ai.search.informed
Class RandomizedAStarSearcherForIntCostFn<S extends SearchStateForIntCostFn>

java.lang.Object
  ai.search.SearchEngine
      ai.search.SearchEngineForIntCostFn<S>
          ai.search.informed.BestFirstSearcherForIntCostFn<S>
              ai.search.informed.RandomizedAStarSearcherForIntCostFn<S>

public class RandomizedAStarSearcherForIntCostFn<S extends SearchStateForIntCostFn>
extends BestFirstSearcherForIntCostFn<S>

This class represents a search engine that generates a search space from a given initial state and guided by a heuristic function. The initial state and the heuristic function must be given to this search engine at construction time.

Now the search can be started using the function doSearch(). Successors of the initial state will then be explored in the order determined by the heuristic function plus some random element. The essential strategy is to expand those nodes first for which the current path cost plus the estimated distance to the closest goal state is smallest, i.e. nodes which appear to be on a shortest path to a goal state. The search will halt when a goal state has been expanded. The goal state and the path to it can then be inspected using the appropriate functions described in the SearchEngineForIntCostFn class.

Thus, a typical invocation would look something like this:

     SearchEngineForIntCostFn s = new RandomizedAStarSearcherForIntCostFn(
             initialState, heuristic, 1000, GRAPH);
     s.doSearch();
     if (s.foundGoalState()) { ... }
 

To continue the search call doSearch() again and the search will be continued as far as possible. If the test for repeated states is switched on, continuing the search will not result in the discovery of alternative paths to an already discovered goal state, but in the discovery of alternative goal states, if any. Furthermore, paths through a goal state will never be considered.

Author:

Gerhard Wickler

Nested Class Summary

static class

RandomizedAStarSearcherForIntCostFn.RandomizationType determines the type of randomization performed here; see setRandomizationType(ai.search.informed.RandomizedAStarSearcherForIntCostFn.RandomizationType)

Nested classes/interfaces inherited from class ai.search.SearchEngineForIntCostFn

SearchEngineForIntCostFn.SearchNodeForIntCostFn<S extends SearchStateForIntCostFn>

Nested classes/interfaces inherited from class ai.search.SearchEngine

SearchEngine.GraphType

Field Summary

protected java.util.Map<S,SearchEngineForIntCostFn.SearchNodeForIntCostFn<S>>	foundStates the table of all states generated this far for the test for repeated states
protected inf.util.RandomAccessIntPriorityQueue<SearchEngineForIntCostFn.SearchNodeForIntCostFn<S>>	searchQueue the search queue for unexplored (open) nodes

Fields inherited from class ai.search.informed.BestFirstSearcherForIntCostFn

dfTendency, h, hweight

Fields inherited from class ai.search.SearchEngineForIntCostFn

foundGoalNode, initialState

Fields inherited from class ai.search.SearchEngine

doInterrupt, doRepeatTest, searchLimit, searchStarted, traceStream, yieldFrequency

Constructor Summary

RandomizedAStarSearcherForIntCostFn(S state, IntCostHeuristic<S> heuristic, long limit, SearchEngine.GraphType structure)
This constructor creates a new RandomizedAStarSearcherForIntCostFn for the given initial state, heuristic, search limit, and type of search space.

Method Summary

boolean	continuable() This function tests whether the search may be continued.
void	doSearch() This function must be called to start or continue the search.
long	getNrOfExploredStates() This function returns the number of states that have been explored during this search.
long	getNrOfGeneratedStates() This function returns the number of states that have been generated during this search.
void	setRandomDistribution(inf.util.BiasedRandom bRandom) This function can be used to set the random distribution with which elements of the search queue are selected for expansion.
void	setRandomizationType(RandomizedAStarSearcherForIntCostFn.RandomizationType rt) This function can be used set the type of randomization performed by this search engine.

Methods inherited from class ai.search.informed.BestFirstSearcherForIntCostFn

setBreadthFirstTendency, setDepthFirstTendency, setHeuristicWeight

Methods inherited from class ai.search.SearchEngineForIntCostFn

foundGoalState, getGoalState, getInitialState, getSolutionDepth, getSolutionPath, getSolutionPathCost

Methods inherited from class ai.search.SearchEngine

clone, equals, getSearchLimit, hashCode, interrupt, printTrace, setSearchLimit, setTraceStream, setYieldFrequency, toString

Methods inherited from class java.lang.Object

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

Field Detail

searchQueue

protected inf.util.RandomAccessIntPriorityQueue<SearchEngineForIntCostFn.SearchNodeForIntCostFn<S extends SearchStateForIntCostFn>> searchQueue

the search queue for unexplored (open) nodes

foundStates

protected java.util.Map<S extends SearchStateForIntCostFn,SearchEngineForIntCostFn.SearchNodeForIntCostFn<S extends SearchStateForIntCostFn>> foundStates

the table of all states generated this far for the test for repeated states

Constructor Detail

RandomizedAStarSearcherForIntCostFn

public RandomizedAStarSearcherForIntCostFn(S state,
                                           IntCostHeuristic<S> heuristic,
                                           long limit,
                                           SearchEngine.GraphType structure)

This constructor creates a new RandomizedAStarSearcherForIntCostFn for the given initial state, heuristic, search limit, and type of search space. The search does not start immediately but waits for the function doSearch() to be called.

Parameters:

state - the initial state from which the search space is generated

heuristic - the heuristic function that guides the search

limit - the initial value for the search limit

structure - TREE or GRAPH, depending on what is to be be searched

Method Detail

doSearch

public void doSearch()

This function must be called to start or continue the search. It will only return when one of the following conditions becomes true:

• a goal state has been found: in this case SearchEngineForIntCostFn.foundGoalState() will return true, the search will be continuable (continuable() usually will return true), and SearchEngineForIntCostFn.getGoalState() will return a goal state, not null
• the search limit has been reached: in this case SearchEngineForIntCostFn.foundGoalState() will return false, the search will not be continuable (continuable() will return false), and SearchEngineForIntCostFn.getGoalState() will return null; to make the search continuable the search limit has to be increased using SearchEngine.setSearchLimit(long)
• the search space has been exhausted: in this case SearchEngineForIntCostFn.foundGoalState() will return false, the search will not be continuable (continuable() will return false), and SearchEngineForIntCostFn.getGoalState() will return null

Specified by:

doSearch in class SearchEngine

continuable

public boolean continuable()

This function tests whether the search may be continued. Usually this is the case if there are more search states to be explored and the search limit has not yet been exceeded.

Specified by:

continuable in class SearchEngine

Returns:

true if and only if there are more search states that can be explored

getNrOfExploredStates

public long getNrOfExploredStates()

This function returns the number of states that have been explored during this search. A state counts as explored when the goal test has been performed on the state and its successors have been generated.

Specified by:

getNrOfExploredStates in class SearchEngine

Returns:

the number of nodes that have been closed during this search

getNrOfGeneratedStates

public long getNrOfGeneratedStates()

This function returns the number of states that have been generated during this search. Note that these states may not all be different.

Specified by:

getNrOfGeneratedStates in class SearchEngine

Returns:

the number of states that have been generated during this search

setRandomizationType

public void setRandomizationType(RandomizedAStarSearcherForIntCostFn.RandomizationType rt)

This function can be used set the type of randomization performed by this search engine. The values have the following effects:

• NODE (default): randomization across all search nodes - any search node in the queue may be chosen according to the given biased random distribution; note: since search nodes with equal h-value are kept in a linear list, the access time may be linear in the length of this list
• HVALUE: randomization across first at h-value - only nodes that are first in the queue of all nodes with equal h-value will be considered

Parameters:

rt - the type of randomization to be performed during the search

setRandomDistribution

public void setRandomDistribution(inf.util.BiasedRandom bRandom)

This function can be used to set the random distribution with which elements of the search queue are selected for expansion. The default value is BiasedRandom.POLY8_0.

Parameters:

bRandom - the source of biased randomness