def test(self):
corpus: CategorizedCorpus = DataFetcher.load_corpus(
NLPData.AIVIVN2019_SA_SAMPLE)
# corpus: CategorizedCorpus = DataFetcher.load_corpus(NLPData.AIVIVN2019_SA)
params = {
"vectorizer": CountVectorizer(ngram_range=(1, 2),
max_features=4000),
"svc": SVC(kernel='linear', C=0.3)
}
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.SVC,
**params)
model_trainer = ModelTrainer(classifier, corpus)
tmp_model_folder = mkdtemp()
def negative_f1_score(y_true, y_pred):
score_class_0, score_class_1 = f1_score(y_true,
y_pred,
average=None)
return score_class_1
def macro_f1_score(y_true, y_pred):
return f1_score(y_true, y_pred, average='macro')
score = model_trainer.train(tmp_model_folder,
scoring=negative_f1_score)
print(score)
classifier = TextClassifier.load(tmp_model_folder)
sentence = Sentence('tuyệt vời')
classifier.predict(sentence)
shutil.rmtree(tmp_model_folder)
print(sentence)
def predict(text):
print(f"\nText: {text}")
sentence = Sentence(text)
classifier.predict(sentence)
labels = sentence.labels
print(f"Labels: {labels}")
def read_text_classification_file(path_to_file) -> List[Sentence]:
sentences = []
with open(path_to_file) as f:
lines = f.read().splitlines()
for line in lines:
label_pattern = r"__label__(?P<label>[\w#]+)"
labels = re.findall(label_pattern, line)
labels = [Label(label) for label in labels]
text = re.sub(label_pattern, "", line)
s = Sentence(text, labels)
sentences.append(s)
return sentences
def predict(self, sentence: Sentence):
if self.estimator == TEXT_CLASSIFIER_ESTIMATOR.FAST_TEXT:
values, scores = self.ft.predict(sentence.text)
labels = []
for value, score in zip(values, scores):
value = value.replace("__label__", "")
label = Label(value, score)
labels.append(label)
sentence.add_labels(labels)
if self.estimator == TEXT_CLASSIFIER_ESTIMATOR.SVC:
text = sentence.text
X = self.x_transformer.transform([text])
y = self.svc.predict(X)
y = self.y_transformer.inverse_transform(y)
sentence.add_labels(y)
if self.estimator == TEXT_CLASSIFIER_ESTIMATOR.PIPELINE:
text = sentence.text
y = self.pipeline.predict([text])
if self.multilabel:
y = self.y_encoder.inverse_transform(y)
y = list(y[0])
else:
y = list(y)
sentence.add_labels(y)
def sentiment(text):
global classifier
if not classifier:
if os.path.exists(model_path):
classifier = TextClassifier.load(model_path)
else:
logger.error(
f"Could not load model at {model_path}.\n"
f"Download model with \"underthesea download {UTSModel.sa_bank.value}\"."
)
sys.exit(1)
sentence = Sentence(text)
classifier.predict(sentence)
labels = sentence.labels
return [label.value for label in labels]
def classify(X):
global classifier
if not classifier:
if os.path.exists(model_path):
classifier = TextClassifier.load(model_path)
else:
logger.error(
f"Could not load model at {model_path}.\n"
f"Download model with \"underthesea download {UTSModel.tc_general.value}\".")
sys.exit(1)
sentence = Sentence(X)
classifier.predict(sentence)
labels = sentence.labels
return labels
def sentiment(text):
global classifier
if not classifier:
if os.path.exists(model_path):
classifier = TextClassifier.load(model_path)
else:
logger.error(
f"Could not load model at {model_path}.\n"
f"Download model with \"underthesea download {UTSModel.sa_general.value}\"."
)
sys.exit(1)
sentence = Sentence(text)
classifier.predict(sentence)
label = sentence.labels[0]
if label == "1":
label = "negative"
if label == "0":
label = "positive"
return label
def test_fasttext(self):
corpus: CategorizedCorpus = DataFetcher.load_corpus(
NLPData.AIVIVN2019_SA_SAMPLE)
params = {"lr": 0.01, "epoch": 20, "wordNgrams": 3, "dim": 20}
classifier = TextClassifier(
estimator=TEXT_CLASSIFIER_ESTIMATOR.FAST_TEXT, **params)
model_trainer = ModelTrainer(classifier, corpus)
tmp_model_folder = mkdtemp()
def macro_f1_score(y_true, y_pred):
return f1_score(y_true, y_pred, average='macro')
score = model_trainer.train(tmp_model_folder, scoring=macro_f1_score)
print(score)
classifier = TextClassifier.load(tmp_model_folder)
sentence = Sentence('tuyệt vời')
classifier.predict(sentence)
shutil.rmtree(tmp_model_folder)
print(sentence)
def test(self):
corpus: CategorizedCorpus = DataFetcher.load_corpus(NLPData.UTS2017_BANK_SA_SAMPLE)
# corpus: CategorizedCorpus = DataFetcher.load_corpus(NLPData.AIVIVN2019_SA)
pipeline = Pipeline(
steps=[('features', CountVectorizer(ngram_range=(1, 2), max_features=4000)),
('estimator', OneVsRestClassifier(SVC(kernel='linear', C=0.3)))]
)
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.PIPELINE, pipeline=pipeline, multilabel=True)
model_trainer = ModelTrainer(classifier, corpus)
tmp_model_folder = mkdtemp()
def macro_f1_score(y_true, y_pred):
return f1_score(y_true, y_pred, average='macro')
score = model_trainer.train(tmp_model_folder, scoring=macro_f1_score)
print(score)
classifier = TextClassifier.load(tmp_model_folder)
sentence = Sentence('Dịch vụ tiện dụng quá')
classifier.predict(sentence)
print(sentence)
shutil.rmtree(tmp_model_folder)