def test_clf(self, clf, clf_name):
pre_train = Preprocesser(self.pos_file, self.neg_file)
ngrams = ('unigram', ) # 'bigram'
selector_methods = ('df', )
weight_methods = ('tf_idf', )
for ngram in ngrams:
data, target = pre_train.get_ngram(ngram)
train_data, test_data, train_target, test_target = train_test_split(
data, target, test_size=0.70)
selector = FeatureSelector(train_data, train_target)
for selector_method in selector_methods:
all_features = selector.select(selector_method,
selector.all_features_size)
for count in self.feature_range:
features = all_features[:count]
train_vectorizer = TermWeight(train_data, train_target,
features)
for weight_method in weight_methods:
train_weighted_data = train_vectorizer.weight(
weight_method)
test_vectorizer = TermWeight(test_data, test_target,
features)
test_weighted_data = test_vectorizer.weight(
weight_method)
clf.fit(train_weighted_data, train_target)
test_result = clf.predict(test_weighted_data)
score = compute_aprf(test_target, test_result)
self.write_result(clf_name, ngram, selector_method,
count, weight_method, score)
def test_clf(self, clf, clf_name):
pre_train = Preprocesser(self.pos_file, self.neg_file)
pre_test = Preprocesser(self.pos_test, self.neg_test)
ngrams = ('unigram', ) # 'unigram', 'bigram', 'dictWords'
selector_methods = (
'df',
'mi',
)
weight_methods = ('tf_idf', )
for ngram in ngrams:
train_data, train_target = pre_train.get_ngram(ngram)
test_data, test_target = pre_test.get_ngram(ngram)
selector = FeatureSelector(train_data, train_target)
for selector_method in selector_methods:
all_features = selector.select(selector_method,
selector.all_features_size)
print(len(all_features))
for count in self.feature_range:
print(int(len(all_features) * count))
features = all_features[:int(len(all_features) *
count)] #[:count]
train_vectorizer = TermWeight(train_data, train_target,
features)
for weight_method in weight_methods:
train_weighted_data = train_vectorizer.weight(
weight_method)
test_vectorizer = TermWeight(test_data, test_target,
features)
test_weighted_data = test_vectorizer.weight(
weight_method)
clf.fit(train_weighted_data, train_target)
test_result = clf.predict(test_weighted_data)
score = compute_aprf(test_target, test_result)
self.write_result(clf_name, ngram, selector_method,
len(features), weight_method, score)
def test_unigram(clf, clf_name, pos_file, neg_file, data_type):
pre = Preprocesser(pos_file, neg_file)
ngrams = ('unigram', )
weight_methods = ('tf_idf', )
for ngram in ngrams:
data, target = pre.get_ngram(ngram)
train_data, test_data, train_target, test_target = train_test_split(
data, target)
selector = FeatureSelector(train_data, train_target)
all_features_count = selector.all_features_size
features = selector.all_features
train_vectorizer = TermWeight(train_data, train_target, features)
for weight_method in weight_methods:
train_weighted_data = train_vectorizer.weight(weight_method)
test_vectorizer = TermWeight(test_data, test_target, features)
test_weighted_data = test_vectorizer.weight(weight_method)
clf.fit(train_weighted_data, train_target)
score = clf.score(test_weighted_data, test_target)
result_dir = '../data/result/'
if not os.path.exists(result_dir):
os.mkdir(result_dir)
with open(os.path.join(result_dir, clf_name + '_unigram.txt'),
'at',
encoding='utf-8') as f:
f.write(
'clf={1}\t data_type={2}\t count={3}\t score={4:.2f}\n'.
format(clf_name, data_type, all_features_count,
score * 100))
def test_clf_by_percent(self,
clf,
clf_name,
feature_count_range=[
1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000,
9000, 10000
],
bigram_range=[0.1, 0.3, 0.5, 0.7, 0.9]):
pre = Preprocesser(self.pos_file, self.neg_file)
selector_methods = ('df', )
weight_methods = ('tf_idf', )
unigram_data, unigram_target = pre.get_unigram(is_shuffle=False)
bigram_data, bigram_target = pre.get_bigram(is_shuffle=False)
train_unigram, test_unigram, train_bigram, test_bigram, train_target, \
test_target = self.split_data(unigram_data, bigram_data, unigram_target)
unigram_selector = FeatureSelector(train_unigram, train_target)
bigram_selector = FeatureSelector(train_bigram, train_target)
train_data = [
unigram_line + bigram_line
for unigram_line, bigram_line in zip(train_unigram, train_bigram)
]
test_data = [
unigram_line + bigram_line
for unigram_line, bigram_line in zip(test_unigram, test_bigram)
]
for selector_method in selector_methods:
unigram_features = unigram_selector.select(
selector_method, unigram_selector.all_features_size)
bigram_features = bigram_selector.select(
selector_method, bigram_selector.all_features_size)
for count in feature_count_range:
for bigram_size in bigram_range:
bigram_count = int(count * bigram_size)
bigram_selected_features = bigram_features[:bigram_count]
unigram_selected_features = unigram_features[:(
count - bigram_count)]
features = unigram_selected_features
features.extend(bigram_selected_features)
train_vectorizer = TermWeight(train_data, train_target,
features)
for weight_method in weight_methods:
train_weighted_data = train_vectorizer.weight(
weight_method)
test_vectorizer = TermWeight(test_data, test_target,
features)
test_weighted_data = test_vectorizer.weight(
weight_method)
clf.fit(train_weighted_data, train_target)
test_result = clf.predict(test_weighted_data)
score = compute_aprf(test_target, test_result)
self.write_result(clf_name + '_uni_bi',
'unigram+bigram', selector_method,
str(count) + ':' + str(bigram_size),
weight_method, score)
def test_rule(clf, clf_name, train_unigram, train_bigram, train_target,
test_unigram, test_bigram, test_target, data_type):
feature_count_range = [
500, 1000, 1500, 2000, 2500, 3000, 3500, 4000, 5000, 6000
]
bigram_range = [0.1, 0.3, 0.5, 0.7, 0.9, 1]
selector_methods = ('mi', 'df', 'ig')
weight_methods = ('tf_idf', )
unigram_selector = FeatureSelector(train_unigram, train_target)
bigram_selector = FeatureSelector(train_bigram, train_target)
train_data = [
unigram_line + bigram_line
for unigram_line, bigram_line in zip(train_unigram, train_bigram)
]
test_data = [
unigram_line + bigram_line
for unigram_line, bigram_line in zip(test_unigram, test_bigram)
]
for selector_method in selector_methods:
unigram_features = unigram_selector.select(
selector_method, unigram_selector.all_features_size)
bigram_features = bigram_selector.select(
selector_method, bigram_selector.all_features_size)
for count in feature_count_range:
for bigram_size in bigram_range:
bigram_count = int(count * bigram_size)
bigram_selected_features = bigram_features[:bigram_count]
unigram_selected_features = unigram_features[:(count -
bigram_count)]
features = unigram_selected_features
features.extend(bigram_selected_features)
train_vectorizer = TermWeight(train_data, train_target,
features)
for weight_method in weight_methods:
train_weighted_data = train_vectorizer.weight(
weight_method)
test_vectorizer = TermWeight(test_data, test_target,
features)
test_weighted_data = test_vectorizer.weight(weight_method)
clf.fit(train_weighted_data, train_target)
score = clf.score(test_weighted_data, test_target)
result_dir = '../data/result/'
if not os.path.exists(result_dir):
os.mkdir(result_dir)
with open(os.path.join(result_dir,
clf_name + '_unigram_and_rule.txt'),
'at',
encoding='utf-8') as f:
f.write(
'clf={1}\t data_type={2}\t count={3}\t rule_size={4}\t score={5:.2f}\n'
.format(clf_name, data_type, count, bigram_size,
score * 100))
def test_clf_uni_bi(
self,
clf,
clf_name,
feature_count_range=[
100, 300, 500, 700, 1000, 1500, 2000, 2500, 3000, 4000, 5000
],
uni_bi_range=[0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1]):
# pre = Preprocesser(self.pos_file, self.neg_file)
pre_train = Preprocesser(self.pos_file, self.neg_file)
pre_test = Preprocesser(self.pos_test, self.neg_test)
selector_methods = (
'df',
'ig',
'mi',
'chi',
) # 'df', 'ig', 'mi', 'chi'
weight_methods = ('tf_idf', )
train_unigram, train_target = pre_train.get_unigram(is_shuffle=False)
train_bigram, train_target = pre_train.get_bigram(is_shuffle=False)
test_unigram, test_target = pre_test.get_unigram(
is_shuffle=False) #, 0.7, 0.9
test_bigram, test_target = pre_test.get_bigram(is_shuffle=False)
unigram_selector = FeatureSelector(train_unigram, train_target)
bigram_selector = FeatureSelector(train_bigram, train_target)
train_data = [
unigram_line + bigram_line
for unigram_line, bigram_line in zip(train_unigram, train_bigram)
]
test_data = [
unigram_line + bigram_line
for unigram_line, bigram_line in zip(test_unigram, test_bigram)
]
for selector_method in selector_methods:
unigram_features = unigram_selector.select(
selector_method, unigram_selector.all_features_size)
bigram_features = bigram_selector.select(
selector_method, bigram_selector.all_features_size)
for count in feature_count_range:
for uni_bi_size in uni_bi_range:
bi_size = int(count * uni_bi_size)
bigram_selected_features = bigram_features[:bi_size]
unigram_selected_features = unigram_features[:(count -
bi_size)]
features = unigram_selected_features
features.extend(bigram_selected_features)
train_vectorizer = TermWeight(train_data, train_target,
features)
for weight_method in weight_methods:
train_weighted_data = train_vectorizer.weight(
weight_method)
test_vectorizer = TermWeight(test_data, test_target,
features)
test_weighted_data = test_vectorizer.weight(
weight_method)
clf.fit(train_weighted_data, train_target)
test_result = clf.predict(test_weighted_data)
score = compute_aprf(test_target, test_result)
self.write_result(clf_name + '_uni_bi',
'unigram+bigram', selector_method,
str(count) + ':' + str(uni_bi_size),
weight_method, score)
def test_unigram_rule(pos_file, neg_file, clf, clf_name, data_type):
selector_methods = ('df', )
weight_methods = ('tf_idf', )
feature_count_range = [
1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000, 9000, 10000
]
rule_range = (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1)
pos_data, rule_pos_data = read_word_rule(pos_file)
neg_data, rule_neg_data = read_word_rule(neg_file)
test_size = 0.2
pos_test_index = int(len(pos_data) * test_size)
neg_test_index = int(len(neg_data) * test_size)
pos_test_data = pos_data[:pos_test_index]
pos_train_data = pos_data[pos_test_index:]
neg_test_data = neg_data[:neg_test_index]
neg_train_data = neg_data[neg_test_index:]
train_data = pos_train_data + neg_train_data
test_data = pos_test_data + neg_test_data
train_target = [1] * len(pos_train_data) + [0] * len(neg_train_data)
test_target = [1] * len(pos_test_data) + [0] * len(neg_test_data)
rule_pos_test_data = rule_pos_data[:pos_test_index]
rule_pos_train_data = rule_pos_data[pos_test_index:]
rule_neg_test_data = rule_neg_data[:neg_test_index]
rule_neg_train_data = rule_neg_data[neg_test_index:]
rule_train_data = rule_pos_train_data + rule_neg_train_data
rule_test_data = rule_pos_test_data + rule_neg_test_data
def merge_data(data_a, data_b):
result = []
for i in range(len(data_a)):
result.append(data_a[i] + data_b[i])
return result
all_train_data = merge_data(train_data, rule_train_data)
all_test_data = merge_data(test_data, rule_test_data)
unigram_selector = FeatureSelector(train_data, train_target)
rule_selector = FeatureSelector(rule_train_data, train_target)
for selector_method in selector_methods:
unigram_features = unigram_selector.select(
selector_method, unigram_selector.all_features_size)
rule_features = rule_selector.select(selector_method,
rule_selector.all_features_size)
for count in feature_count_range:
for rule_size in rule_range:
rule_count = int(count * rule_size)
rule_selected_features = rule_features[:rule_count]
unigram_selected_features = unigram_features[:(count -
rule_count)]
features = unigram_selected_features
features.extend(rule_selected_features)
train_vectorizer = TermWeight(all_train_data, train_target,
features)
for weight_method in weight_methods:
train_weighted_data = train_vectorizer.weight(
weight_method)
test_vectorizer = TermWeight(all_test_data, test_target,
features)
test_weighted_data = test_vectorizer.weight(weight_method)
clf.fit(train_weighted_data, train_target)
test_result = clf.predict(test_weighted_data)
score = compute_aprf(test_target, test_result)
write_result(clf_name, data_type, selector_method, count,
rule_size, score)