def test_kbdm_m_none_should_use_default_value(data_brain_sim, dwell):
l = 30
line_lists, info = kbdm(data=data_brain_sim, dwell=dwell, l=l)
assert np.shape(line_lists) == (l, 4)
assert info.m == l
assert info.l == l
def test_kbdm_svd(data_brain_sim, dwell, df_params_brain_sim, columns):
m = 300
params_est, info = kbdm(
data_brain_sim,
dwell,
m=m,
)
assert params_est.shape == (m, 4)
assert info.m == m
assert info.l == m
assert info.q == pytest.approx(0)
assert info.p == 1
df_est = pd.DataFrame(data=params_est, columns=columns)
df_est = df_est[df_est['amplitude'] > 1e-4]
df_est = df_est.sort_values(['frequency'])
assert len(df_est) == 16
for i in range(16):
assert pytest.approx(df_est['amplitude'].iloc[i]) == df_params_brain_sim['amplitude'].iloc[i], \
f'Amplitude does not match at peak #{i}'
assert pytest.approx(df_est['t2'].iloc[i], rel=1e-3) == df_params_brain_sim['t2'].iloc[i], \
f'T2 does not match at peak #{i}'
assert pytest.approx(df_est['frequency'].iloc[i], abs=0.3) == df_params_brain_sim['frequency'].iloc[i], \
f'Frequency does not match at peak #{i}'
assert pytest.approx(df_est['phase'].iloc[i], abs=1e-10) == df_params_brain_sim['phase'].iloc[i], \
f'Phase does not match at peak #{i}'
def test_sample_kbdm(data_brain_sim, dwell):
m_range = range(100, 103, 1)
line_lists, infos = sample_kbdm(data=data_brain_sim,
dwell=dwell,
m_range=m_range,
p=1,
l=None,
q=0,
filter_invalid_features=False)
assert len(line_lists) == len(m_range)
assert len(infos) == len(m_range)
line_list_0_expected, info_expected = kbdm(data=data_brain_sim,
dwell=dwell,
m=m_range[0],
p=1,
l=None)
assert line_lists[0] == pytest.approx(line_list_0_expected)
assert infos[0].m == info_expected.m
assert infos[0].l == info_expected.l
assert infos[0].p == info_expected.p
assert infos[0].q == info_expected.q
assert infos[0].singular_values == pytest.approx(
info_expected.singular_values)
def m_150_kbdm_line_list(data_brain_sim, dwell):
line_list, _ = kbdm(
data=data_brain_sim,
dwell=dwell,
m=150,
)
return line_list
def test_kbdm_svd_with_q_greater_than_zero_should_use_tikhonov_regularization(
data_brain_sim, dwell, caplog):
caplog.set_level('DEBUG')
m = 10
params_est, info = kbdm(
data_brain_sim,
dwell,
m=m,
q=1e-3,
)
assert 'Using Tikhonov Regularization' in caplog.text
assert params_est.shape == (m, 4)
assert info.q == pytest.approx(1e-3)
assert info.m == m
def test_kbdm_should_check_for_m_and_l_values(data_brain_sim, dwell):
with pytest.raises(ValueError) as e_info:
kbdm(data=data_brain_sim, dwell=dwell)
assert "l or m must be specified" in str(e_info.value)
with pytest.raises(ValueError) as e_info:
kbdm(data=data_brain_sim, dwell=dwell, l=30, m=20)
assert "l can't be greater than m" in str(e_info.value)
_, info = kbdm(data=data_brain_sim, dwell=dwell, l=30)
assert info.l == 30
assert info.m == 30
_, info = kbdm(data=data_brain_sim, dwell=dwell, m=30)
assert info.l == 30
assert info.m == 30
def test_kbdm_invalid_m_n_p_constraint_should_raise_value_error(
data_brain_sim, dwell):
with pytest.raises(ValueError) as e_info:
kbdm(data=data_brain_sim,
dwell=dwell,
m=int(len(data_brain_sim) / 2) + 1)
assert "m or l can't be greater than (n + 1 - p)/2." in str(e_info.value)
with pytest.raises(ValueError) as e_info:
kbdm(data=data_brain_sim,
dwell=dwell,
l=int(len(data_brain_sim) / 2) + 1,
m=int(len(data_brain_sim) / 2) + 1)
with pytest.raises(ValueError) as e_info:
kbdm(data=data_brain_sim,
dwell=dwell,
l=10,
m=int(len(data_brain_sim) / 2) + 1)
assert "m or l can't be greater than (n + 1 - p)/2." in str(e_info.value)
def sample_kbdm(data, dwell, m_range, p, l, q=0, filter_invalid_features=True):
"""
Compute samples of multiple computations of KBDM in a given range of m values.
:param numpy.ndarray data:
Complex input data.
:param float dwell:
Dwell time in seconds.
:param list|range m_range:
Range or list with number of columns/rows of U matrices for iteration of KBDM sampling.
Its size must be greater than 2.
:param str gep_solver:
Method used to solve Generalized Eigenvalue Problem. Can be 'svd', for self-implemented solution; or scipy
to use eig function from scipy.linalg.eig.
Default is 'svd'.
:param int p:
Eigenvalue exponent of the generalized eigenvalue equation. It will represent a 'shift' during the construction
of U^p and U^{p-1} matrices.
:param int|TypeNone l:
This is used only with if gep_solver is set to 'svd'.
..see:: llckbdm.kbdm._solve_gep_svd
Default is None.
:param float q:
This is used only with if gep_solver is set to 'svd'.
..see:: llckbdm.kbdm._solve_gep_svd
Default is 0.
:param bool filter_invalid_features:
Apply filter_samples on each sample.
Default is True.
:return: Tuple containing a list with obtained features for each value of m and respective dictionary containing
KBDM computations meta-information.
:rtype: tuple(list(numpy.array), dict)
"""
line_lists = []
infos = []
for m in m_range:
logger.info(f'Computing KBDM with m = {m}')
line_list, info = kbdm(
data=data,
dwell=dwell,
m=m,
p=p,
l=l,
q=q,
)
if filter_invalid_features:
line_list = filter_samples(line_list)
if len(line_list) > 0:
logger.debug(f'Empty line list')
line_lists.append(line_list)
infos.append(info)
return line_lists, infos