def test_on_small_data(self):
# ---------------------------------------------
__X_A = np.array([[0, 1, 2], [1, 0, 0], [3, 4, 0]])
__X_B = np.array([[0, 0, 1], [3, 1, 0]])
__K = minmax_kernel(__X_A)
np.testing.assert_array_equal(np.diag(__K), np.ones((3, )))
np.testing.assert_equal(__K.shape, (3, 3))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
np.testing.assert_equal(__K[0, 1], 0.)
np.testing.assert_equal(__K[1, 0], 0.)
np.testing.assert_equal(__K[0, 2], 1. / 9.)
np.testing.assert_equal(__K[2, 0], 1. / 9.)
np.testing.assert_equal(__K[1, 2], 1. / 7.)
np.testing.assert_equal(__K[2, 1], 1. / 7.)
__K = minmax_kernel(__X_A, __X_B)
np.testing.assert_equal(__K.shape, (3, 2))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
np.testing.assert_equal(__K[0, 1], 1. / 6.)
np.testing.assert_equal(__K[1, 0], 0.)
np.testing.assert_equal(__K[1, 1], 1. / 4.)
np.testing.assert_equal(__K[2, 1], 4. / 7.)
def test_compatibility_with_sparse_matrix(self):
# ---------------------------------------------
__X_A = csr_matrix(np.array([[0, 1, 2], [1, 0, 0], [3, 4, 0]]))
__X_B = csr_matrix(np.array([[0, 0, 1], [3, 1, 0]]))
__K = minmax_kernel(__X_A)
np.testing.assert_array_equal(np.diag(__K), np.ones((3, )))
np.testing.assert_equal(__K.shape, (3, 3))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
np.testing.assert_equal(__K[0, 1], 0.)
np.testing.assert_equal(__K[1, 0], 0.)
np.testing.assert_equal(__K[0, 2], 1. / 9.)
np.testing.assert_equal(__K[2, 0], 1. / 9.)
np.testing.assert_equal(__K[1, 2], 1. / 7.)
np.testing.assert_equal(__K[2, 1], 1. / 7.)
__K = minmax_kernel(__X_A, __X_B)
np.testing.assert_equal(__K.shape, (3, 2))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
np.testing.assert_equal(__K[0, 1], 1. / 6.)
np.testing.assert_equal(__K[1, 0], 0.)
np.testing.assert_equal(__K[1, 1], 1. / 4.)
np.testing.assert_equal(__K[2, 1], 4. / 7.)
def test_on_larger_random_data_with_sparse_matrix(self):
n_A = 45
n_B = 21
d = 121
X_A = np.random.RandomState(5943).randint(0, 43, size=(n_A, d))
X_B = np.random.RandomState(842).randint(0, 43, size=(n_B, d))
# Symmetric kernel
K = minmax_kernel(csr_matrix(X_A))
np.testing.assert_array_equal(K, self._minmax_kernel_slow(X_A))
np.testing.assert_array_equal(np.diag(K), np.ones((n_A, )))
assert (np.max(K) <= 1.)
assert (np.min(K) >= 0.)
# Non-symmetric kernel
K = minmax_kernel(csr_matrix(X_A), csr_matrix(X_B))
np.testing.assert_array_equal(K, self._minmax_kernel_slow(X_A, X_B))
assert (np.max(K) <= 1.)
assert (np.min(K) >= 0.)
def test_corner_cases(self):
# Empty features
# ---------------------------------------------
__X_A = np.array([[0, 1, 2, 0], [1, 0, 0, 0], [3, 4, 0, 0]])
__X_B = np.array([[0, 0, 1, 0], [3, 1, 0, 0]])
__K = minmax_kernel(__X_A, __X_B)
np.testing.assert_equal(__K.shape, (3, 2))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
np.testing.assert_equal(__K[0, 1], 1. / 6.)
np.testing.assert_equal(__K[1, 0], 0.)
np.testing.assert_equal(__K[1, 1], 1. / 4.)
np.testing.assert_equal(__K[2, 1], 4. / 7.)
# Empty features on larger random set
# ---------------------------------------------
n_A = 92
n_B = 29
d = 103
X_A = np.random.RandomState(59).randint(0, 90, size=(n_A, d))
X_B = np.random.RandomState(8443).randint(0, 43, size=(n_B, d))
X_A[:, 12] = 0
X_A[:, 10] = 0
X_A[:, 54] = 0
X_B[:, 12] = 0
X_B[:, 10] = 0
X_B[:, 21] = 0
__K = minmax_kernel(X_A)
np.testing.assert_array_equal(__K, self._minmax_kernel_slow(X_A))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
assert all(np.diag(__K) == 1.0)
__K = minmax_kernel(X_A, X_B)
np.testing.assert_array_equal(__K, self._minmax_kernel_slow(X_A, X_B))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
def test_corner_cases_with_sparse_matrix(self):
# Empty features
# ---------------------------------------------
__X_A = csr_matrix(np.array([[0, 1, 2, 0], [1, 0, 0, 0], [3, 4, 0,
0]]))
__X_B = csr_matrix(np.array([[0, 0, 1, 0], [3, 1, 0, 0]]))
__K = minmax_kernel(__X_A, __X_B)
np.testing.assert_equal(__K.shape, (3, 2))
assert (np.max(__K) <= 1.), "Kernel values must be <= 1"
assert (np.min(__K) >= 0.), "Kernel values must be >= 0"
np.testing.assert_equal(__K[0, 1], 1. / 6.)
np.testing.assert_equal(__K[1, 0], 0.)
np.testing.assert_equal(__K[1, 1], 1. / 4.)
np.testing.assert_equal(__K[2, 1], 4. / 7.)