def test_naive_bayes(test_path):
stream = SEAGenerator(random_state=1)
stream.prepare_for_use()
learner = NaiveBayes()
cnt = 0
max_samples = 5000
y_pred = array('i')
X_batch = []
y_batch = []
y_proba = []
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
X_batch.append(X[0])
y_batch.append(y[0])
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
y_pred.append(learner.predict(X)[0])
y_proba.append(learner.predict_proba(X)[0])
learner.partial_fit(X, y, classes=stream.target_values)
cnt += 1
expected_predictions = array('i', [
1, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 1,
1
])
assert np.alltrue(y_pred == expected_predictions)
test_file = os.path.join(test_path, 'data_naive_bayes_proba.npy')
y_proba_expected = np.load(test_file)
assert np.allclose(y_proba, y_proba_expected)
expected_info = 'NaiveBayes: nominal attributes: [] - '
assert learner.get_info() == expected_info
learner.reset()
learner.fit(X=np.array(X_batch[:4500]), y=np.array(y_batch[:4500]))
expected_score = 0.9378757515030061
assert np.isclose(
expected_score,
learner.score(X=np.array(X_batch[4501:]), y=np.array(y_batch[4501:])))
assert 'estimator' == learner.get_class_type()
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
class LDDDSDA(BaseDistributionDetector):
def __init__(self, batch_size=100, train_size=100, rho=0.1, alpha=0.05, base_learner=NaiveBayes()):
super().__init__()
self.w = batch_size
self.l = base_learner
self.n = train_size
self.alpha = alpha
self.rho = rho
self.trained = False
self.d_train_X, self.d_train_y = [], []
self.d_buffer_X, self.d_buffer_y = [], []
self.reset()
def reset(self):
super().reset()
def add_element(self, X, y):
if self.in_concept_change:
self.reset()
X, y = np.asarray(X), np.asarray(y)
# if X.ndim != 1 or y.ndim != 1:
# raise ValueError("input_value should has one dimension")
if (not self.trained) and len(self.d_train_X) < self.n:
self.d_train_X.append(X)
self.d_train_y.append(y)
if len(self.d_train_X) == self.n:
self.l.partial_fit(np.asarray(self.d_train_X), np.asarray(self.d_train_y))
self.trained = True
return
if len(self.d_train_X) < self.w:
self.d_train_X.append(X)
self.d_train_y.append(y)
return
self.d_buffer_X.append(X)
self.d_buffer_y.append(y)
if len(self.d_buffer_X) < self.w:
return
self.d_train_X, self.d_train_y = self.ldd_dis(np.asarray(self.d_train_X),
np.asarray(self.d_train_y),
np.asarray(self.d_buffer_X),
np.asarray(self.d_buffer_y))
self.l = NaiveBayes()
self.l.fit(self.d_train_X, self.d_train_y)
self.d_train_X = self.d_train_X.tolist()
self.d_train_y = self.d_train_y.tolist()
print(len(self.d_train_X))
self.d_buffer_X = []
self.d_buffer_y = []
return
def predict(self, X):
return self.l.predict(X)
def ldd_dis(self, d1_X, d1_y, d2_X, d2_y):
d = np.append(d1_X, d2_X, axis=0)
d_y = np.append(d1_y, d2_y, axis=0)
d1_dec, d1_sta, d1_inc = [], [], []
d2_dec, d2_sta, d2_inc = [], [], []
kdtree = KDTree(d)
d_knn = []
for i in range(d.shape[0]):
d_knn.append(set(kdtree.query(X=d[i:i+1],
k=int(d.shape[0] * self.rho),
return_distance=False)[0]))
indexes = np.arange(d.shape[0])
np.random.shuffle(indexes)
_d1 = set(indexes[:d1_X.shape[0]])
_d2 = set(indexes[d1_X.shape[0]:])
deltas = []
for i in range(d.shape[0]):
x1 = len(d_knn[indexes[i]] & _d1)
x2 = len(d_knn[indexes[i]] & _d2)
if i < d1_X.shape[0]:
deltas.append(x2 / x1 - 1)
else:
deltas.append(x1 / x2 - 1)
delta_std = np.std(deltas, ddof=1)
theta_dec = stats.norm.ppf(1 - self.alpha, 0, delta_std)
theta_inc = stats.norm.ppf(self.alpha, 0, delta_std)
_d1 = set(np.arange(d1_X.shape[0]))
_d2 = set(np.arange(d1_X.shape[0], d.shape[0]))
for i in range(d.shape[0]):
x1 = len(d_knn[i] & _d1)
x2 = len(d_knn[i] & _d2)
if i < d1_X.shape[0]:
delta = x2 / x1 - 1
if delta < theta_dec:
d1_dec.append(i)
elif delta > theta_inc:
d1_inc.append(i)
else:
d1_sta.append(i)
else:
delta = x1 / x2 - 1
if delta < theta_dec:
d2_dec.append(i)
elif delta > theta_inc:
d2_inc.append(i)
else:
d2_sta.append(i)
if len(d1_dec) == 0 and len(d2_inc) == 0:
return d1_X, d1_y
self.in_concept_change = True
aux = []
if len(d2_dec) != 0:
aux.append(len(d1_inc) / len(d2_dec))
if len(d2_sta) != 0:
aux.append(len(d1_sta) / len(d2_sta))
if len(d2_inc) != 0:
aux.append(len(d1_dec) / len(d2_inc))
k = min(aux)
d2_dec += d1_inc[:int(k * len(d2_dec))]
d2_sta += d1_sta[:int(k * len(d2_sta))]
d2_inc += d1_dec[:int(k * len(d2_inc))]
aux_indexes = d2_inc + d2_sta + d2_dec
r = self.w / len(aux_indexes)
d2_dec = d2_dec[:int(len(d2_dec)*r)]
d2_sta = d2_sta[:int(len(d2_sta)*r)]
d2_inc = d1_inc[:int(len(d2_inc)*r)]
aux_indexes = d2_inc + d2_sta + d2_dec
return d[aux_indexes], d_y[aux_indexes]
