def run_model(model_type,
in_notebook=False,
is_verbose=True,
return_model=False):
if model_type == 'LR':
print(
"\nPredicting Speaker Stance - Multi Label Logistic Regression Baseline Model "
)
model = baseline.Model()
elif model_type == 'MultiClassLR':
print(
"\nPredicting Speaker Stance - Multi Class Logistic Regression Baseline Model "
)
model = multiclassbaseline.Model()
elif model_type == 'MTNN':
print("\nPredicting Speaker Stance - Multi Task Model ")
model = MultiTaskNN.My_Model(is_verbose=is_verbose)
elif model_type == 'MLP':
print("\nPredicting Speaker Stance - Multi Layer Perceptron Model ")
model = MultiLayerPercep.Model(is_verbose=is_verbose)
elif model_type == 'FastText':
print("\nPredicting Speaker Stance - FastText Model ")
model = FastText.My_Model(is_verbose=is_verbose)
elif model_type == 'MultiClassFastText':
print("\nPredicting Speaker Stance - FastText Model ")
model = FastTextMultiClass.My_Model(is_verbose=is_verbose)
elif model_type == 'FastTextMT':
print("\nPredicting Speaker Stance - FastText Multi Task Model ")
model = FastTextMultiTask.My_Model(is_verbose=is_verbose)
elif model_type == 'FastTextAux':
print("\nPredicting Speaker Stance - FastText Aux Task Model ")
model = FastTextAuxTask.My_Model(is_verbose=is_verbose)
else:
raise ValueError(
'Unknown Model Type: {} not supported option'.format(model_type))
print("Loading Data")
if in_notebook is True:
data = Dataset(in_notebook=True)
else:
data = Dataset()
print("Training Model")
model.train(data.train_set())
y, y_pred = model.test(data.test_set())
if return_model is True:
return model, y, y_pred
return y, y_pred
def run_model(model_type,
domain,
is_verbose=True,
probs=False,
get_history=False):
if model_type == 'logistic_regression':
model = LR(domain, probs=probs)
elif model_type == 'fast_text':
model = FastText(domain, probs=probs, is_verbose=is_verbose)
elif model_type == 'mlp':
model = MLP(domain, is_verbose=is_verbose)
else:
raise ValueError('incorrect model type choice')
print("Loading Data")
data = Dataset()
print("Training Model")
hist = model.train(data.train_set())
if get_history is True:
return hist
else:
y, y_pred = model.test(data.test_set())
return y, y_pred
def run_model(language, dataset_name, evaluation_split, detailed_report):
"""Trains and tests the CWI model for a particular dataset of a particular language. Reports results.
Args:
language: The language of the dataset.
dataset_name: The name of the dataset (all files should have it).
evaluation_split: The split of the data to use for evaluating the performance of the model (dev or test).
detailed_report: Whether to display a detailed report or just overall score.
"""
print("\nModel for {} - {}.".format(language, dataset_name))
data = Dataset(language, dataset_name)
#The code below is used for creating unigram probability csv files
# if (language == 'spanish'):
# corpus_words = nltk.corpus.cess_esp.words()
# unigram_counts = Counter(corpus_words)
# total_words = len(corpus_words)
# def calc_unigram_prob(unigram_counts, total_words):
# u_prob = {} #defaultdict
# for word in unigram_counts:
# u_prob[word] = unigram_counts[word]/total_words
# return u_prob
# def save_to_file(u_prob,file_name):
# w = csv.writer(open(file_name, "w"))
# for word, prob in u_prob.items():
# w.writerow([word, prob])
# print('calc unigram prob: ')
# u_prob = calc_unigram_prob(unigram_counts, total_words)
# print('saving file')
# save_to_file(u_prob, 'data/external/spanish_u_prob.csv')
# kdfjei
baseline = MonolingualCWI(language)
baseline.train(data.train_set())
if evaluation_split in ["dev", "both"]:
print("\nResults on Development Data")
predictions_dev = baseline.predict(data.dev_set())
gold_labels_dev = data.dev_set()['gold_label']
print(
report_binary_score(gold_labels_dev, predictions_dev,
detailed_report))
if evaluation_split in ["test", "both"]:
print("\nResults on Test Data")
predictions_test = baseline.predict(data.test_set())
gold_labels_test = data.test_set()['gold_label']
print(
report_binary_score(gold_labels_test, predictions_test,
detailed_report))
print()
def run_model(test_language, evaluation_split, detailed_report):
""" Trains the CWI model in all languages but one. Tests on all datasets of
a particular language. Reports results.
Args:
test_language: The language of the dataset to use for testing.
evaluation_split: The split of the data to use for
evaluating the performance of the model (dev or
test).
detailed_report: Whether to display a detailed report or just overall
score.
"""
# collect the training data for all the languages but one
train_data = []
for language, datasets_names in datasets_per_language.items():
if language != test_language:
for dataset_name in datasets_names:
data = Dataset(language, dataset_name)
lang_train_set = data.train_set()
if lang_train_set is None:
print("No training data found for language {}.".format(
language))
else:
train_data.append(lang_train_set)
train_data = pd.concat(train_data)
# train the CWI model
cwi_model = CrosslingualCWI(list(datasets_per_language.keys()))
cwi_model.train(train_data)
# test the model
test_datasets = datasets_per_language[test_language]
for dataset_name in test_datasets:
data = Dataset(test_language, dataset_name)
print("\nTesting on {} - {}.".format(test_language, dataset_name))
if evaluation_split in ["dev", "both"]:
print("\nResults on Development Data")
predictions_dev = cwi_model.predict(data.dev_set())
gold_labels_dev = data.dev_set()['gold_label']
print(
report_binary_score(gold_labels_dev, predictions_dev,
detailed_report))
if evaluation_split in ["test", "both"]:
print("\nResults on Test Data")
predictions_test = cwi_model.predict(data.test_set())
gold_labels_test = data.test_set()['gold_label']
print(
report_binary_score(gold_labels_test, predictions_test,
detailed_report))
print()
def run_model():
print("\nPredicting Speaker Stance - Baseline Model ")
print("Loading Data")
data = Dataset()
model = Model()
print("Training Model")
model.train(data.train_set())
print("\nResults on Test Data")
y, y_pred = model.test(data.test_set())
print(report_scores(y, y_pred))
def run_model(in_notebook=False):
print("\nPredicting Speaker Stance - Baseline Model ")
print("Loading Data")
if in_notebook is True:
data = Dataset(in_notebook=True)
else:
data = Dataset()
model = Model()
print("Training Model")
model.train(data.train_set())
print("\nResults on Test Data")
y, y_pred = model.test(data.test_set())
print(report_scores(y, y_pred))
def run_model(language, dataset_name, evaluation_split, detailed_report, ablate):
"""Trains and tests the CWI model for a particular dataset of a particular language. Reports results.
Args:
language: The language of the dataset.
dataset_name: The name of the dataset (all files should have it).
evaluation_split: The split of the data to use for evaluating the performance of the model (dev or test).
detailed_report: Whether to display a detailed report or just overall score.
"""
score_only = True if ablate else False
data = Dataset(language, dataset_name)
#The code below is used for creating unigram probability csv files
#corp = nltk.corpus.ConllCorpusReader('.', 'tiger_release_aug07.corrected.16012013.conll09',
# ['ignore', 'words', 'ignore', 'ignore', 'ignore'],
#encoding='utf-8')
# filename = 'europarl-v7.fr-en.fr'
# file = open(filename, mode='rt', encoding='utf-8')
# corpus_words = []
# for line in file:
# #print(line)
# corpus_words += line.strip(',').strip('.').split()
# #print(corpus_words)
# #corpus_words = corp.words()
# unigram_counts = Counter(corpus_words)
# total_words = len(corpus_words)
# def calc_unigram_prob(unigram_counts, total_words):
# u_prob = {} #defaultdict
# for word in unigram_counts:
# u_prob[word] = unigram_counts[word]/total_words
# return u_prob
# def save_to_file(u_prob,file_name):
# w = csv.writer(open(file_name, "w"))
# for word, prob in u_prob.items():
# w.writerow([word, prob])
# print('calc unigram prob: ')
# u_prob = calc_unigram_prob(unigram_counts, total_words)
# print('saving file')
# save_to_file(u_prob, 'data/external/french_u_prob.csv')
# hgiuyo
baseline = MonolingualCWI(language, ablate)
baseline.train(data.train_set())
if evaluation_split in ["dev", "both"]:
if not score_only:
print("\nResults on Development Data")
predictions_dev = baseline.predict(data.dev_set())
gold_labels_dev = data.dev_set()['gold_label']
print(report_binary_score(gold_labels_dev, predictions_dev, detailed_report, score_only))
if evaluation_split in ["test", "both"]:
if not score_only:
print("\nResults on Test Data")
predictions_test = baseline.predict(data.test_set())
gold_labels_test = data.test_set()['gold_label']
print(report_binary_score(gold_labels_test, predictions_test, detailed_report, score_only))
if not score_only:
print()
y = self.target_pipe.fit_transform(trainset)
num_features = X.shape[1]
self.model = self.build_net(num_features)
self.model.fit(X, y, epochs=500, batch_size=32, verbose=self.is_verbose)
def test(self, testset):
X = self.feature_pipe.transform(testset)
y = self.mlb.transform(testset)
print(y.shape)
y_pred_raw = self.model.predict(X)
y_pred = np.column_stack(y_pred_raw) # join the raw outputs into format for sklearn scoring
print(y_pred.shape)
threshold = 0.5
for row in y_pred:
for i, val in enumerate(row):
if val > threshold:
row[i] = 1
elif val < threshold:
row[i] = 0
print(y_pred[0])
return y, y_pred
if __name__ == '__main__':
from src.data.dataset import Dataset
data = Dataset()
model = Model()
model.train(data.train_set())
model.test(data.test_set())
def run_model(selective_testing, translate, test_language, evaluation_split,
detailed_report):
""" Trains the CWI model in all languages but one. Tests on all datasets of
a particular language. Reports results.
Args:
test_language: The language of the dataset to use for testing.
evaluation_split: The split of the data to use for
evaluating the performance of the model (dev or
test).
detailed_report: Whether to display a detailed report or just overall
score.
"""
# collect the training data for all the languages but one
train_data = []
if selective_testing == 'ESG':
for language, datasets_names in datasets_per_language.items():
if language != test_language:
for dataset_name in datasets_names:
data = Dataset(language, dataset_name)
lang_train_set = data.train_set()
if lang_train_set is None:
print("No training data found for language {}.".format(
language))
else:
train_data.append(lang_train_set)
train_data = pd.concat(train_data)
else:
train_data = pd.DataFrame()
if 'E' in selective_testing:
train_data = pd.concat([
train_data,
Dataset('english', 'News').train_set(),
Dataset('english', 'WikiNews').train_set(),
Dataset('english', 'Wikipedia').train_set()
])
if 'S' in selective_testing:
train_data = pd.concat(
[train_data,
Dataset('spanish', 'Spanish').train_set()])
if 'G' in selective_testing:
train_data = pd.concat(
[train_data,
Dataset('german', 'German').train_set()])
# if selective_testing == 'ES':
# train_data = pd.concat([Dataset('english','News').train_set(),
# Dataset('english','WikiNews').train_set(),
# Dataset('english','Wikipedia').train_set(),
# Dataset('spanish','Spanish').train_set()])
# elif selective_testing == 'EG':
# train_data = pd.concat([Dataset('english','News').train_set(),
# Dataset('english','WikiNews').train_set()
# ,Dataset('english','Wikipedia').train_set(),
# Dataset('german','German').train_set()])
# elif selective_testing == 'E':
# train_data = pd.concat([Dataset('english','News').train_set(),
# Dataset('english','WikiNews').train_set()
# ,Dataset('english','Wikipedia').train_set()])
# elif selective_testing == 'G':
# train_data = pd.concat([Dataset('german','German').train_set()])
#
# elif selective_testing == 'S':
# train_data = pd.concat([Dataset('spanish','Spanish').train_set()])
# else:
# train_data = pd.concat([Dataset('spanish','Spanish').train_set(),
# Dataset('german','German').train_set()])
# train the CWI model
cwi_model = CrosslingualCWI(list(datasets_per_language.keys()))
cwi_model.train(train_data)
# test the model
test_datasets = datasets_per_language[test_language]
for dataset_name in test_datasets:
data = Dataset(test_language, dataset_name)
print("\nTesting on {} - {}.".format(test_language, dataset_name))
if evaluation_split in ["dev", "both"]:
print("\nResults on Development Data")
if test_language == 'french':
print("\nNo Dev Data for French, skipping...")
else:
predictions_dev = cwi_model.predict(data.dev_set())
gold_labels_dev = data.dev_set()['gold_label']
print(
report_binary_score(gold_labels_dev, predictions_dev,
detailed_report))
if evaluation_split in ["test", "both"]:
print("\nResults on Test Data")
data.translate = translate
predictions_test = cwi_model.predict(data.test_set())
gold_labels_test = data.test_set()['gold_label']
print(
report_binary_score(gold_labels_test, predictions_test,
detailed_report))
print()