Module org.chocosolver

Package org.chocosolver.solver.constraints.nary.automata

Class PropMultiCostRegular

java.lang.Object

org.chocosolver.solver.constraints.Propagator<IntVar>

org.chocosolver.solver.constraints.nary.automata.PropMultiCostRegular

All Implemented Interfaces:

Comparable<Propagator>, ICause, Identity

public final class PropMultiCostRegular
extends Propagator<IntVar>

Created by IntelliJ IDEA. User: julien S Date: Jul 16, 2008 Time: 5:56:50 PM

Multi-Cost-Regular is a propagator for the constraint ensuring that, given : an automaton Pi; a sequence of domain variables X; a set of bound variables Z; a assignment cost matrix for each bound variable C;

The word formed by the sequence of assigned variables is accepted by Pi; for each z^k in Z, sum_i(C_i(x_k)k) = z^k

AC is NP hard for such a constraint. The propagation is based on a Lagrangian Relaxation approach of the underlying Resource constrained shortest/longest path problems

Field Summary

Fields

Modifier and Type	Field	Description
double	_MCR_DECIMAL_PREC
gnu.trove.map.hash.TObjectIntHashMap<IntVar>	map	Map to retrieve rapidly the index of a given variable.
static int	MAXBOUNDITER	Maximum number of iteration during a bound computation
static int	MAXNONIMPROVEITER	Maximum number of non improving iteration while computing a bound
static double	RO	Lagrangian multiplier decreasing factor
static double	U0	Constant coefficient of the lagrangian relaxation

Fields inherited from class org.chocosolver.solver.constraints.Propagator

ACTIVE, constraint, DEFAULT_EXPL, model, operations, OUTPUT_DEFAULT_EXPL, priority, reactToFineEvt, state, vars

Constructor Summary

Constructors

Constructor	Description
PropMultiCostRegular(IntVar[] variables, IntVar[] costvariables, ICostAutomaton cauto, double precision)	Constructs a multi-cost-regular propagator

Method Summary

Modifier and Type	Method	Description
boolean	check()	Necessary condition : checks whether the constraint is violted or not
boolean	check(int[] word)
protected void	checkWorld()
void	computeSharpBounds()	Iteratively compute upper and lower bound for the underlying RCSPP
protected void	delayedGraphUpdate()	Updates the graphs w.r.t. the caught event during event-based propagation
protected void	filterDown(double realsp)	Filters w.r.t. a given lower bound.
protected void	filterUp(double reallp)	Filters w.r.t. a given upper bound.
void	forcePathRecomputation()
StoredDirectedMultiGraph	getGraph()
double[]	getInstantiatedLayerCosts(int layer)
int[]	getMinMaxPathCostForAssignment(int col, int val, int... resources)
int	getMinPathCost(int... resources)
int	getMinPathCostForAssignment(int col, int val, int... resources)
int	getPropagationConditions(int vIdx)	Returns the specific mask indicating the variable events on which this Propagator object can react. A mask is a bitwise OR operations over IEventType this can react on.
int	getRegret(int layer, int value, int... resources)
protected void	initialize()	Build internal structure of the propagator, if necessary
ESat	isEntailed()	Check wether this is entailed according to the current state of its internal structure.
boolean	isGraphConsistent()
boolean	isSatisfied()
boolean	needPropagation()
protected boolean	prefilter()	Performs cost based filtering w.r.t. each cost dimension.
void	propagate(int evtmask)	Call the main filtering algorithm to apply to the Domain of the Variable objects.
void	propagate(int varIdx, int mask)	Incremental filtering algorithm defined within the Propagator, called whenever the variable of index idxVarInProp has changed.
void	rebuildCostRegInfo()
protected void	updateLowerBound()	Performs a lagrangian relaxation to compute a new Lower bound of the underlying RCSPP problem Each built subproblem is a shortest path one can use to perform cost based filtering
protected void	updateUpperBound()	Performs a lagrangian relaxation to compute a new Upper bound of the underlying RCSPP problem Each built subproblem is a longest path one can use to perform cost based filtering

Methods inherited from class org.chocosolver.solver.constraints.Propagator

addVariable, arity, compareTo, defaultExplain, doFinePropagation, doFlush, doSchedule, doScheduleEvent, dynPriority, equals, explain, fails, forcePropagate, forcePropagationOnBacktrack, forEachIntVar, getConstraint, getId, getModel, getNbVars, getPosition, getPriority, getVar, getVars, getVIndice, getVIndices, hashCode, isActive, isCompletelyInstantiated, isPassive, isReified, isReifiedAndSilent, isScheduled, isStateLess, linkVariables, reactToFineEvent, reifiedWith, setActive, setActive0, setPassive, setPosition, setReifiedSilent, setReifiedTrue, setVIndices, toString, unlinkVariables, unschedule

Methods inherited from class java.lang.Object

clone, finalize, getClass, notify, notifyAll, wait, wait, wait

Field Detail

MAXBOUNDITER

public static final int MAXBOUNDITER

Maximum number of iteration during a bound computation

See Also:

Constant Field Values

MAXNONIMPROVEITER

public static final int MAXNONIMPROVEITER

Maximum number of non improving iteration while computing a bound

See Also:

Constant Field Values

U0

public static final double U0

Constant coefficient of the lagrangian relaxation

See Also:

Constant Field Values

RO

public static final double RO

Lagrangian multiplier decreasing factor

See Also:

Constant Field Values

map

public final gnu.trove.map.hash.TObjectIntHashMap<IntVar> map

Map to retrieve rapidly the index of a given variable.

_MCR_DECIMAL_PREC

public final double _MCR_DECIMAL_PREC

Constructor Detail

PropMultiCostRegular

public PropMultiCostRegular(IntVar[] variables,
                            IntVar[] costvariables,
                            ICostAutomaton cauto,
                            double precision)

Constructs a multi-cost-regular propagator

Parameters:

variables - decision variables

costvariables - cost variables

cauto - finite automaton with costs

Method Detail

getPropagationConditions

public int getPropagationConditions(int vIdx)

Description copied from class: Propagator

Returns the specific mask indicating the variable events on which this Propagator object can react.
A mask is a bitwise OR operations over IEventType this can react on. For example, consider a propagator that can deduce filtering based on the lower bound of the integer variable X. Then, for this variable, the mask should be equal to :

     int mask = IntEventType.INCLOW.getMask() | IntEventType.INSTANTIATE.getMask();
 

or, in a more convenient way:

     int mask = IntEvtType.combine(IntEventType.INCLOW,IntEventType.INSTANTIATE);
 

That indicates the following behavior:

1. if X is instantiated, this propagator will be executed,
2. if the lower bound of X is modified, this propagator will be executed,
3. if the lower bound of X is removed, the event is promoted from REMOVE to INCLOW and this propagator will NOT be executed,
4. otherwise, this propagator will NOT be executed

Some combinations are valid. For example, a propagator which reacts on REMOVE and INSTANTIATE should also declare INCLOW and DECUPP as conditions. Indeed INCLOW (resp. DECUPP), for efficiency purpose, removing the lower bound (resp. upper bound) of an integer variable will automatically be promoted into INCLOW (resp. DECUPP). So, ignoring INCLOW and/or DECUPP in that case may result in a lack of filtering. The same goes with events of other variable types, but most of the time, there are only few combinations. Reacts to any kind of event by default. Alternatively, this method can return IntEventType.VOID which states that this propagator should not be aware of modifications applied to the variable in position vIdx.

Overrides:

getPropagationConditions in class Propagator<IntVar>

Parameters:

vIdx - index of the variable within the propagator

Returns:

an int composed of REMOVE and/or INSTANTIATE and/or DECUPP and/or INCLOW

initialize

protected void initialize()
                   throws ContradictionException

Build internal structure of the propagator, if necessary

Throws:

ContradictionException - if initialisation encounters a contradiction

propagate

public void propagate(int evtmask)
               throws ContradictionException

Description copied from class: Propagator

Call the main filtering algorithm to apply to the Domain of the Variable objects. It considers the current state of this objects to remove some values from domains and/or instantiate some variables. Calling this method is done from 2 (and only 2) steps:
- at the initial propagation step,
- when involved in a reified constraint.
It should initialized the internal data structure and apply filtering algorithm from scratch.

Specified by:

propagate in class Propagator<IntVar>

Parameters:

evtmask - type of propagation event this must consider.

Throws:

ContradictionException - when a contradiction occurs, like domain wipe out or other incoherencies.

propagate

public void propagate(int varIdx,
                      int mask)
               throws ContradictionException

Description copied from class: Propagator

Incremental filtering algorithm defined within the Propagator, called whenever the variable of index idxVarInProp has changed. This method calls a CUSTOM_PROPAGATION (coarse-grained) by default.

This method should be overridden if the argument reactToFineEvt is set to true in the constructor. Otherwise, it executes propagate(PropagatorEventType.CUSTOM_PROPAGATION.getStrengthenedMask());

Overrides:

propagate in class Propagator<IntVar>

Parameters:

varIdx - index of the variable var in this

mask - type of event

Throws:

ContradictionException - if a contradiction occurs

updateUpperBound

protected void updateUpperBound()
                         throws ContradictionException

Performs a lagrangian relaxation to compute a new Upper bound of the underlying RCSPP problem Each built subproblem is a longest path one can use to perform cost based filtering

Throws:

ContradictionException - if a domain becomes empty

updateLowerBound

protected void updateLowerBound()
                         throws ContradictionException

Performs a lagrangian relaxation to compute a new Lower bound of the underlying RCSPP problem Each built subproblem is a shortest path one can use to perform cost based filtering

Throws:

ContradictionException - if a domain becomes empty

prefilter

protected boolean prefilter()
                     throws ContradictionException

Performs cost based filtering w.r.t. each cost dimension.

Throws:

ContradictionException - if a domain is emptied

filterDown

protected void filterDown(double realsp)
                   throws ContradictionException

Filters w.r.t. a given lower bound.

Parameters:

realsp - a given lower bound

Throws:

ContradictionException - if the cost variable domain is emptied

filterUp

protected void filterUp(double reallp)
                 throws ContradictionException

Filters w.r.t. a given upper bound.

Parameters:

reallp - a given upper bound

Throws:

ContradictionException - if the cost variable domain is emptied

checkWorld

protected void checkWorld()
                   throws ContradictionException

Throws:

ContradictionException

delayedGraphUpdate

protected void delayedGraphUpdate()
                           throws ContradictionException

Updates the graphs w.r.t. the caught event during event-based propagation

Throws:

ContradictionException - if removing an edge causes a domain to be emptied

computeSharpBounds

public void computeSharpBounds()
                        throws ContradictionException

Iteratively compute upper and lower bound for the underlying RCSPP

Throws:

ContradictionException - if a domain gets empty

rebuildCostRegInfo

public void rebuildCostRegInfo()
                        throws ContradictionException

Throws:

ContradictionException

needPropagation

public final boolean needPropagation()

isGraphConsistent

public boolean isGraphConsistent()

getGraph

public final StoredDirectedMultiGraph getGraph()

getRegret

public final int getRegret(int layer,
                           int value,
                           int... resources)

isEntailed

public ESat isEntailed()

Description copied from class: Propagator

Check wether this is entailed according to the current state of its internal structure. At least, should check the satisfaction of this (when all is instantiated).

Specified by:

isEntailed in class Propagator<IntVar>

Returns:

ESat.TRUE if entailed, ESat.FALSE if not entailed, ESat.UNDEFINED if unknown

isSatisfied

public boolean isSatisfied()

check

public boolean check(int[] word)

check

public boolean check()

Necessary condition : checks whether the constraint is violted or not

Returns:

true if the constraint is not violated

getMinPathCostForAssignment

public int getMinPathCostForAssignment(int col,
                                       int val,
                                       int... resources)

getMinMaxPathCostForAssignment

public int[] getMinMaxPathCostForAssignment(int col,
                                            int val,
                                            int... resources)

getMinPathCost

public int getMinPathCost(int... resources)

getInstantiatedLayerCosts

public double[] getInstantiatedLayerCosts(int layer)

forcePathRecomputation

public void forcePathRecomputation()
                            throws ContradictionException

Throws:

ContradictionException