def multinomial_nb(which_comments):
print("=> Multinomial Bayes naive classifier")
data_frame = pd.read_excel(r'preprocessed_data\all_comments.xlsx')
sss = StratifiedShuffleSplit(n_splits=10, test_size=0.1)
index = 1
average = 0
for train_index, test_index in sss.split(data_frame['Comment'],
data_frame['Type']):
preprocessing.preprocess_train_test_data(train_index, test_index)
mnb = MultinomialNB()
mnb.fit(preprocessing.get_data_set(), preprocessing.get_data_labels())
score = f1_score(preprocessing.get_test_labels(),
mnb.predict(preprocessing.get_test_set()),
average='weighted')
average = average + score
print("Score {}.: {:.2f}%".format(index, score * 100), end=" ")
if index == 5:
print()
index += 1
print()
print("Average: {:.2f}%".format(average / 10 * 100))
def compare_regularisation_functions(data_frame, rf, c=1.0):
sss = StratifiedShuffleSplit(n_splits=10, test_size=0.1)
index = 1
average = 0
for train_index, test_index in sss.split(data_frame['Comment'], data_frame['Type']):
preprocessing.preprocess_train_test_data(train_index, test_index)
if rf == 'l1':
solver = 'saga'
else:
solver = 'lbfgs'
lr = LogisticRegression(penalty=rf, C=c, solver=solver, max_iter=15000)
lr.fit(preprocessing.get_data_set(), preprocessing.get_data_labels())
score = f1_score(preprocessing.get_test_labels(), lr.predict(preprocessing.get_test_set()), average='weighted')
average = average + score
print("Score({}) {}.: {:.2f}%".format(rf.upper(), index, score * 100), end=" ")
if index == 5:
print()
index += 1
print()
print("Average: {:.2f}%".format(average / 10 * 100))
def optimize_c_parameter():
models_param = {
'max_iter': [15000],
'C': [0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000]
}
nested_cv_search = NestedCV(model=LogisticRegression(), params_grid=models_param,
outer_kfolds=5, inner_kfolds=5,
cv_options={'sqrt_of_score': True, 'randomized_search_iter': 30})
nested_cv_search.fit(preprocessing.get_data_set(), preprocessing.get_data_labels())
optimized_c_value = np.mean([d['C'] for d in nested_cv_search.best_inner_params_list])
print("Optimized C: {:.3f}".format(optimized_c_value))
def compare_regularisation_functions(data_frame, rf, c=1):
sss = StratifiedShuffleSplit(n_splits=10, test_size=0.1)
index = 1
average = 0
for train_index, test_index in sss.split(data_frame['Comment'],
data_frame['Type']):
preprocessing.preprocess_train_test_data(train_index, test_index)
svc = LinearSVC(penalty=rf, C=c, dual=rf == 'l2', max_iter=15000)
svc.fit(preprocessing.get_data_set(), preprocessing.get_data_labels())
score = f1_score(preprocessing.get_test_labels(),
svc.predict(preprocessing.get_test_set()),
average='weighted')
average = average + score
print("Score({}) {}.: {:.2f}%".format(rf.upper(), index, score * 100),
end=" ")
if index == 5:
print()
index += 1
print()
print("Average: {:.2f}%".format(average / 10 * 100))