def test_svr():
# Test Support Vector Regression
diabetes = datasets.load_diabetes()
for clf in (svm.NuSVR(kernel='linear', nu=.4,
C=1.0), svm.NuSVR(kernel='linear', nu=.4, C=10.),
svm.SVR(kernel='linear',
C=10.), svm.LinearSVR(C=10.), svm.LinearSVR(C=10.)):
clf.fit(diabetes.data, diabetes.target)
assert clf.score(diabetes.data, diabetes.target) > 0.02
# non-regression test; previously, BaseLibSVM would check that
# len(np.unique(y)) < 2, which must only be done for SVC
svm.SVR().fit(diabetes.data, np.ones(len(diabetes.data)))
svm.LinearSVR().fit(diabetes.data, np.ones(len(diabetes.data)))
def test_sparse_fit_support_vectors_empty():
# Regression test for #14893
X_train = sparse.csr_matrix([[0, 1, 0, 0], [0, 0, 0, 1], [0, 0, 1, 0],
[0, 0, 0, 1]])
y_train = np.array([0.04, 0.04, 0.10, 0.16])
model = svm.SVR(kernel='linear')
model.fit(X_train, y_train)
assert not model.support_vectors_.data.size
assert not model.dual_coef_.data.size
def test_svr_errors():
X = [[0.0], [1.0]]
y = [0.0, 0.5]
# Bad kernel
clf = svm.SVR(kernel=lambda x, y: np.array([[1.0]]))
clf.fit(X, y)
with pytest.raises(ValueError):
clf.predict(X)
def test_svr_predict():
# Test SVR's decision_function
# Sanity check, test that predict implemented in python
# returns the same as the one in libsvm
X = iris.data
y = iris.target
# linear kernel
reg = svm.SVR(kernel='linear', C=0.1).fit(X, y)
dec = np.dot(X, reg.coef_.T) + reg.intercept_
assert_array_almost_equal(dec.ravel(), reg.predict(X).ravel())
# rbf kernel
reg = svm.SVR(kernel='rbf', gamma=1).fit(X, y)
rbfs = rbf_kernel(X, reg.support_vectors_, gamma=reg.gamma)
dec = np.dot(rbfs, reg.dual_coef_.T) + reg.intercept_
assert_array_almost_equal(dec.ravel(), reg.predict(X).ravel())
def test_linearsvr():
# check that SVR(kernel='linear') and LinearSVC() give
# comparable results
diabetes = datasets.load_diabetes()
lsvr = svm.LinearSVR(C=1e3).fit(diabetes.data, diabetes.target)
score1 = lsvr.score(diabetes.data, diabetes.target)
svr = svm.SVR(kernel='linear', C=1e3).fit(diabetes.data, diabetes.target)
score2 = svr.score(diabetes.data, diabetes.target)
assert_allclose(np.linalg.norm(lsvr.coef_), np.linalg.norm(svr.coef_), 1,
0.0001)
assert_almost_equal(score1, score2, 2)
def test_immutable_coef_property():
# Check that primal coef modification are not silently ignored
svms = [
svm.SVC(kernel='linear').fit(iris.data, iris.target),
svm.NuSVC(kernel='linear').fit(iris.data, iris.target),
svm.SVR(kernel='linear').fit(iris.data, iris.target),
svm.NuSVR(kernel='linear').fit(iris.data, iris.target),
svm.OneClassSVM(kernel='linear').fit(iris.data),
]
for clf in svms:
with pytest.raises(AttributeError):
clf.__setattr__('coef_', np.arange(3))
with pytest.raises((RuntimeError, ValueError)):
clf.coef_.__setitem__((0, 0), 0)
def test_svr_coef_sign():
# Test that SVR(kernel="linear") has coef_ with the right sign.
# Non-regression test for #2933.
X = np.random.RandomState(21).randn(10, 3)
y = np.random.RandomState(12).randn(10)
for svr in [
svm.SVR(kernel='linear'),
svm.NuSVR(kernel='linear'),
svm.LinearSVR()
]:
svr.fit(X, y)
assert_array_almost_equal(
svr.predict(X),
np.dot(X, svr.coef_.ravel()) + svr.intercept_)
def test_n_support_oneclass_svr():
# Make n_support is correct for oneclass and SVR (used to be
# non-initialized)
# this is a non regression test for issue #14774
X = np.array([[0], [0.44], [0.45], [0.46], [1]])
clf = svm.OneClassSVM()
assert not hasattr(clf, 'n_support_')
clf.fit(X)
assert clf.n_support_ == clf.support_vectors_.shape[0]
assert clf.n_support_.size == 1
assert clf.n_support_ == 3
y = np.arange(X.shape[0])
reg = svm.SVR().fit(X, y)
assert reg.n_support_ == reg.support_vectors_.shape[0]
assert reg.n_support_.size == 1
assert reg.n_support_ == 4
def test_ovr_single_label_predict_proba():
base_clf = MultinomialNB(alpha=1)
X, Y = iris.data, iris.target
X_train, Y_train = X[:80], Y[:80]
X_test = X[80:]
clf = OneVsRestClassifier(base_clf).fit(X_train, Y_train)
# Decision function only estimator.
decision_only = OneVsRestClassifier(svm.SVR()).fit(X_train, Y_train)
assert not hasattr(decision_only, 'predict_proba')
Y_pred = clf.predict(X_test)
Y_proba = clf.predict_proba(X_test)
assert_almost_equal(Y_proba.sum(axis=1), 1.0)
# predict assigns a label if the probability that the
# sample has the label is greater than 0.5.
pred = np.array([l.argmax() for l in Y_proba])
assert not (pred - Y_pred).any()
def test_ovr_multilabel_predict_proba():
base_clf = MultinomialNB(alpha=1)
for au in (False, True):
X, Y = datasets.make_multilabel_classification(n_samples=100,
n_features=20,
n_classes=5,
n_labels=3,
length=50,
allow_unlabeled=au,
random_state=0)
X_train, Y_train = X[:80], Y[:80]
X_test = X[80:]
clf = OneVsRestClassifier(base_clf).fit(X_train, Y_train)
# Decision function only estimator.
decision_only = OneVsRestClassifier(svm.SVR()).fit(X_train, Y_train)
assert not hasattr(decision_only, 'predict_proba')
# Estimator with predict_proba disabled, depending on parameters.
decision_only = OneVsRestClassifier(svm.SVC(probability=False))
assert not hasattr(decision_only, 'predict_proba')
decision_only.fit(X_train, Y_train)
assert not hasattr(decision_only, 'predict_proba')
assert hasattr(decision_only, 'decision_function')
# Estimator which can get predict_proba enabled after fitting
gs = GridSearchCV(svm.SVC(probability=False),
param_grid={'probability': [True]})
proba_after_fit = OneVsRestClassifier(gs)
assert not hasattr(proba_after_fit, 'predict_proba')
proba_after_fit.fit(X_train, Y_train)
assert hasattr(proba_after_fit, 'predict_proba')
Y_pred = clf.predict(X_test)
Y_proba = clf.predict_proba(X_test)
# predict assigns a label if the probability that the
# sample has the label is greater than 0.5.
pred = Y_proba > .5
assert_array_equal(pred, Y_pred)
y_pred = estimator.decision_function([[-1., 1.]])
assert y_pred == pytest.approx(0)
# give more weights to opposed samples
sample_weight = [10., .1, .1, .1, .1, 10]
estimator.fit(X, Y, sample_weight=sample_weight)
y_pred = estimator.decision_function([[-1., 1.]])
assert y_pred < 0
sample_weight = [1., .1, 10., 10., .1, .1]
estimator.fit(X, Y, sample_weight=sample_weight)
y_pred = estimator.decision_function([[-1., 1.]])
assert y_pred > 0
@pytest.mark.parametrize("estimator", [svm.SVR(C=1e-2), svm.NuSVR(C=1e-2)])
def test_svm_regressor_sided_sample_weight(estimator):
# similar test to test_svm_classifier_sided_sample_weight but for
# SVM regressors
X = [[-2, 0], [-1, -1], [0, -2], [0, 2], [1, 1], [2, 0]]
estimator.set_params(kernel='linear')
# check that with unit weights, a sample is supposed to be predicted on
# the boundary
sample_weight = [1] * 6
estimator.fit(X, Y, sample_weight=sample_weight)
y_pred = estimator.predict([[-1., 1.]])
assert y_pred == pytest.approx(1.5)
# give more weights to opposed samples
sample_weight = [10., .1, .1, .1, .1, 10]
