def get_explained(subject_no, n_comp, n_bads):
subject = subjects[subject_no]
raw_fname = root_dir + '%s/passive_raw.fif' % subject
room_fname = root_dir + '%s/emptyroom_%s.fif' % (subject, subject)
raw = mne.io.read_raw_fif(raw_fname, preload=True)
raw_room = mne.io.read_raw_fif(room_fname, preload=True)
raw.filter(0.5, 70)
raw_room.filter(1, 70)
mne.channels.fix_mag_coil_types(raw.info)
mne.channels.fix_mag_coil_types(raw_room.info)
picks = mne.pick_types(raw.info, meg='mag', eeg=False, eog=False,
stim=False, exclude='bads')
events = mne.find_events(raw)
tmin, tmax = -0.1, 0.3
expes = list(event_id.keys())
expe = expes[2]
# reject = dict(grad=4000e-13, mag=1e-12, eog=150e-6)
reject = None
# Fit smica
freqs = np.linspace(1, 70, 41)
smica = ICA(n_components=n_comp,
freqs=freqs, rng=0).fit(raw, picks, tol=1e-9,
verbose=100,
n_it_min=3000, em_it=3000,
n_it_lbfgs=3000)
ica = ICA_mne(n_comp).fit(raw, picks)
X_room = raw_room.get_data(picks=picks)
X_room /= np.linalg.norm(X_room)
Ss = smica.compute_sources(X_room)
As = smica.A
del smica
Ai = ica.get_components()
Si = np.linalg.pinv(Ai).dot(X_room)
def greedy_comp(X_room, A, S, n_comp=5):
bads = []
vals = []
Xc = X_room.copy()
idxs = np.arange(S.shape[0])
for i in range(n_comp):
print(i)
diffs = [np.linalg.norm(Xc - a[:, None] * s)
for a, s in zip(A.T[idxs], S[idxs])]
j = idxs[np.argmin(diffs)]
bads.append(j)
Xc -= A[:, j][:, None] * S[j]
vals.append(np.linalg.norm(Xc))
idxs = np.delete(idxs, np.where(idxs == j))
return bads, vals
bads, vals = greedy_comp(X_room, As, Ss, n_bads)
badi, vali = greedy_comp(X_room, Ai, Si, n_bads)
np.save('results/expl_%d_%d_%d.npy' % (subject_no, n_comp, n_bads),
np.array((vals, vali)))
return vals, vali
def compute_decomposition(algo_name, subject, n_components, decomp_num,
n_decomp):
runs = [6, 10, 14] # motor imagery: hands vs feet
filename = '/home/pierre/work/smica/dataset/%s.set' % SUBJECTS[subject]
epochs = mne.io.read_epochs_eeglab(filename, verbose='CRITICAL')
epochs.set_channel_types({'LEYE': 'eog', 'REYE': 'eog'})
epochs.filter(2, 60)
picks = mne.pick_types(epochs.info,
meg=False,
eeg=True,
eog=False,
stim=False,
exclude='bads')
n_bins = 40
freqs = np.linspace(2, 60, n_bins + 1)
if algo_name == 'smica':
algorithm = ICA(n_components=n_components, freqs=freqs, rng=0)
algo_args = dict(em_it=100000, tol=1e-8, n_it_min=100000)
mixing = standard_mixing(epochs, picks, algorithm, algo_args)
if algo_name == 'jdiag':
algorithm = JDIAG_mne(n_components=n_components, freqs=freqs, rng=0)
algo_args = dict(max_iter=1000, tol=1e-10)
mixing = standard_mixing(epochs, picks, algorithm, algo_args)
if algo_name == 'sobi':
algorithm = SOBI_mne(1000,
n_components=n_components,
freqs=freqs,
rng=0)
algo_args = dict()
mixing = standard_mixing(epochs, picks, algorithm, algo_args)
if algo_name == 'infomax':
jdiag = JDIAG_mne(n_components=n_components, freqs=freqs, rng=0)
jdiag.fit(epochs, picks, max_iter=10)
sources = jdiag.compute_sources(method='pinv')
K, W, _ = picard(sources, max_iter=1000, ortho=False)
picard_mix = np.linalg.pinv(np.dot(W, K))
mixing = np.dot(jdiag.A, picard_mix)
if algo_name in ['pinv_infomax', 'wiener_infomax']:
smica = ICA(n_components=n_components, freqs=freqs, rng=0)
algo_args = dict(em_it=100000, tol=1e-8, n_it_min=100000)
smica.fit(epochs, picks, corr=False, **algo_args)
method = {
'pinv_infomax': 'pinv',
'wiener_infomax': 'wiener'
}[algo_name]
sources = smica.compute_sources(method=method)
K, W, _ = picard(sources, max_iter=1000, ortho=False)
picard_mix = np.linalg.pinv(np.dot(W, K))
mixing = np.dot(smica.A, picard_mix)
gof, _, _ = dipolarity(mixing, epochs, picks)
print(decomp_num, n_decomp)
return gof
epochs_data = epochs.get_data()
l_list = []
for event_id in [dict(hands=2), dict(feet=3)]:
epochs = Epochs(raw,
events,
event_id,
tmin,
tmax,
proj=True,
picks=picks,
baseline=None,
preload=True)
n_components = 20
freqs = np.linspace(7, 30, 31)
ica = ICA(n_components=n_components, freqs=freqs, rng=0)
ica.fit(epochs, max_iter=5000, verbose=100)
l_list.append([ica.compute_loss(x) for x in epochs_data])
plt.figure()
likelihoods = np.array(l_list)
for i, label in enumerate(['hand', 'feet']):
plt.scatter(*(likelihoods[j, labels == i] for j in [0, 1]), label=label)
plt.xlabel('negative loglik of model hand')
plt.ylabel('negative loglik of model feet')
l_min = np.min(likelihoods)
l_max = np.max(likelihoods)
t = np.linspace(l_min, l_max)
plt.title(subject)
plt.plot(t, t, color='k', label='y=x')
plt.legend()
def process_one(dipole_no, amp, n_components, crop_idx, crop_len=None):
raw_fname = "%sAmp%d_Dip%d_IASoff.fif" % (get_root_dir(), amp, dipole_no)
raw = read_raw_fif(raw_fname)
raw.load_data()
raw.crop(5, None)
if crop_len is not None:
raw.crop(crop_idx * crop_len, (crop_idx + 1) * crop_len)
raw.fix_mag_coil_types()
events = find_events(raw, "SYS201")
picks = mne.pick_types(raw.info,
meg="mag",
eeg=False,
eog=False,
stim=False,
exclude="bads")
n_bins = 40
freqs = np.linspace(1, 70, n_bins + 1)
smica = ICA(n_components, freqs, rng=0).fit(
raw,
picks=picks,
em_it=50000,
n_it_lbfgs=200,
tol=1e-8,
verbose=100,
n_it_min=10000,
)
ica = ICA_mne(n_components, method="picard").fit(raw.copy().filter(1, 70),
picks=picks)
jdiag = JDIAG_mne(n_components, freqs, rng=0).fit(raw, picks=picks)
tmin, tmax = -0.1, 0.1
event_id = [dipole_no]
# pows = np.linalg.norm(smica.A, axis=0) *
# np.linalg.norm(smica.powers, axis=0) ** .5
raw_max = raw.copy().filter(1, 40)
raw_max = maxwell_filter(raw_max, origin=(0.0, 0.0, 0.0))
epochs = mne.Epochs(
raw_max,
events,
event_id,
tmin,
tmax,
baseline=(None, -0.01),
preload=True,
picks=picks,
)
cov = mne.compute_covariance(epochs, tmax=0)
e1 = get_error(epochs, cov)
e_2 = get_error_A(smica.A, cov, epochs.info)
good = np.argmin(e_2[0])
e2 = [e_2[0][good], e_2[1][good]]
e_3 = get_error_A(jdiag.A, cov, epochs.info)
good = np.argmin(e_3[0])
e3 = [e_3[0][good], e_3[1][good]]
e_4 = get_error_A(ica.get_components(), cov, epochs.info)
good = np.argmin(e_4[0])
e4 = [e_4[0][good], e_4[1][good]]
crop_str = crop_len if crop_len is not None else 0
np.save(
"results/phantom_%d_%d_%d_%d_%d.npy" %
(dipole_no, amp, n_components, crop_idx, crop_str),
np.array([e1, e2, e3, e4]),
)
print(e1, e2, e3, e4)
return e1, e2, e3, e4
raw.info, meg="mag", eeg=False, eog=False, stim=False, exclude="bads"
)
raw.set_eeg_reference("average", projection=True) # set EEG average reference
events = mne.find_events(raw)
tmin, tmax = -0.2, 0.5
event_id = {
"Auditory/Left": 1,
"Auditory/Right": 2,
"Visual/Left": 3,
"Visual/Right": 4,
}
reject = dict(grad=4000e-13, mag=4e-12, eog=150e-6)
# Fit smica
freqs = np.linspace(1, 70, 41)
smica = ICA(n_components=20, freqs=freqs, rng=0).fit(raw, picks)
colors = ["k", "darkred", "forestgreen", "royalblue", "orange"]
# Plot the clusters
# smica.plot_extended(sort=False)
bad_sources = [17, 18, 19]
X_filtered = smica.filter(
raw._data[picks], bad_sources=bad_sources, method="wiener"
)
raw_filtered = raw.copy()
raw_filtered._data[picks] = X_filtered
import mne
from mne.datasets import sample
data_path = sample.data_path()
raw_fname = data_path + '/MEG/sample/sample_audvis_raw.fif'
raw = mne.io.read_raw_fif(raw_fname, preload=True)
picks = mne.pick_types(raw.info,
meg=False,
eeg=True,
eog=False,
stim=False,
exclude='bads')
n_components = 59
n_bins = 40
freqs = np.linspace(1, 60, n_bins + 1)
loss_list = []
for n_component in np.arange(1, n_components):
print(n_component)
smica = ICA(n_components=n_component, freqs=freqs, rng=0)
smica.fit(raw, picks=picks, verbose=2000, tol=1e-7, em_it=10000)
loss_list.append(smica.compute_loss())
plt.plot(np.arange(1, n_components), loss_list - np.min(loss_list))
plt.yscale('log')
plt.xlabel('Number of sources')
plt.ylabel('Negative log likelihood')
plt.show()
# fetch data
raw = mne.io.read_raw_fif(raw_fname, preload=True)
picks = mne.pick_types(raw.info,
meg='mag',
eeg=False,
eog=False,
stim=False,
exclude='bads')
# Compute ICA on raw: chose the frequency decomposition. Here, uniform between
# 2 - 35 Hz.
#
jdiag = JDIAG_mne(n_components=n_components, freqs=freqs, rng=0)
jdiag.fit(raw, picks=picks, verbose=True, tol=1e-9, max_iter=1000)
smica = ICA(n_components=n_components, freqs=freqs, rng=0)
smica.fit(raw, picks=picks, verbose=100, tol=1e-10, em_it=100000)
# Plot powers
# noise_sources = [6, 8, 9]
# muscle_source = [7]
# f, ax = plt.subplots(figsize=(4, 2))
# plot_powers(smica.powers, noise_sources, muscle_source, ax, 'smica')
# plt.show()
# #
# #
# sobi = SOBI_mne(100, n_components, freqs, rng=0)
# sobi.fit(raw, picks=picks)
raw.filter(2, 70)
ica = ICA_mne(n_components=n_components, method='fastica', random_state=0)
sources = set(range(n_s))
free = set(range(n_s))
while sources:
i, j = np.unravel_index(np.argmax(corr, axis=None), corr.shape)
corr[i, j] = -1
if i in sources and j in free:
sources.remove(i)
free.remove(j)
order[i] = j
return order
n_bins = 40
freqs = np.linspace(1, 70, n_bins + 1)
n_comp = 40
smica_room = ICA(10, freqs=freqs, rng=0).fit(raw_room, picks)
jdiag_room = JDIAG_mne(10, freqs=freqs, rng=0).fit(raw_room, picks=picks)
smica = ICA(n_comp, freqs=freqs, rng=0).fit(raw, picks=picks)
to_plot_smica = [37, 38, 39, 20, 4, 3, 19, 8, 12, 2]
raw.filter(1, 70)
jdiag = JDIAG_mne(n_comp, freqs=freqs, rng=0).fit(raw, picks=picks)
to_plot_jdiag = [2, 1, 0, 13, 34, 3, 20, 30, 5, 10]
sobi = SOBI_mne(100, n_comp, freqs).fit(raw, picks=picks)
to_plot_sobi = [2, 0, 8, 12, 20, 31, 3, 17, 9, 10]
ifmx_ = ICA_mne(n_comp, random_state=0).fit(raw, picks=picks)
ifmx = transfer_to_ica(raw, picks, freqs,
ifmx_.get_sources(raw).get_data(),
ifmx_.get_components())
to_plot_ifmx = [1, 2, 0, 35, 33, 19, 3, 32, 25, 9]
plot_args = dict(number=True, t_min=2, t_max=4)
data_path = sample.data_path()
raw_fname = data_path + '/MEG/sample/ernoise_raw.fif'
raw = mne.io.read_raw_fif(raw_fname, preload=True)
picks = mne.pick_types(raw.info,
meg='mag',
eeg=False,
eog=False,
stim=False,
exclude='bads')
# Compute ICA on raw: chose the frequency decomposition. Here, uniform between
# 2 - 35 Hz.
n_bins = 40
n_components = 10
freqs = np.linspace(1, 70, n_bins + 1)
smica = ICA(n_components=n_components, freqs=freqs, rng=0)
smica.fit(raw, picks=picks, verbose=100, tol=1e-15, em_it=100000)
jdiag = JDIAG_mne(n_components=n_components, freqs=freqs, rng=0)
jdiag.fit(raw, picks=picks, verbose=True, tol=1e-7, max_iter=10000)
raw.filter(1, 70)
sobi = SOBI_mne(p=2000, n_components=n_components, freqs=freqs, rng=0)
sobi.fit(raw, picks=picks, verbose=True, tol=1e-7, max_iter=2000)
raw_fname = data_path + '/MEG/sample/sample_audvis_raw.fif'
raw = mne.io.read_raw_fif(raw_fname, preload=True)
picks = mne.pick_types(raw.info,
meg='mag',
eeg=False,
eog=False,
stim=False,
exclude='bads')
eeg=True,
eog=False,
stim=False,
exclude='bads')
n_bins = 40
n_components = 20
freqs = np.linspace(1, 60, n_bins + 1)
#
jdiag = JDIAG_mne(n_components=n_components, freqs=freqs, rng=0)
jdiag.fit(raw, picks=picks, verbose=True, tol=1e-9, max_iter=1000)
# jdiag2 = JDIAG_mne(n_components=n_components, freqs=freqs, rng=0)
# jdiag2.fit(raw, pca='other', picks=picks, verbose=True, tol=1e-9, max_iter=1000)
smica = ICA(n_components=n_components, freqs=freqs, rng=0)
smica.fit(raw, picks=picks, verbose=100, tol=1e-10, em_it=100000, corr=True)
# Plot powers
# noise_sources = [6, 8, 9]
# muscle_source = [7]
# f, ax = plt.subplots(figsize=(4, 2))
# plot_powers(smica.powers, noise_sources, muscle_source, ax, 'smica')
# plt.show()
# #
# #
raw.filter(2, 70)
ica = ICA_mne(n_components=n_components, method='fastica', random_state=0)
ica.fit(raw, picks=picks)
