def __init__(self, stream=AGRAWALGenerator(random_state=112),
drift_stream=AGRAWALGenerator(random_state=112, classification_function=2),
position=5000,
width=1000,
random_state=None,
alpha=0.0):
super(ConceptDriftStream, self).__init__()
self.n_samples = stream.n_samples
self.n_targets = stream.n_targets
self.n_features = stream.n_features
self.n_num_features = stream.n_num_features
self.n_cat_features = stream.n_cat_features
self.n_classes = stream.n_classes
self.cat_features_idx = stream.cat_features_idx
self.feature_names = stream.feature_names
self.target_names = stream.target_names
self.target_values = stream.target_values
self.n_targets = stream.n_targets
self.name = 'Drifting' + stream.name
self._original_random_state = random_state
self.random_state = None
self.alpha = alpha
if self.alpha != 0.0:
if 0 < self.alpha <= 90.0:
w = int(1 / np.tan(self.alpha * np.pi / 180))
self.width = w if w > 0 else 1
else:
raise ValueError('Invalid alpha value: {}'.format(alpha))
else:
self.width = width
self.position = position
self._input_stream = stream
self._drift_stream = drift_stream
def get_conceptdrift_data_generated(self,
classification_function=0,
noise_percentage=0.1,
random_state=112,
drift_classification_function=3,
drift_random_state=112,
drift_noise_percentage=0.0,
drift_start_position=5000,
drift_width=1000,
n_num_features=2,
n_cat_features=0):
from skmultiflow.data import ConceptDriftStream
from skmultiflow.data import AGRAWALGenerator
stream = AGRAWALGenerator(
classification_function=classification_function,
perturbation=noise_percentage,
random_state=random_state
#,n_num_features = n_num_features, n_cat_features = n_cat_features
)
drift_stream = AGRAWALGenerator(
classification_function=drift_classification_function,
perturbation=drift_noise_percentage,
random_state=drift_random_state
#,n_num_features = n_num_features, n_cat_features = n_cat_features
)
return ConceptDriftStream(stream=stream,
drift_stream=drift_stream,
position=drift_start_position,
width=drift_width)
def test_vfdr():
learner = VFDR(ordered_rules=True,
rule_prediction='first_hit',
nominal_attributes=[3,4,5],
expand_criterion='info_gain',
remove_poor_atts=True,
min_weight=100,
nb_prediction=False)
stream = AGRAWALGenerator(random_state=11)
stream.prepare_for_use()
cnt = 0
max_samples = 5000
predictions = array('i')
proba_predictions = []
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
proba_predictions.append(learner.predict_proba(X)[0])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0,
0, 1, 0, 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0,
0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0])
assert np.alltrue(predictions == expected_predictions)
expected_info = 'VFDR: ordered_rules: True - grace_period: 200 - split_confidence: 1e-07 ' + \
'- tie_threshold: 0.05 - remove_poor_atts: True - rule_prediction: first_hit ' + \
'- nb_threshold: 0 - nominal_attributes: [3, 4, 5] - drift_detector: NoneType ' + \
'- Predict using Naive Bayes: False'
assert learner.get_info() == expected_info
expected_model_description = 'Rule 0 :Att (2) <= 39.550| class :0 {0: 1365.7101742993455}\n' + \
'Rule 1 :Att (2) <= 58.180| class :1 {1: 1269.7307449971418}\n' + \
'Rule 2 :Att (2) <= 60.910| class :0 {0: 66.24158839706533, 1: 54.0}\n' + \
'Default Rule :| class :0 {0: 1316.7584116029348}'
expected_model_description_ = 'Rule 0 :Att (2) <= 39.550| class :0 {0: 1365.7101742993455}\n' + \
'Rule 1 :Att (2) <= 58.180| class :1 {1: 1269.7307449971418}\n' + \
'Rule 2 :Att (2) <= 60.910| class :0 {0: 66.241588397065328, 1: 54.0}\n' + \
'Default Rule :| class :0 {0: 1316.7584116029348}'
assert (learner.get_model_description() == expected_model_description) or \
(learner.get_model_description() == expected_model_description_)
expected_model_measurements = {'Number of rules: ': 3, 'model_size in bytes': 62295}
expected_model_measurements_ = {'Number of rules: ': 3, 'model_size in bytes': 73167}
if sys.version_info.minor != 6:
assert (learner.get_model_measurements() == expected_model_measurements) or\
(learner.get_model_measurements() == expected_model_measurements_)
def test_vfdr_info_gain():
learner = VeryFastDecisionRulesClassifier(ordered_rules=True,
rule_prediction='first_hit',
nominal_attributes=[3, 4, 5],
expand_criterion='info_gain',
remove_poor_atts=True,
min_weight=100,
nb_prediction=False)
stream = AGRAWALGenerator(random_state=11)
cnt = 0
max_samples = 5000
predictions = array('i')
proba_predictions = []
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
proba_predictions.append(learner.predict_proba(X)[0])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0,
0, 1, 0, 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0,
0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0])
assert np.alltrue(predictions == expected_predictions)
expected_info = "VeryFastDecisionRulesClassifier(drift_detector=None, expand_confidence=1e-07, " \
"expand_criterion='info_gain', grace_period=200, max_rules=1000, min_weight=100, " \
"nb_prediction=False, nb_threshold=0, nominal_attributes=[3, 4, 5], ordered_rules=True, " \
"remove_poor_atts=True, rule_prediction='first_hit', tie_threshold=0.05)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
expected_model_description_1 = 'Rule 0 :Att (2) <= 39.550| class :0 {0: 1365.7101742993455}\n' + \
'Rule 1 :Att (2) <= 58.180| class :1 {1: 1269.7307449971418}\n' + \
'Rule 2 :Att (2) <= 60.910| class :0 {0: 66.24158839706533, 1: 54.0}\n' + \
'Default Rule :| class :0 {0: 1316.7584116029348}'
expected_model_description_2 = 'Rule 0 :Att (2) <= 39.550| class :0 {0: 1365.7101742993455}\n' + \
'Rule 1 :Att (2) <= 58.180| class :1 {1: 1269.7307449971418}\n' + \
'Rule 2 :Att (2) <= 60.910| class :0 {0: 66.241588397065328, 1: 54.0}\n' + \
'Default Rule :| class :0 {0: 1316.7584116029348}'
assert (learner.get_model_description() == expected_model_description_1) or \
(learner.get_model_description() == expected_model_description_2)
# Following test only covers 'Number of rules' since 'model_size in bytes' is calculated using
# the 'calculate_object_size' utility function which is validated in its own test
expected_number_of_rules = 3
assert learner.get_model_measurements()['Number of rules: '] == expected_number_of_rules
def __init__(self,
stream=AGRAWALGenerator(random_state=112),
drift_stream=AGRAWALGenerator(random_state=112,
classification_function=2),
pause=1000,
random_state=None,
alpha=0.0,
position=0,
width=1):
self.n_samples = stream.n_samples
self.n_targets = stream.n_targets
self.n_features = stream.n_features
self.n_num_features = stream.n_num_features
self.n_cat_features = stream.n_cat_features
self.n_classes = stream.n_classes
self.cat_features_idx = stream.cat_features_idx
self.feature_names = stream.feature_names
self.target_names = ['target'] if self.n_targets == 1 else [
'target_' + i for i in range(self.n_targets)
]
self.target_values = stream.target_values
self.name = stream.name + "_" + drift_stream.name + "_" + str(
pause) + "_" + str(width)
self.probability_function = "sigmoid_prob"
self.pause = pause
self.counter = -1
self._original_random_state = random_state
self.random_state = None
self.alpha = alpha
if self.alpha != 0.0:
if 0 < self.alpha <= 90.0:
w = int(1 / np.tan(self.alpha * np.pi / 180))
self.width = w if w > 0 else 1
else:
raise ValueError('Invalid alpha value: {}'.format(alpha))
else:
self.width = width
if self.width < 0:
raise ValueError("Width must be greater than 0")
if self.pause < 0:
raise ValueError("Pause must be greater than 0")
self.position = position
self._input_stream = stream
self._drift_stream = drift_stream
self.n_targets = stream.n_targets
self._prepare_for_use()
def test_vfdr_hellinger():
learner = VFDR(ordered_rules=False,
rule_prediction='weighted_sum',
nominal_attributes=[3, 4, 5],
expand_criterion='hellinger',
remove_poor_atts=True,
min_weight=100,
nb_prediction=True)
stream = AGRAWALGenerator(random_state=11)
stream.prepare_for_use()
cnt = 0
max_samples = 5000
predictions = array('i')
proba_predictions = []
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
proba_predictions.append(learner.predict_proba(X)[0])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [
0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0, 0, 1,
0, 1, 0, 1, 0, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0, 1, 1, 0, 0, 0, 0,
0
])
assert np.alltrue(predictions == expected_predictions)
expected_model_description = 'Rule 0 :Att (2) > 58.180 and Att (5) = 4.000| class :0 {0: 202.0, 1: 3.0}\n' + \
'Rule 1 :Att (2) <= 41.820| class :0 {0: 1387.1186637804824, 1: 151.83928023717402}\n' + \
'Default Rule :| class :1 {0: 512.8813362195176, 1: 1356.160719762826}'
expected_model_description_ = 'Rule 0 :Att (2) > 58.180 and Att (5) = 4.000| class :0 {0: 202.0, 1: 3.0}\n' + \
'Rule 1 :Att (2) <= 41.820| class :0 {0: 1387.1186637804824, 1: 151.83928023717402}\n' + \
'Default Rule :| class :1 {0: 512.8813362195176, 1: 1356.1607197628259}'
if sys.version_info.minor != 6:
assert (learner.get_model_description() == expected_model_description) or \
(learner.get_model_description() == expected_model_description_)
def __init__(self,
stream=AGRAWALGenerator(random_state=112),
drift_stream=AGRAWALGenerator(random_state=112,
classification_function=2),
position=5000,
width=1000,
random_state=None,
alpha=None):
super(ConceptDriftStream, self).__init__()
self.n_samples = stream.n_samples
self.n_targets = stream.n_targets
self.n_features = stream.n_features
self.n_num_features = stream.n_num_features
self.n_cat_features = stream.n_cat_features
self.n_classes = stream.n_classes
self.cat_features_idx = stream.cat_features_idx
self.feature_names = stream.feature_names
self.target_names = stream.target_names
self.target_values = stream.target_values
self.n_targets = stream.n_targets
self.name = 'Drifting' + stream.name
self.random_state = random_state
self._random_state = None # This is the actual random_state object used internally
self.alpha = alpha
if self.alpha == 0:
warnings.warn(
"Default value for 'alpha' has changed from 0 to None. 'alpha=0' will "
"throw an error from v0.7.0",
category=FutureWarning)
self.alpha = None
if self.alpha is not None:
if 0 < self.alpha <= 90.0:
w = int(1 / np.tan(self.alpha * np.pi / 180))
self.width = w if w > 0 else 1
else:
raise ValueError('Invalid alpha value: {}'.format(alpha))
else:
self.width = width
self.position = position
self.stream = stream
self.drift_stream = drift_stream
self._prepare_for_use()
def test_hoeffding_adaptive_tree_alternate_tree():
stream = AGRAWALGenerator(random_state=7)
learner = HoeffdingAdaptiveTreeClassifier(random_state=1)
cnt = 0
change_point1 = 1500
change_point2 = 2500
change_point3 = 4000
max_samples = 5000
while cnt < max_samples:
X, y = stream.next_sample()
learner.partial_fit(X, y)
cnt += 1
if cnt > change_point1:
stream.generate_drift()
change_point1 = float('Inf')
expected_description = "if Attribute 2 <= 63.63636363636363:\n" \
" if Attribute 2 <= 39.54545454545455:\n" \
" Leaf = Class 0 | {0: 397.5023676194098}\n" \
" if Attribute 2 > 39.54545454545455:\n" \
" if Attribute 2 <= 58.81818181818181:\n" \
" Leaf = Class 1 | {1: 299.8923824199619}\n" \
" if Attribute 2 > 58.81818181818181:\n" \
" Leaf = Class 0 | {0: 54.0, 1: 20.107617580038095}\n" \
"if Attribute 2 > 63.63636363636363:\n" \
" Leaf = Class 0 | {0: 512.5755895049351}\n"
assert expected_description == learner.get_model_description()
if cnt > change_point2:
stream.generate_drift()
change_point2 = float('Inf')
expected_description = "if Attribute 8 <= 268547.7178694747:\n" \
" Leaf = Class 0 | {0: 446.18690518790413, 1: 80.6180778406834}\n" \
"if Attribute 8 > 268547.7178694747:\n" \
" Leaf = Class 1 | {0: 36.8130948120959, 1: 356.38192215931656}\n"
assert expected_description == learner.get_model_description()
if cnt > change_point3:
stream.generate_drift()
change_point3 = float('Inf')
expected_description = "Leaf = Class 0 | {0: 1083.0, 1: 2.0}\n"
assert expected_description == learner.get_model_description()
def test_vfdr_foil():
learner = VFDR(ordered_rules=False,
rule_prediction='weighted_sum',
nominal_attributes=[3, 4, 5],
expand_criterion='foil_gain',
remove_poor_atts=True,
min_weight=100,
nb_prediction=True)
stream = AGRAWALGenerator(random_state=11)
stream.prepare_for_use()
cnt = 0
max_samples = 5000
predictions = array('i')
proba_predictions = []
wait_samples = 100
while cnt < max_samples:
X, y = stream.next_sample()
# Test every n samples
if (cnt % wait_samples == 0) and (cnt != 0):
predictions.append(learner.predict(X)[0])
proba_predictions.append(learner.predict_proba(X)[0])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [
0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0,
0, 1, 0, 1, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0,
0
])
assert np.alltrue(predictions == expected_predictions)
expected_model_description = 'Rule 0 :Att (2) <= 25.450 | class: 1| class :0 {0: 464.44730579120136}\n' + \
'Rule 1 :Att (4) = 3.000 | class: 0| class :0 {0: 95.0, 1: 45.0}\n' + \
'Rule 2 :Att (2) <= 30.910 | class: 1| class :0 {0: 330.68821225514125}\n' + \
'Default Rule :| class :0 {0: 573.0, 1: 336.0}'
assert (learner.get_model_description() == expected_model_description)
def prepare_for_use(self):
if self.generator in ['sea', 'sine']:
self.concepts = [v for v in range(0, 4)]
elif self.generator in ['stagger']:
self.concepts = [v for v in range(0, 3)]
elif self.generator in ['mixed']:
self.concepts = [v for v in range(0, 2)]
elif self.generator in ['led']:
self.concepts = [v for v in range(0, 7)]
elif self.generator in ['tree']:
self.concepts = [2, 3, 4, 5, 6, 7, 8, 9, 10]
if self.concept_shift_step > 0:
for concept in self.all_concepts:
stream = AGRAWALGenerator(classification_function=concept,
random_state=self.random_state,
balance_classes=False,
perturbation=0.05)
stream.prepare_for_use()
self.streams.append(stream)
else:
for concept in self.concepts:
if self.generator == 'agrawal':
stream = AGRAWALGenerator(classification_function=concept,
random_state=self.random_state,
balance_classes=False,
perturbation=0.05)
elif self.generator == 'sea':
stream = SEAGenerator(classification_function=concept,
random_state=self.random_state,
balance_classes=False,
noise_percentage=0.05)
elif self.generator == 'sine':
stream = SineGenerator(classification_function=concept,
random_state=self.random_state,
balance_classes=False,
has_noise=False)
elif self.generator == 'stagger':
stream = STAGGERGenerator(classification_function=concept,
random_state=self.random_state,
balance_classes=False)
elif self.generator == 'mixed':
stream = MIXEDGenerator(classification_function=concept,
random_state=self.random_state,
balance_classes=False)
elif self.generator == 'led':
stream = LEDGeneratorDrift(random_state=self.random_state,
has_noise=True,
n_drift_features=concept)
elif self.generator == 'tree':
stream = RandomTreeGenerator(tree_random_state=concept,
sample_random_state=concept,
max_tree_depth=concept+2,
min_leaf_depth=concept,
n_classes=2)
else:
print(f"unknown stream generator {self.generator}")
exit()
stream.prepare_for_use()
self.streams.append(stream)
self.cur_stream = self.streams[0]
self.drift_stream = self.streams[1]
stream = self.cur_stream
self.n_samples = stream.n_samples
self.n_targets = stream.n_targets
self.n_features = stream.n_features
self.n_num_features = stream.n_num_features
self.n_cat_features = stream.n_cat_features
self.n_classes = stream.n_classes
self.cat_features_idx = stream.cat_features_idx
self.feature_names = stream.feature_names
self.target_names = stream.target_names
self.target_values = stream.target_values
self.n_targets = stream.n_targets
self.name = 'drifting' + stream.name
print(f"len: {len(self.concepts)}")
self.concept_probs = \
self.__get_poisson_probs(len(self.concepts), self.lam)
#2 0.091587 0.977452 0.411501 0.458305 ... 0.181444 0.303406 0.174454 0.0
#3 0.635272 0.496203 0.014126 0.627222 ... 0.517752 0.570683 0.546333 1.0
#4 0.450078 0.876507 0.537356 0.495684 ... 0.606895 0.217841 0.912944 1.0
#
#[5 rows x 11 columns]
# Store it in csv
data.to_csv('data_stream_hyperplane.csv', index=False)
#####################################################################################
### Agarwal generator using scikit-multiflow
from skmultiflow.data import AGRAWALGenerator
import pandas as pd
import numpy as np
create = AGRAWALGenerator(random_state=333)
create.prepare_for_use()
X, Y = create.next_sample(10000)
data = pd.DataFrame(np.hstack((X, np.array([Y]).T)))
data.shape
# output- (1000, 10)
print(data.head())
# Output:
# 0 1 2 ... 7 8 9
#0 90627.841313 0.000000 33.0 ... 20.0 24151.832875 0.0
#1 33588.924462 17307.813671 72.0 ... 29.0 315025.363876 0.0
#2 24375.065287 12426.917711 39.0 ... 4.0 363158.576720 0.0
#3 82949.727691 0.000000 68.0 ... 2.0 35758.528073 0.0
#4 149423.790417 0.000000 52.0 ... 29.0 98440.362484 1.0
#
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/1uHKbJ3KLUITTHJRxegzbTvA_-6M7eO5v
"""
!pip install -U scikit-multiflow
from skmultiflow.data import AGRAWALGenerator
from skmultiflow.trees import HoeffdingTree
from skmultiflow.evaluation import EvaluatePrequential
import numpy as np
# 1. Create a stream
stream = AGRAWALGenerator()
stream.prepare_for_use()
# 2. Instantiate the HoeffdingTree classifier
ht = HoeffdingTree()
# # 3. Setup the evaluator
# evaluator = EvaluatePrequential(show_plot=False,
# pretrain_size=500,
# max_samples=500)
# # 4. Run evaluation
# evaluator.evaluate(stream=stream, model=ht)
def base_classifier(e, U, I, L, D, wd, ws):
return print("I am here")
from skmultiflow.data.file_stream import FileStream
import numpy as np
from Goowe import Goowe
from skmultiflow.data import ConceptDriftStream
from skmultiflow.data import AGRAWALGenerator
import logging
from GooweMS import GooweMS
import random
logger = logging.getLogger()
logger.setLevel(logging.INFO)
# Prepare the data stream
stream_1 = ConceptDriftStream(
stream=AGRAWALGenerator(balance_classes=False,
classification_function=1,
perturbation=0.0,
random_state=112),
drift_stream=AGRAWALGenerator(balance_classes=False,
classification_function=2,
perturbation=0.0,
random_state=112),
position=3000,
width=1000,
random_state=None,
alpha=0.0)
stream_2 = ConceptDriftStream(
stream=AGRAWALGenerator(balance_classes=False,
classification_function=3,
perturbation=0.0,
random_state=21),
drift_stream=AGRAWALGenerator(balance_classes=False,