Package opennlp.tools.ml

Class BeamSearch

java.lang.Object
opennlp.tools.ml.BeamSearch
All Implemented Interfaces:
AutoCloseable, opennlp.tools.ml.model.SequenceClassificationModel
@ThreadSafe public class BeamSearch extends Object implements opennlp.tools.ml.model.SequenceClassificationModel, AutoCloseable
Performs k-best search over a sequence.

This is based on the description in Ratnaparkhi (1998), PhD diss, Univ. of Pennsylvania.

This implementation is thread-safe. The contexts cache and probability buffer are maintained per-thread via ThreadLocal.

Note: In container environments with classloader isolation (e.g. Jakarta EE), ThreadLocal state may pin the classloader. Ensure instances do not outlive the application's lifecycle, or call ThreadLocal.remove() on pooled threads.

See Also:
  • Sequence
  • SequenceValidator
  • BeamSearchContextGenerator

  • Field Summary

    Fields
    Modifier and Type
    Field
    Description
    static final String
    BEAM_SIZE_PARAMETER
     

  • Constructor Summary

    Constructors
    Constructor
    Description
    BeamSearch(int size, opennlp.tools.ml.model.MaxentModel model)
    Initializes a BeamSearch instance.
    BeamSearch(int size, opennlp.tools.ml.model.MaxentModel model, int cacheSize)
    Initializes a BeamSearch instance with an optional per-thread contexts cache.

  • Method Summary

    Modifier and Type
    Method
    Description
    <T> opennlp.tools.util.Sequence
    bestSequence(T[] sequence, Object[] additionalContext, opennlp.tools.util.BeamSearchContextGenerator<T> cg, opennlp.tools.util.SequenceValidator<T> validator)
    <T> opennlp.tools.util.Sequence[]
    bestSequences(int numSequences, T[] sequence, Object[] additionalContext, double minSequenceScore, opennlp.tools.util.BeamSearchContextGenerator<T> cg, opennlp.tools.util.SequenceValidator<T> validator)
    <T> opennlp.tools.util.Sequence[]
    bestSequences(int numSequences, T[] sequence, Object[] additionalContext, opennlp.tools.util.BeamSearchContextGenerator<T> cg, opennlp.tools.util.SequenceValidator<T> validator)
    void
    close()
    Clears ThreadLocal state for the current thread only.
    String[]
    getOutcomes()
     

    Methods inherited from class java.lang.Object

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

  • Field Details

  • Constructor Details

    • BeamSearch

      public BeamSearch(int size, opennlp.tools.ml.model.MaxentModel model)
      Initializes a BeamSearch instance.
      Parameters:
      size - The size of the beam (k).
      model - The MaxentModel for assigning probabilities to the sequence outcomes.

    • BeamSearch

      public BeamSearch(int size, opennlp.tools.ml.model.MaxentModel model, int cacheSize)
      Initializes a BeamSearch instance with an optional per-thread contexts cache.
      Parameters:
      size - The size of the beam (k).
      model - The MaxentModel for assigning probabilities to the sequence outcomes.
      cacheSize - The capacity of the per-thread contexts cache. Use 0 to disable only that cache; per-thread score buffers are still allocated so evaluation stays thread-safe (see BeamSearch.CacheState).

  • Method Details

    • bestSequences

      public <T> opennlp.tools.util.Sequence[] bestSequences(int numSequences, T[] sequence, Object[] additionalContext, double minSequenceScore, opennlp.tools.util.BeamSearchContextGenerator<T> cg, opennlp.tools.util.SequenceValidator<T> validator)
      Specified by:
      bestSequences in interface opennlp.tools.ml.model.SequenceClassificationModel

    • bestSequences

      public <T> opennlp.tools.util.Sequence[] bestSequences(int numSequences, T[] sequence, Object[] additionalContext, opennlp.tools.util.BeamSearchContextGenerator<T> cg, opennlp.tools.util.SequenceValidator<T> validator)
      Specified by:
      bestSequences in interface opennlp.tools.ml.model.SequenceClassificationModel

    • bestSequence

      public <T> opennlp.tools.util.Sequence bestSequence(T[] sequence, Object[] additionalContext, opennlp.tools.util.BeamSearchContextGenerator<T> cg, opennlp.tools.util.SequenceValidator<T> validator)
      Specified by:
      bestSequence in interface opennlp.tools.ml.model.SequenceClassificationModel

    • getOutcomes

      public String[] getOutcomes()
      Specified by:
      getOutcomes in interface opennlp.tools.ml.model.SequenceClassificationModel

    • close

      public void close()
      Clears ThreadLocal state for the current thread only. This is intentionally not a "shut down the BeamSearch instance" operation: a single BeamSearch is typically shared across many pool threads, and each one owns an independent BeamSearch.CacheState entry.

      Typical usage patterns:

      • Worker thread returning to a pool: call close() (or wrap a single decode call in try-with-resources) on each pool thread that has touched the instance.
      • Application shutdown / classloader unload: close() on a single thread is not sufficient to release every per-thread slot — those die with their owning threads, or must be cleared on each thread before the application classloader is released.

      Same lifecycle contract as clearThreadLocalState() on the seven ME classes.

      Specified by:
      close in interface AutoCloseable