def test_apply_forward():
"""Test projection of source space data to sensor space."""
start = 0
stop = 5
n_times = stop - start - 1
sfreq = 10.0
t_start = 0.123
fwd = read_forward_solution(fname_meeg)
fwd = convert_forward_solution(fwd, surf_ori=True, force_fixed=True,
use_cps=True)
fwd = pick_types_forward(fwd, meg=True)
assert isinstance(fwd, Forward)
vertno = [fwd['src'][0]['vertno'], fwd['src'][1]['vertno']]
stc_data = np.ones((len(vertno[0]) + len(vertno[1]), n_times))
stc = SourceEstimate(stc_data, vertno, tmin=t_start, tstep=1.0 / sfreq)
gain_sum = np.sum(fwd['sol']['data'], axis=1)
# Evoked
evoked = read_evokeds(fname_evoked, condition=0)
evoked.pick_types(meg=True)
with pytest.warns(RuntimeWarning, match='only .* positive values'):
evoked = apply_forward(fwd, stc, evoked.info, start=start, stop=stop)
data = evoked.data
times = evoked.times
# do some tests
assert_array_almost_equal(evoked.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
assert_array_almost_equal(times[0], t_start)
assert_array_almost_equal(times[-1], t_start + (n_times - 1) / sfreq)
# vector
stc_vec = VectorSourceEstimate(
fwd['source_nn'][:, :, np.newaxis] * stc.data[:, np.newaxis],
stc.vertices, stc.tmin, stc.tstep)
with pytest.warns(RuntimeWarning, match='very large'):
evoked_2 = apply_forward(fwd, stc_vec, evoked.info)
assert np.abs(evoked_2.data).mean() > 1e-5
assert_allclose(evoked.data, evoked_2.data, atol=1e-10)
# Raw
with pytest.warns(RuntimeWarning, match='only .* positive values'):
raw_proj = apply_forward_raw(fwd, stc, evoked.info, start=start,
stop=stop)
data, times = raw_proj[:, :]
# do some tests
assert_array_almost_equal(raw_proj.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
atol = 1. / sfreq
assert_allclose(raw_proj.first_samp / sfreq, t_start, atol=atol)
assert_allclose(raw_proj.last_samp / sfreq,
t_start + (n_times - 1) / sfreq, atol=atol)
def test_load_fiff_mne():
data_path = mne.datasets.sample.data_path()
fwd_path = os.path.join(data_path, 'MEG', 'sample', 'sample-ico-4-fwd.fif')
evoked_path = os.path.join(data_path, 'MEG', 'sample',
'sample_audvis-no-filter-ave.fif')
cov_path = os.path.join(data_path, 'MEG', 'sample', 'sample_audvis-cov.fif')
mri_sdir = os.path.join(data_path, 'subjects')
mne_evoked = mne.read_evokeds(evoked_path, 'Left Auditory')
mne_fwd = mne.read_forward_solution(fwd_path)
mne_fwd = mne.convert_forward_solution(mne_fwd, force_fixed=True, use_cps=True)
cov = mne.read_cov(cov_path)
picks = mne.pick_types(mne_evoked.info, 'mag')
channels = [mne_evoked.ch_names[i] for i in picks]
mne_evoked = mne_evoked.pick_channels(channels)
mne_fwd = mne.pick_channels_forward(mne_fwd, channels)
cov = mne.pick_channels_cov(cov, channels)
mne_inv = mne.minimum_norm.make_inverse_operator(mne_evoked.info, mne_fwd,
cov, 0, None, True)
mne_stc = mne.minimum_norm.apply_inverse(mne_evoked, mne_inv, 1., 'MNE')
meg = load.fiff.evoked_ndvar(mne_evoked)
inv = load.fiff.inverse_operator(mne_inv, 'ico-4', mri_sdir)
stc = inv.dot(meg)
assert_array_almost_equal(stc.get_data(('source', 'time')), mne_stc.data)
fwd = load.fiff.forward_operator(mne_fwd, 'ico-4', mri_sdir)
reconstruct = fwd.dot(stc)
mne_reconstruct = mne.apply_forward(mne_fwd, mne_stc, mne_evoked.info)
assert_array_almost_equal(reconstruct.get_data(('sensor', 'time')),
mne_reconstruct.data)
def apply_foward_model(self, x):
out = []
for z in x:
info = mne.io.read_info(self.raw_fname)
stc = mne.SourceEstimate(z.cpu().detach().numpy(),
self.vertices,
tmin=0.,
tstep=self.time_step)
# print("STC", stc)
# print("Info", info)
leadfield = mne.apply_forward(self.fwd_fixed, stc,
info).data[0:self.num_nodes]
# leadfield = torch.from_numpy(leadfield)
# leadfield = leadfield.transpose(0,1)
# print("YOUYOUYOUYOUY")
# array = leadfield.to_data_frame().values
# print("array", leadfield.shape)
out += [leadfield]
out = np.asarray(out)
out = torch.from_numpy(out).transpose(1, 2)
if x.is_cuda:
out = out.type(torch.cuda.FloatTensor)
else:
out = out.type(torch.FloatTensor)
# print("OUTOUTOUT", out.shape)
# (n sensors, n samples)
return out
def generate_eeg(self, source_index):
stc = mne.SourceEstimate(self.preloaded_examples_source[source_index],
self.vertices,
tmin=0.,
tstep=1 / 250)
leadfield = mne.apply_forward(self.fwd_fixed, stc,
self.info).data / 1e-9
return list(leadfield[:self.num_channels])
def test_apply_forward():
"""Test projection of source space data to sensor space
"""
start = 0
stop = 5
n_times = stop - start - 1
sfreq = 10.0
t_start = 0.123
fwd = read_forward_solution(fname_meeg)
fwd = convert_forward_solution(fwd,
surf_ori=True,
force_fixed=True,
use_cps=True)
fwd = pick_types_forward(fwd, meg=True)
assert_true(isinstance(fwd, Forward))
vertno = [fwd['src'][0]['vertno'], fwd['src'][1]['vertno']]
stc_data = np.ones((len(vertno[0]) + len(vertno[1]), n_times))
stc = SourceEstimate(stc_data, vertno, tmin=t_start, tstep=1.0 / sfreq)
gain_sum = np.sum(fwd['sol']['data'], axis=1)
# Evoked
with warnings.catch_warnings(record=True) as w:
evoked = read_evokeds(fname_evoked, condition=0)
evoked.pick_types(meg=True)
evoked = apply_forward(fwd, stc, evoked.info, start=start, stop=stop)
assert_equal(len(w), 2)
data = evoked.data
times = evoked.times
# do some tests
assert_array_almost_equal(evoked.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
assert_array_almost_equal(times[0], t_start)
assert_array_almost_equal(times[-1], t_start + (n_times - 1) / sfreq)
# Raw
raw_proj = apply_forward_raw(fwd,
stc,
evoked.info,
start=start,
stop=stop)
data, times = raw_proj[:, :]
# do some tests
assert_array_almost_equal(raw_proj.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
atol = 1. / sfreq
assert_allclose(raw_proj.first_samp / sfreq, t_start, atol=atol)
assert_allclose(raw_proj.last_samp / sfreq,
t_start + (n_times - 1) / sfreq,
atol=atol)
def test_apply_forward():
"""Test projection of source space data to sensor space
"""
start = 0
stop = 5
n_times = stop - start - 1
sfreq = 10.0
t_start = 0.123
fwd = read_forward_solution(fname_meeg)
fwd = convert_forward_solution(fwd, surf_ori=True, force_fixed=True,
use_cps=True)
fwd = pick_types_forward(fwd, meg=True)
assert_true(isinstance(fwd, Forward))
vertno = [fwd['src'][0]['vertno'], fwd['src'][1]['vertno']]
stc_data = np.ones((len(vertno[0]) + len(vertno[1]), n_times))
stc = SourceEstimate(stc_data, vertno, tmin=t_start, tstep=1.0 / sfreq)
gain_sum = np.sum(fwd['sol']['data'], axis=1)
# Evoked
with warnings.catch_warnings(record=True) as w:
evoked = read_evokeds(fname_evoked, condition=0)
evoked.pick_types(meg=True)
evoked = apply_forward(fwd, stc, evoked.info, start=start, stop=stop)
assert_equal(len(w), 2)
data = evoked.data
times = evoked.times
# do some tests
assert_array_almost_equal(evoked.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
assert_array_almost_equal(times[0], t_start)
assert_array_almost_equal(times[-1], t_start + (n_times - 1) / sfreq)
# Raw
raw_proj = apply_forward_raw(fwd, stc, evoked.info, start=start,
stop=stop)
data, times = raw_proj[:, :]
# do some tests
assert_array_almost_equal(raw_proj.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
atol = 1. / sfreq
assert_allclose(raw_proj.first_samp / sfreq, t_start, atol=atol)
assert_allclose(raw_proj.last_samp / sfreq,
t_start + (n_times - 1) / sfreq, atol=atol)
def make_surrogates_empty_room(raw, fwd, inverse_operator, step=10000):
"""Create spatially structured noise from empty room MEG
.. note::
Convert MEG empty room to spatially structured noise by applying
the inverse solution and then the forward solution.
.. note::
To safe memory, projection proceeds in non-overlapping sliding windows.
Parameters
----------
raw : instance of mne.io.Raw
The raw data (empty room).
fwd : instance of mne.Forward
The forward solution
inverse_operator : mne.minimum_norm.InverseOperator
The inverse solution.
step : int
The step size (in samples) when iterating over the raw object.
Returns
-------
raw_surr : instance of mne.io.Raw
The surrogate MEG data.
"""
index = np.arange(len(raw.times)).astype(int)
out = np.empty(raw.get_data().shape, dtype=raw.get_data().dtype)
picks = mne.pick_types(raw.info, meg=True, eeg=True, ref_meg=False)
other_picks = [ii for ii in range(len(raw.ch_names)) if ii not in picks]
out[other_picks] = raw.get_data()[other_picks]
last = len(index)
for start in index[::step]:
stop = start + min(step, last - start)
stc = mne.minimum_norm.apply_inverse_raw(raw,
inverse_operator,
lambda2=1.0,
method='MNE',
start=start,
stop=stop,
pick_ori="normal")
reprojected = mne.apply_forward(fwd=fwd, stc=stc, info=raw.info)
out[picks, start:stop] = reprojected.data
out = mne.io.RawArray(out, info=copy.deepcopy(raw.info))
return out
def generate_data(num_examples, data_dir, fwd_fixed, n_sensors=60, fs_gen=200, n_times=10, frequency=200, prefix="generated-"):
leadfield = fwd_fixed['sol']['data']
n_dipoles = leadfield.shape[1]
vertices = [src_hemi['vertno'] for src_hemi in fwd_fixed['src']]
time_step = 1.0/fs_gen # sample freq = 200 was 0.5
n_times = 10 * fs_gen # try 10 s of generation
channel_names = [str(i) for i in range(60)]
for i in range(num_examples):
z = np.dot(np.random.randn(n_dipoles, n_sensors), np.random.randn(n_sensors, n_times)) * 1e-9
filename = data_dir + "/" + prefix + str(i) + "raw.fif"
stc = mne.SourceEstimate(z, vertices, tmin=0., tstep=time_step)
leadfield = mne.apply_forward(fwd_fixed, stc, info).data
# save_data(leadfield, filename, channel_names, frequency)
print("i: " + str(i) + " - " + str(int((i+1) * 100/num_examples)) + "%")
print("finished")
def test_apply_forward():
"""Test projection of source space data to sensor space
"""
start = 0
stop = 5
n_times = stop - start - 1
sfreq = 10.0
t_start = 0.123
fwd = read_forward_solution(fname, force_fixed=True)
fwd = pick_types_forward(fwd, meg=True)
vertno = [fwd['src'][0]['vertno'], fwd['src'][1]['vertno']]
stc_data = np.ones((len(vertno[0]) + len(vertno[1]), n_times))
stc = SourceEstimate(stc_data, vertno, tmin=t_start, tstep=1.0 / sfreq)
gain_sum = np.sum(fwd['sol']['data'], axis=1)
# Evoked
with warnings.catch_warnings(record=True) as w:
evoked = Evoked(fname_evoked, setno=0)
evoked = apply_forward(fwd, stc, evoked, start=start, stop=stop)
assert_equal(len(w), 2)
data = evoked.data
times = evoked.times
# do some tests
assert_array_almost_equal(evoked.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
assert_array_almost_equal(times[0], t_start)
assert_array_almost_equal(times[-1], t_start + (n_times - 1) / sfreq)
# Raw
raw = Raw(fname_raw)
raw_proj = apply_forward_raw(fwd, stc, raw, start=start, stop=stop)
data, times = raw_proj[:, :]
# do some tests
assert_array_almost_equal(raw_proj.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
assert_array_almost_equal(times[0], t_start)
assert_array_almost_equal(times[-1], t_start + (n_times - 1) / sfreq)
def test_apply_forward():
"""Test projection of source space data to sensor space
"""
start = 0
stop = 5
n_times = stop - start - 1
sfreq = 10.0
t_start = 0.123
fwd = read_forward_solution(fname, force_fixed=True)
fwd = pick_types_forward(fwd, meg=True)
vertno = [fwd['src'][0]['vertno'], fwd['src'][1]['vertno']]
stc_data = np.ones((len(vertno[0]) + len(vertno[1]), n_times))
stc = SourceEstimate(stc_data, vertno, tmin=t_start, tstep=1.0 / sfreq)
gain_sum = np.sum(fwd['sol']['data'], axis=1)
# Evoked
with warnings.catch_warnings(record=True) as w:
evoked = Evoked(fname_evoked, setno=0)
evoked = apply_forward(fwd, stc, evoked, start=start, stop=stop)
assert_equal(len(w), 2)
data = evoked.data
times = evoked.times
# do some tests
assert_array_almost_equal(evoked.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
assert_array_almost_equal(times[0], t_start)
assert_array_almost_equal(times[-1], t_start + (n_times - 1) / sfreq)
# Raw
raw = Raw(fname_raw)
raw_proj = apply_forward_raw(fwd, stc, raw, start=start, stop=stop)
data, times = raw_proj[:, :]
# do some tests
assert_array_almost_equal(raw_proj.info['sfreq'], sfreq)
assert_array_almost_equal(np.sum(data, axis=1), n_times * gain_sum)
assert_array_almost_equal(times[0], t_start)
assert_array_almost_equal(times[-1], t_start + (n_times - 1) / sfreq)
true_locs = list()
for v in vertices:
true_locs.append(np.where(vertno == v[0])[0][0])
# Construct SourceEstimates that describe the signals at the cortical level.
stc_signal = mne.SourceEstimate(q,
vertices,
tmin=0,
tstep=1. / sfreq,
subject='sample')
###############################################################################
# Now we run the signal through the forward model to obtain simulated sensor
# data. We then corrupt the resulting simulated gradiometer recordings
# by empty room noise.
evoked = mne.apply_forward(fwd, stc_signal, info)
mne.simulation.add_noise(evoked, noise_cov, random_state=rand)
###############################################################################
# Visualize the data
evoked.plot()
###############################################################################
# Define the parameters and apply SESAME. We convert the data to the frequency
# domain by setting ``Fourier_transf=True``.
n_parts = 100
noise_std = None
dip_mom_std = None
freq_min = 9.5
freq_max = 10.5
# To extract the numpy array containing the forward operator corresponding to
# the source space `fwd['src']` with cortical orientation constraint
# we can use the following:
fwd_fixed = mne.convert_forward_solution(fwd, surf_ori=True, force_fixed=True,
use_cps=True)
leadfield = fwd_fixed['sol']['data']
print("Leadfield size : %d sensors x %d dipoles" % leadfield.shape)
###############################################################################
# This is equivalent to the following code that explicitly applies the
# forward operator to a source estimate composed of the identity operator:
n_dipoles = leadfield.shape[1]
vertices = [src_hemi['vertno'] for src_hemi in fwd_fixed['src']]
stc = mne.SourceEstimate(1e-9 * np.eye(n_dipoles), vertices, tmin=0., tstep=1)
leadfield = mne.apply_forward(fwd_fixed, stc, info).data / 1e-9
###############################################################################
# To save to disk a forward solution you can use
# :func:`mne.write_forward_solution` and to read it back from disk
# :func:`mne.read_forward_solution`. Don't forget that FIF files containing
# forward solution should end with *-fwd.fif*.
#
# To get a fixed-orientation forward solution, use
# :func:`mne.convert_forward_solution` to convert the free-orientation
# solution to (surface-oriented) fixed orientation.
###############################################################################
# Exercise
# --------
#
def data_fun(times):
"""Function to generate random source time courses"""
rng = np.random.RandomState(42)
return (50e-9 * np.sin(30. * times) *
np.exp(-(times - 0.15 + 0.05 * rng.randn(1))**2 / 0.01))
simulated_stc = simulate_sparse_stc(
fwd['src'], n_dipoles=2, times=times, random_state=42, data_fun=data_fun
) # Gives an error later if n_dipoles = 1, as it puts a source only to the right hemisphere.
evoked = apply_forward(fwd=fwd_fixed,
stc=simulated_stc,
info=fwd_fixed['info'],
use_cps=True,
verbose=True)
"""Visualize dipole locations."""
surf = 'inflated'
brain = Brain(subject_id=subject,
subjects_dir=subjects_dir,
hemi='both',
surf=surf)
brain.add_foci(coords=simulated_stc.lh_vertno,
coords_as_verts=True,
hemi='lh',
map_surface=surf,
color='red')
brain.add_foci(coords=simulated_stc.rh_vertno,
z[stimulate_sites, :] = z[stimulate_sites, :] + source_scale * np.sin(
2 * np.pi * 10.0 * t)
# determine vertex number of ??? something in the fwd_fixed solution
# vertices = [src_hemi['vertno'] for src_hemi in fwd_fixed['src']]
#len(vertices)
#vertices
# SourceEstimate(data, vertices=None, tmin=None, tstep=None,
# subject=None, verbose=None
# data: array of shape (n_dipoles, n_times)
# | 2-tuple (kernel, sens_data)
# vertices: list of two arrays with vertex numbers corresponding to the data
srcest = mne.SourceEstimate(z, vertices, tmin=0., tstep=time_step)
gen_eeg = mne.apply_forward(fwd_fixed, srcest, info) # / np.sum(z, axis=1)
gen_eeg.data.shape
fig = gen_eeg.plot(exclude=(), time_unit='s')
picks = mne.pick_types(gen_eeg.info, meg=False, eeg=True, eog=False)
gen_eeg.plot(spatial_colors=True, gfp=True, picks=picks, time_unit='s')
gen_eeg.plot_topomap(time_unit='s')
signals = gen_eeg.data
electrode_labels = list(range(n_sensors))
ch0, ch1 = (0, 59)
DE = 1 # how many 10s epochs to display
epoch = 0
ptepoch = 10 * int(fs_gen)
# Random source # In[75]: z = np.dot(np.random.randn(n_dipoles, n_sensors), np.random.randn(n_sensors, n_times)) * 1e-9 # In[76]: # n_dipoles = leadfield.shape[1] vertices = [src_hemi['vertno'] for src_hemi in fwd_fixed['src']] # z = np.random.randn(n_dipoles, n_dipoles) * 1e-9 stc = mne.SourceEstimate(z, vertices, tmin=0., tstep=time_step) leadfield_data = mne.apply_forward(fwd_fixed, stc, info).data # / np.sum(z, axis=1) # In[13]: leadfield_data.shape # In[77]: leadfield = mne.apply_forward(fwd_fixed, stc, info) # In[97]:
# we can use the following:
fwd_fixed = mne.convert_forward_solution(fwd,
surf_ori=True,
force_fixed=True,
use_cps=True)
leadfield = fwd_fixed['sol']['data']
print("Leadfield size : %d sensors x %d dipoles" % leadfield.shape)
#%%##############################################################################
# This is equivalent to the following code that explicitly applies the
# forward operator to a source estimate composed of the identity operator:
n_dipoles = leadfield.shape[1]
vertices = [src_hemi['vertno'] for src_hemi in fwd_fixed['src']]
stc = mne.SourceEstimate(1e-9 * np.eye(n_dipoles), vertices, tmin=0., tstep=1)
leadfield = mne.apply_forward(fwd_fixed, stc, info).data / 1e-9
#%%##############################################################################
# To save to disk a forward solution you can use
# :func:`mne.write_forward_solution` and to read it back from disk
# :func:`mne.read_forward_solution`. Don't forget that FIF files containing
# forward solution should end with *-fwd.fif*.
#
# To get a fixed-orientation forward solution, use
# :func:`mne.convert_forward_solution` to convert the free-orientation
# solution to (surface-oriented) fixed orientation.
###############################################################################
# Exercise
# --------
#
def generate_source_then_eeg(self):
source = np.random.randn(self.n_dipoles, self.length) * 1e-9
stc = mne.SourceEstimate(source, self.vertices, tmin=0., tstep=1 / 250)
leadfield = mne.apply_forward(self.fwd_fixed, stc,
self.info).data / 1e-9
return list(leadfield[:self.num_channels])
