Source Code for Module nltk.classify.tadm

  1  # Natural Language Toolkit: Interface to TADM Classifier 
  2  # 
  3  # Copyright (C) 2001-2011 NLTK Project  
  4  # Author: Joseph Frazee <jfrazee@mail.utexas.edu> 
  5  # URL: <http://www.nltk.org/> 
  6  # For license information, see LICENSE.TXT 
  7   
  8  import sys 
  9  import subprocess 
 10   
 11  from nltk.internals import find_binary 
 12  try: 
 13      import numpy 
 14  except ImportError: 
 15      numpy = None 
 16   
 17  _tadm_bin = None 
 18 -def config_tadm(bin=None): 
 19      global _tadm_bin 
 20      _tadm_bin = find_binary( 
 21          'tadm', bin, 
 22          env_vars=['TADM_DIR'], 
 23          binary_names=['tadm'], 
 24          url='http://tadm.sf.net') 
 25           
 26 -def write_tadm_file(train_toks, encoding, stream): 
 27      """ 
 28      Generate an input file for C{tadm} based on the given corpus of 
 29      classified tokens. 
 30   
 31      @type train_toks: C{list} of C{tuples} of (C{dict}, C{str}) 
 32      @param train_toks: Training data, represented as a list of 
 33          pairs, the first member of which is a feature dictionary, 
 34          and the second of which is a classification label. 
 35   
 36      @type encoding: L{TadmEventMaxentFeatureEncoding} 
 37      @param encoding: A feature encoding, used to convert featuresets 
 38          into feature vectors. 
 39   
 40      @type stream: C{stream} 
 41      @param stream: The stream to which the C{tadm} input file should be 
 42          written. 
 43      """ 
 44      # See the following for a file format description: 
 45      # 
 46      # http://sf.net/forum/forum.php?thread_id=1391502&forum_id=473054 
 47      # http://sf.net/forum/forum.php?thread_id=1675097&forum_id=473054 
 48      labels = encoding.labels() 
 49      for featureset, label in train_toks: 
 50          stream.write('%d\n' % len(labels)) 
 51          for known_label in labels: 
 52              v = encoding.encode(featureset, known_label) 
 53              stream.write('%d %d %s\n' % (int(label == known_label), len(v),  
 54                  ' '.join('%d %d' % u for u in v))) 
 55   
 56 -def parse_tadm_weights(paramfile): 
 57      """ 
 58      Given the stdout output generated by C{tadm} when training a 
 59      model, return a C{numpy} array containing the corresponding weight 
 60      vector. 
 61      """ 
 62      weights = [] 
 63      for line in paramfile: 
 64          weights.append(float(line.strip())) 
 65      return numpy.array(weights, 'd')    
 66   
 67 -def call_tadm(args): 
 68      """ 
 69      Call the C{tadm} binary with the given arguments. 
 70      """ 
 71      if isinstance(args, basestring): 
 72          raise TypeError('args should be a list of strings') 
 73      if _tadm_bin is None: 
 74          config_tadm() 
 75   
 76      # Call tadm via a subprocess 
 77      cmd = [_tadm_bin] + args 
 78      p = subprocess.Popen(cmd, stdout=sys.stdout) 
 79      (stdout, stderr) = p.communicate() 
 80   
 81      # Check the return code. 
 82      if p.returncode != 0: 
 83          print 
 84          print stderr 
 85          raise OSError('tadm command failed!') 
 86   
 87 -def names_demo(): 
 88      from nltk.classify.util import names_demo 
 89      from nltk.classify.maxent import TadmMaxentClassifier 
 90      classifier = names_demo(TadmMaxentClassifier.train)                
 91   
 92 -def encoding_demo(): 
 93      import sys 
 94      from nltk.classify.maxent import TadmEventMaxentFeatureEncoding 
 95      from nltk.classify.tadm import write_tadm_file 
 96      tokens = [({'f0':1, 'f1':1, 'f3':1}, 'A'), 
 97                ({'f0':1, 'f2':1, 'f4':1}, 'B'), 
 98                ({'f0':2, 'f2':1, 'f3':1, 'f4':1}, 'A')] 
 99      encoding = TadmEventMaxentFeatureEncoding.train(tokens) 
100      write_tadm_file(tokens, encoding, sys.stdout) 
101      print 
102      for i in range(encoding.length()): 
103          print '%s --> %d' % (encoding.describe(i), i) 
104      print 
105           
106  if __name__ == '__main__': 
107      encoding_demo() 
108      names_demo() 
109