def test_calc_mahalanobis(self):
rdm = rsr.calc_rdm_unbalanced(
self.test_data, descriptor='conds',
method='mahalanobis')
assert rdm.n_cond == 6
rdm_bal = rsr.calc_rdm_unbalanced(
self.test_data, descriptor='conds',
method='mahalanobis')
rdm_check = rsr.calc_rdm(
self.test_data, descriptor='conds',
method='mahalanobis')
assert_array_almost_equal(
rdm_bal.dissimilarities.flatten(),
rdm_check.dissimilarities.flatten()
)
def test_calc_euclid_nconds(self):
d = self.test_data
rdm = rsr.calc_rdm_unbalanced(
[d, d], descriptor='conds',
method='euclidean')
assert rdm.n_cond == 6
assert rdm.n_rdm == 2
def test_calc_poisson_cv(self):
rdm = rsr.calc_rdm_unbalanced(
self.test_data,
descriptor='conds',
cv_descriptor='fold',
method='poisson_cv')
assert rdm.n_cond == 6
def test_calc_poisson(self):
""" for the poisson-KL the dissimilarities differ! This is explained
in more detail in the demo on this computation"""
rdm = rsr.calc_rdm_unbalanced(
self.test_data,
descriptor='conds',
method='poisson'
)
assert rdm.n_cond == 6
def test_calc_crossnobis_noise(self):
noise = np.random.randn(10, 5)
noise = np.matmul(noise.T, noise)
rdm = rsr.calc_rdm_unbalanced(self.test_data,
descriptor='conds', cv_descriptor='fold',
noise=noise,
method='crossnobis')
assert rdm.n_cond == 6
rdm_bal = rsr.calc_rdm_unbalanced(
self.test_data_balanced, descriptor='conds',
cv_descriptor='fold',
noise=noise,
method='crossnobis')
rdm_check = rsr.calc_rdm(
self.test_data_balanced, descriptor='conds',
cv_descriptor='fold',
noise=noise,
method='crossnobis')
assert_array_almost_equal(
rdm_bal.dissimilarities.flatten(),
rdm_check.dissimilarities.flatten()
)
def test_calc_poisson_extreme_pairs(self):
"""Check the dissimilarities computed with the 'poisson' method
The closest pair should be that between the condition 1 and itself
The furthest pair should be that between condition 1 and condition 3
"""
rdm = rsr.calc_rdm_unbalanced(
self.test_data_deterministic,
descriptor='conds',
method='poisson'
)
rdm_array = squareform(rdm.get_vectors()[0, :])
closest_pair_index = np.argmin(rdm_array)
furthest_pair_index = np.argmax(rdm_array)
self.assertEqual(closest_pair_index, 0)
self.assertEqual(furthest_pair_index, 2)
def test_calc_euclid_as_scipy(self, _parse_input):
from rsatoolbox.rdm import calc_rdm_unbalanced
data = rsa.data.Dataset(
np.random.rand(6, 5),
descriptors={'session': 0, 'subj': 0},
obs_descriptors={'conds': [0, 1, 2, 3, 4, 5]}
)
rdm_expected = pdist(data.measurements) ** 2 / 5
rdms = calc_rdm_unbalanced(
data,
descriptor='conds',
method='euclidean'
)
self.assertIsNone(
assert_array_almost_equal(
rdm_expected,
rdms.dissimilarities.flatten()
)
)
def test_calc_correlation(self, _parse_input):
from rsatoolbox.rdm import calc_rdm_unbalanced
data = rsa.data.Dataset(
np.random.rand(6, 5),
descriptors={'session': 0, 'subj': 0},
obs_descriptors={'conds': [0, 1, 2, 3, 4, 5]}
)
rdm_expected = 1 - np.corrcoef(data.measurements)
rdme = rsr.RDMs(
dissimilarities=np.array([rdm_expected]),
dissimilarity_measure='correlation',
descriptors=data.descriptors)
rdm = calc_rdm_unbalanced(
data,
descriptor='conds',
method='correlation'
)
self.assertIsNone(
assert_array_almost_equal(
rdme.dissimilarities.flatten(),
rdm.dissimilarities.flatten()
)
)
def test_calc_crossnobis_no_descriptor(self):
rdm = rsr.calc_rdm_unbalanced(self.test_data,
descriptor='conds',
method='crossnobis')
assert rdm.n_cond == 6
def test_calc_list_descriptors(self):
rdm = rsr.calc_rdm_unbalanced(
[self.test_data, self.test_data, self.test_data],
descriptor='conds',
method='euclidean')
assert np.all(rdm.rdm_descriptors['subj'] == np.array([0, 0, 0]))
