org.chocosolver.solver.search.loop.lns.neighbors

Class ExplainingCut

java.lang.Object

org.chocosolver.solver.search.loop.lns.neighbors.ExplainingCut

All Implemented Interfaces:

INeighbor

public class ExplainingCut
extends Object
implements INeighbor

a specific neighborhood for LNS based on the explanation of the cut imposed by a new solution.

This neighborhood is specific in the sense that it needs to compute explanation after a new solution has been found. Furthermore, the fixSomeVariables method creates and applies decisions, so that the explanation recorder can infer.
It works as follow: - on a solution: force the application of the cut together with the decision path which leads to the solution, explain the failure - then, on a call to fixSomeVariables, it selects randomly K decisions explaining the cut, and relax them from the decision path.

Unrelated decisions are never relaxed, the idea here is to work only on the decisions which lead to a failure.

Since:

03/07/13

TODO: fix some variables TODO: catch up case when the sub tree is closed and this imposes a fragment

Author:

Charles Prud'homme

Field Summary

Fields

Modifier and Type	Field and Description
protected boolean	forceCft Indicates if the cut has already been explained
protected List<IntDecision>	mDecisionPath A copy of the decision path which led to the solution
protected IExplanationEngine	mExplanationEngine The explanation engine that computes explanation on solution
protected Model	mModel Reference to the model
protected int	nbCall Number of times this neighbor is called
protected double	nbFixedVariables Number of decisions to fix in the fragment
protected BitSet	notFrozen Indicates which decisions are frozen in the current fragment
protected Random	random For randomness purpose
protected BitSet	related Indicates which decisions are related to the explanation in mDecisionPath
protected BitSet	unrelated Indicates which decisions are not related to the explanation in mDecisionPath

Constructor Summary

Constructors

Constructor and Description
ExplainingCut(Model aModel, int level, long seed) Create a neighborhood which analyses the explanation of the cut wrt decision path to focus on decisions more prone to not interact with the cut

Method Summary

Modifier and Type	Method and Description
protected void	_fixVar() Regarding the explanation, pick decisions to create a fragment
protected void	clonePath() Compute the initial fragment, ie set of decisions to keep.
protected void	explain() Force the failure, apply decisions to the last solution + cut => failure!
void	fixSomeVariables(DecisionPath decisionPath) Freezes some variables in order to have a fast computation.
protected void	increaseLimit() Increase the relaxing limit
void	init() Initialize this neighbor
boolean	isSearchComplete()
void	loadFromSolution(Solution solution) Load a solution and record it
void	recordSolution() Record values of decision variables to freeze some ones during the next LNS run
void	restrictLess() Use less restriction at the beginning of a LNS run in order to get better solutions Called when no solution was found during a LNS run (trapped into a local optimum)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

mExplanationEngine

protected IExplanationEngine mExplanationEngine

The explanation engine that computes explanation on solution

random

protected final Random random

For randomness purpose

mDecisionPath

protected List<IntDecision> mDecisionPath

A copy of the decision path which led to the solution

related

protected BitSet related

Indicates which decisions are related to the explanation in mDecisionPath

unrelated

protected BitSet unrelated

Indicates which decisions are not related to the explanation in mDecisionPath

notFrozen

protected BitSet notFrozen

Indicates which decisions are frozen in the current fragment

forceCft

protected boolean forceCft

Indicates if the cut has already been explained

nbFixedVariables

protected double nbFixedVariables

Number of decisions to fix in the fragment

nbCall

protected int nbCall

Number of times this neighbor is called

mModel

protected Model mModel

Reference to the model

Constructor Detail

ExplainingCut

public ExplainingCut(Model aModel,
                     int level,
                     long seed)

Create a neighborhood which analyses the explanation of the cut wrt decision path to focus on decisions more prone to not interact with the cut

Parameters:

aModel - the model

level - the relaxing factor

seed - a seed for randomness

Method Detail

init

public void init()

Description copied from interface: INeighbor

Initialize this neighbor

Specified by:

init in interface INeighbor

recordSolution

public void recordSolution()

Description copied from interface: INeighbor

Record values of decision variables to freeze some ones during the next LNS run

Specified by:

recordSolution in interface INeighbor

loadFromSolution

public void loadFromSolution(Solution solution)

Description copied from interface: INeighbor

Load a solution and record it

Specified by:

loadFromSolution in interface INeighbor

Parameters:

solution - a solution to record

fixSomeVariables

public void fixSomeVariables(DecisionPath decisionPath)

Description copied from interface: INeighbor

Freezes some variables in order to have a fast computation. The fixed variables are declared as decisions in the decision path.

Specified by:

fixSomeVariables in interface INeighbor

Parameters:

decisionPath - the decision path in which declaring variable to freeze

_fixVar

protected void _fixVar()

Regarding the explanation, pick decisions to create a fragment

restrictLess

public void restrictLess()

Description copied from interface: INeighbor

Use less restriction at the beginning of a LNS run in order to get better solutions Called when no solution was found during a LNS run (trapped into a local optimum)

Specified by:

restrictLess in interface INeighbor

isSearchComplete

public boolean isSearchComplete()

Specified by:

isSearchComplete in interface INeighbor

Returns:

true iff the search is in a complete mode (no fixed variable)

increaseLimit

protected void increaseLimit()

Increase the relaxing limit

clonePath

protected void clonePath()

Compute the initial fragment, ie set of decisions to keep.

explain

protected void explain()

Force the failure, apply decisions to the last solution + cut => failure!