ViterbiAlgorithm_HMMModel

##Author: Panther (Pavan Yadiki)

• Version: 1.0
• Date of release: March 20, 2015
• Development Environment:
  • Language- Python (2.7.3)
  • Operating System - Windows 8.1

######This project is developed during the Machine Learning course study at University of Texas at Dallas
####This project implements a Viterbi algorithm for a given HMM model
####To know more about Viterbi Algorithm, please read http://en.wikipedia.org/wiki/Viterbi_algorithm

• Input Format:

  • A text file that provides details of the HMM Model
    • Number of states in HMM - N
    • Initial probabilities of each state - N values separated by spaces
    • Transition probabilities (N*N values) Separated by spaces
    • Number of observations - M
    • Observation values - M values separated by spaces
    • Observation Probability Matrix -(N*M values)
  • A text file that contains the Test data
    • Each observation sequence in one line
  • Sample Input to HMM model is available with title, Input_Model.txt
  • Sample Input to Test data is available with title, Input_Test.txt

• Output:

  • Possible most likely paths for each of the given sequence

• Project descriptions file is titled ProjectDescription.pdf

• How to run?

  • Clone the project
  • python Viterbi.py path_to_model_data path_to_test_data

• Implementation Design Details:

  • ViterbiState.py - Contains a class 'State' which reflects the states of a HMM model. State Class contains the following:
    • Transition probability matrix - Defines the prob of moving to other states of a HMM model
    • Observation Probability matrix - Defines the prob of observing 'o' in a given state
    • Timestamp - A subclass that is evaluated during the evaluation phase.
      • Contains the value of delta of a given state and lists the state it has been reached from
        • Methods to return various parameters of a given state
    • Two static variables
      • number_of_states = Holds the number of states of this HMM mode. A list.
      • number_of_observations = Holds the number of observations possible in this HMM model. A dictionary object
    • Two static methods
      • set_observation_values() - Sets the unique observation values of this HMM model
      • get_observation_values() - Returns the observation values of this HMM model
        *Viterbi.py - Contains the driver program to build a HMM model for the provided input

• For any questions, please send an email to [email protected]