def test_warm_start(self):
X = np.array([0.50, 0.75, 1.00])
y = np.array([0, 1, 2])
X = X[:, np.newaxis]
clf = LogisticRegression(data_norm=1.0, warm_start=True)
clf.fit(X, y)
self.assertIsNotNone(clf.fit(X, y))
def test_one_class(self):
X = [[1]]
y = [0]
clf = LogisticRegression(data_norm=1)
with self.assertRaises(ValueError):
clf.fit(X, y)
def test_large_norm(self):
X = np.array(
[0.50, 0.75, 1.00, 1.25, 1.50, 1.75, 1.75, 2.00, 2.25, 2.50, 2.75, 3.00, 3.25, 3.50, 4.00, 4.25, 4.50, 4.75,
5.00, 5.50])
y = np.array([0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1])
X = X[:, np.newaxis]
clf = LogisticRegression(data_norm=1.0)
with self.assertWarns(PrivacyLeakWarning):
clf.fit(X, y)
def test_no_params(self):
clf = LogisticRegression()
X = np.array(
[0.50, 0.75, 1.00, 1.25, 1.50, 1.75, 1.75, 2.00, 2.25, 2.50, 2.75, 3.00, 3.25, 3.50, 4.00, 4.25, 4.50, 4.75,
5.00, 5.50])
y = np.array([0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1])
X = X[:, np.newaxis]
with self.assertWarns(PrivacyLeakWarning):
clf.fit(X, y)
def test_sample_weight_warning(self):
X = np.array([
0.50, 0.75, 1.00, 1.25, 1.50, 1.75, 1.75, 2.00, 2.25, 2.50, 2.75,
3.00, 3.25, 3.50, 4.00, 4.25, 4.50, 4.75, 5.00, 5.50
])
y = np.array(
[0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1])
X = X[:, np.newaxis]
clf = LogisticRegression(data_norm=5.5)
with self.assertWarns(DiffprivlibCompatibilityWarning):
clf.fit(X, y, sample_weight=np.ones_like(y))
def test_trinomial(self):
X = np.array([0.50, 0.75, 1.00])
y = np.array([0, 1, 2])
X = X[:, np.newaxis]
clf = LogisticRegression(data_norm=1.0)
self.assertIsNotNone(clf.fit(X, y))
def test_simple(self):
X = np.array(
[0.50, 0.75, 1.00, 1.25, 1.50, 1.75, 1.75, 2.00, 2.25, 2.50, 2.75, 3.00, 3.25, 3.50, 4.00, 4.25, 4.50, 4.75,
5.00, 5.50])
y = np.array([0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1])
X = X[:, np.newaxis]
X -= 3.0
X /= 2.5
clf = LogisticRegression(epsilon=2, data_norm=1.0)
clf.fit(X, y)
# print(clf.predict(np.array([0.5, 2, 5.5])))
self.assertIsNotNone(clf)
self.assertFalse(clf.predict(np.array([(0.5 - 3) / 2.5]).reshape(-1, 1)))
self.assertTrue(clf.predict(np.array([(5.5 - 3) / 2.5]).reshape(-1, 1)))
def test_same_results(self):
from sklearn import datasets
from sklearn.model_selection import train_test_split
from sklearn import linear_model
dataset = datasets.load_iris()
X_train, X_test, y_train, y_test = train_test_split(dataset.data, dataset.target, test_size=0.2)
clf = LogisticRegression(data_norm=12, epsilon=float("inf"))
clf.fit(X_train, y_train)
predict1 = clf.predict(X_test)
clf = linear_model.LogisticRegression(solver="lbfgs", multi_class="ovr")
clf.fit(X_train, y_train)
predict2 = clf.predict(X_test)
self.assertTrue(np.all(predict1 == predict2))
def test_large_norm(self):
X = np.array([
0.50, 0.75, 1.00, 1.25, 1.50, 1.75, 1.75, 2.00, 2.25, 2.50, 2.75,
3.00, 3.25, 3.50, 4.00, 4.25, 4.50, 4.75, 5.00, 5.50
])
y = np.array(
[0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1])
X = X[:, np.newaxis]
clf = LogisticRegression(data_norm=1.0)
self.assertIsNotNone(clf.fit(X, y))
def test_accountant(self):
from diffprivlib.accountant import BudgetAccountant
acc = BudgetAccountant()
X = np.array([
0.50, 0.75, 1.00, 1.25, 1.50, 1.75, 1.75, 2.00, 2.25, 2.50, 2.75,
3.00, 3.25, 3.50, 4.00, 4.25, 4.50, 4.75, 5.00, 5.50
])
y = np.array(
[0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1])
X = X[:, np.newaxis]
X -= 3.0
X /= 2.5
clf = LogisticRegression(epsilon=2, data_norm=1.0, accountant=acc)
clf.fit(X, y)
self.assertEqual((2, 0), acc.total())
with BudgetAccountant(3, 0) as acc2:
clf = LogisticRegression(epsilon=2, data_norm=1.0)
clf.fit(X, y)
self.assertEqual((2, 0), acc2.total())
with self.assertRaises(BudgetError):
clf.fit(X, y)
def test_different_results(self):
from sklearn import datasets
from sklearn import linear_model
from sklearn.model_selection import train_test_split
dataset = datasets.load_iris()
X_train, X_test, y_train, y_test = train_test_split(dataset.data,
dataset.target,
test_size=0.2)
clf = LogisticRegression(data_norm=12)
clf.fit(X_train, y_train)
predict1 = clf.predict(X_test)
clf = LogisticRegression(data_norm=12)
clf.fit(X_train, y_train)
predict2 = clf.predict(X_test)
clf = linear_model.LogisticRegression(solver="lbfgs",
multi_class="ovr")
clf.fit(X_train, y_train)
predict3 = clf.predict(X_test)
self.assertFalse(np.all(predict1 == predict2))
self.assertFalse(
np.all(predict3 == predict1) and np.all(predict3 == predict2))
# Convert dataframe to numpy array
X_train = np.array(X_train, dtype=np.float32)
X_test = np.array(X_test, dtype=np.float32)
y_train = np.array(y_train, dtype=np.int32)
y_test = np.array(y_test, dtype=np.int32)
# # define list of epsilon
epsilons = [1] # np.logspace(-2, 2, 50)
acc_w_dp = list()
# output = open("ibm_mnist_dp_data_norm_18_70k_images.txt", "w+")
for epsilon in epsilons:
logreg_w_dp = LogisticRegression(epsilon=epsilon,
data_norm=500,
max_iter=1000)
# # l2 norm = sqrt(all_columns), there are 64 columns -> data_norm = 8
logreg_w_dp.fit(X_train, y_train.ravel())
# score_dp = logreg_w_dp.score(X_test, y_test)
accuracy, recall, precision, auc = utils.predict_score(
logreg_w_dp, X_test, y_test)
# output.write("%.3f \t %.3f \t %.3f \n" % (epsilon, recall, precision))
print(accuracy, recall, precision, auc)
# acc_w_dp.append(score_dp)
# output.write("%.3f \t %.3f\n" % (epsilon, score_dp))
print('Total time = ', time.perf_counter() - start_time, ' seconds')
# plt.plot(epsilons, acc_w_dp)
# output.close()
# plt.show()
df_X_train_col_filtered = df_X_train[keep_cols]
df_X_tests_col_filtered = [df[keep_cols] for df in df_X_tests]
# Modify sample values without modifying the shape of the dataset
X_train_processed, X_tests_processed = sample_preprocessing(
df_X_train_col_filtered, df_X_tests_col_filtered)
return X_train_processed, X_tests_processed, list(keep_cols)
X_train, [X_test, X_eval], keep_cols = preprocessing(df_X_train,
[df_X_test, df_X_eval])
print(X_train.shape, X_test.shape)
#epsilons = [float('inf')]
epsilons = [5] #np.logspace(-2, 2, 50)
# output = open("aps_dataset_ibm_dp_data_norm_1000_70k_images.txt", "w+")
for epsilon in epsilons:
# can't make a new scoring function, this option is not allow in IBM diffprivlib
logreg_w_dp = LogisticRegression(epsilon=epsilon,
data_norm=500,
max_iter=20)
logreg_w_dp.fit(X_train, y_train)
# score_dp = logreg_w_dp.score(X_test, y_test)
# print('Accuracy = ', score_dp)
accuracy, recall, precision, auc = utils.predict_score(
logreg_w_dp, X_test, y_test)
# output.write("%.3f \t %.3f \t %.3f \n" % (epsilon, recall, precision))
print(accuracy, recall, precision, auc)
print('Total time = ', time.perf_counter() - start_time, ' seconds')
# output.close()