def predict(m, fn):
""" returns a list of 0s and 1s, corresponding to the lines in the specified file.
:param m: the trained model
:type m: BaseClassifier
:param fn: the full path to a file in the same format as the test set
:type fn: str
:return: a list of 0s and 1s, corresponding to the lines in the specified file
:rtype: list
"""
dm_test = DataManager(fn, is_train=False, algorithm_name=m.clf_name)
dm_test.run_first_preprocessing_flow()
X_test, _ = dm_test.complete_preprocessing_flow()
return m.predict(X_test)
def train_best_model():
""" training a XGBoost classifier from scratch with it's best hyper-parameters.
:return: a trained XGBoost classifier built with the best performing hyper-parameters.
:rtype: XGBoostClassifier
"""
clf = XGBoostClassifier()
clf.set_best_hyper_parameters() # sets the best hyper-parameters that were found in the optimization stage.
dm_train = DataManager('trump_train.tsv', is_train=True, algorithm_name=clf.clf_name)
# we had two stages of the preprocessing flow because of the padding of the text features used in the NN algorithms:
dm_train.run_first_preprocessing_flow()
X_train, y_train = dm_train.complete_preprocessing_flow()
# fit the classifier using all of the training data:
clf.fit(X_train, y_train)
return clf
for clf in classifiers:
print(f'Initialized classifier {clf.clf_name}')
evaluator = Evaluator(clf)
print('Initialized evaluator')
dm_train = DataManager('trump_train.tsv',
is_train=True,
algorithm_name=clf.clf_name)
dm_test = DataManager('trump_test.tsv',
is_train=False,
algorithm_name=clf.clf_name)
print('Initialized Data manager')
dm_train.run_first_preprocessing_flow()
dm_test.run_first_preprocessing_flow()
fix_max_length(dm_train, dm_test)
X_train, y_train = dm_train.complete_preprocessing_flow()
X_test, _ = dm_test.complete_preprocessing_flow()
if clf.clf_name == 'RNN':
clf.sequence_length = dm_train.max_length
print('Cleaned X, y')
best_score = evaluator.optimize_hyper_parameters(X_train,
y_train,
cv=3,
scoring='f1')
print('Done')
print(
f'The classifier {clf.clf_name} gave us a best 3-fold score of: {best_score}'
)
print(clf.hyper_parameters)
best_hyper_parameters_solver[clf.clf_name] = clf.hyper_parameters
best_score_solver[clf.clf_name] = best_score