def test_half_space_trees(test_path):
stream = SEAGenerator(classification_function=0,
noise_percentage=0.1,
random_state=1)
learner = HalfSpaceTrees(n_estimators=13,
size_limit=75,
anomaly_threshold=0.90,
depth=10,
random_state=5)
cnt = 0
max_samples = 5000
y_pred = array('i')
y_proba = []
wait_samples = 500
while cnt < max_samples:
X, y = stream.next_sample()
# Scale inputs between 0 and 1
X = X / 10
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)
cnt += 1
expected_predictions = array('i', [1, 0, 0, 0, 1, 0, 0, 1, 0])
assert np.alltrue(y_pred == expected_predictions)
test_file = os.path.join(test_path, 'test_half_space_trees.npy')
expected_proba = np.load(test_file)
assert np.allclose(y_proba, expected_proba)
def test_clone():
stream = SEAGenerator(random_state=1)
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=[0, 1])
cnt += 1
cloned = clone(learner)
assert learner._observed_class_distribution != {} and cloned._observed_class_distribution == {}
def test_perceptron(test_path):
stream = SEAGenerator(random_state=1)
learner = PerceptronMask(random_state=1)
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=[0, 1])
cnt += 1
expected_predictions = array('i', [
1, 1, 1, 0, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
0, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1, 0,
1
])
assert np.alltrue(y_pred == expected_predictions)
test_file = os.path.join(test_path, 'data_perceptron_proba.npy')
y_proba_expected = np.load(test_file)
assert np.allclose(y_proba, y_proba_expected)
expected_info = "PerceptronMask(alpha=0.0001, class_weight=None, early_stopping=False, " \
"eta0=1.0, fit_intercept=True, max_iter=1000, n_iter_no_change=5, " \
"n_jobs=None, penalty=None, random_state=1, shuffle=True, tol=0.001, " \
"validation_fraction=0.1, verbose=0, warm_start=False)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
# Coverage tests
learner.reset()
if not sklearn_version.startswith("0.21"):
learner.fit(X=np.asarray(X_batch[:4500]),
y=np.asarray(y_batch[:4500], dtype=int))
else:
# Root cause of failure (TypeError: an integer is required) is in the fit() method
# in sklearn 0.21.0. This is a workaround until a fix is made available in sklearn
learner.partial_fit(X=np.asarray(X_batch[:4500]),
y=np.asarray(y_batch[:4500]),
classes=stream.target_values)
learner.predict(X=X_batch[4501:]) # Run for coverage
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
def test_leverage_bagging():
stream = SEAGenerator(classification_function=1,
noise_percentage=0.067,
random_state=112)
knn = KNNClassifier(n_neighbors=8,
leaf_size=40,
max_window_size=2000)
learner = LeveragingBaggingClassifier(base_estimator=knn,
n_estimators=3,
random_state=112)
first = True
cnt = 0
max_samples = 5000
predictions = []
wait_samples = 100
correct_predictions = 0
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])
if y[0] == predictions[-1]:
correct_predictions += 1
if first:
learner.partial_fit(X, y, classes=[0, 1])
first = False
else:
learner.partial_fit(X, y)
cnt += 1
performance = correct_predictions / len(predictions)
expected_predictions = [1, 0, 1, 0, 0, 0, 0, 1, 0, 1,
0, 0, 1, 0, 1, 1, 1, 0, 1, 0,
0, 1, 0, 0, 1, 1, 0, 1, 0, 1,
1, 1, 1, 1, 0, 0, 1, 1, 1, 0,
0, 1, 1, 0, 1, 0, 0, 1, 1]
assert np.alltrue(predictions == expected_predictions)
expected_performance = 0.8571428571428571
assert np.isclose(expected_performance, performance)
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
expected_info = "LeveragingBaggingClassifier(base_estimator=KNNClassifier(leaf_size=40, " \
"max_window_size=2000, metric='euclidean', n_neighbors=8), " \
"delta=0.002, enable_code_matrix=False, leverage_algorithm='leveraging_bag'," \
" n_estimators=3, random_state=112, w=6)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
def test_online_csb2():
stream = SEAGenerator(1, noise_percentage=0.067, random_state=112)
nb = NaiveBayes()
learner = OnlineCSB2Classifier(base_estimator=nb,
n_estimators=3,
cost_positive=1,
cost_negative=0.9,
random_state=112)
first = True
cnt = 0
max_samples = 5000
predictions = []
wait_samples = 100
correct_predictions = 0
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])
if y[0] == predictions[-1]:
correct_predictions += 1
if first:
learner.partial_fit(X, y, classes=[0, 1])
first = False
else:
learner.partial_fit(X, y)
cnt += 1
performance = correct_predictions / len(predictions)
expected_predictions = [
1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 0,
1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
1
]
expected_correct_predictions = 43
expected_performance = 0.8775510204081632
assert np.alltrue(predictions == expected_predictions)
assert np.isclose(expected_performance, performance)
assert correct_predictions == expected_correct_predictions
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
expected_info = "OnlineCSB2Classifier(base_estimator=NaiveBayes(nominal_attributes=None), cost_negative=0.9, " \
"cost_positive=1, drift_detection=True, n_estimators=3, random_state=112)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
def test_naive_bayes(test_path):
stream = SEAGenerator(random_state=1)
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=[0, 1])
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=None)'
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 is_classifier(learner)
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
def test_oza_bagging_adwin():
stream = SEAGenerator(1, noise_percentage=0.067, random_state=112)
knn = KNNClassifier(n_neighbors=8, leaf_size=40, max_window_size=2000)
learner = OzaBaggingADWINClassifier(base_estimator=knn,
n_estimators=3,
random_state=112)
first = True
cnt = 0
max_samples = 5000
predictions = []
wait_samples = 100
correct_predictions = 0
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])
if y[0] == predictions[-1]:
correct_predictions += 1
if first:
learner.partial_fit(X, y, classes=[0, 1])
first = False
else:
learner.partial_fit(X, y)
cnt += 1
performance = correct_predictions / len(predictions)
expected_predictions = [
1, 0, 1, 1, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 1, 1, 0, 1, 0, 0, 1, 0, 0,
1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
1
]
assert np.alltrue(predictions == expected_predictions)
expected_performance = 0.8979591836734694
assert np.isclose(expected_performance, performance)
expected_correct_predictions = 44
assert correct_predictions == expected_correct_predictions
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
expected_info = "OzaBaggingADWINClassifier(base_estimator=KNNClassifier(leaf_size=40, " \
"max_window_size=2000, metric='euclidean', n_neighbors=8), n_estimators=3, " \
"random_state=112)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
def test_knn_adwin():
stream = ConceptDriftStreamGenerator(stream=SEAGenerator(random_state=1),
drift_stream=SEAGenerator(
random_state=2,
classification_function=2),
random_state=1,
position=250,
width=10)
learner = KNNADWINClassifier(n_neighbors=8,
leaf_size=40,
max_window_size=200)
cnt = 0
max_samples = 1000
predictions = array('i')
correct_predictions = 0
wait_samples = 20
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])
if y[0] == predictions[-1]:
correct_predictions += 1
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [
1, 0, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 0,
1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 0, 1,
1
])
assert np.alltrue(predictions == expected_predictions)
expected_correct_predictions = 46
assert correct_predictions == expected_correct_predictions
learner.reset()
assert learner.data_window.size == 0
expected_info = "KNNADWINClassifier(leaf_size=40, max_window_size=200, " \
"metric='euclidean', n_neighbors=8)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
def test_hoeffding_adaptive_tree_nb(test_path):
stream = ConceptDriftStreamGenerator(
stream=SEAGenerator(random_state=1, noise_percentage=0.05),
drift_stream=SEAGenerator(random_state=2,
classification_function=2,
noise_percentage=0.05),
random_state=1,
position=250,
width=10)
learner = HoeffdingAdaptiveTreeClassifier(leaf_prediction='nb',
random_state=1)
cnt = 0
max_samples = 1000
y_pred = array('i')
y_proba = []
wait_samples = 20
while cnt < max_samples:
X, y = stream.next_sample()
# 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)
cnt += 1
expected_predictions = array('i', [
1, 0, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 0,
1, 0, 0, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1,
1
])
assert np.alltrue(y_pred == expected_predictions)
test_file = os.path.join(test_path, 'test_hoeffding_adaptive_tree_nb.npy')
data = np.load(test_file)
assert np.allclose(y_proba, data)
expected_info = "HoeffdingAdaptiveTreeClassifier(binary_split=False, bootstrap_sampling=True, grace_period=200, " \
"leaf_prediction='nb', max_byte_size=33554432, memory_estimate_period=1000000, nb_threshold=0, " \
"no_preprune=False, nominal_attributes=None, random_state=1, remove_poor_atts=False, " \
"split_confidence=1e-07, split_criterion='info_gain', stop_mem_management=False, tie_threshold=0.05)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
def test_hoeffding_tree_model_information():
stream = SEAGenerator(random_state=1, noise_percentage=0.05)
X, y = get_next_n_samples(stream, 5000)
nominal_attr_idx = [x for x in range(5, stream.n_features)]
learner = HoeffdingTreeClassifier(nominal_attributes=nominal_attr_idx)
learner.partial_fit(X, y, classes=[0, 1])
expected_info = {
'Tree size (nodes)': 5,
'Tree size (leaves)': 3,
'Active learning nodes': 3,
'Tree depth': 2,
'Active leaf byte size estimate': 0.0,
'Inactive leaf byte size estimate': 0.0,
'Byte size estimate overhead': 1.0
}
observed_info = learner.model_measurements
for k in expected_info:
assert k in observed_info
assert expected_info[k] == observed_info[k]
expected_description = "if Attribute 0 <= 4.549969620513424:\n" \
" if Attribute 1 <= 5.440182925299016:\n" \
" Leaf = Class 0 | {0: 345.54817975126275, 1: 44.43855503614928}\n" \
" if Attribute 1 > 5.440182925299016:\n" \
" Leaf = Class 1 | {0: 54.451820248737235, 1: 268.5614449638507}\n" \
"if Attribute 0 > 4.549969620513424:\n" \
" Leaf = Class 1 | {0: 390.5845685762964, 1: 2372.3747376855454}\n" \
assert expected_description == learner.get_model_description()
def test_extremely_fast_decision_tree_coverage():
# Cover memory management
max_size_kb = 20
stream = SEAGenerator(random_state=1, noise_percentage=0.05)
X, y = get_next_n_samples(stream, 5000)
# Unconstrained model has over 50 kB
learner = ExtremelyFastDecisionTreeClassifier(leaf_prediction='mc',
memory_estimate_period=200,
max_byte_size=max_size_kb *
2**10,
min_samples_reevaluate=2500)
learner.partial_fit(X, y, classes=[0, 1])
assert calculate_object_size(learner, 'kB') <= max_size_kb
learner.reset()
# Cover nominal attribute observer
stream = RandomTreeGenerator(tree_random_state=23,
sample_random_state=12,
n_classes=2,
n_cat_features=2,
n_categories_per_cat_feature=4,
n_num_features=1,
max_tree_depth=30,
min_leaf_depth=10,
fraction_leaves_per_level=0.45)
X, y = get_next_n_samples(stream, 5000)
learner = ExtremelyFastDecisionTreeClassifier(
leaf_prediction='nba', nominal_attributes=[i for i in range(1, 9)])
learner.partial_fit(X, y, classes=[0, 1])
def test_dynamic_weighted_majority():
stream = SEAGenerator(1, noise_percentage=0.067, random_state=112)
learner = DynamicWeightedMajorityClassifier(3,
NaiveBayes(),
beta=0.5,
theta=0.01)
cnt = 0
max_samples = 5000
predictions = []
wait_samples = 100
correct_predictions = 0
first = True
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])
if y[0] == predictions[-1]:
correct_predictions += 1
if first:
learner.partial_fit(X, y, classes=[0, 1])
first = False
else:
learner.partial_fit(X, y)
cnt += 1
performance = correct_predictions / len(predictions)
expected_predictions = [
1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 1, 1, 0, 1, 0, 0, 1, 0, 0,
1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1,
1
]
expected_correct_predictions = 44
expected_performance = 0.8979591836734694
assert np.alltrue(predictions == expected_predictions)
assert np.isclose(expected_performance, performance)
assert correct_predictions == expected_correct_predictions
assert type(learner.predict(X)) == np.ndarray
expected_info = 'DynamicWeightedMajorityClassifier(base_estimator=NaiveBayes(nominal_attributes=None),\n' \
' beta=0.5, n_estimators=3, period=50,\n' \
' theta=0.01)'
assert learner.get_info() == expected_info
def test_online_rus_3():
stream = SEAGenerator(1, noise_percentage=0.067, random_state=112)
nb = NaiveBayes()
learner = OnlineRUSBoostClassifier(base_estimator=nb,
n_estimators=3,
sampling_rate=5,
algorithm=3,
random_state=112)
first = True
cnt = 0
max_samples = 5000
predictions = []
wait_samples = 100
correct_predictions = 0
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])
if y[0] == predictions[-1]:
correct_predictions += 1
if first:
learner.partial_fit(X, y, classes=[0, 1])
first = False
else:
learner.partial_fit(X, y)
cnt += 1
performance = correct_predictions / len(predictions)
expected_predictions = [
1, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 0,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 1, 1,
1
]
expected_correct_predictions = 35
expected_performance = 0.7142857142857143
assert np.alltrue(predictions == expected_predictions)
assert np.isclose(expected_performance, performance)
assert correct_predictions == expected_correct_predictions
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
def test_sam_knn_coverage():
stream = SEAGenerator(random_state=1)
hyperParams = {
'maxSize': 50,
'n_neighbors': 3,
'weighting': 'uniform',
'stm_size_option': 'maxACC',
'min_stm_size': 10,
'use_ltm': True
}
learner = SAMKNNClassifier(n_neighbors=hyperParams['n_neighbors'],
max_window_size=hyperParams['maxSize'],
weighting=hyperParams['weighting'],
stm_size_option=hyperParams['stm_size_option'],
min_stm_size=hyperParams['min_stm_size'],
use_ltm=hyperParams['use_ltm'])
cnt = 0
max_samples = 1000
predictions = array('i')
wait_samples = 20
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])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [
1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1,
1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1,
1
])
assert np.alltrue(predictions == expected_predictions)
expected_info = "SAMKNNClassifier(ltm_size=0.4, max_window_size=None, min_stm_size=10, n_neighbors=3, " \
"stm_size_option='maxACC', use_ltm=True, weighting='uniform')"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
def test_additive_expert_ensemble_weakest():
stream = SEAGenerator(1, noise_percentage=0.067, random_state=112)
learner = AdditiveExpertEnsembleClassifier(3, NaiveBayes(), beta=0.5, gamma=0.1,
pruning='weakest')
cnt = 0
max_samples = 5000
predictions = []
wait_samples = 100
correct_predictions = 0
first = True
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])
if y[0] == predictions[-1]:
correct_predictions += 1
if first:
learner.partial_fit(X, y, classes=[0, 1])
first = False
else:
learner.partial_fit(X, y)
cnt += 1
performance = correct_predictions / len(predictions)
expected_predictions = [1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 1, 1,
0, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1,
0, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 0, 0, 1, 1]
expected_correct_predictions = 45
expected_performance = 0.9183673469387755
assert np.alltrue(predictions == expected_predictions)
assert np.isclose(expected_performance, performance)
assert correct_predictions == expected_correct_predictions
assert type(learner.predict(X)) == np.ndarray
expected_info = "AdditiveExpertEnsembleClassifier(base_estimator=NaiveBayes(nominal_attributes=None),\n" \
" beta=0.5, gamma=0.1, n_estimators=3,\n" \
" pruning='weakest')"
assert learner.get_info() == expected_info
def test_sam_knn():
stream = SEAGenerator(random_state=1)
hyperParams = {
'maxSize': 1000,
'nNeighbours': 5,
'knnWeights': 'distance',
'STMSizeAdaption': 'maxACCApprox',
'use_ltm': False
}
learner = SAMKNNClassifier(n_neighbors=hyperParams['nNeighbours'],
max_window_size=hyperParams['maxSize'],
weighting=hyperParams['knnWeights'],
stm_size_option=hyperParams['STMSizeAdaption'],
use_ltm=hyperParams['use_ltm'])
cnt = 0
max_samples = 5000
predictions = array('d')
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])
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [
1, 1, 1, 0, 1, 1, 0, 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, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 0,
1
])
assert np.alltrue(predictions == expected_predictions)
assert type(learner.predict(X)) == np.ndarray
with pytest.raises(NotImplementedError):
learner.predict_proba(X)
def test_sea_generator(test_path):
stream = SEAGenerator(classification_function=2,
random_state=112,
balance_classes=False,
noise_percentage=0.28)
# 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']
for j in range(0, 10):
X, y = stream.next_sample()
assert np.alltrue(np.isclose(X, X_expected[j]))
assert np.alltrue(np.isclose(y[0], y_expected[j]))
expected_info = "SEAGenerator(balance_classes=False, classification_function=2, noise_percentage=0.28, random_state=112)"
assert stream.get_info() == expected_info
def test_leverage_bagging_half():
knn = KNNClassifier(n_neighbors=8, leaf_size=40, max_window_size=2000)
# leveraging_bag_half
learner = LeveragingBaggingClassifier(base_estimator=knn, n_estimators=3, random_state=112, leverage_algorithm='leveraging_bag_half')
y_expected = np.asarray([0, 1, 1, 0, 1, 0, 1, 0, 1, 0,
1, 0, 0, 0, 1, 0, 1, 1, 1, 1,
1, 1, 1, 1, 0, 1, 0, 1, 1, 0,
0, 0, 1, 1, 1, 0, 1, 1, 0, 0,
1, 0, 0, 1, 0, 0, 0, 1, 1, 0], dtype=np.int)
run_prequential_supervised(SEAGenerator(classification_function=1, noise_percentage=0.067, random_state=112),
learner, max_samples=2000, n_wait=40, target_values=[0,1], y_expected=y_expected)
def test_leverage_bagging_coverage():
# Invalid leverage_algorithm
with pytest.raises(ValueError):
LeveragingBaggingClassifier(leverage_algorithm='invalid')
estimator = LeveragingBaggingClassifier(random_state=4321)
stream = SEAGenerator(random_state=4321)
X, y = stream.next_sample()
# classes not passed in partial_fit
with pytest.raises(ValueError):
estimator.partial_fit(X, y, classes=None)
estimator.partial_fit(X, y, classes=[0, 1])
# different observed classes
with pytest.raises(ValueError):
estimator.partial_fit(X, y, classes=[0, 1] + [-1])
# Invalid leverage_algorithm, changed after initialization
with pytest.raises(RuntimeError):
estimator.leverage_algorithm = 'invalid'
estimator.partial_fit(X, y, classes=[0, 1])
# Reset ensemble
estimator.reset()
assert estimator.classes is None
def test_learn_nse():
stream = SEAGenerator(random_state=2212)
estimator = GaussianNB()
corrects, acc, classifier = run_classifier(estimator, stream)
expected_correct_predictions = 1754
expected_acc = 0.877
assert np.isclose(expected_acc, acc)
assert corrects == expected_correct_predictions
# Test reset method
classifier.reset()
assert len(classifier.ensemble) == 0
assert len(classifier.ensemble_weights) == 0
assert len(classifier.bkts) == 0
assert len(classifier.wkts) == 0
assert len(classifier.X_batch) == 0
assert len(classifier.y_batch) == 0
expected_info = 'LearnPPNSEClassifier(base_estimator=GaussianNB(), crossing_point=10, ' \
'n_estimators=15, pruning=None, slope=0.5, window_size=250)'
info = " ".join([line.strip() for line in classifier.get_info().split()])
assert info == expected_info
# test pruning error
corrects, acc, classifier = run_classifier(estimator,
stream,
pruning="error",
ensemble_size=5)
expected_correct_predictions = 1751
expected_acc = 0.8755
assert np.isclose(expected_acc, acc)
assert corrects == expected_correct_predictions
# test pruning age
corrects, acc, classifier = run_classifier(estimator,
stream,
pruning="age",
ensemble_size=5)
expected_correct_predictions = 1774
expected_acc = 0.887
assert np.isclose(expected_acc, acc)
assert corrects == expected_correct_predictions
stream = SEAGenerator(random_state=2212)
estimator = HoeffdingTreeClassifier()
classifier = LearnPPNSEClassifier(base_estimator=estimator)
# Keeping track of sample count and correct prediction count
sample_count = 0
corrects = 0
m = 250
# Pre training the classifier
X, y = get_next_n_samples(stream, m)
classifier.partial_fit(X, y, classes=[0, 1])
# print(classifier.ensemble_weights)
for i in range(10):
X, y = get_next_n_samples(stream, m)
pred = classifier.predict(X)
classifier.partial_fit(X, y)
if pred is not None:
# print(pred)
corrects += np.sum(y == pred)
sample_count += m
acc = corrects / sample_count
expected_acc = 0.9436
assert acc == expected_acc
def test_knn():
stream = SEAGenerator(random_state=1)
learner = KNNClassifier(n_neighbors=8, max_window_size=2000, leaf_size=40)
cnt = 0
max_samples = 5000
predictions = array('i')
correct_predictions = 0
wait_samples = 100
X_batch = []
y_batch = []
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):
predictions.append(learner.predict(X)[0])
if y[0] == predictions[-1]:
correct_predictions += 1
learner.partial_fit(X, y)
cnt += 1
expected_predictions = array('i', [
1, 1, 1, 0, 1, 1, 0, 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, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 0,
1
])
assert np.alltrue(predictions == expected_predictions)
expected_correct_predictions = 49
assert correct_predictions == expected_correct_predictions
expected_info = "KNNClassifier(leaf_size=40, max_window_size=2000, " \
"metric='euclidean', n_neighbors=8)"
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
learner.reset()
info = " ".join([line.strip() for line in learner.get_info().split()])
assert info == expected_info
X_batch = np.array(X_batch)
y_batch = np.array(y_batch)
learner.fit(X_batch[:4500], y_batch[:4500], classes=[0, 1])
predictions = learner.predict(X_batch[4501:4550])
expected_predictions = array('i', [
1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 0, 1, 0, 1, 1,
1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1,
0
])
assert np.alltrue(predictions == expected_predictions)
correct_predictions = sum(predictions == y_batch[4501:4550])
expected_correct_predictions = 49
assert correct_predictions == expected_correct_predictions
assert type(learner.predict(X)) == np.ndarray
assert type(learner.predict_proba(X)) == np.ndarray
def test_rslvq():
stream = SEAGenerator(random_state=1)
learner_adadelta = RSLVQ(gradient_descent='adadelta')
learner_vanilla = RSLVQ(gradient_descent='vanilla')
cnt = 0
max_samples = 5000
y_pred_vanilla = array('i')
y_pred_adadelta = array('i')
X_batch = []
y_batch = []
wait_samples = 100
# Check if predicted labels are as expected
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_vanilla.append(learner_vanilla.predict(X)[0])
y_pred_adadelta.append(learner_adadelta.predict(X)[0])
learner_adadelta.partial_fit(X, y, classes=[0, 1])
learner_vanilla.partial_fit(X, y, classes=[0, 1])
cnt += 1
expected_predictions_vanilla = array('i', [
1, 1, 1, 0, 0, 1, 1, 1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 0, 1, 1, 1,
0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 0,
1
])
expected_predictions_adadelta = array('i', [
1, 0, 1, 0, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
0, 1, 1, 0, 0, 1, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 1, 0, 1, 0, 0, 1, 0,
1
])
assert np.alltrue(y_pred_vanilla == expected_predictions_vanilla)
assert np.alltrue(y_pred_adadelta == expected_predictions_adadelta)
# Check get_info method
expected_info = "RobustSoftLearningVectorQuantization(gamma=0.9, gradient_descent='vanilla',\n" \
" initial_prototypes=None,\n" \
" prototypes_per_class=1, random_state=None,\n" \
" sigma=1.0)"
assert learner_vanilla.get_info() == expected_info
# Check reset method
learner_vanilla.reset()
learner_vanilla.fit(X=np.array(X_batch[:4500]), y=np.array(y_batch[:4500]))
learner_adadelta.reset()
learner_adadelta.fit(X=np.array(X_batch[:4500]),
y=np.array(y_batch[:4500]))
# Check classifiers performance
learner_w_init_ppt = RSLVQ(
initial_prototypes=[[2.59922826, 2.57368134, 4.92501, 0],
[6.05801971, 6.01383352, 5.02135783, 1]],
gradient_descent='adadelta')
learner_w_init_ppt.fit(X=np.array(X_batch[:4500]),
y=np.array(y_batch[:4500]))
expected_score_ppt = .9539078156312625
assert np.isclose(
expected_score_ppt,
learner_w_init_ppt.score(X=np.array(X_batch[4501:]),
y=np.array(y_batch[4501:])))
expected_score_vanilla = .8897795591182365
assert np.isclose(
expected_score_vanilla,
learner_vanilla.score(X=np.array(X_batch[4501:]),
y=np.array(y_batch[4501:])))
expected_score_adadelta = .9458917835671342
assert np.isclose(
expected_score_adadelta,
learner_adadelta.score(X=np.array(X_batch[4501:]),
y=np.array(y_batch[4501:])))
# Check types
assert is_classifier(learner_vanilla)
assert is_classifier(learner_adadelta)
assert type(learner_vanilla.predict(X)) == np.ndarray
assert type(learner_adadelta.predict(X)) == np.ndarray
# Check properties after learning
expected_prototypes = np.array([[2.59922826, 2.57368134, 4.92501],
[6.05801971, 6.01383352, 5.02135783]])
assert np.allclose(learner_adadelta.prototypes, expected_prototypes)
expected_prototypes_classes = np.array([0, 1])
assert np.allclose(learner_adadelta.prototypes_classes,
expected_prototypes_classes)
expected_class_labels = np.array([0, 1])
assert np.allclose(learner_adadelta.class_labels, expected_class_labels)