def lr_predict(train, test, preprocessing_type):
# separate class label (last column)
train, labels = preprocessing.split_labels(train)
if preprocessing_type == 'smote':
train, labels = preprocessing.apply_smote(train, labels)
classifier = linear_model.LogisticRegression()
validation.cross_validate(classifier, train, labels)
classifier.fit(train, labels)
# test
test, test_labels = preprocessing.split_labels(test)
validation.test(classifier, test, test_labels)
def multiple_balanced_sets():
train, test = preprocessing.prepare_data()
train_list = preprocessing.multiple_balanced_samples(train, 5)
# separate class label (last column)
for i in range(5):
train, labels = preprocessing.split_labels(train_list[i])
classifier = linear_model.LogisticRegression()
validation.cross_validate_set(classifier, train, labels)
validation.cross_validate(classifier, train, labels)
classifier.fit(train, labels)
# test
test, test_labels = preprocessing.split_labels(test)
validation.test(classifier, test, test_labels)
def svm_predict(train, test, preprocessing_type):
# separate class label (last column)
train, labels = preprocessing.split_labels(train)
if preprocessing_type == 'smote':
train, labels = preprocessing.apply_smote(train, labels)
# Classifier
# Class weight parameter: weights positive class more strongly than negative class.
# class_weight={1: 2.61, 0: 0.383}
classifier = svm.SVC(kernel='rbf')
validation.cross_validate(classifier, train, labels)
classifier.fit(train, labels)
# test
test, test_labels = preprocessing.split_labels(test)
validation.test(classifier, test, test_labels)
def rf_predict(train, test, preprocessing_type, results_file):
# separate class label
train, labels = preprocessing.split_labels(train)
if preprocessing_type == 'smote':
train, labels = preprocessing.apply_smote(train, labels)
classifier = ensemble.RandomForestClassifier(class_weight={
0: 0.75,
1: 1.5
},
min_samples_split=40,
n_estimators=15)
classifier.fit(train, labels)
vresult = validation.cross_validate(classifier, train, labels)
# test
test, test_labels = preprocessing.split_labels(test)
tresult = validation.test(classifier, test, test_labels)
# save results
results = []
results.append("low_skew (0=0.75, 1=1.5)")
results.append(40)
results.append(15)
results.append(vresult['roc_auc'])
results.append(vresult['precision'])
results.append(vresult['recall'])
results.append(vresult['f1'])
results.append(vresult['fp'])
results.append(vresult['fn'])
results.append(tresult['roc_auc'])
results.append(tresult['precision'])
results.append(tresult['recall'])
results.append(tresult['f1'])
results.append(tresult['fp'])
results.append(tresult['fn'])
results_file.writerow(results)
def basic():
train, test = preprocessing.prepare_data(True)
with open('nn_resultsfeaturedrop_nosmote.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
# add header
# vary each parameter of random forest
split_train, split_labels = preprocessing.split_labels(train)
split_train, split_labels = preprocessing.apply_smote(
split_train, split_labels)
nn_predict(split_train, split_labels, test, writer, {})
def basic():
train, test = preprocessing.prepare_data(True)
train = train.drop('Amount', axis=1)
test = test.drop('Amount', axis=1)
with open('nn_results.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
split_train, split_labels = preprocessing.split_labels(train)
nn_predict(split_train, split_labels, test, writer)
split_train, split_labels = preprocessing.apply_smote(
split_train, split_labels)
nn_predict(split_train, split_labels, test, writer)
def basic():
train, test = preprocessing.prepare_data(True)
train = train.drop('Amount', axis=1)
test = test.drop('Amount', axis=1)
with open('nn_resultslayer2_nosmote.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
# add header
# vary each parameter of random forest
num_layers_1 = 20
num_layers_2 = 0
num_layers_3 = 0
alpha = 1.e-3
for num_layers_2 in range(1, 30):
if num_layers_2 == 0 and num_layers_3 != 0:
continue
split_train, split_labels = preprocessing.split_labels(train)
params = {'layer_2': num_layers_2}
nn_predict(split_train, split_labels, test, writer, params)
def basic():
train, test = preprocessing.prepare_data(True)
train = train.drop('Amount', axis=1)
test = test.drop('Amount', axis=1)
with open('nn_resultsepsilon_nosmote.csv', 'w') as csvfile:
writer = csv.writer(csvfile)
# add header
# vary each parameter of random forest
num_layers_1 = 20
num_layers_2 = 0
num_layers_3 = 0
alpha = 1.e-3
for epsilon in [1.e0, 1.e-1, 1.e-2, 1.e-3, 1.e-4, 1.e-5]:
if num_layers_2 == 0 and num_layers_3 != 0:
continue
split_train, split_labels = preprocessing.split_labels(train)
params = {'epsilon': epsilon}
nn_predict(split_train, split_labels, test, writer, params)