def test_mxne_inverse_sure():
"""Tests SURE criterion for automatic alpha selection on MEG data."""
def data_fun(times):
data = np.zeros(times.shape)
data[times >= 0] = 50e-9
return data
n_dipoles = 2
raw = mne.io.read_raw_fif(fname_raw)
info = mne.io.read_info(fname_data)
with info._unlock():
info['projs'] = []
noise_cov = mne.make_ad_hoc_cov(info)
label_names = ['Aud-lh', 'Aud-rh']
labels = [mne.read_label(data_path + '/MEG/sample/labels/%s.label' % ln)
for ln in label_names]
fname_fwd = op.join(data_path, 'MEG', 'sample',
'sample_audvis_trunc-meg-eeg-oct-4-fwd.fif')
forward = mne.read_forward_solution(fname_fwd)
forward = mne.pick_types_forward(forward, meg="grad", eeg=False,
exclude=raw.info['bads'])
times = np.arange(100, dtype=np.float64) / raw.info['sfreq'] - 0.1
stc = simulate_sparse_stc(forward['src'], n_dipoles=n_dipoles, times=times,
random_state=1, labels=labels, data_fun=data_fun)
nave = 30
evoked = simulate_evoked(forward, stc, info, noise_cov, nave=nave,
use_cps=False, iir_filter=None)
evoked = evoked.crop(tmin=0, tmax=10e-3)
stc_ = mixed_norm(evoked, forward, noise_cov, loose=0.9, n_mxne_iter=5,
depth=0.9)
assert_array_equal(stc_.vertices, stc.vertices)
def test_mxne_inverse():
"""Test MxNE inverse computation"""
alpha = 60 # spatial regularization parameter
stc_prox = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=0.9, maxit=1000, tol=1e-8, active_set_size=10,
solver='prox')
stc_cd = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=0.9, maxit=1000, tol=1e-8, active_set_size=10,
solver='cd')
assert_array_almost_equal(stc_prox.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_cd.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_prox.data, stc_cd.data, 5)
assert_true(stc_prox.vertno[1][0] in label.vertices)
assert_true(stc_cd.vertno[1][0] in label.vertices)
stc, _ = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=depth, maxit=500, tol=1e-4, active_set_size=10,
weights=stc_dspm, weights_min=weights_min,
return_residual=True)
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert_true(stc.vertno[1][0] in label.vertices)
def test_mxne_inverse_empty():
"""Tests solver with too high alpha."""
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.pick("grad", exclude="bads")
fname_fwd = op.join(data_path, 'MEG', 'sample',
'sample_audvis_trunc-meg-eeg-oct-4-fwd.fif')
forward = mne.read_forward_solution(fname_fwd)
forward = mne.pick_types_forward(forward, meg="grad", eeg=False,
exclude=evoked.info['bads'])
cov = read_cov(fname_cov)
with pytest.warns(RuntimeWarning, match='too big'):
stc, residual = mixed_norm(
evoked, forward, cov, n_mxne_iter=3, alpha=99,
return_residual=True)
assert stc.data.size == 0
assert stc.vertices[0].size == 0
assert stc.vertices[1].size == 0
assert_allclose(evoked.data, residual.data)
def test_mxne_vol_sphere():
"""(TF-)MxNE with a sphere forward and volumic source space"""
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.05, tmax=0.2)
cov = read_cov(fname_cov)
evoked_l21 = evoked.copy()
evoked_l21.crop(tmin=0.081, tmax=0.1)
info = evoked.info
sphere = mne.make_sphere_model(r0=(0., 0., 0.), head_radius=0.080)
src = mne.setup_volume_source_space(subject=None, pos=15., mri=None,
sphere=(0.0, 0.0, 0.0, 80.0),
bem=None, mindist=5.0,
exclude=2.0)
fwd = mne.make_forward_solution(info, trans=None, src=src,
bem=sphere, eeg=False, meg=True)
alpha = 80.
assert_raises(ValueError, mixed_norm, evoked, fwd, cov, alpha,
loose=None, return_residual=False,
maxit=3, tol=1e-8, active_set_size=10)
assert_raises(ValueError, mixed_norm, evoked, fwd, cov, alpha,
loose=0.2, return_residual=False,
maxit=3, tol=1e-8, active_set_size=10)
# irMxNE tests
stc = mixed_norm(evoked_l21, fwd, cov, alpha,
n_mxne_iter=1, maxit=30, tol=1e-8,
active_set_size=10)
assert_true(isinstance(stc, VolSourceEstimate))
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
# Do with TF-MxNE for test memory savings
alpha_space = 60. # spatial regularization parameter
alpha_time = 1. # temporal regularization parameter
stc, _ = tf_mixed_norm(evoked, fwd, cov, alpha_space, alpha_time,
maxit=3, tol=1e-4,
tstep=16, wsize=32, window=0.1,
return_residual=True)
assert_true(isinstance(stc, VolSourceEstimate))
assert_array_almost_equal(stc.times, evoked.times, 5)
def test_mxne_vol_sphere():
"""(TF-)MxNE with a sphere forward and volumic source space"""
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.05, tmax=0.2)
cov = read_cov(fname_cov)
evoked_l21 = evoked.copy()
evoked_l21.crop(tmin=0.081, tmax=0.1)
info = evoked.info
sphere = mne.make_sphere_model(r0=(0., 0., 0.), head_radius=0.080)
src = mne.setup_volume_source_space(subject=None, pos=15., mri=None,
sphere=(0.0, 0.0, 0.0, 80.0),
bem=None, mindist=5.0,
exclude=2.0)
fwd = mne.make_forward_solution(info, trans=None, src=src,
bem=sphere, eeg=False, meg=True)
alpha = 80.
assert_raises(ValueError, mixed_norm, evoked, fwd, cov, alpha,
loose=0.0, return_residual=False,
maxit=3, tol=1e-8, active_set_size=10)
assert_raises(ValueError, mixed_norm, evoked, fwd, cov, alpha,
loose=0.2, return_residual=False,
maxit=3, tol=1e-8, active_set_size=10)
# irMxNE tests
stc = mixed_norm(evoked_l21, fwd, cov, alpha,
n_mxne_iter=1, maxit=30, tol=1e-8,
active_set_size=10)
assert_true(isinstance(stc, VolSourceEstimate))
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
# Do with TF-MxNE for test memory savings
alpha_space = 60. # spatial regularization parameter
alpha_time = 1. # temporal regularization parameter
stc, _ = tf_mixed_norm(evoked, fwd, cov, alpha_space, alpha_time,
maxit=3, tol=1e-4,
tstep=16, wsize=32, window=0.1,
return_residual=True)
assert_true(isinstance(stc, VolSourceEstimate))
assert_array_almost_equal(stc.times, evoked.times, 5)
alpha = 55 # regularization parameter between 0 and 100 (100 is high)
loose, depth = 0.2, 0.9 # loose orientation & depth weighting
n_mxne_iter = 10 # if > 1 use L0.5/L2 reweighted mixed norm solver
# if n_mxne_iter > 1 dSPM weighting can be avoided.
# Compute dSPM solution to be used as weights in MxNE
inverse_operator = make_inverse_operator(evoked.info, forward, cov,
depth=depth, fixed=True,
use_cps=True)
stc_dspm = apply_inverse(evoked, inverse_operator, lambda2=1. / 9.,
method='dSPM')
# Compute (ir)MxNE inverse solution with dipole output
dipoles, residual = mixed_norm(
evoked, forward, cov, alpha, loose=loose, depth=depth, maxit=3000,
tol=1e-4, active_set_size=10, debias=True, weights=stc_dspm,
weights_min=8., n_mxne_iter=n_mxne_iter, return_residual=True,
return_as_dipoles=True)
###############################################################################
# Plot dipole activations
plot_dipole_amplitudes(dipoles)
# Plot dipole location of the strongest dipole with MRI slices
idx = np.argmax([np.max(np.abs(dip.amplitude)) for dip in dipoles])
plot_dipole_locations(dipoles[idx], forward['mri_head_t'], 'sample',
subjects_dir=subjects_dir, mode='orthoview',
idx='amplitude')
# Plot dipole locations of all dipoles with MRI slices
for dip in dipoles:
def test_mxne_inverse():
"""Test (TF-)MxNE inverse computation"""
# Read noise covariance matrix
cov = read_cov(fname_cov)
# Handling average file
loose = 0.0
depth = 0.9
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.05, tmax=0.2)
evoked_l21 = evoked.copy()
evoked_l21.crop(tmin=0.081, tmax=0.1)
label = read_label(fname_label)
forward = read_forward_solution(fname_fwd)
forward = convert_forward_solution(forward, surf_ori=True)
# Reduce source space to make test computation faster
inverse_operator = make_inverse_operator(evoked_l21.info,
forward,
cov,
loose=loose,
depth=depth,
fixed=True,
use_cps=True)
stc_dspm = apply_inverse(evoked_l21,
inverse_operator,
lambda2=1. / 9.,
method='dSPM')
stc_dspm.data[np.abs(stc_dspm.data) < 12] = 0.0
stc_dspm.data[np.abs(stc_dspm.data) >= 12] = 1.
weights_min = 0.5
# MxNE tests
alpha = 70 # spatial regularization parameter
stc_prox = mixed_norm(evoked_l21,
forward,
cov,
alpha,
loose=loose,
depth=depth,
maxit=300,
tol=1e-8,
active_set_size=10,
weights=stc_dspm,
weights_min=weights_min,
solver='prox')
stc_cd = mixed_norm(evoked_l21,
forward,
cov,
alpha,
loose=loose,
depth=depth,
maxit=300,
tol=1e-8,
active_set_size=10,
weights=stc_dspm,
weights_min=weights_min,
solver='cd')
stc_bcd = mixed_norm(evoked_l21,
forward,
cov,
alpha,
loose=loose,
depth=depth,
maxit=300,
tol=1e-8,
active_set_size=10,
weights=stc_dspm,
weights_min=weights_min,
solver='bcd')
assert_array_almost_equal(stc_prox.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_cd.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_bcd.times, evoked_l21.times, 5)
assert_allclose(stc_prox.data, stc_cd.data, rtol=1e-3, atol=0.0)
assert_allclose(stc_prox.data, stc_bcd.data, rtol=1e-3, atol=0.0)
assert_allclose(stc_cd.data, stc_bcd.data, rtol=1e-3, atol=0.0)
assert_true(stc_prox.vertices[1][0] in label.vertices)
assert_true(stc_cd.vertices[1][0] in label.vertices)
assert_true(stc_bcd.vertices[1][0] in label.vertices)
dips = mixed_norm(evoked_l21,
forward,
cov,
alpha,
loose=loose,
depth=depth,
maxit=300,
tol=1e-8,
active_set_size=10,
weights=stc_dspm,
weights_min=weights_min,
solver='cd',
return_as_dipoles=True)
stc_dip = make_stc_from_dipoles(dips, forward['src'])
assert_true(isinstance(dips[0], Dipole))
_check_stcs(stc_cd, stc_dip)
stc, _ = mixed_norm(evoked_l21,
forward,
cov,
alpha,
loose=loose,
depth=depth,
maxit=300,
tol=1e-8,
active_set_size=10,
return_residual=True,
solver='cd')
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert_true(stc.vertices[1][0] in label.vertices)
# irMxNE tests
stc = mixed_norm(evoked_l21,
forward,
cov,
alpha,
n_mxne_iter=5,
loose=loose,
depth=depth,
maxit=300,
tol=1e-8,
active_set_size=10,
solver='cd')
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert_true(stc.vertices[1][0] in label.vertices)
assert_equal(stc.vertices, [[63152], [79017]])
# Do with TF-MxNE for test memory savings
alpha = 60. # overall regularization parameter
l1_ratio = 0.01 # temporal regularization proportion
stc, _ = tf_mixed_norm(evoked,
forward,
cov,
None,
None,
loose=loose,
depth=depth,
maxit=100,
tol=1e-4,
tstep=4,
wsize=16,
window=0.1,
weights=stc_dspm,
weights_min=weights_min,
return_residual=True,
alpha=alpha,
l1_ratio=l1_ratio)
assert_array_almost_equal(stc.times, evoked.times, 5)
assert_true(stc.vertices[1][0] in label.vertices)
with warnings.catch_warnings(record=True) as w:
assert_raises(ValueError, tf_mixed_norm, evoked, forward, cov, 101.,
3.)
assert_raises(ValueError, tf_mixed_norm, evoked, forward, cov, 50,
101.)
assert_true(len(w) == 2)
assert_raises(ValueError,
tf_mixed_norm,
evoked,
forward,
cov,
None,
None,
alpha=101,
l1_ratio=0.03)
assert_raises(ValueError,
tf_mixed_norm,
evoked,
forward,
cov,
None,
None,
alpha=50.,
l1_ratio=1.01)
def test_mxne_vol_sphere():
"""(TF-)MxNE with a sphere forward and volumic source space"""
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.05, tmax=0.2)
cov = read_cov(fname_cov)
evoked_l21 = evoked.copy()
evoked_l21.crop(tmin=0.081, tmax=0.1)
info = evoked.info
sphere = mne.make_sphere_model(r0=(0., 0., 0.), head_radius=0.080)
src = mne.setup_volume_source_space(subject=None,
pos=15.,
mri=None,
sphere=(0.0, 0.0, 0.0, 80.0),
bem=None,
mindist=5.0,
exclude=2.0)
fwd = mne.make_forward_solution(info,
trans=None,
src=src,
bem=sphere,
eeg=False,
meg=True)
alpha = 80.
assert_raises(ValueError,
mixed_norm,
evoked,
fwd,
cov,
alpha,
loose=0.0,
return_residual=False,
maxit=3,
tol=1e-8,
active_set_size=10)
assert_raises(ValueError,
mixed_norm,
evoked,
fwd,
cov,
alpha,
loose=0.2,
return_residual=False,
maxit=3,
tol=1e-8,
active_set_size=10)
# irMxNE tests
stc = mixed_norm(evoked_l21,
fwd,
cov,
alpha,
n_mxne_iter=1,
maxit=30,
tol=1e-8,
active_set_size=10)
assert_true(isinstance(stc, VolSourceEstimate))
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
# Compare orientation obtained using fit_dipole and gamma_map
# for a simulated evoked containing a single dipole
stc = mne.VolSourceEstimate(50e-9 * np.random.RandomState(42).randn(1, 4),
vertices=stc.vertices[:1],
tmin=stc.tmin,
tstep=stc.tstep)
evoked_dip = mne.simulation.simulate_evoked(fwd,
stc,
info,
cov,
nave=1e9,
use_cps=True)
dip_mxne = mixed_norm(evoked_dip,
fwd,
cov,
alpha=80,
n_mxne_iter=1,
maxit=30,
tol=1e-8,
active_set_size=10,
return_as_dipoles=True)
amp_max = [np.max(d.amplitude) for d in dip_mxne]
dip_mxne = dip_mxne[np.argmax(amp_max)]
assert_true(dip_mxne.pos[0] in src[0]['rr'][stc.vertices])
dip_fit = mne.fit_dipole(evoked_dip, cov, sphere)[0]
assert_true(np.abs(np.dot(dip_fit.ori[0], dip_mxne.ori[0])) > 0.99)
# Do with TF-MxNE for test memory savings
alpha = 60. # overall regularization parameter
l1_ratio = 0.01 # temporal regularization proportion
stc, _ = tf_mixed_norm(evoked,
fwd,
cov,
maxit=3,
tol=1e-4,
tstep=16,
wsize=32,
window=0.1,
alpha=alpha,
l1_ratio=l1_ratio,
return_residual=True)
assert_true(isinstance(stc, VolSourceEstimate))
assert_array_almost_equal(stc.times, evoked.times, 5)
def test_mxne_inverse():
"""Test (TF-)MxNE inverse computation."""
# Read noise covariance matrix
cov = read_cov(fname_cov)
# Handling average file
loose = 0.0
depth = 0.9
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.05, tmax=0.2)
evoked_l21 = evoked.copy()
evoked_l21.crop(tmin=0.081, tmax=0.1)
label = read_label(fname_label)
forward = read_forward_solution(fname_fwd)
forward = convert_forward_solution(forward, surf_ori=True)
# Reduce source space to make test computation faster
inverse_operator = make_inverse_operator(evoked_l21.info, forward, cov,
loose=loose, depth=depth,
fixed=True, use_cps=True)
stc_dspm = apply_inverse(evoked_l21, inverse_operator, lambda2=1. / 9.,
method='dSPM')
stc_dspm.data[np.abs(stc_dspm.data) < 12] = 0.0
stc_dspm.data[np.abs(stc_dspm.data) >= 12] = 1.
weights_min = 0.5
# MxNE tests
alpha = 70 # spatial regularization parameter
stc_prox = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8,
active_set_size=10, weights=stc_dspm,
weights_min=weights_min, solver='prox')
with pytest.warns(None): # CD
stc_cd = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8,
active_set_size=10, weights=stc_dspm,
weights_min=weights_min, solver='cd',
pca=False) # pca=False deprecated, doesn't matter
stc_bcd = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8, active_set_size=10,
weights=stc_dspm, weights_min=weights_min,
solver='bcd')
assert_array_almost_equal(stc_prox.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_cd.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_bcd.times, evoked_l21.times, 5)
assert_allclose(stc_prox.data, stc_cd.data, rtol=1e-3, atol=0.0)
assert_allclose(stc_prox.data, stc_bcd.data, rtol=1e-3, atol=0.0)
assert_allclose(stc_cd.data, stc_bcd.data, rtol=1e-3, atol=0.0)
assert stc_prox.vertices[1][0] in label.vertices
assert stc_cd.vertices[1][0] in label.vertices
assert stc_bcd.vertices[1][0] in label.vertices
with pytest.warns(None): # CD
dips = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8, active_set_size=10,
weights=stc_dspm, weights_min=weights_min,
solver='cd', return_as_dipoles=True)
stc_dip = make_stc_from_dipoles(dips, forward['src'])
assert isinstance(dips[0], Dipole)
assert stc_dip.subject == "sample"
_check_stcs(stc_cd, stc_dip)
with pytest.warns(None): # CD
stc, _ = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8,
active_set_size=10, return_residual=True,
solver='cd')
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert stc.vertices[1][0] in label.vertices
# irMxNE tests
with pytest.warns(None): # CD
stc = mixed_norm(evoked_l21, forward, cov, alpha,
n_mxne_iter=5, loose=loose, depth=depth,
maxit=300, tol=1e-8, active_set_size=10,
solver='cd')
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert stc.vertices[1][0] in label.vertices
assert stc.vertices == [[63152], [79017]]
# Do with TF-MxNE for test memory savings
alpha = 60. # overall regularization parameter
l1_ratio = 0.01 # temporal regularization proportion
stc, _ = tf_mixed_norm(evoked, forward, cov,
loose=loose, depth=depth, maxit=100, tol=1e-4,
tstep=4, wsize=16, window=0.1, weights=stc_dspm,
weights_min=weights_min, return_residual=True,
alpha=alpha, l1_ratio=l1_ratio)
assert_array_almost_equal(stc.times, evoked.times, 5)
assert stc.vertices[1][0] in label.vertices
pytest.raises(ValueError, tf_mixed_norm, evoked, forward, cov,
alpha=101, l1_ratio=0.03)
pytest.raises(ValueError, tf_mixed_norm, evoked, forward, cov,
alpha=50., l1_ratio=1.01)
def test_mxne_vol_sphere():
"""Test (TF-)MxNE with a sphere forward and volumic source space."""
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.05, tmax=0.2)
cov = read_cov(fname_cov)
evoked_l21 = evoked.copy()
evoked_l21.crop(tmin=0.081, tmax=0.1)
info = evoked.info
sphere = mne.make_sphere_model(r0=(0., 0., 0.), head_radius=0.080)
src = mne.setup_volume_source_space(subject=None, pos=15., mri=None,
sphere=(0.0, 0.0, 0.0, 80.0),
bem=None, mindist=5.0,
exclude=2.0)
fwd = mne.make_forward_solution(info, trans=None, src=src,
bem=sphere, eeg=False, meg=True)
alpha = 80.
pytest.raises(ValueError, mixed_norm, evoked, fwd, cov, alpha,
loose=0.0, return_residual=False,
maxit=3, tol=1e-8, active_set_size=10)
pytest.raises(ValueError, mixed_norm, evoked, fwd, cov, alpha,
loose=0.2, return_residual=False,
maxit=3, tol=1e-8, active_set_size=10)
# irMxNE tests
stc = mixed_norm(evoked_l21, fwd, cov, alpha,
n_mxne_iter=1, maxit=30, tol=1e-8,
active_set_size=10)
assert isinstance(stc, VolSourceEstimate)
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
# Compare orientation obtained using fit_dipole and gamma_map
# for a simulated evoked containing a single dipole
stc = mne.VolSourceEstimate(50e-9 * np.random.RandomState(42).randn(1, 4),
vertices=stc.vertices[:1],
tmin=stc.tmin,
tstep=stc.tstep)
evoked_dip = mne.simulation.simulate_evoked(fwd, stc, info, cov, nave=1e9,
use_cps=True)
dip_mxne = mixed_norm(evoked_dip, fwd, cov, alpha=80,
n_mxne_iter=1, maxit=30, tol=1e-8,
active_set_size=10, return_as_dipoles=True)
amp_max = [np.max(d.amplitude) for d in dip_mxne]
dip_mxne = dip_mxne[np.argmax(amp_max)]
assert dip_mxne.pos[0] in src[0]['rr'][stc.vertices]
dip_fit = mne.fit_dipole(evoked_dip, cov, sphere)[0]
assert np.abs(np.dot(dip_fit.ori[0], dip_mxne.ori[0])) > 0.99
# Do with TF-MxNE for test memory savings
alpha = 60. # overall regularization parameter
l1_ratio = 0.01 # temporal regularization proportion
stc, _ = tf_mixed_norm(evoked, fwd, cov, maxit=3, tol=1e-4,
tstep=16, wsize=32, window=0.1, alpha=alpha,
l1_ratio=l1_ratio, return_residual=True)
assert isinstance(stc, VolSourceEstimate)
assert_array_almost_equal(stc.times, evoked.times, 5)
def test_mxne_inverse():
"""Test (TF-)MxNE inverse computation"""
# Handling forward solution
evoked = fiff.Evoked(fname_data, setno=1, baseline=(None, 0))
# Read noise covariance matrix
cov = read_cov(fname_cov)
# Handling average file
setno = 0
loose = None
depth = 0.9
evoked = fiff.read_evoked(fname_data, setno=setno, baseline=(None, 0))
evoked.crop(tmin=-0.1, tmax=0.4)
evoked_l21 = copy.deepcopy(evoked)
evoked_l21.crop(tmin=0.08, tmax=0.1)
label = read_label(fname_label)
weights_min = 0.5
forward = read_forward_solution(fname_fwd, force_fixed=False,
surf_ori=True)
# Reduce source space to make test computation faster
inverse_operator = make_inverse_operator(evoked.info, forward, cov,
loose=loose, depth=depth,
fixed=True)
stc_dspm = apply_inverse(evoked_l21, inverse_operator, lambda2=1. / 9.,
method='dSPM')
stc_dspm.data[np.abs(stc_dspm.data) < 12] = 0.0
stc_dspm.data[np.abs(stc_dspm.data) >= 12] = 1.
# MxNE tests
alpha = 60 # spatial regularization parameter
stc_prox = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=0.9, maxit=1000, tol=1e-8, active_set_size=10,
solver='prox')
stc_cd = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=0.9, maxit=1000, tol=1e-8, active_set_size=10,
solver='cd')
assert_array_almost_equal(stc_prox.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_cd.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_prox.data, stc_cd.data, 5)
assert_true(stc_prox.vertno[1][0] in label.vertices)
assert_true(stc_cd.vertno[1][0] in label.vertices)
stc, _ = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=depth, maxit=500, tol=1e-4, active_set_size=10,
weights=stc_dspm, weights_min=weights_min,
return_residual=True)
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert_true(stc.vertno[1][0] in label.vertices)
# Do with TF-MxNE for test memory savings
alpha_space = 60. # spatial regularization parameter
alpha_time = 1. # temporal regularization parameter
stc, _ = tf_mixed_norm(evoked, forward, cov, alpha_space, alpha_time,
loose=loose, depth=depth, maxit=100, tol=1e-4,
tstep=4, wsize=16, window=0.1, weights=stc_dspm,
weights_min=weights_min, return_residual=True)
assert_array_almost_equal(stc.times, evoked.times, 5)
assert_true(stc.vertno[1][0] in label.vertices)
import matplotlib.pyplot as plt
plt.figure()
ylim = dict(eeg=[-10, 10], grad=[-400, 400], mag=[-600, 600])
evoked.plot(ylim=ylim, proj=True)
###############################################################################
# Run solver
alpha = 70 # regularization parameter between 0 and 100 (100 is high)
loose, depth = 0.2, 0.9 # loose orientation & depth weighting
# Compute dSPM solution to be used as weights in MxNE
inverse_operator = make_inverse_operator(evoked.info, forward, cov,
loose=None, depth=depth, fixed=True)
stc_dspm = apply_inverse(evoked, inverse_operator, lambda2=1. / 9.,
method='dSPM')
# Compute MxNE inverse solution
stc, residual = mixed_norm(evoked, forward, cov, alpha, loose=loose,
depth=depth, maxit=3000, tol=1e-4,
active_set_size=10, debias=True, weights=stc_dspm,
weights_min=8., return_residual=True)
plt.figure()
residual.plot(ylim=ylim, proj=True)
###############################################################################
# View in 2D and 3D ("glass" brain like 3D plot)
plot_sparse_source_estimates(forward['src'], stc, bgcolor=(1, 1, 1),
opacity=0.1, fig_name="MxNE (cond %s)" % setno)
def test_mxne_inverse_standard():
"""Test (TF-)MxNE inverse computation."""
# Read noise covariance matrix
cov = read_cov(fname_cov)
# Handling average file
loose = 0.0
depth = 0.9
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.05, tmax=0.2)
evoked_l21 = evoked.copy()
evoked_l21.crop(tmin=0.081, tmax=0.1)
label = read_label(fname_label)
assert label.hemi == 'rh'
forward = read_forward_solution(fname_fwd)
forward = convert_forward_solution(forward, surf_ori=True)
# Reduce source space to make test computation faster
inverse_operator = make_inverse_operator(evoked_l21.info, forward, cov,
loose=loose, depth=depth,
fixed=True, use_cps=True)
stc_dspm = apply_inverse(evoked_l21, inverse_operator, lambda2=1. / 9.,
method='dSPM')
stc_dspm.data[np.abs(stc_dspm.data) < 12] = 0.0
stc_dspm.data[np.abs(stc_dspm.data) >= 12] = 1.
weights_min = 0.5
# MxNE tests
alpha = 70 # spatial regularization parameter
stc_prox = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8,
active_set_size=10, weights=stc_dspm,
weights_min=weights_min, solver='prox')
with pytest.warns(None): # CD
stc_cd = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8,
active_set_size=10, weights=stc_dspm,
weights_min=weights_min, solver='cd')
stc_bcd = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8, active_set_size=10,
weights=stc_dspm, weights_min=weights_min,
solver='bcd')
assert_array_almost_equal(stc_prox.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_cd.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_bcd.times, evoked_l21.times, 5)
assert_allclose(stc_prox.data, stc_cd.data, rtol=1e-3, atol=0.0)
assert_allclose(stc_prox.data, stc_bcd.data, rtol=1e-3, atol=0.0)
assert_allclose(stc_cd.data, stc_bcd.data, rtol=1e-3, atol=0.0)
assert stc_prox.vertices[1][0] in label.vertices
assert stc_cd.vertices[1][0] in label.vertices
assert stc_bcd.vertices[1][0] in label.vertices
# vector
with pytest.warns(None): # no convergence
stc = mixed_norm(evoked_l21, forward, cov, alpha, loose=1, maxit=2)
with pytest.warns(None): # no convergence
stc_vec = mixed_norm(evoked_l21, forward, cov, alpha, loose=1, maxit=2,
pick_ori='vector')
assert_stcs_equal(stc_vec.magnitude(), stc)
with pytest.warns(None), pytest.raises(ValueError, match='pick_ori='):
mixed_norm(evoked_l21, forward, cov, alpha, loose=0, maxit=2,
pick_ori='vector')
with pytest.warns(None): # CD
dips = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8, active_set_size=10,
weights=stc_dspm, weights_min=weights_min,
solver='cd', return_as_dipoles=True)
stc_dip = make_stc_from_dipoles(dips, forward['src'])
assert isinstance(dips[0], Dipole)
assert stc_dip.subject == "sample"
assert_stcs_equal(stc_cd, stc_dip)
with pytest.warns(None): # CD
stc, _ = mixed_norm(evoked_l21, forward, cov, alpha, loose=loose,
depth=depth, maxit=300, tol=1e-8,
weights=stc_dspm, # gh-6382
active_set_size=10, return_residual=True,
solver='cd')
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert stc.vertices[1][0] in label.vertices
# irMxNE tests
with pytest.warns(None): # CD
stc = mixed_norm(evoked_l21, forward, cov, alpha,
n_mxne_iter=5, loose=loose, depth=depth,
maxit=300, tol=1e-8, active_set_size=10,
solver='cd')
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert stc.vertices[1][0] in label.vertices
assert stc.vertices == [[63152], [79017]]
# Do with TF-MxNE for test memory savings
alpha = 60. # overall regularization parameter
l1_ratio = 0.01 # temporal regularization proportion
stc, _ = tf_mixed_norm(evoked, forward, cov,
loose=loose, depth=depth, maxit=100, tol=1e-4,
tstep=4, wsize=16, window=0.1, weights=stc_dspm,
weights_min=weights_min, return_residual=True,
alpha=alpha, l1_ratio=l1_ratio)
assert_array_almost_equal(stc.times, evoked.times, 5)
assert stc.vertices[1][0] in label.vertices
# vector
stc_nrm = tf_mixed_norm(
evoked, forward, cov, loose=1, depth=depth, maxit=2, tol=1e-4,
tstep=4, wsize=16, window=0.1, weights=stc_dspm,
weights_min=weights_min, alpha=alpha, l1_ratio=l1_ratio)
stc_vec = tf_mixed_norm(
evoked, forward, cov, loose=1, depth=depth, maxit=2, tol=1e-4,
tstep=4, wsize=16, window=0.1, weights=stc_dspm,
weights_min=weights_min, alpha=alpha, l1_ratio=l1_ratio,
pick_ori='vector')
assert_stcs_equal(stc_vec.magnitude(), stc_nrm)
pytest.raises(ValueError, tf_mixed_norm, evoked, forward, cov,
alpha=101, l1_ratio=0.03)
pytest.raises(ValueError, tf_mixed_norm, evoked, forward, cov,
alpha=50., l1_ratio=1.01)
# Group MNE leads to better accuracy
# ----------------------------------
# To evaluate the effect of the joint inverse solution, we compute the
# individual solutions using `mne.inverse_sparse.mixed_norm` for each subject.
# The group solutions are better located in S1.
from mne.inverse_sparse import mixed_norm # noqa: E402
t = 0.02
t_idx = stcs[0].time_as_index(t)
view = "lateral"
for subject, evoked, fwd, cov in zip(subjects, evoked_s, fwds, noise_cov_s):
ev = evoked.copy()
ev.pick_types(meg="grad")
ev.crop(0.015, 0.025)
stc = mixed_norm(ev, fwd, cov, alpha=60., loose=0., depth=0.9)
stc.subject = subject
g_post_central = mne.read_labels_from_annot(subject,
"aparc.a2009s",
subjects_dir=subjects_dir,
regexp="G_postcentral-lh")[0]
m = abs(stc.data[:group_info["n_sources"][0], t_idx]).max()
surfer_kwargs = dict(clim=dict(kind='value', pos_lims=[0., 0.1 * m, m]),
hemi='lh',
subjects_dir=subjects_dir,
initial_time=t * 1e3,
time_unit='ms',
size=(500, 500),
smoothing_steps=5)
brain = stc.plot(**surfer_kwargs, views=view)
brain.add_text(0.1, 0.9, subject + "_mxne", "title")
inverse_operator,
lambda2=1. / 9.,
method='dSPM')
# Compute (ir)MxNE inverse solution with dipole output
dipoles, residual = mixed_norm(
evoked,
forward,
cov,
alpha,
loose=loose,
depth=depth,
maxit=3000,
tol=1e-4,
active_set_size=10,
debias=False,
weights=stc_dspm,
weights_min=8.,
n_mxne_iter=n_mxne_iter,
return_residual=True,
return_as_dipoles=True,
verbose=True,
random_state=0,
# for this dataset we know we should use a high alpha, so avoid some
# of the slower (lower) alpha values
sure_alpha_grid=np.linspace(100, 40, 10),
)
t = 0.083
tidx = evoked.time_as_index(t)
for di, dip in enumerate(dipoles, 1):
print(f'Dipole #{di} GOF at {1000 * t:0.1f} ms: '
def test_mxne_inverse():
"""Test (TF-)MxNE inverse computation"""
# Handling forward solution
evoked = read_evokeds(fname_data, condition=1, baseline=(None, 0))
# Read noise covariance matrix
cov = read_cov(fname_cov)
# Handling average file
loose = None
depth = 0.9
evoked = read_evokeds(fname_data, condition=0, baseline=(None, 0))
evoked.crop(tmin=-0.1, tmax=0.4)
evoked_l21 = copy.deepcopy(evoked)
evoked_l21.crop(tmin=0.08, tmax=0.1)
label = read_label(fname_label)
weights_min = 0.5
forward = read_forward_solution(fname_fwd, force_fixed=False,
surf_ori=True)
# Reduce source space to make test computation faster
inverse_operator = make_inverse_operator(evoked.info, forward, cov,
loose=loose, depth=depth,
fixed=True)
stc_dspm = apply_inverse(evoked_l21, inverse_operator, lambda2=1. / 9.,
method='dSPM')
stc_dspm.data[np.abs(stc_dspm.data) < 12] = 0.0
stc_dspm.data[np.abs(stc_dspm.data) >= 12] = 1.
# MxNE tests
alpha = 60 # spatial regularization parameter
stc_prox = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=0.9, maxit=1000, tol=1e-8, active_set_size=10,
solver='prox')
stc_cd = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=0.9, maxit=1000, tol=1e-8, active_set_size=10,
solver='cd')
assert_array_almost_equal(stc_prox.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_cd.times, evoked_l21.times, 5)
assert_array_almost_equal(stc_prox.data, stc_cd.data, 5)
assert_true(stc_prox.vertno[1][0] in label.vertices)
assert_true(stc_cd.vertno[1][0] in label.vertices)
stc, _ = mixed_norm(evoked_l21, forward, cov, alpha, loose=None,
depth=depth, maxit=500, tol=1e-4, active_set_size=10,
weights=stc_dspm, weights_min=weights_min,
return_residual=True)
assert_array_almost_equal(stc.times, evoked_l21.times, 5)
assert_true(stc.vertno[1][0] in label.vertices)
# Do with TF-MxNE for test memory savings
alpha_space = 60. # spatial regularization parameter
alpha_time = 1. # temporal regularization parameter
stc, _ = tf_mixed_norm(evoked, forward, cov, alpha_space, alpha_time,
loose=loose, depth=depth, maxit=100, tol=1e-4,
tstep=4, wsize=16, window=0.1, weights=stc_dspm,
weights_min=weights_min, return_residual=True)
assert_array_almost_equal(stc.times, evoked.times, 5)
assert_true(stc.vertno[1][0] in label.vertices)
