Python NLTK Classifier.train（trainfeats）…ValueError：需要多个值才能解压

Question

def word_feats(words):
     return dict([(word, True) for word in words])

for tweet in negTweets:
     words = re.findall(r"[\w']+|[.,!?;]", tweet) #splits the tweet into words
     negwords = [(word_feats(words), 'neg')] #tag the words with feature
     negfeats.append(negwords) #add the words to the feature list
for tweet in posTweets:
     words = re.findall(r"[\w']+|[.,!?;]", tweet)
     poswords = [(word_feats(words), 'pos')]
     posfeats.append(poswords)

negcutoff = len(negfeats)*3/4 #take 3/4ths of the words
poscutoff = len(posfeats)*3/4

trainfeats = negfeats[:negcutoff] + posfeats[:poscutoff] #assemble the train set
testfeats = negfeats[negcutoff:] + posfeats[poscutoff:]

classifier = NaiveBayesClassifier.train(trainfeats)
print 'accuracy:', nltk.classify.util.accuracy(classifier, testfeats)
classifier.show_most_informative_features()

I am getting the following error when running this code...

File "C:\Python27\lib\nltk\classify\naivebayes.py", line 191, in train

for featureset, label in labeled_featuresets:

ValueError: need more than 1 value to unpack

The error is coming from the classifier = NaiveBayesClassifier.train(trainfeats) line and I'm not sure why. I have done something like this before, and my trainfeats seams to be in the same format as then... a sample from the format is listed below...

[[({'me': True, 'af': True, 'this': True, 'joy': True, 'high': True, 'hookah': True, 'got': True}, 'pos')]]

what other value does my trainfeats need to create the classifier? emphasized text

Answer 1

The comment by @Prune is right: Your labeled_featuresets should be a sequence of pairs (two-element lists or tuples): A feature dict and a category for each data point. Instead, each element in your trainfeats is a list containing one element: A tuple of those two things. Lose the square brackets in both feature-building loops and this part should work correctly. Eg,

negwords = (word_feats(words), 'neg')
negfeats.append(negwords)

Two more things: Consider using nltk.word_tokenize() instead of doing your own tokenization. And you should randomize the order of your training data, eg with random.scramble(trainfeats) .