def to_weka_arff(ngram, number_of_features):
count_vect = TfidfVectorizer(ngram_range=(1, ngram),
norm='l2',
sublinear_tf=True)
label_list = get_labels()
tweet_list = get_labelled_tweets()
features = count_vect.fit_transform(tweet_list)
features = SelectKBest(chi2, k=number_of_features).fit_transform(
features, label_list)
print features.shape
arff_data = []
arff_data.append("@RELATION sport")
for i in range(features.shape[1]):
arff_data.append("@ATTRIBUTE feature" + str(i) + " REAL")
arff_data.append("@ATTRIBUTE sportclass {neutral,neg,pos}")
arff_data.append("@DATA")
array_features = features.toarray()
for i in range(len(array_features)):
feature = array_features[i]
label = label_list[i]
csv_feature = ",".join(str(x) for x in feature)
csv_feature = csv_feature + "," + label
arff_data.append(csv_feature)
with open('data/sport.arff', 'w') as file:
for item in arff_data:
file.write("%s\n" % item)
def to_weka_arff(ngram, number_of_features):
count_vect = TfidfVectorizer(ngram_range=(1, ngram), norm='l2', sublinear_tf=True)
label_list = get_labels()
tweet_list = get_labelled_tweets()
features = count_vect.fit_transform(tweet_list)
features = SelectKBest(chi2, k=number_of_features).fit_transform(features, label_list)
print features.shape
arff_data = []
arff_data.append("@RELATION sport")
for i in range(features.shape[1]):
arff_data.append("@ATTRIBUTE feature" + str(i) + " REAL")
arff_data.append("@ATTRIBUTE sportclass {neutral,neg,pos}")
arff_data.append("@DATA")
array_features = features.toarray()
for i in range(len(array_features)):
feature = array_features[i]
label = label_list[i]
csv_feature = ",".join(str(x) for x in feature)
csv_feature = csv_feature + "," + label
arff_data.append(csv_feature)
with open('data/sport.arff', 'w') as file:
for item in arff_data:
file.write("%s\n" % item)
def find_and_save_timings():
tweet_list = get_labelled_tweets()
num_tweets = len(tweet_list)
setup = """
from data_source import get_labelled_tweets, get_labels;
from sklearn.externals import joblib;
tweet_list = get_labelled_tweets();
# do transformation into vector;
vectoriser = joblib.load('model/tfidf_vectoriser.pkl');
vectorised_tweet_list = vectoriser.transform(tweet_list);
svm_model = joblib.load('model/tfidf_linsvc.pkl');
svm_model.predict(vectorised_tweet_list);
"""
test_statement = 'svm_model.predict(vectorised_tweet_list)'
REPETITIONS = 100
# check timing of svm
# time in micro seconds
svm_time = timeit.timeit(stmt=test_statement, setup=setup, number=REPETITIONS)
svm_time_dataset = get_dataset_time(svm_time, REPETITIONS)
svm_time_record = get_record_time(svm_time_dataset, num_tweets)
setup_ensemble = """
import cPickle;
from data_source import get_labelled_tweets;
from sklearn.externals import joblib;
tweet_list = get_labelled_tweets();
vectoriser = joblib.load('model/tfidf_vectoriser.pkl');
vectorised_tweet_list = vectoriser.transform(tweet_list);
with open('model/tfidf_ada.pickle', 'rb') as f:
ensemble_model = cPickle.load(f);
ensemble_model.predict(vectorised_tweet_list);
"""
test_statement_ensemble = 'ensemble_model.predict(vectorised_tweet_list)'
ensemble_time = timeit.timeit(stmt=test_statement_ensemble,
setup=setup_ensemble,
number=REPETITIONS)
ens_time_dataset = get_dataset_time(ensemble_time, REPETITIONS)
ens_time_record = get_record_time(ens_time_dataset, num_tweets)
# save results in a txt file
create_directory('metric_result')
with open("metric_result/" + 'timings' + ".txt", "w") as text_file:
text_file.write("Number of records in dataset: {0}\n".format(num_tweets))
text_file.write("Svm dataset time: {0}\n".format(svm_time_dataset))
text_file.write("Svm record time: {0}\n".format(svm_time_record))
text_file.write("Ensemble dataset time: {0}\n".format(ens_time_dataset))
text_file.write("Ensemble record time: {0}\n".format(ens_time_record))
def gensim_classifier():
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s',
level=logging.INFO)
label_list = get_labels()
tweet_list = get_labelled_tweets()
# split all sentences to list of words
sentences = []
for tweet in tweet_list:
temp_doc = tweet.split()
sentences.append(temp_doc)
# parameters for model
num_features = 100
min_word_count = 1
num_workers = 4
context = 2
downsampling = 1e-3
# Initialize and train the model
w2v_model = Word2Vec(sentences, workers=num_workers, \
size=num_features, min_count = min_word_count, \
window = context, sample = downsampling, seed=1)
index_value, train_set, test_set = train_test_split(0.80, sentences)
train_vector = getAvgFeatureVecs(train_set, w2v_model, num_features)
test_vector = getAvgFeatureVecs(test_set, w2v_model, num_features)
train_vector = Imputer().fit_transform(train_vector)
test_vector = Imputer().fit_transform(test_vector)
# train model and predict
model = LinearSVC()
classifier_fitted = OneVsRestClassifier(model).fit(
train_vector, label_list[:index_value])
result = classifier_fitted.predict(test_vector)
# output result to csv
create_directory('data')
result.tofile("data/w2v_linsvc.csv", sep=',')
# store the model to mmap-able files
create_directory('model')
joblib.dump(model, 'model/%s.pkl' % 'w2v_linsvc')
# evaluation
label_score = classifier_fitted.decision_function(test_vector)
binarise_result = label_binarize(result, classes=class_list)
binarise_labels = label_binarize(label_list, classes=class_list)
evaluate(binarise_result, binarise_labels[index_value:], label_score,
'w2v_linsvc')
def gensim_classifier():
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.INFO)
label_list = get_labels()
tweet_list = get_labelled_tweets()
# split all sentences to list of words
sentences = []
for tweet in tweet_list:
temp_doc = tweet.split()
sentences.append(temp_doc)
# parameters for model
num_features = 100
min_word_count = 1
num_workers = 4
context = 2
downsampling = 1e-3
# Initialize and train the model
w2v_model = Word2Vec(sentences, workers=num_workers, \
size=num_features, min_count = min_word_count, \
window = context, sample = downsampling, seed=1)
index_value, train_set, test_set = train_test_split(0.80, sentences)
train_vector = getAvgFeatureVecs(train_set, w2v_model, num_features)
test_vector = getAvgFeatureVecs(test_set, w2v_model, num_features)
train_vector = Imputer().fit_transform(train_vector)
test_vector = Imputer().fit_transform(test_vector)
# train model and predict
model = LinearSVC()
classifier_fitted = OneVsRestClassifier(model).fit(train_vector, label_list[:index_value])
result = classifier_fitted.predict(test_vector)
# output result to csv
create_directory('data')
result.tofile("data/w2v_linsvc.csv", sep=',')
# store the model to mmap-able files
create_directory('model')
joblib.dump(model, 'model/%s.pkl' % 'w2v_linsvc')
# evaluation
label_score = classifier_fitted.decision_function(test_vector)
binarise_result = label_binarize(result, classes=class_list)
binarise_labels = label_binarize(label_list, classes=class_list)
evaluate(binarise_result, binarise_labels[index_value:], label_score, 'w2v_linsvc')
def lin_svc():
label_list = get_labels()
tweet_list = get_labelled_tweets()
# vectorise using tf-idf
vectoriser = TfidfVectorizer(
min_df=3,
max_features=None,
strip_accents='unicode',
analyzer='word',
token_pattern=r'\w{1,}',
ngram_range=(1, 2),
use_idf=1,
smooth_idf=1,
sublinear_tf=1,
)
## do transformation into vector
fitted_vectoriser = vectoriser.fit(tweet_list)
vectorised_tweet_list = fitted_vectoriser.transform(tweet_list)
train_vector, test_vector, train_labels, test_labels = train_test_split(
vectorised_tweet_list, label_list, test_size=0.8, random_state=42)
# train model and predict
model = LinearSVC()
ovr_classifier = OneVsRestClassifier(model).fit(train_vector, train_labels)
result = ovr_classifier.predict(test_vector)
# output result to csv
create_directory('data')
save_to_csv("data/testset_labels.csv", test_labels)
result.tofile("data/tfidf_linsvc.csv", sep=',')
save_model(ovr_classifier, 'tfidf_linsvc')
save_vectoriser(fitted_vectoriser, 'tfidf_vectoriser')
# evaluation
label_score = ovr_classifier.decision_function(test_vector)
binarise_result = label_binarize(result, classes=class_list)
binarise_labels = label_binarize(test_labels, classes=class_list)
evaluate(binarise_result, binarise_labels, label_score, 'tfidf_linsvc')
def lin_svc():
label_list = get_labels()
tweet_list = get_labelled_tweets()
# vectorise using tf-idf
vectoriser = TfidfVectorizer(min_df=3,
max_features=None,
strip_accents='unicode',
analyzer='word',
token_pattern=r'\w{1,}',
ngram_range=(1, 2),
use_idf=1,
smooth_idf=1,
sublinear_tf=1,)
## do transformation into vector
fitted_vectoriser = vectoriser.fit(tweet_list)
vectorised_tweet_list = fitted_vectoriser.transform(tweet_list)
train_vector, test_vector, train_labels, test_labels = train_test_split(vectorised_tweet_list,
label_list,
test_size=0.8,
random_state=42)
# train model and predict
model = LinearSVC()
ovr_classifier = OneVsRestClassifier(model).fit(train_vector, train_labels)
result = ovr_classifier.predict(test_vector)
# output result to csv
create_directory('data')
save_to_csv("data/testset_labels.csv", test_labels)
result.tofile("data/tfidf_linsvc.csv", sep=',')
save_model(ovr_classifier, 'tfidf_linsvc')
save_vectoriser(fitted_vectoriser, 'tfidf_vectoriser')
# evaluation
label_score = ovr_classifier.decision_function(test_vector)
binarise_result = label_binarize(result, classes=class_list)
binarise_labels = label_binarize(test_labels, classes=class_list)
evaluate(binarise_result, binarise_labels, label_score, 'tfidf_linsvc')
def ensemble_classify():
label_list = get_labels()
tweet_list = get_labelled_tweets()
# vectorise using tf-idf
vectoriser = TfidfVectorizer(min_df=3,
max_features=None,
strip_accents='unicode',
analyzer='word',
token_pattern=r'\w{1,}',
ngram_range=(1, 2),
use_idf=1,
smooth_idf=1,
sublinear_tf=1,)
## do transformation into vector
vectoriser.fit(tweet_list)
vectorised_tweet_list = vectoriser.transform(tweet_list)
train_vector, test_vector, train_labels, test_labels = train_test_split(vectorised_tweet_list,
label_list,
test_size=0.8,
random_state=42)
n_estimators = 10 # number of weak learners
model = AdaBoostClassifier(n_estimators=n_estimators)
ada_classifier = model.fit(train_vector, train_labels)
result = ada_classifier.predict(test_vector)
# output result to csv
create_directory('data')
result.tofile("data/tfidf_ada.csv", sep=',')
save_model(ada_classifier, 'tfidf_ada')
# evaluation
binarise_result = label_binarize(result, classes=class_list)
binarise_labels = label_binarize(test_labels, classes=class_list)
generate_eval_metrics(binarise_result, 'tfidf_ada', binarise_labels)
def ensemble_classify():
label_list = get_labels()
tweet_list = get_labelled_tweets()
# vectorise using tf-idf
vectoriser = TfidfVectorizer(
min_df=3,
max_features=None,
strip_accents='unicode',
analyzer='word',
token_pattern=r'\w{1,}',
ngram_range=(1, 2),
use_idf=1,
smooth_idf=1,
sublinear_tf=1,
)
## do transformation into vector
vectoriser.fit(tweet_list)
vectorised_tweet_list = vectoriser.transform(tweet_list)
train_vector, test_vector, train_labels, test_labels = train_test_split(
vectorised_tweet_list, label_list, test_size=0.8, random_state=42)
n_estimators = 10 # number of weak learners
model = AdaBoostClassifier(n_estimators=n_estimators)
ada_classifier = model.fit(train_vector, train_labels)
result = ada_classifier.predict(test_vector)
# output result to csv
create_directory('data')
result.tofile("data/tfidf_ada.csv", sep=',')
save_model(ada_classifier, 'tfidf_ada')
# evaluation
binarise_result = label_binarize(result, classes=class_list)
binarise_labels = label_binarize(test_labels, classes=class_list)
generate_eval_metrics(binarise_result, 'tfidf_ada', binarise_labels)
