def fit(self, debate_data, y, with_meta_grid_search=True):
for predictor, grid_search in self.stack:
if grid_search:
predictor.fit(debate_data, y, True)
debate_estimator_sample_probs = []
debates_len = len(debate_data)
for index, (test_debate_data, test_labels) in enumerate(zip(debate_data, y)):
train_debate_data = debate_data[0:index] + debate_data[index + 1:debates_len + 1]
train_labels = y[0:index] + y[index + 1:debates_len + 1]
estimator_sample_probs = self.stack_fit_predict(train_debate_data, train_labels, [test_debate_data], [test_labels])
debate_estimator_sample_probs.append(estimator_sample_probs)
predictions_zero = list(zip(*debate_estimator_sample_probs))
predictions_zero = [flatten(x) for x in predictions_zero]
base_predictions = self.convert_to_meta_input(predictions_zero)
y_meta = flatten(y)
self.meta_learner = Pipeline([('log_reg', LogisticRegression(**self.params['meta_params']))])
if with_meta_grid_search:
self.meta_learner = grid_search_log_reg_c(self.meta_learner, base_predictions, y_meta, None, 'macro_recall')
else:
self.meta_learner.fit(base_predictions, y_meta)
self.stack_fit_predict(debate_data, y)
def fit(self, debate_data, y, with_grid_search=False):
y = flatten(y)
train_x = IVectorTransformer(debate_data).features
pipe = Pipeline([('log_reg',
LogisticRegression(**self.params['log_reg']))])
if with_grid_search:
self.model = grid_search_log_reg_c(pipe, train_x, y, debate_data,
'macro_recall')
else:
self.model = pipe
self.model.fit(train_x, y)
def fit(self, debate_data, y, with_grid_search=False):
x_text = flatten([x.texts for x in debate_data])
train_x = x_text
pipe = Pipeline([
('tfidf', TfidfVectorizer(**self.params['tfidf_params'])),
('log_reg', LogisticRegression(**self.params['log_reg']))
])
if with_grid_search:
self.model = grid_search_log_reg_c(pipe, train_x, y, debate_data, 'macro_recall')
else:
self.model = pipe
self.model.fit(train_x, y)
def fit(self, debate_data, y, with_grid_search=True):
debates = [x.debate for x in debate_data]
x_texts = flatten([x.texts for x in debate_data])
x_authors = flatten([x.authors for x in debate_data])
y = flatten(y)
self.ivectors_scaler = MinMaxScaler(
**self.params['min_max_scaler_params'])
self.opensmile_scaler = MinMaxScaler(
**self.params['min_max_scaler_params'])
self.bert_cls_token_scaler = MinMaxScaler(
**self.params['min_max_scaler_params'])
self.tfidf_scaler = MinMaxScaler(
**self.params['min_max_scaler_params'])
self.liwc_scaler = MinMaxScaler(**self.params['min_max_scaler_params'])
self.authors_scaler = MinMaxScaler(
**self.params['min_max_scaler_params'])
self.vectorizer = TfidfVectorizer(**self.params['tfidf_params'])
ivectors = IVectorTransformer(debate_data).features
ivectors = self.ivectors_scaler.fit_transform(ivectors)
opensmile = OpensmileTransformer(debates).features
opensmile = self.opensmile_scaler.fit_transform(opensmile)
bert_cls_token = self.bert_client.encode(x_texts)
bert_cls_token = self.bert_cls_token_scaler.fit_transform(
bert_cls_token)
tfidf_ngrams = self.vectorizer.fit_transform(x_texts).toarray()
tfidf_ngrams = self.tfidf_scaler.fit_transform(tfidf_ngrams)
liwc = LIWCTransformer(x_texts).features
liwc = self.liwc_scaler.fit_transform(liwc)
authors = AuthorsTransformer(x_authors).features
authors = self.authors_scaler.fit_transform(authors)
train_x = merge_features(ivectors, opensmile, bert_cls_token,
tfidf_ngrams, liwc, authors)
pipe = Pipeline([('log_reg',
LogisticRegression(**self.params['log_reg']))])
if with_grid_search:
self.model = grid_search_log_reg_c(pipe, train_x, y, debate_data,
'macro_recall')
else:
self.model = pipe
self.model.fit(train_x, y)
def fit(self, debate_data, y, with_grid_search=False):
x_texts = flatten([x.texts for x in debate_data])
y = flatten(y)
train_x = self.bert_client.encode(x_texts)
pipe = Pipeline([
('log_reg', LogisticRegression(**self.params['log_reg']))
])
if with_grid_search:
self.model = grid_search_log_reg_c(pipe, train_x, y, debate_data, 'macro_recall')
else:
self.model = pipe
self.model.fit(train_x, y)
def fit(self, debate_data, y, with_grid_search=False):
x_texts = flatten([x.texts for x in debate_data])
y = flatten(y)
train_x = TextstatReadabilityTransformer(x_texts).features
pipe = Pipeline([('log_reg',
LogisticRegression(**self.params['log_reg']))])
if with_grid_search:
self.model = grid_search_log_reg_c(pipe, train_x, y, debate_data,
'macro_recall')
else:
self.model = pipe
self.model.fit(train_x, y)
def fit(self, debate_data, y, with_grid_search=False):
x_texts = flatten([x.texts for x in debate_data])
x_authors = flatten([x.authors for x in debate_data])
y = flatten(y)
train_x = merge_features(
LIWCTransformer(x_texts).features,
AuthorsTransformer(x_authors).features)
pipe = Pipeline([('log_reg',
LogisticRegression(**self.params['log_reg']))])
if with_grid_search:
self.model = grid_search_log_reg_c(pipe, train_x, y, debate_data,
'macro_recall')
else:
self.model = pipe
self.model.fit(train_x, y)