def test_ovr_partial_fit():
# Test if partial_fit is working as intended
X, y = shuffle(iris.data, iris.target, random_state=0)
ovr = OneVsRestClassifier(MultinomialNB())
ovr.partial_fit(X[:100], y[:100], np.unique(y))
ovr.partial_fit(X[100:], y[100:])
pred = ovr.predict(X)
ovr2 = OneVsRestClassifier(MultinomialNB())
pred2 = ovr2.fit(X, y).predict(X)
assert_almost_equal(pred, pred2)
assert len(ovr.estimators_) == len(np.unique(y))
assert np.mean(y == pred) > 0.65
# Test when mini batches doesn't have all classes
# with SGDClassifier
X = np.abs(np.random.randn(14, 2))
y = [1, 1, 1, 1, 2, 3, 3, 0, 0, 2, 3, 1, 2, 3]
ovr = OneVsRestClassifier(SGDClassifier(max_iter=1, tol=None,
shuffle=False, random_state=0))
ovr.partial_fit(X[:7], y[:7], np.unique(y))
ovr.partial_fit(X[7:], y[7:])
pred = ovr.predict(X)
ovr1 = OneVsRestClassifier(SGDClassifier(max_iter=1, tol=None,
shuffle=False, random_state=0))
pred1 = ovr1.fit(X, y).predict(X)
assert np.mean(pred == y) == np.mean(pred1 == y)
# test partial_fit only exists if estimator has it:
ovr = OneVsRestClassifier(SVC())
assert not hasattr(ovr, "partial_fit")
def test_ovr_coef_exceptions():
# Not fitted exception!
ovr = OneVsRestClassifier(LinearSVC(random_state=0))
# lambda is needed because we don't want coef_ to be evaluated right away
assert_raises(ValueError, lambda x: ovr.coef_, None)
# Doesn't have coef_ exception!
ovr = OneVsRestClassifier(DecisionTreeClassifier())
ovr.fit(iris.data, iris.target)
assert_raises(AttributeError, lambda x: ovr.coef_, None)
def test_ovr_pipeline():
# Test with pipeline of length one
# This test is needed because the multiclass estimators may fail to detect
# the presence of predict_proba or decision_function.
clf = Pipeline([("tree", DecisionTreeClassifier())])
ovr_pipe = OneVsRestClassifier(clf)
ovr_pipe.fit(iris.data, iris.target)
ovr = OneVsRestClassifier(DecisionTreeClassifier())
ovr.fit(iris.data, iris.target)
assert_array_equal(ovr.predict(iris.data), ovr_pipe.predict(iris.data))
def test_ovr_fit_predict():
# A classifier which implements decision_function.
ovr = OneVsRestClassifier(LinearSVC(random_state=0))
pred = ovr.fit(iris.data, iris.target).predict(iris.data)
assert len(ovr.estimators_) == n_classes
clf = LinearSVC(random_state=0)
pred2 = clf.fit(iris.data, iris.target).predict(iris.data)
assert np.mean(iris.target == pred) == np.mean(iris.target == pred2)
# A classifier which implements predict_proba.
ovr = OneVsRestClassifier(MultinomialNB())
pred = ovr.fit(iris.data, iris.target).predict(iris.data)
assert np.mean(iris.target == pred) > 0.65
def test_ovr_coef_():
for base_classifier in [SVC(kernel='linear', random_state=0),
LinearSVC(random_state=0)]:
# SVC has sparse coef with sparse input data
ovr = OneVsRestClassifier(base_classifier)
for X in [iris.data, sp.csr_matrix(iris.data)]:
# test with dense and sparse coef
ovr.fit(X, iris.target)
shape = ovr.coef_.shape
assert shape[0] == n_classes
assert shape[1] == iris.data.shape[1]
# don't densify sparse coefficients
assert (sp.issparse(ovr.estimators_[0].coef_) ==
sp.issparse(ovr.coef_))
def test_ovr_ovo_regressor():
# test that ovr and ovo work on regressors which don't have a decision_
# function
ovr = OneVsRestClassifier(DecisionTreeRegressor())
pred = ovr.fit(iris.data, iris.target).predict(iris.data)
assert len(ovr.estimators_) == n_classes
assert_array_equal(np.unique(pred), [0, 1, 2])
# we are doing something sensible
assert np.mean(pred == iris.target) > .9
ovr = OneVsOneClassifier(DecisionTreeRegressor())
pred = ovr.fit(iris.data, iris.target).predict(iris.data)
assert len(ovr.estimators_) == n_classes * (n_classes - 1) / 2
assert_array_equal(np.unique(pred), [0, 1, 2])
# we are doing something sensible
assert np.mean(pred == iris.target) > .9
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)
def test_ovr_fit_predict_svc():
ovr = OneVsRestClassifier(svm.SVC())
ovr.fit(iris.data, iris.target)
assert len(ovr.estimators_) == 3
assert ovr.score(iris.data, iris.target) > .9