def test_minmax_query_mah(self):
qs_kwargs = {'metric': 'mahalanobis'}
qs = QS.MinMax(**qs_kwargs)
scores = qs.score(*self.args)
self.assertEqual(scores.shape, self.unlabeled.y.shape)
self.assertNotIn(np.NaN, scores)
choice = qs.choose(scores)
# A difference in sklearn versions cause
# a different result to be computed.
if sklearn_ver == version.parse("0.19.0"):
self.assertEqual(choice, 0)
else:
self.assertEqual(choice, 4)
def test_combined_dynamic_beta(self):
qs1 = QS.Entropy()
qs2 = QS.MinMax()
beta = 'dynamic'
qs = QS.CombinedSampler(qs1=qs1,
qs2=qs2,
beta=beta,
choice_metric=np.argmax)
self.assertEqual(qs.beta, 'dynamic')
scores = qs.score(*self.args)
self.assertEqual(scores.shape, self.unlabeled.y.shape)
self.assertNotIn(np.NaN, scores)
choice = qs.choose(scores)
self.assertEqual(choice, 3)
def test_minmax_query_mah_singular(self):
"""Singular matrix"""
qs_kwargs = {'metric': 'mahalanobis'}
XU = np.array([[5, 5, 5, 5, 5]] * 5)
XL = np.array([[3, 3, 3, 3, 3]] * 3)
yU = np.array([0, 1, 0, 1, 0])
yL = np.array([1, 0, 1])
U = Data(X=XU, y=yU)
L = Data(X=XL, y=yL)
args = [U, L, self.clf]
qs = QS.MinMax(**qs_kwargs)
scores = qs.score(*args)
self.assertEqual(scores.shape, self.unlabeled.y.shape)
self.assertNotIn(np.NaN, scores)
choice = qs.choose(scores)
self.assertEqual(choice, 0)
def test_minmax_metric(self):
qs = QS.MinMax()
self.assertEqual(qs.distance_metric, 'euclidean')
def test_distdiv_dyn(self):
ent = QS.Entropy()
mm = QS.MinMax()
qs = QS.DistDivSampler(qs1=ent, qs2=mm, lam="dynamic")
self.run_test(qs)
def test_distdiv_even(self):
ent = QS.Entropy()
mm = QS.MinMax()
qs = QS.DistDivSampler(qs1=ent, qs2=mm, lam=0.5)
self.run_test(qs)
def test_combined_dyn(self):
ent = QS.Entropy()
mm = QS.MinMax()
qs = QS.CombinedSampler(qs1=ent, qs2=mm, beta="dynamic")
self.run_test(qs)
def test_minmax(self):
qs = QS.MinMax()
self.run_test(qs)