Source Code for Module jazzparser.utils.nltk.hmm

 1  """Tools to extend NLTK's implementation of HMMs. 
 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   
29  # The NLTK import has already been checked in our __init__.py 
30  from nltk.tag.hmm import HiddenMarkovModelTrainer, HiddenMarkovModelTagger 
31  from jazzparser.utils.nltk.storage import ObjectStorer 
32   
33 -class PicklableHmmTrainer(HiddenMarkovModelTrainer): 
34      """ 
35      We override HiddenMarkovModelTrainer to overcome the fact that it  
36      produces HMMs that can't be pickled. We only make supervised  
37      trained HMMs picklable at the moment. 
38       
39      This is quite a nasty hack to overcome the fact that NLTK HMMs  
40      can't be stored and also can't be pickled if constructed using the  
41      default trainer. However, this is not very stable, since someone  
42      could, for example, set some attribute of the model to be a Python  
43      lambda and pickling would once again fail. 
44       
45      """ 
46 -    def train_supervised(self, *args, **kwargs): 
47          """ 
48          If you set 'estimator' in the kwargs, make sure it's a top-level  
49          named function, not a lambda, or else you won't be able to pickle 
50          your HMM. 
51           
52          """ 
53          from jazzparser.utils.nltk.probability import mle_estimator 
54          estimator = kwargs.get('estimator') 
55          if estimator is None: 
56              estimator = mle_estimator 
57          kwargs['estimator'] = estimator 
58          return super(PicklableHmmTrainer, self).train_supervised(*args, **kwargs) 
59   
60 -class HiddenMarkovModelTaggerStorer(ObjectStorer): 
61      STORED_CLASS = HiddenMarkovModelTagger 
62       
63      @staticmethod 
64 -    def _object_to_dict(obj): 
65          from .storage import object_to_dict 
66          data = {} 
67          # The states and symbols are just dicts of strings, so we're  
68          #  fine to leave them as they are 
69          data['states'] = obj._states 
70          data['symbols'] = obj._symbols 
71          # The prior distribution is a FreqDist, which needs to be processed 
72          data['priors'] = object_to_dict(obj._priors) 
73          # The outputs and transitions are ConditionalProbDists 
74          data['outputs'] = object_to_dict(obj._outputs) 
75          data['transitions'] = object_to_dict(obj._transitions) 
76          return data 
77           
78      @staticmethod 
79 -    def _dict_to_object(dic): 
80          from .storage import dict_to_object 
81          states = dic['states'] 
82          symbols = dic['symbols'] 
83          priors = dict_to_object(dic['priors']) 
84          outputs = dict_to_object(dic['outputs']) 
85          transitions = dict_to_object(dic['transitions']) 
86          return HiddenMarkovModelTagger(symbols, states, transitions, outputs, priors) 
87