def test_regression_corrupted_weights(weighting):
reg = RobustWeightedRegressor(
max_iter=100,
weighting=weighting,
k=5,
c=1,
burn_in=0,
random_state=rng,
)
reg.fit(X_rc, y_rc)
assert reg.weights_[0] < np.mean(reg.weights_[1:])
def test_corrupted_regression(loss, weighting, k, c):
reg = RobustWeightedRegressor(
loss=loss,
max_iter=50,
weighting=weighting,
k=k,
c=None,
random_state=rng,
)
reg.fit(X_rc, y_rc)
score = median_absolute_error(reg.predict(X_rc), y_rc)
assert score < 0.2
def test_corrupted_regression(loss, weighting, k, c):
reg = RobustWeightedRegressor(
loss=loss,
max_iter=50,
weighting=weighting,
k=k,
c=c,
random_state=rng,
n_iter_no_change=20,
)
reg.fit(X_rc, y_rc)
assert np.abs(reg.coef_[0] - 1) < 0.1
assert np.abs(reg.intercept_[0]) < 0.1
def test_vs_huber():
reg1 = RobustWeightedRegressor(
max_iter=100,
weighting="huber",
k=5,
c=1,
burn_in=0,
sgd_args={"learning_rate": "adaptive"}, # test sgd_args
random_state=rng,
)
reg2 = HuberRegressor()
reg1.fit(X_rcy, y_rcy)
reg2.fit(X_rcy, y_rcy)
assert np.abs(reg1.coef_[0] - reg2.coef_[0]) < 1e-2
def test_not_robust_regression(loss, weighting):
clf = RobustWeightedRegressor(
loss=loss,
max_iter=100,
weighting=weighting,
k=0,
c=1e7,
burn_in=0,
random_state=rng,
)
clf_not_rob = SGDRegressor(loss=loss, random_state=rng)
clf.fit(X_r, y_r)
clf_not_rob.fit(X_r, y_r)
pred1 = clf.predict(X_r)
pred2 = clf_not_rob.predict(X_r)
difference = [
np.linalg.norm(pred1[i] - pred2[i]) for i in range(len(pred1))
]
assert np.mean(difference) < 1e-1
def test_not_robust_regression(loss, weighting):
reg = RobustWeightedRegressor(
loss=loss,
max_iter=100,
weighting=weighting,
k=0,
c=1e7,
burn_in=0,
random_state=rng,
)
reg_not_rob = SGDRegressor(loss=loss, random_state=rng)
reg.fit(X_r, y_r)
reg_not_rob.fit(X_r, y_r)
pred1 = reg.predict(X_r)
pred2 = reg_not_rob.predict(X_r)
difference = [
np.linalg.norm(pred1[i] - pred2[i]) for i in range(len(pred1))
]
assert np.mean(difference) < 1
assert_almost_equal(reg.score(X_r, y_r), r2_score(y_r, reg.predict(X_r)))
y[-5:] = -1
# Shuffle the data so that we don't know where the outlier is.
X, y = shuffle(X, y, random_state=rng)
estimators = [
("OLS", LinearRegression()),
("Theil-Sen", TheilSenRegressor(random_state=rng)),
("RANSAC", RANSACRegressor(random_state=rng)),
("HuberRegressor", HuberRegressor()),
(
"SGD epsilon loss",
SGDRegressor(loss="epsilon_insensitive", random_state=rng),
),
(
"RobustWeightedRegressor",
RobustWeightedRegressor(weighting="mom", k=7, random_state=rng),
# The parameter k is set larger to the number of outliers
# because here we know it.
),
]
colors = {
"OLS": "turquoise",
"Theil-Sen": "gold",
"RANSAC": "lightgreen",
"HuberRegressor": "black",
"RobustWeightedRegressor": "magenta",
"SGD epsilon loss": "purple",
}
linestyle = {
"OLS": "-",
# Using GridSearchCV, we do a light tuning of the parameters for SGDRegressor
# and RobustWeightedEstimator. A fine tune is possible but not necessary to
# illustrate the problem of outliers in the output.
estimators = [
(
"SGD",
SGDRegressor(learning_rate="adaptive", eta0=1e-2),
),
(
"RobustWeightedRegressor",
RobustWeightedRegressor(
weighting="huber",
c=0.01,
eta0=1e-2,
sgd_args={
"learning_rate": "adaptive",
"eta0": 1e-3,
},
),
),
("RANSAC", RANSACRegressor()),
("TheilSen", TheilSenRegressor()),
]
M = 10
res = np.zeros(shape=[len(estimators), M, 2])
for f in range(M):
print("\r Progress: %s / %s" % (f + 1, M), end="")
X = RobustScaler().fit_transform(X)
# Using GridSearchCV, we do a light tuning of the parameters for SGDRegressor
# and RobustWeightedEstimator. A fine tune is possible but not necessary to
# illustrate the problem of outliers in the output.
estimators = [
(
"SGD",
SGDRegressor(learning_rate="adaptive", eta0=1e-2),
),
(
"RobustWeightedRegressor",
RobustWeightedRegressor(
weighting="huber",
c=0.1,
eta0=1e-2,
sgd_args={
"learning_rate": "invscaling",
},
),
),
("RANSAC", RANSACRegressor()),
("TheilSen", TheilSenRegressor()),
]
M = 10
res = np.zeros(shape=[len(estimators), M, 2])
for f in range(M):
print("\r Progress: %s / %s" % (f + 1, M), end="")
rng = np.random.RandomState(f)