def test_kaggle():
N = config.N
reviews = parse_data.parse_training_file()
popular_words = features.get_popular_words(reviews)
word_lists = features.create_word_lists()
sentiment_counts = likelihood.get_sentiment_counts(reviews)
states = v.get_states(sentiment_counts)
ngram_counts = prior_prob.gen_ngram_counts(reviews, N)
l = prior_prob.train_interpolation(ngram_counts, N)
vectors = likelihood.gen_sentiment_vectors(reviews, word_lists, popular_words)
test_reviews = parse_data.parse_training_file('data/test_data_no_true_labels.txt')
f = open('kaggle.txt', 'w')
f.write("Id,answer\n")
rid = 0
for i in test_reviews:
lines = []
answer = []
tags = []
for line in test_reviews[i]['reviews']:
lines.append(line[1])
tags.append(test_reviews[i]['title'])
M = v.calculate_viterbi_matrix(N, states, lines, ngram_counts, l, vectors, sentiment_counts, word_lists, popular_words, tags)
result = v.backtrace_viterbi(M, states, ngram_counts, l)
for res in result:
if res == "pos":
f.write(str(rid) + ",1\n")
elif res == "neu":
f.write(str(rid) + ",0\n")
elif res == "neg":
f.write(str(rid) + ",-1\n")
else:
raise Error("Unexpected error: unidentified state.")
rid += 1
print "On row " + str(rid)
f.close()
print "Done. Output saved to kaggle.txt."
def test_training():
N = config.N
reviews = parse_data.parse_training_file('data/training_data.txt', 0, 150)
popular_words = features.get_popular_words(reviews)
word_lists = features.create_word_lists()
sentiment_counts = likelihood.get_sentiment_counts(reviews)
states = v.get_states(sentiment_counts)
ngram_counts = prior_prob.gen_ngram_counts(reviews, N)
l = prior_prob.train_interpolation(ngram_counts, N)
vectors = likelihood.gen_sentiment_vectors(reviews, word_lists, popular_words)
test_reviews = parse_data.parse_training_file('data/training_data.txt', 150)
correct = 0
false = 0
for i in test_reviews:
lines = []
answer = []
tags = []
for line in test_reviews[i]['reviews']:
lines.append(line[1])
tags.append(test_reviews[i]['title'])
answer.append(line[0])
M = v.calculate_viterbi_matrix(N, states, lines, ngram_counts, l, vectors, sentiment_counts, word_lists, popular_words, tags)
result = v.backtrace_viterbi(M, states, ngram_counts, l)
for j in range(len(answer)):
if answer[j] == result[j]:
correct += 1
else:
false += 1
percent = float(correct) / float(correct + false)
return percent
import parse_data
import features
f = open('kaggle_baseline.txt', 'w')
f.write("Id,answer\n")
test_reviews = parse_data.parse_training_file("data/training_data.txt")
neg_words = features.get_negative_words()
pos_words = features.get_positive_words()
lines = []
answer = []
tags = []
for i in test_reviews:
for line in test_reviews[i]['reviews']:
tags.append(line[0])
lines.append(line[1])
for line in lines:
word_list = line.split(" ")
pos_words_count = 0
neg_words_count = 0
for word in word_list:
if word in pos_words:
pos_words_count += 1
if word in neg_words:
neg_words_count += 1
if pos_words_count > neg_words_count + 1:
answer.append("pos")