def conduct_test(test_set, lexicon_only = False, svm_only = False, number_of_tweets = 3700):
scores = []
if svm_only:
for i in range(0,number_of_tweets,100):
transformed = svm.vectorize_tweets(test_set.keys()[i:i+100], False)
predicted = svm.score(transformed, test_set.values()[i:i+100])
scores.append(predicted)
transformed = svm.vectorize_tweets(test_set.keys()[number_of_tweets:len(test_set.keys())], False)
predicted = svm.score(transformed, test_set.values()[number_of_tweets:len(test_set.keys())])
scores.append(predicted)
else:
for i in range(0,number_of_tweets,100):
predicted = predict(test_set.keys()[i:i+100], lexicon_only)
expected = test_set.items()[i:i+100]
scores.append(calculate_accuracy(predicted, dict(expected)))
predicted = predict(test_set.keys()[number_of_tweets:len(test_set.keys())],lexicon_only)
expected = test_set.items()[number_of_tweets:len(test_set.keys())]
scores.append(calculate_accuracy(predicted, dict(expected)))
return scores
def predict(test_tweets, lexicon_only = False):
global lexicon
predicted_from_lexicon = sentlex.predict(test_tweets, lexicon)
if lexicon_only:
return predicted_from_lexicon
undecided = sentlex.get_unclassified(predicted_from_lexicon)
#undecided_labels = get_labels(undecided, test_set)
undecided_transformed = svm.vectorize_tweets(undecided, False)
predicted_from_svm = svm.predict(undecided_transformed)
final = {}
for i in range(len(undecided)):
final[undecided[i]] = predicted_from_svm[i]
predicted_from_lexicon.update(final)
return predicted_from_lexicon
def load_sets(training_file, test_file):
global train_set, lexicon_set, test_set, train_labels, train_tweets, lexicon
#Loading the training and lexicon sets
train_set = loader.tweets_to_tweetlist(training_file, neutral = True)
lexicon_set = loader.tweets_to_wordlist(training_file, neutral = False)
#Loading the test set
test_set = loader.tweets_to_tweetlist(test_file, neutral = True)
#training the svm anlyzer
train_labels = train_set.values()
train_tweets = svm.vectorize_tweets(train_set.keys())
svm.train(train_tweets, train_labels)
#Building the lexicon
lexicon = sentlex.get_ratioDict(lexicon_set['positive'], lexicon_set['negative'], 10)
