def test_baseline(src, tar):
prefix = '../dataset/Sentiment/'
suffix = ['/positive.review', '/negative.review', '/unlabeled.review']
src_pos, src_pos_label = data_helper.read_file(prefix + src + suffix[0])
src_neg, src_neg_label = data_helper.read_file(prefix + src + suffix[1])
tar_pos, tar_pos_label = data_helper.read_file(prefix + tar + suffix[0])
tar_neg, tar_neg_label = data_helper.read_file(prefix + tar + suffix[1])
transformer = TfidfTransformer()
cv_src = CountVectorizer(min_df=20)
# x_src = cv_src.fit_transform(src_pos + src_neg).toarray()
x_src = transformer.fit_transform(cv_src.fit_transform(src_pos +
src_neg)).toarray()
y_src = src_pos_label + src_neg_label
lr = LogisticRegression(solver='lbfgs', C=C)
lr.fit(x_src, y_src)
# x_tar = cv_src.transform(tar_pos + tar_neg).toarray()
x_tar = transformer.fit_transform(cv_src.transform(tar_pos +
tar_neg)).toarray()
y_tar = tar_pos_label + tar_neg_label
acc = lr.score(x_tar, y_tar)
return acc
def _readData(self, filePath):
"""
:param filePath:
:return:
"""
labels, s = data_helper.read_file(filePath)
contents = []
for i in range(len(labels)):
content = self.cleanReview(s[i])
contents.append(content)
return labels, contents
def features_stub():
datafile = "restaurant-training.data"
raw_data = data_helper.read_file(os.path.join(DATA_DIR, datafile))
positive_texts, negative_texts = data_helper.get_reviews(raw_data)
category_texts = {"positive": positive_texts, "negative": negative_texts}
feature_set = "word_features"
features_category_tuples, texts = get_features_category_tuples(
category_texts, feature_set)
raise NotImplemented
filename = "???"
write_features_category(features_category_tuples, filename)
def features_stub():
datafile = "imdb-training.data"
raw_data = data_helper.read_file(datafile)
positive_texts, negative_texts = data_helper.get_reviews(raw_data)
category_texts = {"positive": positive_texts, "negative": negative_texts}
# FEATURE_SETS = {"word_pos_features", "word_features", "word_pos_liwc_features", "word_pos_opinion_features"}
feature_set = "word_pos_opinion_features"
features_category_tuples, texts = get_features_category_tuples(
category_texts, feature_set)
# raise NotImplemented
filename = feature_set + "-testing-features.txt"
write_features_category(features_category_tuples, filename)
def build_features(data_file, feat_name, save_feats=None, binning=False):
# read text data
raw_data = data_helper.read_file(data_file)
positive_texts, negative_texts = data_helper.get_reviews(raw_data)
category_texts = {"positive": positive_texts, "negative": negative_texts}
# build features
features_category_tuples, texts = get_features_category_tuples(
category_texts, feat_name)
# save features to file
if save_feats is not None:
write_features_category(features_category_tuples, save_feats)
return features_category_tuples, texts
def test(src, tar, pivot_num, pivot_min_times, dim, C):
weight_path = './weight/' + src + '2' + tar + '_svd_' + str(dim) + '.npy'
W = np.load(weight_path)
pivot_path = './pivot/' + src + '2' + tar + '_pivot_' + str(
pivot_num) + '_' + str(pivot_min_times)
with open(pivot_path, 'rb') as f:
pivot = pickle.load(f)
prefix = '../dataset/Sentiment/'
suffix = ['/positive.review', '/negative.review', '/unlabeled.review']
src_pos, src_pos_label = data_helper.read_file(prefix + src + suffix[0])
src_neg, src_neg_label = data_helper.read_file(prefix + src + suffix[1])
src_unl, src_unl_label = data_helper.read_file(prefix + src + suffix[2])
tar_pos, tar_pos_label = data_helper.read_file(prefix + tar + suffix[0])
tar_neg, tar_neg_label = data_helper.read_file(prefix + tar + suffix[1])
tar_unl, tar_unl_label = data_helper.read_file(prefix + tar + suffix[2])
# transformer_src = TfidfTransformer()
cv_src = CountVectorizer(min_df=20)
x_src = cv_src.fit_transform(src_pos + src_neg).toarray()
# x_src = transformer_src.fit_transform(cv_src.fit_transform(src_pos + src_neg)).toarray()
cv_lab_unl = CountVectorizer(min_df=40)
x_lab_unl = cv_lab_unl.fit_transform(src_unl + src_pos + src_neg +
tar_unl).toarray()
# transformer_lab_unl = TfidfTransformer()
# x_lab_unl = transformer_lab_unl.fit_transform(cv_lab_unl.fit_transform(src_unl + src_pos + src_neg + tar_unl)).toarray()
# x_src_transform = transformer_lab_unl.transform(cv_lab_unl.transform(src_pos + src_neg)).toarray()
x_src_transform = cv_lab_unl.transform(src_pos + src_neg).toarray()
x_src_transform = np.delete(x_src_transform, pivot, 1)
x_src_transform = x_src_transform.dot(W)
x_src = np.concatenate((x_src, x_src_transform), axis=1)
y_src = src_pos_label + src_neg_label
lr = LogisticRegression(solver='lbfgs', C=C)
lr.fit(x_src, y_src)
# x_tar_transform = transformer_lab_unl.transform(cv_lab_unl.transform(tar_pos + tar_neg)).toarray()
x_tar_transform = cv_lab_unl.transform(tar_pos + tar_neg).toarray()
x_tar_transform = np.delete(x_tar_transform, pivot, 1)
x_tar_transform = x_tar_transform.dot(W)
# x_tar = transformer_src.transform(cv_src.transform(tar_pos + tar_neg)).toarray()
x_tar = cv_src.transform(tar_pos + tar_neg).toarray()
x_tar = np.concatenate((x_tar, x_tar_transform), axis=1)
y_tar = tar_pos_label + tar_neg_label
acc = lr.score(x_tar, y_tar)
return acc
def features_stub():
feature_sets = [
"word_features", "word_pos_features", "word_pos_liwc_features",
"word_pos_opinion_features"
]
datasets = ["training", "development", "testing"]
for dataset in datasets:
for feature_set in feature_sets:
datafile = "data/imdb-" + dataset + ".data"
raw_data = data_helper.read_file(datafile)
positive_texts, negative_texts = data_helper.get_reviews(raw_data)
category_texts = {
"positive": positive_texts,
"negative": negative_texts
}
features_category_tuples, texts = get_features_category_tuples(
category_texts, feature_set)
filename = "best_features/" + feature_set + "-" + dataset + "-features.txt"
write_features_category(features_category_tuples, filename)
def features_stub():
datafiles = [
"imdb-training.data", "imdb-testing.data", "imdb-development.data"
]
featuresets = [
"word_pos_features", "word_features", "word_pos_liwc_features",
"word_pos_opinion_features"
]
for datafile in datafiles:
for feature_set in featuresets:
raw_data = data_helper.read_file(datafile)
positive_texts, negative_texts = data_helper.get_reviews(raw_data)
category_texts = {
"positive": positive_texts,
"negative": negative_texts
}
features_category_tuples, texts = get_features_category_tuples(
category_texts, feature_set)
data_set = re.search(r'-[a-z]+', datafile).group()
filename = f'{feature_set}{data_set}-features.txt'
write_features_category(features_category_tuples, filename)
def features_stub(filename):
# open restaurant-training.data
# calls data_helper.py to put file in pos or neg category list
# here is where I would call other files as well
datafile = filename
raw_data = data_helper.read_file(datafile)
positive_texts, negative_texts = data_helper.get_reviews(raw_data)
# category_texts creates
# { posive, [... all positive reviews ] , negative, [...all neg ...] }
#
#category_texts = {"positive": positive_texts, "negative": negative_texts}
#feature_set = "word_features"
positive_toks = []
positive_pos_toks = []
negative_toks = []
negative_pos_toks = []
print('begin tokenize')
# get word and pos tokens not the most
# efficient but easier to trace
for documents in positive_texts:
positive_toks += get_words(documents)
for documents in negative_texts:
negative_toks += get_words(documents)
for documents in positive_texts:
positive_pos_toks += get_pos(documents)
for documents in negative_texts:
negative_pos_toks += get_pos(documents)
print('tokenizing compl')
# get ngrams for positive and negative categories
posi_word_ngram = {}
posi_pos_ngram = {}
neg_word_ngram = {}
neg_pos_ngram = {}
print('begin word ngram')
#for tokens in positive_toks:
# posi_word_ngram.update( get_ngram_features( tokens ) )
posi_word_ngram.update(get_ngram_features(positive_toks))
print('all positive word ngram completed')
print('begin negative word ngram')
#for tokens in negative_toks:
# neg_word_ngram.update( get_ngram_features( tokens ) )
neg_word_ngram.update(get_ngram_features(negative_toks))
print('all negative word ngram completed')
print('end word ngram')
print('begin pos ngram')
#for tokens in positive_toks:
# posi_pos_ngram.update( get_ngram_features( tokens ) )
posi_pos_ngram.update(get_ngram_features(positive_pos_toks))
print('all pos pos ngram completed')
print('begin negative ngram')
#for tokens in negative_toks:
# neg_pos_ngram.update( get_ngram_features( tokens ) )
neg_pos_ngram.update(get_ngram_features(negative_pos_toks))
print('all negative pos ngram completed')
print('end pos ngram')
print('begin liwc')
# get LIWC features
posi_liwc_feat = get_liwc_features(positive_toks)
neg_liwc_feat = get_liwc_features(negative_toks)
print('end liwc')
print('begin file write')
print(posi_liwc_feat)
print(neg_liwc_feat)
fwrite_feature_vectors(filename, posi_word_ngram, neg_word_ngram,
posi_pos_ngram, neg_pos_ngram, posi_liwc_feat,
neg_liwc_feat)
