def test_accuracy_stream(self):
stream = SEAGenerator(random_state=42)
stream.prepare_for_use()
clf = ARSLVQ(sigma=0.5,
prototypes_per_class=2,
batch_size=5,
decay_rate=0.999)
evaluator = EvaluatePrequential(show_plot=False,
max_samples=20000,
batch_size=5)
evaluator.evaluate(stream, clf, model_names=['ARSLVQ'])
measurements = np.asarray(evaluator.get_measurements()[0])[0]
self.assertTrue(
measurements.get_accuracy() >= 0.84,
msg='Accuracy was {} but has to be greater than 0.84'.format(
measurements.get_accuracy()))
self.assertTrue(
measurements.get_kappa() >= 0.68,
msg='Kappa was {} but has to be greater than 0.68'.format(
measurements.get_kappa()))
def test_accuracy_stream(self):
stream = ConceptDriftStream(stream=SEAGenerator(random_state=112,
noise_percentage=0.1),
drift_stream=SEAGenerator(
random_state=112,
classification_function=1,
noise_percentage=0.1),
random_state=None,
position=20000,
width=50000)
stream.prepare_for_use()
clf = GLVQ(prototypes_per_class=6,
beta=2,
C=None,
decay_rate=0.9,
gradient_descent="SGD")
evaluator = EvaluatePrequential(pretrain_size=1,
show_plot=False,
max_samples=20000,
batch_size=1)
evaluator.evaluate(stream, clf, model_names=['GLVQ'])
measurements = np.asarray(evaluator.get_measurements()[0])[0]
self.assertTrue(
measurements.get_accuracy() >= 0.93,
msg='Accuracy was {} but has to be greater than 0.93'.format(
measurements.get_accuracy()))
self.assertTrue(
measurements.get_kappa() >= 0.84,
msg='Kappa was {} but has to be greater than 0.84'.format(
measurements.get_kappa()))
def test_stream(self):
stream = SEAGenerator(classification_function = 2, random_state = 112, balance_classes = False, noise_percentage = 0.28)
stream.prepare_for_use()
evaluator = EvaluatePrequential(show_plot=False,max_samples=5000,
restart_stream=True,batch_size=10,metrics=['kappa', 'kappa_m', 'accuracy'])
evaluator.evaluate(stream=stream, model=RRSLVQ(prototypes_per_class=4,sigma=10))
measurements = np.asarray(evaluator.get_measurements()[0])[0]
self.assertIsNotNone(eval)
self.assertTrue(measurements.get_accuracy() >= 0.5,
msg='Accuracy was {} but has to be greater than 0.5'.
format(measurements.get_accuracy()))
def test_anwin(self):
anwin = ANWIN(alpha=0.001)
stream = SEAGenerator(classification_function=2,
random_state=112,
balance_classes=False,
noise_percentage=0.28)
stream.prepare_for_use()
stream.restart()
detections, mean = [], []
print("\n--------------------\n")
for i in range(10000):
data = stream.next_sample(10)
batch = data[0][0][0]
mean.append(batch)
anwin.add_element(batch)
if anwin.detected_change():
print("\rIteration {}".format(i))
print("\r ANWINReject Null Hypotheses")
print(np.mean(mean))
mean = []
detections.append(i)
print("----- Number of detections: " + str(len(detections)) + " -----")
self.assertGreaterEqual(len(detections), 5)
def stream_example():
"""Create stream"""
stream = SEAGenerator(noise_percentage=0.1)
stream.prepare_for_use()
"""Init BRSLVQ"""
clf = [
RSLVQ(sigma=5.0, batch_size=1, n_epochs=1),
RSLVQ(sigma=5.0, batch_size=5, n_epochs=1),
RSLVQ(sigma=5.0, batch_size=10, n_epochs=1)
]
"""Evaluate"""
evaluator = EvaluatePrequential(max_samples=10000,
batch_size=100,
show_plot=True)
"""Start evaluation"""
evaluator.evaluate(stream=stream,
model=clf,
model_names=['bs=1', 'bs=5', 'bs=10'])
def test_accuracy_stream(self):
stream = SEAGenerator(random_state=42)
stream.prepare_for_use()
clf = GRLVQ(prototypes_per_class=2, regularization=5.0, beta=2, C=None)
evaluator = EvaluatePrequential(pretrain_size=1,
show_plot=False,
max_samples=20000,
batch_size=1)
evaluator.evaluate(stream, clf, model_names=['GRLVQ'])
measurements = np.asarray(evaluator.get_measurements()[0])[0]
self.assertTrue(
measurements.get_accuracy() >= 0.7,
msg='Accuracy was {} but has to be greater than 0.7'.format(
measurements.get_accuracy()))
self.assertTrue(
measurements.get_kappa() >= 0.3,
msg='Kappa was {} but has to be greater than 0.3'.format(
measurements.get_kappa()))
random_state=112,
balance_classes=False)
mixed_ra = ReoccuringDriftStream(stream=s1,
drift_stream=s2,
random_state=None,
alpha=90.0,
position=2000,
width=100,
pause=1000)
mixed_a = ConceptDriftStream(stream=s1,
drift_stream=s2,
alpha=90.0,
random_state=None,
position=int(study_size / 2),
width=1)
sea_a = ConceptDriftStream(stream=SEAGenerator(random_state=112,
noise_percentage=0.1),
drift_stream=SEAGenerator(random_state=112,
classification_function=2,
noise_percentage=0.1),
alpha=90.0,
random_state=None,
position=int(study_size / 2),
width=1)
sea_ra = ReoccuringDriftStream(stream=SEAGenerator(random_state=112,
noise_percentage=0.1),
drift_stream=SEAGenerator(
random_state=112,
classification_function=2,
noise_percentage=0.1),
alpha=90.0,
# -*- coding: utf-8 -*-
"""
Created on Tue Oct 9 17:08:11 2018
@author: Moritz Heusinger <*****@*****.**>
"""
from skmultiflow.data.sea_generator import SEAGenerator
from rslvq import RSLVQ
from adaptive_rslvqs_batch import RSLVQ as BARSLVQ
from skmultiflow.evaluation.evaluate_prequential import EvaluatePrequential
from arslvq import RSLVQ as ARSLVQ
import numpy as np
from rslvq_stream import RSLVQ as MasterRSLVQ
stream = SEAGenerator()
stream.prepare_for_use()
clf = [
RSLVQ(batch_size=5),
BARSLVQ(gradient_descent='Adadelta',
batch_size=5,
decay_rate=0.9,
sigma=1.0),
RSLVQ(),
MasterRSLVQ(gradient_descent='Adadelta', decay_rate=0.999, sigma=1.0)
]
evaluator = EvaluatePrequential(max_samples=100000,
batch_size=5,
show_plot=True)
def test_sea_generator(test_path):
stream = SEAGenerator(classification_function=2,
random_state=112,
balance_classes=False,
noise_percentage=0.28)
stream.prepare_for_use()
assert stream.n_remaining_samples() == -1
expected_names = ['att_num_0', 'att_num_1', 'att_num_2']
assert stream.feature_names == expected_names
expected_targets = [0, 1]
assert stream.target_values == expected_targets
assert stream.target_names == ['target_0']
assert stream.n_features == 3
assert stream.n_cat_features == 0
assert stream.n_num_features == 3
assert stream.n_targets == 1
assert stream.get_data_info(
) == 'SEA Generator - 1 target(s), 2 classes, 3 features'
assert stream.has_more_samples() is True
assert stream.is_restartable() is True
# Load test data corresponding to first 10 instances
test_file = os.path.join(test_path, 'sea_stream.npz')
data = np.load(test_file)
X_expected = data['X']
y_expected = data['y']
X, y = stream.next_sample()
assert np.alltrue(X[0] == X_expected[0])
assert np.alltrue(y[0] == y_expected[0])
X, y = stream.last_sample()
assert np.alltrue(X[0] == X_expected[0])
assert np.alltrue(y[0] == y_expected[0])
stream.restart()
X, y = stream.next_sample(10)
assert np.alltrue(X == X_expected)
assert np.alltrue(y == y_expected)
assert stream.n_targets == np.array(y).ndim
assert stream.n_features == X.shape[1]
assert 'stream' == stream._estimator_type
expected_info = "SEAGenerator(balance_classes=False, classification_function=2,\n" \
" noise_percentage=0.28, random_state=112)"
assert stream.get_info() == expected_info
def test_correct_init(self):
T_sne_stream_visualizer(stream=SEAGenerator(),
normalize=True,
path=None,
drift_handling='KS',
confidence=0.0001)
def test_wrong_drift_handler(self):
with self.assertRaises(TypeError):
T_sne_stream_visualizer(stream=SEAGenerator(),
drift_handling='TSWIN')
def init_standard_streams_naive_bayes(
): # RBF Stream beinhaltet negative Werte daher muss dieser beim Naive Bayes Algortihmus weggelassen werden
"""Initialize standard data streams
Standard streams are inspired by the experiment settings of
Gomes, Heitor Murilo & Bifet, Albert & Read, Jesse & Barddal, Jean Paul &
Enembreck, Fabrício & Pfahringer, Bernhard & Holmes, Geoff &
Abdessalem, Talel. (2017). Adaptive random forests for evolving data
stream classification. Machine Learning. 1-27. 10.1007/s10994-017-5642-8.
"""
agrawal_a = ConceptDriftStream(stream=AGRAWALGenerator(random_state=112,
perturbation=0.1),
drift_stream=AGRAWALGenerator(
random_state=112,
classification_function=2,
perturbation=0.1),
random_state=None,
alpha=90.0,
position=21000000)
agrawal_a.name = "agrawal_a"
agrawal_g = ConceptDriftStream(stream=AGRAWALGenerator(random_state=112,
perturbation=0.1),
drift_stream=AGRAWALGenerator(
random_state=112,
classification_function=1,
perturbation=0.1),
random_state=None,
position=21000000,
width=1000000)
agrawal_g.name = "agrawal_g"
hyper = HyperplaneGenerator(mag_change=0.001, noise_percentage=0.1)
led_a = ConceptDriftStream(
stream=LEDGeneratorDrift(has_noise=False,
noise_percentage=0.0,
n_drift_features=3),
drift_stream=LEDGeneratorDrift(has_noise=False,
noise_percentage=0.0,
n_drift_features=7),
random_state=None,
alpha=90.0, # angle of change grade 0 - 90
position=21000000,
width=1)
led_a.name = "led_a"
led_g = ConceptDriftStream(stream=LEDGeneratorDrift(has_noise=False,
noise_percentage=0.0,
n_drift_features=3),
drift_stream=LEDGeneratorDrift(
has_noise=False,
noise_percentage=0.0,
n_drift_features=7),
random_state=None,
position=21000000,
width=1000000)
led_g.name = "led_g"
rand_tree = RandomTreeGenerator()
rand_tree.name = "rand_tree"
#rbf_if = RandomRBFGeneratorDrift(change_speed=0.001)
#rbf_if.name = "rbf_if"
#rbf_im = RandomRBFGeneratorDrift(change_speed=0.0001)
#rbf_im.name = "rbf_im"
sea_a = ConceptDriftStream(stream=SEAGenerator(random_state=112,
noise_percentage=0.1),
drift_stream=SEAGenerator(
random_state=112,
classification_function=2,
noise_percentage=0.1),
alpha=90.0,
random_state=None,
position=21000000,
width=1)
sea_a.name = "sea_a"
sea_g = ConceptDriftStream(stream=SEAGenerator(random_state=112,
noise_percentage=0.1),
drift_stream=SEAGenerator(
random_state=112,
classification_function=1,
noise_percentage=0.1),
random_state=None,
position=21000000,
width=1000000)
sea_g.name = "sea_g"
return [agrawal_a, agrawal_g, hyper, led_a, led_g, rand_tree, sea_a, sea_g]