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 my_run(estimator__C, features__lower_pipe__tfidf__ngram_range,
features__with_tone_char__ngram_range,
features__remove_tone__tfidf__ngram_range):
params = locals().copy()
start = time.time()
print(params)
corpus: CategorizedCorpus = DataFetcher.load_corpus(NLPData.VLSP2016_SA)
pipeline = Pipeline(
steps=[(
'features',
FeatureUnion([(
'lower_pipe',
Pipeline([('lower', Lowercase()),
('tfidf',
TfidfVectorizer(
ngram_range=(1, 4), norm='l2', min_df=2))])),
('with_tone_char',
TfidfVectorizer(ngram_range=(1, 6),
norm='l2',
min_df=2,
analyzer='char')),
('remove_tone',
Pipeline([('remove_tone',
RemoveTone()), ('lower', Lowercase()),
('tfidf',
TfidfVectorizer(ngram_range=(1, 4),
norm='l2',
min_df=2))])
), ('emoticons', CountEmoticons())])),
('estimator',
SVC(kernel='linear', C=0.2175, class_weight=None, verbose=True)
)])
pipeline.set_params(**params)
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.PIPELINE,
pipeline=pipeline)
model_trainer = ModelTrainer(classifier, corpus)
tmp_model_folder = mkdtemp()
def negative_f1_score(y_true, y_pred):
score_class_0, score_class_1, score_class_2 = 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)
ex.log_scalar('dev_score', score['dev_score'])
ex.log_scalar('test_score', score['test_score'])
print(time.time() - start)
return score['dev_score']
def run(estimator, features):
corpus: CategorizedCorpus = DataFetcher.load_corpus(NLPData.AIVIVN2019_SA)
pipeline = Pipeline(
steps=[(
'features',
FeatureUnion([(
'lower_tfidf',
Pipeline([('lower', Lowercase()),
('tfidf',
TfidfVectorizer(
ngram_range=(1, 4), norm='l2', min_df=2))])),
('with_tone_char',
TfidfVectorizer(ngram_range=(1, 6),
norm='l2',
min_df=2,
analyzer='char')),
('remove_tone',
Pipeline([('remove_tone',
RemoveTone()), ('lower', Lowercase()),
('tfidf',
TfidfVectorizer(ngram_range=(1, 4),
norm='l2',
min_df=2))]))])),
('estimator',
SVC(kernel='linear', C=0.2175, class_weight=None, verbose=True)
)])
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.PIPELINE,
pipeline=pipeline)
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)
ex.log_scalar('dev_score', score['dev_score'])
ex.log_scalar('test_score', score['test_score'])
return score['test_score']
def my_run(estimator__C, features__lower_pipe__tfidf__max_features,
features__lower_pipe__tfidf__ngram_range,
features__with_tone_char__ngram_range,
features__remove_tone__tfidf__ngram_range):
params = locals().copy()
start = time.time()
print(params)
corpus: CategorizedCorpus = DataFetcher.load_corpus(
NLPData.UTS2017_BANK_TC)
pipeline = Pipeline(steps=[(
'features',
FeatureUnion([('lower_pipe',
Pipeline([('lower', Lowercase()),
('tfidf',
TfidfVectorizer(norm='l2', min_df=2))])),
('with_tone_char',
TfidfVectorizer(norm='l2', min_df=2, analyzer='char')),
('remove_tone',
Pipeline([('remove_tone',
RemoveTone()), ('lower', Lowercase()),
('tfidf', TfidfVectorizer(norm='l2', min_df=2)
)])), ('emoticons', CountEmoticons())])
), ('estimator', SVC(kernel='linear', class_weight=None, verbose=True))])
pipeline.set_params(**params)
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.PIPELINE,
pipeline=pipeline)
model_trainer = ModelTrainer(classifier, corpus)
tmp_model_folder = "tmp/tmp_model"
def micro_f1_score(y_true, y_pred):
return f1_score(y_true, y_pred, average='micro')
score = model_trainer.train(tmp_model_folder, scoring=micro_f1_score)
tmp_files = listdir(tmp_model_folder)
for file in tmp_files:
if "gitignore" in file:
continue
os.remove(f"{tmp_model_folder}/{file}")
ex.log_scalar('dev_score', score['dev_score'])
ex.log_scalar('test_score', score['test_score'])
print(f"Time: {round(time.time() - start, 2)} s")
return score['dev_score']
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 my_run(features__max_df, features__ngram_range):
params = locals().copy()
start = time.time()
print(params)
from languageflow.data_fetcher import DataFetcher
corpus: CategorizedCorpus = DataFetcher.load_corpus(NLPData.VNTC)
pipeline = Pipeline(steps=[('features',
TfidfVectorizer()), ('estimator',
LinearSVC())])
pipeline.set_params(**params)
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.PIPELINE,
pipeline=pipeline)
model_trainer = ModelTrainer(classifier, corpus)
tmp_model_folder = mkdtemp()
def micro_f1_score(y_true, y_pred):
return f1_score(y_true, y_pred, average='micro')
score = model_trainer.train(tmp_model_folder, scoring=micro_f1_score)
ex.log_scalar('dev_score', score['dev_score'])
ex.log_scalar('test_score', score['test_score'])
print(time.time() - start)
return score['dev_score']
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)
from languageflow.data_fetcher import DataFetcher, NLPData
from languageflow.models.text_classifier import TextClassifier, TEXT_CLASSIFIER_ESTIMATOR
from languageflow.trainers.model_trainer import ModelTrainer
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.metrics import f1_score
from sklearn.pipeline import Pipeline
from sklearn.svm import SVC
corpus: CategorizedCorpus = DataFetcher.load_corpus(NLPData.AIVIVN2019_SA)
pipeline = Pipeline(
steps=[('features', CountVectorizer(ngram_range=(
1, 2), max_features=4000)), ('estimator',
SVC(kernel='linear', C=0.3))])
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.PIPELINE,
pipeline=pipeline)
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)
norm='l2',
min_df=2,
analyzer='char')),
('remove_tone',
Pipeline([('remove_tone',
RemoveTone()), ('lower', Lowercase()),
('tfidf',
TfidfVectorizer(
ngram_range=remove_tone__tfidf__ngram_range,
norm='l2',
min_df=2))])
), ('emoticons',
CountEmoticons())])),
('estimator',
SVC(kernel='linear', C=estimator_C, class_weight=None, verbose=True))
])
print("\n\n>>> Start training")
classifier = TextClassifier(estimator=TEXT_CLASSIFIER_ESTIMATOR.PIPELINE,
pipeline=pipeline)
model_trainer = ModelTrainer(classifier, corpus)
def macro_f1_score(y_true, y_pred):
return f1_score(y_true, y_pred, average='macro')
model_trainer.train(model_folder, scoring=macro_f1_score)
print("\n\n>>> Finish training")
print(f"Your model is saved in {model_folder}")
print(f"Running in {time() - start}")