def test_EasyMKL(self):
self.base_evaluation(algorithms.EasyMKL())
self.base_evaluation(algorithms.EasyMKL(learner=SVC(C=10)))
self.base_evaluation(
algorithms.EasyMKL(learner=algorithms.KOMD(lam=1)))
self.base_evaluation(
algorithms.EasyMKL(solver='libsvm', learner=SVC(C=10)))
def test_multiclass_ovo(self):
mkl = algorithms.EasyMKL(multiclass_strategy='ovo',
learner=SVC()).fit(self.KLtr, self.Ytr)
clf = multiclass.OneVsOneMKLClassifier(mkl).fit(self.KLtr, self.Ytr)
classes = np.unique(self.Ytr)
n = len(classes)
self.assertEqual(len(mkl.solution), (n * (n - 1) / 2))
c1, c2 = classes[:2]
self.assertListEqual(clf.solution[(c1, c2)].weights.tolist(),
mkl.solution[(c1, c2)].weights.tolist())
def test_parameters(self):
self.assertRaises(ValueError, algorithms.EasyMKL, lam=2)
self.assertRaises(ValueError, algorithms.EasyMKL, lam=1.01)
self.assertRaises(ValueError, algorithms.EasyMKL, lam=-0.1)
self.assertRaises(ValueError, algorithms.EasyMKL, solver=0.1)
algorithms.EasyMKL(solver='libsvm', lam=0.2)