def classify():
#corpus = 'Cornell_text_polarity'
#corpus = 'BingLiu_selected_sentences'
corpus = 'Cornell_sentence_polarity'
cases = load_corpus(corpus)
features = get_word_features(cases)
train_feats = []
test_feats = []
for polarity, feats in features.items():
#cutoff = len(feats) * 2 / 4
cutoff = 1000
print polarity, 'number of train:', cutoff
#train_feats += feats[:cutoff]
#test_feats += feats[cutoff:]
temp_feats = feats[:]
random.shuffle(temp_feats)
train_feats += temp_feats[:cutoff]
test_feats += temp_feats[cutoff:]
print 'train on %d instances, test on %d instances' % (len(train_feats), len(test_feats))
classifier = SvmClassifier.train(train_feats)
print 'Test classify:', classifier.classify(dict([('I', 0.0),('love',1.0), ('you', 0.0)]))
print 'accuracy:', nltk.classify.util.accuracy(classifier, test_feats)
classifier.show_most_informative_features()
# filter the feature vectors:
fvecs = [(tweet_features.get_tweet_features(t, worstfeaturesfilter), s) for (t, s) in tweets]
v_train = fvecs[0:num_train]
# v_train = fvecs
v_test = fvecs[num_train : len(tweets)]
# dump tweets which our feature selector found nothing
# for i in range(0,len(tweets)):
# if tweet_features.is_zero_dict( fvecs[i][0] ):
# print tweets[i][1] + ': ' + tweets[i][0]
# train classifier
# classifier = nltk.NaiveBayesClassifier.train(v_train);
# classifier = nltk.classify.maxent.train_maxent_classifier_with_gis(v_train);
classifier = SvmClassifier.train(v_train)
# classifier = nltk.classify.maxent.train_maxent_classifier_with_gis(v_train,count_cutoff=2)
# classifier = nltk.classify.maxent.train_maxent_classifier_with_iis(v_train,count_cutoff=4)
# classifier = nltk.classify.maxent.train_maxent_classifier_with_scipy(v_train, algorithm='BFGS');
pickle.dump(worstfeaturesfilter, open("worstfeaturesfilter.pickle", "w"))
print "WARNING: NOT PICKELING CLASSIFIER ANYMORE"
# pickle.dump(classifier, open('classifier.pickle', 'w'))
# print classifier.classify(tweet_features.get_tweet_features("Christmas Eve without my, with cold feet and nobody to.", worstfeaturesfilter))
# print classifier.classify(tweet_features.get_tweet_features("Nochebuena sin mi @ tdomhan, con los pies fríos y nadie a quien abrazar.", worstfeaturesfilter))
# print classifier.classify(tweet_features.get_tweet_features("Nochebuena sin mi, con los pies fríos y nadie a quien abrazar.", worstfeaturesfilter))
# print classifier.explain(tweet_features.get_tweet_features("Nochebuena sin mi @ tdomhan, con los pies fríos y nadie a quien abrazar.", worstfeaturesfilter))
# classify and dump results for interpretation
#filter the feature vectors:
fvecs = [(tweet_features.get_tweet_features(t, worstfeaturesfilter), s)
for (t, s) in tweets]
v_train = fvecs[0:num_train]
#v_train = fvecs
v_test = fvecs[num_train:len(tweets)]
# dump tweets which our feature selector found nothing
#for i in range(0,len(tweets)):
# if tweet_features.is_zero_dict( fvecs[i][0] ):
# print tweets[i][1] + ': ' + tweets[i][0]
# train classifier
#classifier = nltk.NaiveBayesClassifier.train(v_train);
#classifier = nltk.classify.maxent.train_maxent_classifier_with_gis(v_train);
classifier = SvmClassifier.train(v_train)
#classifier = nltk.classify.maxent.train_maxent_classifier_with_gis(v_train,count_cutoff=2)
#classifier = nltk.classify.maxent.train_maxent_classifier_with_iis(v_train,count_cutoff=4)
#classifier = nltk.classify.maxent.train_maxent_classifier_with_scipy(v_train, algorithm='BFGS');
pickle.dump(worstfeaturesfilter, open('worstfeaturesfilter.pickle', 'w'))
print "WARNING: NOT PICKELING CLASSIFIER ANYMORE"
#pickle.dump(classifier, open('classifier.pickle', 'w'))
#print classifier.classify(tweet_features.get_tweet_features("Christmas Eve without my, with cold feet and nobody to.", worstfeaturesfilter))
#print classifier.classify(tweet_features.get_tweet_features("Nochebuena sin mi @ tdomhan, con los pies fríos y nadie a quien abrazar.", worstfeaturesfilter))
#print classifier.classify(tweet_features.get_tweet_features("Nochebuena sin mi, con los pies fríos y nadie a quien abrazar.", worstfeaturesfilter))
#print classifier.explain(tweet_features.get_tweet_features("Nochebuena sin mi @ tdomhan, con los pies fríos y nadie a quien abrazar.", worstfeaturesfilter))
# classify and dump results for interpretation
tweets.append(item)
if sentiment == "positive":
pos_tweets += item
else:
neg_tweets += item
negcutoff, poscutoff = len(neg_tweets) * 4 / 5, len(pos_tweets) * 4 / 5
pos_train, pos_test = pos_tweets[:poscutoff], pos_tweets[poscutoff:]
neg_train, neg_test = neg_tweets[:negcutoff], neg_tweets[negcutoff:]
neg_feats_train = get_train_features_from_tweets(neg_train, 'neg')
pos_feats_train = get_train_features_from_tweets(pos_train, 'pos')
train_feats = neg_feats_train + pos_feats_train
classifier = SvmClassifier.train(train_feats)
# classifier = nltk.NaiveBayesClassifier.train(train_feats)
# Evaluation
correct, wrong = 0, 0
for tweet in neg_test:
features = get_features_from_tweet(tweet)
result = classifier.classify(features)
if result == "neg":
correct += 1
else:
wrong += 1
