Package de.jungblut.nlp

Class MarkovChain

  • public final class MarkovChain
extends java.lang.Object
    Markov chain, that can "learn" the state transition probabilities by a given input and returns the probability for a given sequence of states.
    Author:
    thomas.jungblut

    • Method Summary

      All Methods Static Methods Instance Methods Concrete Methods 
      Modifier and Type Method Description
      double averageTransitionProbability​(int[] sequence)  
      int[] completeStateSequence​(com.google.common.base.Optional<java.util.Random> optionalRandom, int[] stateSequence, int... unsuppliedStateIndices)
      Completes the given state sequence by picking the best next state on the transition probabilities (so a transition with a high probability is picked more often).
      static MarkovChain create​(int numStates)
      Creates a new markov chain with the supplied number of states.
      static MarkovChain create​(int numStates, de.jungblut.math.DoubleMatrix mat)
      Creates a new markov chain with the supplied number of states and its predefined transition matrix.
      int getNumStates()  
      double getProbabilityForSequence​(int[] stateSequence)
      Calculates the probability that the given sequence occurs.
      de.jungblut.math.DoubleMatrix getTransitionProbabilities()  
      de.jungblut.math.DoubleVector getTransitionProbabilities​(int[] stateSequence)  
      void train​(java.util.stream.Stream<int[]> states)
      Trains the transition probabilities of the markov chain.

      • Methods inherited from class java.lang.Object

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

    • Method Detail

      • train

        public void train​(java.util.stream.Stream<int[]> states)
        Trains the transition probabilities of the markov chain.
        Each list element contains a set of states. The values of the element-states are nominal and should be lower than the number of provided states (each nominal will be a index, so it's from 0 to numStates-1). So each element can be arbitrary sized, because in markov chains we are considering the transition between two states, thus it will measure the occurrence of each following two state pairs. e.G. [ 1, 2, 3, 4 ] will measure the probabilities of [1,2],[2,3],[3,4].

      • getProbabilityForSequence

        public double getProbabilityForSequence​(int[] stateSequence)
        Calculates the probability that the given sequence occurs.
        Returns:
        value between 0d and 1d, where 1d is very likely that the sequence is happening.

      • averageTransitionProbability

        public double averageTransitionProbability​(int[] sequence)
        Returns:
        the average transition probability of the given sequence.

      • getTransitionProbabilities

        public de.jungblut.math.DoubleVector getTransitionProbabilities​(int[] stateSequence)
        Returns:
        the transition probabilities for the states.

      • completeStateSequence

        public int[] completeStateSequence​(com.google.common.base.Optional<java.util.Random> optionalRandom,
                                   int[] stateSequence,
                                   int... unsuppliedStateIndices)
        Completes the given state sequence by picking the best next state on the transition probabilities (so a transition with a high probability is picked more often). If the optional random is not provided, it picks the next state by the highest transition probability between the states (it predicts the next state based on the previous).

      • getTransitionProbabilities

        public de.jungblut.math.DoubleMatrix getTransitionProbabilities()
        Returns:
        the state transition probability matrix to export/serialize.

      • getNumStates

        public int getNumStates()
        Returns:
        how many states were defined?

      • create

        public static MarkovChain create​(int numStates)
        Creates a new markov chain with the supplied number of states.

      • create

        public static MarkovChain create​(int numStates,
                                 de.jungblut.math.DoubleMatrix mat)
        Creates a new markov chain with the supplied number of states and its predefined transition matrix.