def test_multiclass_kernel_sgd():
for fit_intercept in (True, False):
clf = KernelSGDClassifier(kernel="rbf", gamma=0.1,
fit_intercept=fit_intercept,
random_state=0)
clf.fit(mult_dense, mult_target)
assert_equal(clf.score(mult_dense, mult_target), 1.0)
def test_n_components_multiclass_natural():
for loss in ("hinge", "log"):
clf = KernelSGDClassifier(loss=loss, multiclass="natural",
kernel="rbf", gamma=0.1, n_components=50,
random_state=0)
clf.fit(mult_dense, mult_target)
assert_equal(clf.n_support_vectors(), 50)
assert_greater(clf.score(mult_dense, mult_target), 0.38)
def test_multiclass_hinge_kernel_sgd():
for fit_intercept in (True, False):
clf = KernelSGDClassifier(kernel="rbf", gamma=0.1,
loss="hinge", multiclass="natural",
fit_intercept=fit_intercept,
random_state=0)
clf.fit(mult_dense, mult_target)
assert_greater(clf.score(mult_dense, mult_target), 0.90)
def test_multiclass_sgd_equivalence():
clf = KernelSGDClassifier(kernel="linear",
random_state=0)
clf.fit(mult_dense, mult_target)
decisions = clf.decision_function(mult_dense)
predictions = clf.predict(mult_dense)
clf = SGDClassifier(random_state=0)
clf.fit(mult_dense, mult_target)
decisions2 = clf.decision_function(mult_dense)
predictions2 = clf.predict(mult_dense)
assert_array_almost_equal(decisions, decisions2)
assert_array_almost_equal(predictions, predictions2)
def test_multiclass_natural_kernel_sgd_equivalence():
for loss in ("hinge", "log"):
clf = KernelSGDClassifier(kernel="linear",
loss=loss, multiclass="natural",
random_state=0)
clf.fit(mult_dense, mult_target)
decisions = clf.decision_function(mult_dense)
predictions = clf.predict(mult_dense)
clf = SGDClassifier(random_state=0, loss=loss, multiclass="natural")
clf.fit(mult_dense, mult_target)
decisions2 = clf.decision_function(mult_dense)
predictions2 = clf.predict(mult_dense)
assert_array_almost_equal(decisions, decisions2)
assert_array_almost_equal(predictions, predictions2)
def test_n_components_multiclass():
clf = KernelSGDClassifier(kernel="rbf", gamma=1.0, loss="hinge",
random_state=0, n_components=50)
clf.fit(mult_dense, mult_target)
assert_equal(clf.n_support_vectors(), 55)
assert_greater(clf.score(mult_dense, mult_target), 0.3)
def test_n_components_binary():
clf = KernelSGDClassifier(kernel="rbf", gamma=0.1, loss="hinge",
random_state=0, n_components=50)
clf.fit(bin_dense, bin_target)
assert_equal(clf.n_support_vectors(), 50)
assert_greater(clf.score(bin_dense, bin_target), 0.6)