the original space AND to consult their values in solutions. Below are methods for looking up information about variables. Each of these methods can either take a variable as argument, or a list of variables, and returns resp. either a value, or a list of values:
Val := assigned(Space,X)
Val := min(Space,X)
Val := max(Space,X)
Val := med(Space,X)
Val := val(Space,X)
Val := size(Space,X)
Val := width(Space,X)
Val := regret_min(Space,X)
Val := regret_max(Space,X)
Val := glbSize(Space,V)
Val := lubSize(Space,V)
Val := unknownSize(Space,V)
Val := cardMin(Space,V)
Val := cardMax(Space,V)
Val := lubMin(Space,V)
Val := lubMax(Space,V)
Val := glbMin(Space,V)
Val := glbMax(Space,V)
Val := glb_ranges(Space,V)
Val := lub_ranges(Space,V)
Val := unknown_ranges(Space,V)
Val := glb_values(Space,V)
Val := lub_values(Space,V)
Val := unknown_values(Space,V)
DISJUNCTORS
Disjunctors provide support for disjunctions of clauses, where each clause is a conjunction of constraints:
C1 or C2 or ... or Cn
Each clause is executed "speculatively": this means it does not affect the main space. When a clause becomes failed, it is discarded. When only one clause remains, it is committed: this means that it now affects the main space.
Example:
Consider the problem where either X=Y=0 or X=Y+(1 or 2) for variable X and Y that take values in 0..3.
Space := space,
[X,Y] := intvars(Space,2,0,3),
First, we must create a disjunctor as a manager for our 2 clauses:
Disj := disjunctor(Space),
We can now create our first clause:
C1 := clause(Disj),
This clause wants to constrain X and Y to 0. However, since it must be executed "speculatively", it must operate on new variables X1 and Y1 that shadow X and Y:
[X1,Y1] := intvars(C1,2,0,3),
C1 += forward([X,Y],[X1,Y1]),
The forward(...) stipulation indicates which global variable is shadowed by which clause-local variable. Now we can post the speculative clause-local constraints for X=Y=0:
C1 += rel(X1,'IRT_EQ',0),
C1 += rel(Y1,'IRT_EQ',0),
We now create the second clause which uses X2 and Y2 to shadow X and Y:
C2 := clause(Disj),
[X2,Y2] := intvars(C2,2,0,3),
C2 += forward([X,Y],[X2,Y2]),
However, this clause also needs a clause-local variable Z2 taking values 1 or 2 in order to post the clause-local constraint X2=Y2+Z2:
Z2 := intvar(C2,1,2),
C2 += linear([-1,1,1],[X2,Y2,Z2],'IRT_EQ',0),
Finally, we can branch and search:
Space += branch([X,Y],'INT_VAR_SIZE_MIN','INT_VAL_MIN'),
SolSpace := search(Space),
and lookup values of variables in each solution:
[X_,Y_] := val(SolSpace,[X,Y]).