def test_reoccuring(self):
s1 = MIXEDGenerator(classification_function = 1, random_state= 112, balance_classes = False)
s2 = MIXEDGenerator(classification_function = 0, random_state= 112, balance_classes = False)
stream = ReoccuringDriftStream(stream=s1,
drift_stream=s2,
random_state=None,
alpha=90.0, # angle of change grade 0 - 90
position=2000,
width=500)
stream.prepare_for_use()
evaluator = EvaluatePrequential(show_plot=False,batch_size=10,
max_samples=1000,
metrics=['accuracy', 'kappa_t', 'kappa_m', 'kappa'],
output_file=None)
eval = evaluator.evaluate(stream=stream, model=OzaBaggingAdwin(base_estimator=KNN()))
measurements = np.asarray(evaluator.get_measurements()[0])[0]
self.assertIsNotNone(eval)
self.assertTrue(measurements.get_accuracy() >= 0.6,
msg='Accuracy was {} but has to be greater than 0.6'.
format(measurements.get_accuracy()))
def test_alpha(self):
with self.assertRaises(ValueError):
ReoccuringDriftStream(alpha=-1)
with self.assertRaises(ValueError):
ReoccuringDriftStream(alpha=91)
ReoccuringDriftStream(alpha=1)
def test_pause(self):
with self.assertRaises(ValueError):
ReoccuringDriftStream(pause=-1)
ReoccuringDriftStream(pause=1)
def test_width(self):
with self.assertRaises(ValueError):
ReoccuringDriftStream(width=-1)
ReoccuringDriftStream(width=1)
def test_missing_streams():
grid = {
"sigma": np.append(1, np.arange(2, 11, 2)),
"prototypes_per_class": np.append(1, np.arange(2, 11, 2)),
"gamma": np.array([0.7, 0.9, 0.999]),
"confidence": np.array([0.01, 0.001]),
"window_size": np.array([800])
}
clf = ARSLVQ()
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=250000,
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=250000,
width=50000)
led_g.name = "led_g"
led_fa = ReoccuringDriftStream(
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=2000,
width=1)
led_fg = ReoccuringDriftStream(
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=2000,
width=1000)
covertype = FileStream(os.path.realpath('covtype.csv')) # Label failure
covertype.name = "covertype"
poker = FileStream(os.path.realpath('poker.csv')) # label failure
poker.name = "poker"
airlines = FileStream(os.path.realpath('airlines.csv'))
airlines.name = "airport"
gs = GridSearch(clf=clf, grid=grid, max_samples=50000)
gs.streams = [led_a, led_g, led_fa, led_fg, covertype, poker, airlines]
gs.search()
gs.save_summary()
def test_led():
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=250000,
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=250000,
width=50000)
led_g.name = "led_g"
led_fa = ReoccuringDriftStream(
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=2000,
width=1)
led_fg = ReoccuringDriftStream(
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=2000,
width=1000)
np = 2
sigma = 3
clfs = [
ARSLVQ(prototypes_per_class=np,
sigma=sigma,
confidence=0.0001,
window_size=1500),
OzaBaggingAdwin(),
AdaptiveRandomForest(),
HAT(),
RSLVQ(prototypes_per_class=np, sigma=sigma),
SAMKNN()
]
cv = CrossValidation(clfs=clfs, parallel=1)
cv.streams = [led_a, led_g, led_fa, led_fg]
cv.search()
cv.save_summary()
from skmultiflow.data import SEAGenerator
from bix.data.reoccuringdriftstream import ReoccuringDriftStream
import numpy as np
from bix.detectors.kswin import KSWIN
from sklearn.random_projection import SparseRandomProjection
"""Look into if a CD Detector detects the same drifts in high and in low dim space"""
n_samples = 20000
drift_detected_high = []
drift_detected_low = []
stream = ReoccuringDriftStream(SEAGenerator(classification_function=0),
SEAGenerator(classification_function=2),
position=2000,
width=1000,
pause=1000)
stream.prepare_for_use()
stream.next_sample()
"""Init KSWIN for every dimension"""
kswin_high = [
KSWIN(alpha=1e-5, w_size=300, stat_size=30, data=None)
for i in range(10000)
]
kswin_low = [
KSWIN(alpha=1e-5, w_size=300, stat_size=30, data=None) for i in range(1000)
]
"""Calc amount of random features we need"""
from skmultiflow.drift_detection import ADWIN, EDDM, DDM from bix.detectors.kswin import KSWIN from bix.data.reoccuringdriftstream import ReoccuringDriftStream #from mebwin import MEBWIN from sw_mebwin import SWMEBWIN from kernel_swmebwin import Kernel_SWMEBWIN import time import numpy as np import typing from skmultiflow.bayes import NaiveBayes streams = [] """CD at 2000 and then every 1000""" # MIXED stream = ReoccuringDriftStream(stream=MIXEDGenerator(classification_function=0), drift_stream=MIXEDGenerator(classification_function=1), width=1, alpha=90, pause=1000, position=2000) stream.name = 'MIXED Abrupt Reoccuring' streams.append(stream) # SEA stream = ReoccuringDriftStream(stream=SEAGenerator(classification_function=0), drift_stream=SEAGenerator(classification_function=1), width=1, alpha=90, pause=1000, position=2000) stream.name = 'SEA Abrupt Reoccuring' streams.append(stream) # SINE stream = ReoccuringDriftStream(stream=SineGenerator(classification_function=0), drift_stream=SineGenerator(classification_function=1), width=1, alpha=90, pause=1000, position=2000) stream.name = 'Sine Abrupt Reoccuring' streams.append(stream)