Package ec

Class EvolutionState

java.lang.Object
ec.EvolutionState
All Implemented Interfaces:
Setup, Singleton, Serializable
Direct Known Subclasses:
SimpleEvolutionState, SteadyStateEvolutionState
public class EvolutionState extends Object implements Singleton
An EvolutionState object is a singleton object which holds the entire state of an evolutionary run. By serializing EvolutionState, the entire run can be checkpointed out to a file.

The EvolutionState instance is passed around in a lot of methods, so objects can read from the parameter database, pick random numbers, and write to the output facility.

EvolutionState is a unique object in that it calls its own setup(...) method, from run(...).

An EvolutionState object contains quite a few objects, including:

  • Objects you may safely manipulate during the multithreaded sections of a run:
    • MersenneTwisterFast random number generators (one for each evaluation or breeding thread -- use the thread number you were provided to determine which random number generator to use)
    • The ParameterDatabase
    • The Output facility for writing messages and logging
  • Singleton objects:
    • The Initializer.
    • The Finisher.
    • The Breeder.
    • The Evaluator.
    • The Statistics facility.
    • The Exchanger.
  • The current evolution state:
    • The generation number.
    • The population.
    • The maximal number of generations.
  • Auxillary read-only information:
    • The prefix to begin checkpoint file names with.
    • Whether to quit upon finding a perfect individual.
    • The number of breeding threads to spawn.
    • The number of evaluation threads to spawn.
  • A place to stash pointers to static variables so they'll get serialized:
    • Statics

Parameters

generations
int >= 1 or undefined		 (maximal number of generations to run. Either this or evaluations must be set, but not both.)
evaluations
int >= 1 or undefined		 (maximal number of evaluations to run (in subpopulation 0). Either this or generations must be set, but not both.)
checkpoint-modulo
int >= 1		 (how many generations should pass before we do a checkpoint? The definition of "generations" depends on the particular EvolutionState implementation you're using)
checkpoint
bool = true or false (default)		 (should we checkpoint?)
prefix
String		 (the prefix to prepend to checkpoint files -- see ec.util.Checkpoint)
checkpoint-directory
File (default is empty)		 (directory where the checkpoint files should be located)
quit-on-run-complete
bool = true or false (default)		 (do we prematurely quit the run when we find a perfect individual?)
init
classname, inherits and != ec.Initializer		 (the class for initializer)
finish
classname, inherits and != ec.Finisher		 (the class for finisher)
breed
classname, inherits and != ec.Breeder		 (the class for breeder)
eval
classname, inherits and != ec.Evaluator		 (the class for evaluator)
stat
classname, inherits or = ec.Statistics		 (the class for statistics)
exch
classname, inherits and != ec.Exchanger		 (the class for exchanger)

Parameter bases

init initializer
finish finisher
breed breeder
eval evaluator
stat statistics
exch exchanger
See Also:

  • Field Details

    • parameters

      public ParameterDatabase parameters
      The parameter database (threadsafe). Parameter objects are also threadsafe. Nonetheless, you should generally try to treat this database as read-only.

    • random

      public MersenneTwisterFast[] random
      An array of random number generators, indexed by the thread number you were given (or, if you're not in a multithreaded area, use 0). These generators are not threadsafe in and of themselves, but if you only use the random number generator assigned to your thread, as was intended, then you get random numbers in a threadsafe way. These generators must each have a different seed, of course.

    • data

      public HashMap[] data
      An array of HashMaps, indexed by the thread number you were given (or, if you're not in a multithreaded area, use 0). This allows you to store per-thread specialized information (typically keyed with a string).

    • output

      public Output output
      The output and logging facility (threadsafe). Keep in mind that output in Java is expensive.

    • breedthreads

      public int breedthreads
      The requested number of threads to be used in breeding, excepting perhaps a "parent" thread which gathers the other threads. If breedthreads = 1, then the system should not be multithreaded during breeding. Don't modify this during a run.

    • evalthreads

      public int evalthreads
      The requested number of threads to be used in evaluation, excepting perhaps a "parent" thread which gathers the other threads. If evalthreads = 1, then the system should not be multithreaded during evaluation. Don't modify this during a run.

    • checkpoint

      public boolean checkpoint
      Should we checkpoint at all?

    • checkpointDirectory

      public File checkpointDirectory
      The requested directory where checkpoints should be located. This must be a directory, not a file. You probably shouldn't modify this during a run.

    • checkpointPrefix

      public String checkpointPrefix
      The requested prefix to start checkpoint filenames, not including a following period. You probably shouldn't modify this during a run.

    • checkpointModulo

      public int checkpointModulo
      The requested number of generations that should pass before we write out a checkpoint file.

    • randomSeedOffset

      public int randomSeedOffset
      An amount to add to each random number generator seed to "offset" it -- often this is simply the job number. If you are using more random number generators internally than the ones initially created for you in the EvolutionState, you might want to create them with the seed value of seedParameter+randomSeedOffset. At present the only such class creating additional generators is ec.eval.MasterProblem.

    • quitOnRunComplete

      public boolean quitOnRunComplete
      Whether or not the system should prematurely quit when Evaluator returns true for runComplete(...) (that is, when the system found an ideal individual.

    • job

      public Object[] job
      Current job iteration variables, set by Evolve. The default version simply sets this to a single Object[1] containing the current job iteration number as an Integer (for a single job, it's 0). You probably should not modify this inside an evolutionary run.

    • runtimeArguments

      public String[] runtimeArguments
      The original runtime arguments passed to the Java process. You probably should not modify this inside an evolutionary run.

    • UNDEFINED

      public static final int UNDEFINED
      See Also:

    • generation

      public int generation
      The current generation of the population in the run. For non-generational approaches, this probably should represent some kind of incrementing value, perhaps the number of individuals evaluated so far. You probably shouldn't modify this.

    • evaluations

      public int evaluations
      The current number of evaluations which have transpired so far in the run. This is only updated on a generational boundary.

    • numGenerations

      public int numGenerations
      The number of generations the evolutionary computation system will run until it ends, or UNDEFINED

    • numEvaluations

      public long numEvaluations
      The number of evaluations the evolutionary computation system will run until it ends (up to the next generation boundary), or UNDEFINED

    • population

      public Population population
      The current population. This is not a singleton object, and may be replaced after every generation in a generational approach. You should only access this in a read-only fashion.

    • initializer

      public Initializer initializer
      The population initializer, a singleton object. You should only access this in a read-only fashion.

    • finisher

      public Finisher finisher
      The population finisher, a singleton object. You should only access this in a read-only fashion.

    • breeder

      public Breeder breeder
      The population breeder, a singleton object. You should only access this in a read-only fashion.

    • evaluator

      public Evaluator evaluator
      The population evaluator, a singleton object. You should only access this in a read-only fashion.

    • statistics

      public Statistics statistics
      The population statistics, a singleton object. You should generally only access this in a read-only fashion.

    • exchanger

      public Exchanger exchanger
      The population exchanger, a singleton object. You should only access this in a read-only fashion.

    • innovationNumber

      public long innovationNumber
      Global birthday tracker number for genes in representations such as NEAT. Accessed and modified during run time

    • C_STARTED_FRESH

      public static final int C_STARTED_FRESH
      "The population has started fresh (not from a checkpoint)."
      See Also:

    • C_STARTED_FROM_CHECKPOINT

      public static final int C_STARTED_FROM_CHECKPOINT
      "The population started from a checkpoint."
      See Also:

    • R_SUCCESS

      public static final int R_SUCCESS
      "The evolution run has quit, finding a perfect individual."
      See Also:

    • R_FAILURE

      public static final int R_FAILURE
      "The evolution run has quit, failing to find a perfect individual."
      See Also:

    • R_NOTDONE

      public static final int R_NOTDONE
      "The evolution run has not quit
      See Also:

    • P_INITIALIZER

      public static final String P_INITIALIZER
      See Also:

    • P_FINISHER

      public static final String P_FINISHER
      See Also:

    • P_BREEDER

      public static final String P_BREEDER
      See Also:

    • P_EVALUATOR

      public static final String P_EVALUATOR
      See Also:

    • P_STATISTICS

      public static final String P_STATISTICS
      See Also:

    • P_EXCHANGER

      public static final String P_EXCHANGER
      See Also:

    • P_GENERATIONS

      public static final String P_GENERATIONS
      See Also:

    • P_EVALUATIONS

      public static final String P_EVALUATIONS
      See Also:

    • P_QUITONRUNCOMPLETE

      public static final String P_QUITONRUNCOMPLETE
      See Also:

    • P_CHECKPOINTPREFIX

      public static final String P_CHECKPOINTPREFIX
      See Also:

    • P_CHECKPOINTMODULO

      public static final String P_CHECKPOINTMODULO
      See Also:

    • P_CHECKPOINTDIRECTORY

      public static final String P_CHECKPOINTDIRECTORY
      See Also:

    • P_CHECKPOINT

      public static final String P_CHECKPOINT
      See Also:

    • P_INNOVATIONNUMBER

      public static final String P_INNOVATIONNUMBER
      See Also:

  • Constructor Details

    • EvolutionState

      public EvolutionState()
      This will be called to create your evolution state; immediately after the constructor is called, the parameters, random, and output fields will be set for you. The constructor probably won't be called ever if restoring (deserializing) from a checkpoint.

  • Method Details

    • setup

      public void setup(EvolutionState state, Parameter base)
      Unlike for other setup() methods, ignore the base; it will always be null.
      Specified by:
      setup in interface Setup
      See Also:

    • resetFromCheckpoint

      public void resetFromCheckpoint() throws IOException
      This method is called after a checkpoint is restored from but before the run starts up again. You might use this to set up file pointers that were lost, etc.
      Throws:
      IOException

    • finish

      public void finish(int result)

    • startFromCheckpoint

      public void startFromCheckpoint()

    • startFresh

      public void startFresh()

    • evolve

      public int evolve() throws InternalError
      Throws:
      InternalError

    • incrementEvaluations

      public void incrementEvaluations(int val)

    • run

      public void run(int condition)
      Starts the run. condition indicates whether or not the run was restarted from a checkpoint (C_STARTED_FRESH vs C_STARTED_FROM_CHECKPOINT). At the point that run(...) has been called, the parameter database has already been set up, as have the random number generators, the number of threads, and the Output facility. This method should call this.setup(...) to set up the EvolutionState object if condition equals C_STARTED_FRESH.