Source Code for Module jptests.utils.nltk.ngram.baumwelch

  1  """Unit tests for jazzparser.utils.nltk.ngram.baumwelch 
  2   
  3  """ 
  4  """ 
  5  ============================== License ======================================== 
  6   Copyright (C) 2008, 2010-12 University of Edinburgh, Mark Granroth-Wilding 
  7    
  8   This file is part of The Jazz Parser. 
  9    
 10   The Jazz Parser is free software: you can redistribute it and/or modify 
 11   it under the terms of the GNU General Public License as published by 
 12   the Free Software Foundation, either version 3 of the License, or 
 13   (at your option) any later version. 
 14    
 15   The Jazz Parser is distributed in the hope that it will be useful, 
 16   but WITHOUT ANY WARRANTY; without even the implied warranty of 
 17   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the 
 18   GNU General Public License for more details. 
 19    
 20   You should have received a copy of the GNU General Public License 
 21   along with The Jazz Parser.  If not, see <http://www.gnu.org/licenses/>. 
 22   
 23  ============================ End license ====================================== 
 24   
 25  """ 
 26  __author__ = "Mark Granroth-Wilding <mark.granroth-wilding@ed.ac.uk>"  
 27   
 28  import unittest 
 29  from jazzparser.utils.nltk.ngram import DictionaryHmmModel 
 30  from jazzparser.utils.nltk.ngram.baumwelch import BaumWelchTrainer 
 31  from nltk.probability import DictionaryConditionalProbDist, DictionaryProbDist 
 32   
 33 -class TestTrain(unittest.TestCase): 
 34 -    def setUp(self): 
 35          """ 
 36          Prepare some training data. 
 37           
 38          """ 
 39          self.TRAINING_DATA = [ 
 40              [0, 5, 5, 7, 6, 7, 8, 5, 2, 0, 3, 1, 2, 2, 2, 9, 9, 8, 0, 8, 9, 9, 1, 3, 2, 2, 1], 
 41              [3, 3, 1, 2, 1, 1, 0, 1, 9, 7, 8, 7, 7, 9, 0], 
 42              [7, 8, 6, 9, 8, 9, 9, 1, 3, 0, 1, 3, 0, 1, 1, 0, 5, 7, 5, 4, 5, 7, 7] 
 43          ] 
 44          self.TEST_DATA = [0, 1, 2, 3, 4, 3, 5, 6, 7, 8, 8, 9, 7, 7, 0, 0, 1] 
 45           
 46          ems = list(range(10)) 
 47          states = ['H', 'M', 'L'] 
 48          # Construct some initial distributions 
 49          # Emission 
 50          hprobs = { 
 51              0:0.0, 1:0.0, 2:0.0, 3:0.0, 4:0.0, 5:0.0, 6:0.1, 7:0.3,  
 52              8:0.3, 9:0.3 } 
 53          mprobs = { 
 54              0:0.0, 1:0.0, 2:0.0, 3:0.1, 4:0.3, 5:0.3, 6:0.3, 7:0.0,  
 55              8:0.0, 9:0.0 } 
 56          lprobs = { 
 57              0:0.2, 1:0.2, 2:0.2, 3:0.2, 4:0.2, 5:0.0, 6:0.0, 7:0.0, 
 58              8:0.0, 9:0.0 } 
 59          conddist = { 
 60              'H' : DictionaryProbDist(hprobs), 
 61              'M' : DictionaryProbDist(mprobs), 
 62              'L' : DictionaryProbDist(lprobs), 
 63          } 
 64          emdist = DictionaryConditionalProbDist(conddist) 
 65          # And transition 
 66          conddist = {} 
 67          for first in states+[None]: 
 68              probs = dict([(second, 1.0/3) for second in states+[None]]) 
 69              dist = DictionaryProbDist(probs) 
 70              conddist[(first,)] = dist 
 71          transdist = DictionaryConditionalProbDist(conddist) 
 72           
 73          # Initialize an ngram model with these distributions 
 74          self.model = DictionaryHmmModel(transdist, emdist, states, ems) 
 75       
 76 -    def test_init_model(self): 
 77          """ 
 78          Check that the initialized model is doing something sensible. 
 79           
 80          """ 
 81          # We don't check these, just that they're not generating errors:  
 82          #  that would really confuse the rest of it 
 83          self.model.emission_probability(2, 'H') 
 84          self.model.emission_probability(6, 'H') 
 85          self.model.transition_probability('H','L') 
 86          self.model.transition_probability('H','H') 
 87       
 88 -    def test_init_decode(self): 
 89          """ 
 90          Try running the viterbi decoder using the initial model. 
 91           
 92          """ 
 93          self.model.viterbi_decode(self.TEST_DATA) 
 94           
 95 -    def test_baum_welch(self): 
 96          """ 
 97          Runs the Baum Welch trainer using the training data. 
 98           
 99          """ 
100          options = BaumWelchTrainer.process_option_dict({}) 
101          trainer = BaumWelchTrainer(self.model, options) 
102          # Train the model with Baum Welch 
103          trainer.train(self.TRAINING_DATA) 
104          model = trainer.model 
105          # Try decoding using the trained model to check it still works 
106          model.viterbi_decode(self.TEST_DATA) 
107       
108 -    def test_baum_welch_mp(self): 
109          """ 
110          Does the same as L{test_baum_welch}, but uses multiprocessing. 
111           
112          """ 
113          options = BaumWelchTrainer.process_option_dict({'trainprocs':-1}) 
114          trainer = BaumWelchTrainer(self.model, options) 
115          # Train the model with Baum Welch 
116          trainer.train(self.TRAINING_DATA) 
117