def test_random_state_uniform(nl):
y1 = flip_labels_uniform(y, nl, random_state=SEED, exact=False)
y2 = flip_labels_uniform(y, nl, random_state=SEED, exact=False)
assert_array_equal(y1, y2)
_, ya = UniformNoise(nl, random_state=SEED,
exact=False).simulate_noise(X, y)
_, yb = UniformNoise(nl, random_state=SEED,
exact=False).simulate_noise(X, y)
assert_array_equal(ya, yb)
assert_array_equal(ya, y1)
def test_uniform_exact(nl):
yn = flip_labels_uniform(y, nl, exact=True)
assert (y != yn).sum() / len(y) == nl
_, ya = UniformNoise(nl, random_state=SEED,
exact=True).simulate_noise(X, y)
assert (y != ya).sum() / len(y) == nl
def test_stack_sim():
p = Pipeline([
('a', UniformNoise(.3)),
('b', CCNoise(lcm)),
('c', DecisionTreeClassifier()),
])
p.fit(X, y)
p.predict(X)
def test_example_two():
from skclean.simulate_noise import UniformNoise
from skclean.detectors import KDN
from skclean.handlers import Filter
from skclean.pipeline import Pipeline, make_pipeline # Importing from skclean, not sklearn
from skclean.utils import load_data
X, y = load_data('breast_cancer')
clf = Pipeline([
('scale', StandardScaler()), # Scale features
('feat_sel', VarianceThreshold(.2)), # Feature selection
('detector', KDN()), # Detect mislabeled samples
('handler', Filter(SVC())
), # Filter out likely mislabeled samples and then train a SVM
])
clf_g = GridSearchCV(clf, {'detector__n_neighbors': [2, 5, 10]})
n_clf_g = make_pipeline(UniformNoise(.3),
clf_g) # Create label noise at the very first step
print(cross_val_score(n_clf_g, X, y,
cv=5).mean()) # 5-fold cross validation
('b', StandardScaler()),
('c', VarianceThreshold(.2)),
('d', KDN()),
('e', skclean.handlers.SampleWeight(dummy))
])
# Inside Pipeline
tmp_Handlers = []
for h in NOISE_HANDLERS:
if h.iterative: # Exclude iterative handlers
continue
ch = clone(h)
ch.detector = None
tmp_Handlers.append(ch)
preli_steps = [UniformNoise(.2, random_state=SEED), StandardScaler()]
all_comb = product(NOISE_DETECTORS, tmp_Handlers)
INSIDE_PIPE = [make_pipeline(*preli_steps + list(comb)) for comb in all_comb]
# Outside Pipe
OUTSIDE_PIPE = []
for h in NOISE_HANDLERS:
for d in NOISE_DETECTORS:
ch, d = clone(h), clone(d)
ch.detector = d
if 'random_state' in ch.get_params(): # trying to avoid flaky tests
ch.set_params(random_state=42)
OUTSIDE_PIPE.append(ch)
ALL_COMBS = INSIDE_PIPE + OUTSIDE_PIPE
def test_pipe_cv():
p = Pipeline([('a', UniformNoise(.3)), ('p', StandardScaler()),
('d', KDN(n_neighbors=5)),
('c', Filter(DecisionTreeClassifier()))])
p = cross_val_score(p, X, y, cv=CV1, error_score='raise')
print(p)
def test_detector_init():
p = Pipeline([('a', UniformNoise(.3)), ('p', StandardScaler()),
('d', RandomForestDetector(n_estimators=50)),
('c', Filter(DecisionTreeClassifier()))])
p.fit(X, y)
p.predict(X)
from sklearn.tree import DecisionTreeClassifier
from sklearn.datasets import load_iris, make_classification
from sklearn.model_selection import cross_val_score, GridSearchCV, ShuffleSplit
from skclean.detectors.neighbors import KDN
from skclean.detectors.ensemble import RandomForestDetector
from skclean.handlers.filters import Filter
from skclean.pipeline import Pipeline, make_pipeline
from skclean.simulate_noise import UniformNoise, CCNoise
X, y = load_iris(return_X_y=True)
lcm = [[.7, .1, .2], [.3, .6, .1], [.1, .21, .69]]
CV1 = ShuffleSplit(n_splits=1, test_size=.23)
@pytest.mark.parametrize('sim', [UniformNoise(.3), CCNoise(lcm)])
def test_noise_sim(sim):
p = Pipeline([('s', sim), ('c', DecisionTreeClassifier())])
p.fit(X, y)
p.predict(X)
p = make_pipeline(sim, DecisionTreeClassifier())
p.fit(X, y)
p.predict(X)
def test_stack_sim():
p = Pipeline([
('a', UniformNoise(.3)),
('b', CCNoise(lcm)),
('c', DecisionTreeClassifier()),