def test_ctps():
"""Test basic ctps functionality."""
for ii, (n_trials, j_extent, pk_max) in iter_test_ctps:
data = get_data(n_trials, j_extent)
ks_dyn, pk_dyn, phase_trial = ctps(data)
data2 = _compute_normalized_phase(data)
ks_dyn2, pk_dyn2, phase_trial2 = ctps(data2, is_raw=False)
for a, b in zip([ks_dyn, pk_dyn, phase_trial],
[ks_dyn2, pk_dyn2, data2]):
assert_array_equal(a, b)
assert (a.min() >= 0)
assert (a.max() <= 1)
assert (b.min() >= 0)
assert (b.max() <= 1)
# test for normalization
assert ((pk_dyn.min() > 0.0) or (pk_dyn.max() < 1.0))
# test shapes
assert (phase_trial.shape == data.shape)
assert (pk_dyn.shape == data.shape[1:])
# tets ground_truth + random + jittered case
assert (pk_dyn[0].max() == 1.0)
assert (len(np.unique(pk_dyn[0])) == 1.0)
assert (pk_dyn[1].max() < pk_max)
assert (pk_dyn[2].max() > 0.3)
if ii < 1:
pytest.raises(ValueError, ctps, data[:, :, :, None])
assert (_prob_kuiper(1.0, 400) == 1.0)
# test vecrosization
assert_array_equal(_prob_kuiper(np.array([1.0, 1.0]), 400),
_prob_kuiper(np.array([1.0, 1.0]), 400))
assert (_prob_kuiper(0.1, 400) < 0.1)
def test_ctps():
""" Test basic ctps functionality
"""
for ii, (n_trials, j_extent, pk_max) in iter_test_ctps:
data = get_data(n_trials, j_extent)
ks_dyn, pk_dyn, phase_trial = ctps(data)
data2 = _compute_normalized_phase(data)
ks_dyn2, pk_dyn2, phase_trial2 = ctps(data2, is_raw=False)
for a, b in zip([ks_dyn, pk_dyn, phase_trial],
[ks_dyn2, pk_dyn2, data2]):
assert_array_equal(a, b)
assert_true(a.min() >= 0)
assert_true(a.max() <= 1)
assert_true(b.min() >= 0)
assert_true(b.max() <= 1)
# test for normalization
assert_true((pk_dyn.min() > 0.0) or (pk_dyn.max() < 1.0))
# test shapes
assert_true(phase_trial.shape == data.shape)
assert_true(pk_dyn.shape == data.shape[1:])
# tets ground_truth + random + jittered case
assert_true(pk_dyn[0].max() == 1.0)
assert_true(len(np.unique(pk_dyn[0])) == 1.0)
assert_true(pk_dyn[1].max() < pk_max)
assert_true(pk_dyn[2].max() > 0.3)
if ii < 1:
assert_raises(ValueError, ctps,
data[:, :, :, None])
assert_true(_prob_kuiper(1.0, 400) == 1.0)
# test vecrosization
assert_array_equal(_prob_kuiper(np.array([1.0, 1.0]), 400),
_prob_kuiper(np.array([1.0, 1.0]), 400))
assert_true(_prob_kuiper(0.1, 400) < 0.1)
def get_temporal_envelope(self, origdata, W_orig, average=True):
"""
Returns the temporal envelope of the independent
components after FourierICA decomposition. Note, the
'fit()' function must be applied before this routine
can be used (to get W_orig).
Parameters
----------
origdata: array of data to be decomposed [nchan, ntsl].
W_orig: estimated de-mixing matrix
average: if set the temporal envelopes are averaged
over all epochs
default: average=True
Returns
-------
temporal_envelope: temporal envelop of the independent
components
pk_max: pk-values of the independent component
"""
# chop data into epochs and apply short-time Fourier transform (STFT)
X, _ = apply_stft(origdata, events=self.events, tpre=self.tpre, sfreq=self.sfreq,
flow=self.flow, fhigh=self.fhigh, win_length_sec=self.win_length_sec,
overlap_fac=self.overlap_fac, hamming_data=self.hamming_data,
remove_outliers=False, fcnoutliers=self.fcnoutliers,
verbose=False)
# get some size information from data
fftsize, nwindows, nchan = X.shape
ntsl = int(np.floor(self.win_length_sec*self.sfreq))
ncomp = W_orig.shape[0]
temporal_envelope = np.zeros((nwindows, ncomp, ntsl))
startfftind = int(np.floor(self.flow*self.win_length_sec))
endfftind = int(startfftind+fftsize)
fft_act = np.zeros((ncomp, ntsl), dtype=np.complex)
act = np.zeros((ncomp, nwindows, fftsize), dtype=np.complex)
# loop over all windows
for iwin in range(0, nwindows):
# transform data into FourierICA-space
X_norm = (X[:, iwin, :] - np.dot(np.ones((fftsize, 1)), self.dmean)) / \
np.dot(np.ones((fftsize, 1)), self.dstd)
act[:, iwin, :] = np.dot(W_orig, X_norm.transpose())
# act = np.dot(W_orig, X[:, iwin, :].transpose())
# apply inverse STFT to get temporal envelope
fft_act[:, startfftind:endfftind] = act[:, iwin, :]
temporal_envelope[iwin, :, :] = sc.fftpack.ifft(fft_act, n=ntsl, axis=1).real
from mne.preprocessing.ctps_ import ctps
ks_dynamics_orig, pk_dynamics_orig, _ = ctps(temporal_envelope)
pk_max = np.max(pk_dynamics_orig, axis=1)
# average data if required
if average:
temporal_envelope = np.mean(temporal_envelope, axis=0)
return temporal_envelope, pk_max
def get_temporal_envelope(self, origdata, W_orig, average=True):
"""
Returns the temporal envelope of the independent
components after FourierICA decomposition. Note, the
'fit()' function must be applied before this routine
can be used (to get W_orig).
Parameters
----------
origdata: array of data to be decomposed [nchan, ntsl].
W_orig: estimated de-mixing matrix
average: if set the temporal envelopes are averaged
over all epochs
default: average=True
Returns
-------
temporal_envelope: temporal envelop of the independent
components
pk_max: pk-values of the independent component
"""
# import necessary modules
from mne.preprocessing import ctps_ as ctps
# chop data into epochs and apply short-time Fourier transform (STFT)
X, _ = apply_stft(origdata, events=self.events, tpre=self.tpre, sfreq=self.sfreq,
flow=self.flow, fhigh=self.fhigh, win_length_sec=self.win_length_sec,
overlap_fac=self.overlap_fac, hamming_data=self.hamming_data,
remove_outliers=False, fcnoutliers=self.fcnoutliers,
verbose=False)
# get some size information from data
fftsize, nwindows, nchan = X.shape
ntsl = int(np.floor(self.win_length_sec*self.sfreq))
ncomp = W_orig.shape[0]
temporal_envelope = np.zeros((nwindows, ncomp, ntsl))
startfftind = int(np.floor(self.flow*self.win_length_sec))
endfftind = int(startfftind+fftsize)
fft_act = np.zeros((ncomp, ntsl), dtype=np.complex)
# loop over all windows
for iwin in range(0, nwindows):
# transform data into FourierICA-space
act = np.dot(W_orig, X[:, iwin, :].transpose())
# apply inverse STFT to get temporal envelope
fft_act[:, startfftind:endfftind] = act
temporal_envelope[iwin, :, :] = sc.fftpack.ifft(fft_act, n=ntsl, axis=1).real
# estimate pk-values
_, pk, _ = ctps.ctps(temporal_envelope)
pk_max = np.max(pk, axis=1)
# average data if required
if average:
temporal_envelope = np.mean(temporal_envelope, axis=0)
return temporal_envelope, pk_max
def get_ics_cardiac(meg_raw, ica, flow=10, fhigh=20, tmin=-0.3, tmax=0.3,
name_ecg='ECG 001', use_CTPS=True, proj=False,
score_func='pearsonr', thresh=0.3):
'''
Identify components with cardiac artefacts
'''
from mne.preprocessing import find_ecg_events
event_id_ecg = 999
if name_ecg in meg_raw.ch_names:
# get and filter ICA signals
ica_raw = ica.get_sources(meg_raw)
ica_raw.filter(l_freq=flow, h_freq=fhigh, n_jobs=2, method='fft')
# get R-peak indices in ECG signal
idx_R_peak, _, _ = find_ecg_events(meg_raw, ch_name=name_ecg,
event_id=event_id_ecg, l_freq=flow,
h_freq=fhigh, verbose=False)
# -----------------------------------
# default method: CTPS
# else: correlation
# -----------------------------------
if use_CTPS:
# create epochs
picks = np.arange(ica.n_components_)
ica_epochs = mne.Epochs(ica_raw, events=idx_R_peak,
event_id=event_id_ecg, tmin=tmin,
tmax=tmax, baseline=None,
proj=False, picks=picks, verbose=False)
# compute CTPS
_, pk, _ = ctps.ctps(ica_epochs.get_data())
pk_max = np.max(pk, axis=1)
idx_ecg = np.where(pk_max >= thresh)[0]
else:
# use correlation
idx_ecg = [meg_raw.ch_names.index(name_ecg)]
ecg_filtered = mne.filter.band_pass_filter(meg_raw[idx_ecg, :][0],
meg_raw.info['sfreq'],
Fp1=flow, Fp2=fhigh)
ecg_scores = ica.score_sources(meg_raw, target=ecg_filtered,
score_func=score_func)
idx_ecg = np.where(np.abs(ecg_scores) >= thresh)[0]
else:
print ">>>> NOTE: No ECG channel found!"
idx_ecg = np.array([0])
return idx_ecg
def get_ics_cardiac(meg_raw, ica, flow=10, fhigh=20, tmin=-0.3, tmax=0.3,
name_ecg='ECG 001', use_CTPS=True, proj=False,
score_func='pearsonr', thresh=0.3):
'''
Identify components with cardiac artefacts
'''
from mne.preprocessing import find_ecg_events
event_id_ecg = 999
if name_ecg in meg_raw.ch_names:
# get and filter ICA signals
ica_raw = ica.get_sources(meg_raw)
ica_raw.filter(l_freq=flow, h_freq=fhigh, n_jobs=2, method='fft')
# get R-peak indices in ECG signal
idx_R_peak, _, _ = find_ecg_events(meg_raw, ch_name=name_ecg,
event_id=event_id_ecg, l_freq=flow,
h_freq=fhigh, verbose=False)
# -----------------------------------
# default method: CTPS
# else: correlation
# -----------------------------------
if use_CTPS:
# create epochs
picks = np.arange(ica.n_components_)
ica_epochs = mne.Epochs(ica_raw, events=idx_R_peak,
event_id=event_id_ecg, tmin=tmin,
tmax=tmax, baseline=None,
proj=False, picks=picks, verbose=False)
# compute CTPS
_, pk, _ = ctps.ctps(ica_epochs.get_data())
pk_max = np.max(pk, axis=1)
idx_ecg = np.where(pk_max >= thresh)[0]
else:
# use correlation
idx_ecg = [meg_raw.ch_names.index(name_ecg)]
ecg_filtered = mne.filter.band_pass_filter(meg_raw[idx_ecg, :][0],
meg_raw.info['sfreq'],
Fp1=flow, Fp2=fhigh)
ecg_scores = ica.score_sources(meg_raw, target=ecg_filtered,
score_func=score_func)
idx_ecg = np.where(np.abs(ecg_scores) >= thresh)[0]
else:
print ">>>> NOTE: No ECG channel found!"
idx_ecg = np.array([0])
return idx_ecg
def apply_ctps(fname_ica, freqs=[(1, 4), (4, 8), (8, 12), (12, 16), (16, 20)],
tmin=-0.2, tmax=0.4, name_stim='STI 014', event_id=None,
baseline=(None, 0), proj=False):
''' Applies CTPS to a list of ICA files. '''
from jumeg.filter import jumeg_filter
fiws = jumeg_filter(filter_method="bw")
fiws.filter_type = 'bp' # bp, lp, hp
fiws.dcoffset = True
fiws.filter_attenuation_factor = 1
nfreq = len(freqs)
print '>>> CTPS calculation on: ', freqs
# Trigger or Response ?
if name_stim == 'STI 014': # trigger
trig_name = 'trigger'
else:
if name_stim == 'STI 013': # response
trig_name = 'response'
else:
trig_name = 'auxillary'
fnlist = get_files_from_list(fname_ica)
# loop across all filenames
for fnica in fnlist:
name = os.path.split(fnica)[1]
#fname = fnica[0:len(fnica)-4]
basename = fnica[:fnica.rfind(ext_ica)]
fnraw = basename + ext_raw
#basename = os.path.splitext(os.path.basename(fnica))[0]
# load cleaned data
raw = mne.io.Raw(fnraw, preload=True)
picks = mne.pick_types(raw.info, meg=True, ref_meg=False, exclude='bads')
# read (second) ICA
print ">>>> working on: " + basename
ica = mne.preprocessing.read_ica(fnica)
ica_picks = np.arange(ica.n_components_)
ncomp = len(ica_picks)
# stim events
stim_events = mne.find_events(raw, stim_channel=name_stim, consecutive=True)
nevents = len(stim_events)
if (nevents > 0):
# for a specific event ID
if event_id:
ix = np.where(stim_events[:, 2] == event_id)[0]
stim_events = stim_events[ix, :]
else:
event_id = stim_events[0, 2]
# create ctps dictionary
dctps = {'fnica': fnica,
'basename': basename,
'stim_channel': name_stim,
'trig_name': trig_name,
'ncomp': ncomp,
'nevent': nevents,
'event_id': event_id,
'nfreq': nfreq,
'freqs': freqs,
}
# loop across all filenames
pkarr = []
ptarr = []
pkmax_arr = []
for ifreq in range(nfreq):
ica_raw = ica.get_sources(raw)
flow, fhigh = freqs[ifreq][0], freqs[ifreq][1]
bp = str(flow) + '_' + str(fhigh)
# filter ICA data and create epochs
#tw=0.1
# ica_raw.filter(l_freq=flow, h_freq=fhigh, picks=ica_picks,
# method='fft',l_trans_bandwidth=tw, h_trans_bandwidth=tw)
# ica_raw.filter(l_freq=flow, h_freq=fhigh, picks=ica_picks,
# method='fft')
# filter ws settings
# later we will make this as a one line call
data_length = raw._data[0, :].size
fiws.sampling_frequency = raw.info['sfreq']
fiws.fcut1 = flow
fiws.fcut2 = fhigh
#fiws.init_filter_kernel(data_length)
#fiws.init_filter(data_length)
for ichan in ica_picks:
fiws.apply_filter(ica_raw._data[ichan, :])
ica_epochs = mne.Epochs(ica_raw, events=stim_events,
event_id=event_id, tmin=tmin,
tmax=tmax, verbose=False,
picks=ica_picks, baseline=baseline,
proj=proj)
# compute CTPS
_, pk, pt = ctps.ctps(ica_epochs.get_data())
pkmax = pk.max(1)
times = ica_epochs.times * 1e3
pkarr.append(pk)
ptarr.append(pt)
pkmax_arr.append(pkmax)
pkarr = np.array(pkarr)
ptarr = np.array(ptarr)
pkmax_arr = np.array(pkmax_arr)
dctps['pk'] = np.float32(pkarr)
dctps['pt'] = np.float32(ptarr)
dctps['pkmax'] = np.float32(pkmax_arr)
dctps['nsamp'] = len(times)
dctps['times'] = np.float32(times)
dctps['tmin'] = np.float32(ica_epochs.tmin)
dctps['tmax'] = np.float32(ica_epochs.tmax)
fnctps = basename + prefix_ctps + trig_name
np.save(fnctps, dctps)
# Note; loading example: dctps = np.load(fnctps).items()
else:
event_id = None
def ctps_ica_brain_responses_update(self,fname,raw=None,fname_ica=None,ica_raw=None,template_name=None,condition_list=None,
filter_method="bw",remove_dcoffset=False,njobs=None,
freq_ctps=np.array([]),fmin=4,fmax=32,fstep=8,proj=False,exclude_events=None,
ctps_parameter = {'time_pre':None,'time_post':None,'baseline':None},
save_phase_angles=False,fif_extention=".fif",fif_postfix="ctps"):
"""
:param fname:
:param raw:
:param fname_ica:
:param ica_raw:
:param template_name:
:param condition_list:
:param filter_method:
:param remove_dcoffset:
:param njobs:
:param freq_ctps:
:param fmin:
:param fmax:
:param fstep:
:param proj:
:param exclude_events:
:param ctps_parameter:
:param save_phase_angles:
:param fif_extention:
:param fif_postfix:
:return:
"""
self.ctps_init_brain_response_data(fname,raw=raw,fname_ica=fname_ica,ica_raw=ica_raw,template_name=template_name)
self.ctps_init_freq_bands(freq_ctps=freq_ctps,fmin=fmin,fmax=fmax,fstep=fstep)
artifact_events = self.ctps_update_artifact_time_window(aev=exclude_events)
#--- init/define bw-bp-filter-obj for ctps to filter ica_raw data within freq bands
jfi_bw = jumeg_filter(filter_method=filter_method,filter_type='bp',fcut1=None,fcut2=None,remove_dcoffset=remove_dcoffset,njobs=njobs)
jfi_bw.sampling_frequency = self.ica_raw.info['sfreq']
epocher_condition_list = self.ctps_update_hdf_condition_list(condition_list)
for condi in epocher_condition_list:
self.hdf_obj_reset_key('/ctps/'+ condi)
#--- get fresh IC's data & filter inplace
print " ---> get ica sources ...\n"
ica_orig = self.ica_raw.get_sources(self.raw)
#---for filter bands
for idx_freq in range( self.ctps_freq_bands.shape[0] ):
print " ---> START CTPS Filter Band ==> %d / %d\n" % (idx_freq+1, self.ctps_freq_bands.shape[0]+1 )
print self.ctps_freq_bands[idx_freq]
#--- get fresh IC's data & filter inplace
print " ---> copy ica sources ...\n"
ica = ica_orig.copy() # self.ica_raw.get_sources(self.raw)
print " ---> apply filter ...\n"
jfi_bw.fcut1 = self.ctps_freq_bands[idx_freq][0]
jfi_bw.fcut2 = self.ctps_freq_bands[idx_freq][1]
jfi_bw.verbose = self.verbose
jfi_bw.apply_filter(ica._data)
#--- for epocher condition
for condi in epocher_condition_list:
print " ---> START condition : " + condi + " CTPS Filter Band ==> %d / %d \n" % (idx_freq+1, self.ctps_freq_bands.shape[0] )
ctps_key = '/ctps/' + condi
#stim,ep_param,info_param = self.ctps_update_condition_parameter(condi,artifact_events)
#self.ctps_update_ctps_hdf_parameter_time(ctps_parameter=ctps_parameter,ep_param=ep_param)
if not( ctps_key in self.HDFobj.keys() ):
print"---> NEW HDF key: " + ctps_key
stim,ep_param,info_param = self.ctps_update_condition_parameter(condi,artifact_events)
self.ctps_update_ctps_hdf_parameter_time(ctps_parameter=ctps_parameter,ep_param=ep_param)
self.HDFobj[ctps_key] = pd.DataFrame( self.ctps_freq_bands ).astype(np.int16)
Hstorer = self.HDFobj.get_storer(ctps_key)
Hstorer.attrs['ctps_hdf_parameter'] = self.ctps_hdf_parameter
self.HDFobj[ctps_key+'/events'] = pd.Series( stim['events'][:,0] ).astype(np.int32)
# d=np.zeros( [ len(self.ctps_freq_bands_list ),248,1000 ] ).astype(np.int16)
self.HDFobj.flush()
print"--->done update storer: " + ctps_key
else:
ev = self.HDFobj.get(ctps_key+'/events')
Hstorer = self.HDFobj.get_storer(ctps_key)
self.ctps_hdf_parameter = Hstorer.attrs.ctps_hdf_parameter
stim['events'] = np.zeros(( ev.size, 3), dtype=np.float64)
stim['events'][:,0]= ev
stim['events'][:,2]= self.idx_hit
#pk_dynamics_key = ctps_key +'/pk_dynamics'
pk_dynamics_key = ctps_key +'/pk_dynamics/'+ self.ctps_freq_bands_list[idx_freq]
#--- make epochs
# print self.ctps_hdf_parameter
ica_epochs = mne.Epochs(ica,events=stim['events'],picks=self.ica_picks,
event_id=self.ctps_hdf_parameter['event_id'],
tmin=self.ctps_hdf_parameter['time_pre'],
tmax=self.ctps_hdf_parameter['time_post'],
baseline=self.ctps_hdf_parameter['baseline'],verbose=self.verbose,proj=proj)
#--- compute CTPS
#-------
#--- ks_dynamics : ndarray, shape (n_sources, n_times)
# The kuiper statistics.
#--- pk_dynamics : ndarray, shape (n_sources, n_times)
# The normalized kuiper index for ICA sources and
# time slices.
#--- phase_angles : ndarray, (n_epochs, n_sources, n_times) | None
# The phase values for epochs, sources and time slices. If ``assume_raw``
# is False, None is returned.
print " ---> apply compute_ctps ...\n"
if save_phase_angles :
phase_angles_key = ctps_key +'/phase_angle/'+ self.ctps_freq_bands_list[idx_freq]
_,pk_dynamics_f64,phase_angles_f64 = ctps( ica_epochs.get_data() )
#_,pk_dynamics_f64,phase_angles_f64 = ctps.compute_ctps( ica_epochs.get_data() )
self.HDFobj[ phase_angles_key ]= pd.Panel( (phase_angles_f64 * self.ctps_hdf_parameter['scale_factor'])).astype( np.int16 )
else :
_,pk_dynamics_f64,_ = ctps( ica_epochs.get_data() )
#_,pk_dynamics_f64,_ = ctps.compute_ctps( ica_epochs.get_data() )
self.HDFobj[pk_dynamics_key] = pd.DataFrame( (pk_dynamics_f64 * self.ctps_hdf_parameter['scale_factor']) ).astype( np.int16 )
self.HDFobj.flush()
# print self.HDFobj[pk_dynamics_key].transpose().max()
fhdr = self.HDFobj.filename
self.HDFobj.close()
return fhdr
def ctps_ica_steady_state_artifacts_update(self,fname,raw=None,fname_ica=None,ica_raw=None,template_name=None,condition_list=None,
filter_method="bw",remove_dcoffset=False,jobs=4,
freq_ctps=None,proj=False,njobs=None,
ctps_parameter = {'time_pre':-1.0,'time_post':1.0,'baseline':None},
save_phase_angles=False,verbose=False):
self.ctps_init_brain_response_data(fname,raw=raw,fname_ica=fname_ica,ica_raw=ica_raw,template_name=template_name)
if not freq_ctps:
self.ctps_init_freq_bands(freq_ctps=self.steady_state_artifact_bands)
else:
self.ctps_init_freq_bands(freq_ctps=freq_ctps)
#--- init/define bw-bp-filter-obj for ctps to filter ica_raw data within freq bands
jfi_bw = jumeg_filter(filter_method=filter_method,filter_type='bp',fcut1=None,fcut2=None,remove_dcoffset=remove_dcoffset,njobs=njobs)
jfi_bw.sampling_frequency = self.ica_raw.info['sfreq']
#--- init ctps_hdf_parameter
ctps_parameter['dt'] = 4 # ~every 4 sec
ctps_parameter['t0'] = 1
ctps_parameter['t1'] = int ( self.raw.index_as_time( self.raw.n_times)[0] )
self.ctps_update_ctps_hdf_parameter_time(ctps_parameter=ctps_parameter)
#--- make evs
ev_id = 2048
tpoints = np.linspace(ctps_parameter['t0'],ctps_parameter['t1'],int(ctps_parameter['t1']/ctps_parameter['dt']),endpoint=False)
# dtp= tp[1:]-tp[0:-1]
ev = np.zeros(( tpoints.size, 3), dtype=np.float64)
ev[:,0]= self.raw.time_as_index(tpoints)
ev[:,2]= ev_id
# dtsls=tsls[1:]-tsls[0:-1]
#--- make HDF node for steady-state artifacsts
storer_attrs = {'ctps_parameter': self.ctps_hdf_parameter}
stst_key='/artifacts/steady-state'
self.hdf_obj_update_dataframe(pd.DataFrame( self.ctps_freq_bands ).astype(np.int16),key=stst_key,**storer_attrs )
#--- get fresh IC's data & filter inplace
print " ---> get ica sources ...\n"
ica_orig = self.ica_raw.get_sources(self.raw)
#---for filter bands
for idx_freq in range( self.ctps_freq_bands.shape[0] ):
print " ---> START CTPS Steady-State Artifact Detection Filter Band ==> %d / %d\n" % (idx_freq+1, self.ctps_freq_bands.shape[0]+1 )
print self.ctps_freq_bands[idx_freq]
#--- get fresh IC's data & filter inplace
print " ---> copy ica sources ...\n"
ica = ica_orig.copy() # self.ica_raw.get_sources(self.raw)
print " ---> apply filter ...\n"
jfi_bw.fcut1 = self.ctps_freq_bands[idx_freq][0]
jfi_bw.fcut2 = self.ctps_freq_bands[idx_freq][1]
jfi_bw.verbose = self.verbose
jfi_bw.apply_filter(ica._data)
pk_dynamics_key = stst_key +'/pk_dynamics/'+ self.ctps_freq_bands_list[idx_freq]
ica_epochs = mne.Epochs(ica,events=ev,event_id=ev_id,
tmin=self.ctps_hdf_parameter['time_pre'],
tmax=self.ctps_hdf_parameter['time_post'],
baseline=self.ctps_hdf_parameter['baseline'],
verbose=self.verbose,proj=proj)
print " ---> Steady-State Artifact -> apply compute_ ctps ...\n"
#--- compute CTPS
#-------
#--- ks_dynamics : ndarray, shape (n_sources, n_times)
# The kuiper statistics.
#--- pk_dynamics : ndarray, shape (n_sources, n_times)
# The normalized kuiper index for ICA sources and
# time slices.
#--- phase_angles : ndarray, (n_epochs, n_sources, n_times) | None
# The phase values for epochs, sources and time slices. If ``assume_raw``
# is False, None is returned.
if save_phase_angles :
phase_angles_key = stst_key +'/phase_angle/'+ self.ctps_freq_bands_list[idx_freq]
_,pk_dynamics_f64,phase_angles_f64 = ctps( ica_epochs.get_data() )
self.HDFobj[ phase_angles_key ]= pd.Panel( (phase_angles_f64 * self.ctps_hdf_parameter['scale_factor'])).astype( np.int16 )
else :
_,pk_dynamics_f64,_ = ctps( ica_epochs.get_data() )
self.HDFobj[pk_dynamics_key] = pd.DataFrame( (pk_dynamics_f64 * self.ctps_hdf_parameter['scale_factor']) ).astype( np.int16 )
print " ---> done Steady-State Artifact -> "+ pk_dynamics_key
print "Max : %f" % ( pk_dynamics_f64.max() )
print "\n"
self.HDFobj.flush()
fhdr = self.HDFobj.filename
self.HDFobj.close()
return fhdr
def apply_ctps(fname_ica, freqs=[(1, 4), (4, 8), (8, 12), (12, 16), (16, 20)],
tmin=-0.2, tmax=0.4, name_stim='STI 014', event_id=None,
baseline=(None, 0), proj=False):
''' Applies CTPS to a list of ICA files. '''
from jumeg.filter import jumeg_filter
fiws = jumeg_filter(filter_method="bw")
fiws.filter_type = 'bp' # bp, lp, hp
fiws.dcoffset = True
fiws.filter_attenuation_factor = 1
nfreq = len(freqs)
print '>>> CTPS calculation on: ', freqs
# Trigger or Response ?
if name_stim == 'STI 014': # trigger
trig_name = 'trigger'
else:
if name_stim == 'STI 013': # response
trig_name = 'response'
else:
trig_name = 'auxillary'
fnlist = get_files_from_list(fname_ica)
# loop across all filenames
for fnica in fnlist:
name = os.path.split(fnica)[1]
#fname = fnica[0:len(fnica)-4]
basename = fnica[:fnica.rfind(ext_ica)]
fnraw = basename + ext_raw
#basename = os.path.splitext(os.path.basename(fnica))[0]
# load cleaned data
raw = mne.io.Raw(fnraw, preload=True)
picks = mne.pick_types(raw.info, meg=True, ref_meg=False, exclude='bads')
# read (second) ICA
print ">>>> working on: " + basename
ica = mne.preprocessing.read_ica(fnica)
ica_picks = np.arange(ica.n_components_)
ncomp = len(ica_picks)
# stim events
stim_events = mne.find_events(raw, stim_channel=name_stim, consecutive=True)
nevents = len(stim_events)
if (nevents > 0):
# for a specific event ID
if event_id:
ix = np.where(stim_events[:, 2] == event_id)[0]
stim_events = stim_events[ix, :]
else:
event_id = stim_events[0, 2]
# create ctps dictionary
dctps = {'fnica': fnica,
'basename': basename,
'stim_channel': name_stim,
'trig_name': trig_name,
'ncomp': ncomp,
'nevent': nevents,
'event_id': event_id,
'nfreq': nfreq,
'freqs': freqs,
}
# loop across all filenames
pkarr = []
ptarr = []
pkmax_arr = []
for ifreq in range(nfreq):
ica_raw = ica.get_sources(raw)
flow, fhigh = freqs[ifreq][0], freqs[ifreq][1]
bp = str(flow) + '_' + str(fhigh)
# filter ICA data and create epochs
#tw=0.1
# ica_raw.filter(l_freq=flow, h_freq=fhigh, picks=ica_picks,
# method='fft',l_trans_bandwidth=tw, h_trans_bandwidth=tw)
# ica_raw.filter(l_freq=flow, h_freq=fhigh, picks=ica_picks,
# method='fft')
# filter ws settings
# later we will make this as a one line call
data_length = raw._data[0, :].size
fiws.sampling_frequency = raw.info['sfreq']
fiws.fcut1 = flow
fiws.fcut2 = fhigh
#fiws.init_filter_kernel(data_length)
#fiws.init_filter(data_length)
for ichan in ica_picks:
fiws.apply_filter(ica_raw._data[ichan, :])
ica_epochs = mne.Epochs(ica_raw, events=stim_events,
event_id=event_id, tmin=tmin,
tmax=tmax, verbose=False,
picks=ica_picks, baseline=baseline,
proj=proj)
# compute CTPS
_, pk, pt = ctps.ctps(ica_epochs.get_data())
pkmax = pk.max(1)
times = ica_epochs.times * 1e3
pkarr.append(pk)
ptarr.append(pt)
pkmax_arr.append(pkmax)
pkarr = np.array(pkarr)
ptarr = np.array(ptarr)
pkmax_arr = np.array(pkmax_arr)
dctps['pk'] = np.float32(pkarr)
dctps['pt'] = np.float32(ptarr)
dctps['pkmax'] = np.float32(pkmax_arr)
dctps['nsamp'] = len(times)
dctps['times'] = np.float32(times)
dctps['tmin'] = np.float32(ica_epochs.tmin)
dctps['tmax'] = np.float32(ica_epochs.tmax)
fnctps = basename + prefix_ctps + trig_name
np.save(fnctps, dctps)
# Note; loading example: dctps = np.load(fnctps).items()
else:
event_id = None
def ctps_ica_brain_responses_update(self,fname,raw=None,fname_ica=None,ica_raw=None,template_name=None,condition_list=None,
filter_method="bw",remove_dcoffset=False,njobs=None,
freq_ctps=np.array([]),fmin=4,fmax=32,fstep=8,proj=False,exclude_events=None,
ctps_parameter = {'time_pre':None,'time_post':None,'baseline':None},
save_phase_angles=False,fif_extention=".fif",fif_postfix="ctps"):
"""
:param fname:
:param raw:
:param fname_ica:
:param ica_raw:
:param template_name:
:param condition_list:
:param filter_method:
:param remove_dcoffset:
:param njobs:
:param freq_ctps:
:param fmin:
:param fmax:
:param fstep:
:param proj:
:param exclude_events:
:param ctps_parameter:
:param save_phase_angles:
:param fif_extention:
:param fif_postfix:
:return:
"""
self.ctps_init_brain_response_data(fname,raw=raw,fname_ica=fname_ica,ica_raw=ica_raw,template_name=template_name)
self.ctps_init_freq_bands(freq_ctps=freq_ctps,fmin=fmin,fmax=fmax,fstep=fstep)
artifact_events = self.ctps_update_artifact_time_window(aev=exclude_events)
#--- init/define bw-bp-filter-obj for ctps to filter ica_raw data within freq bands
jfi_bw = jumeg_filter(filter_method=filter_method,filter_type='bp',fcut1=None,fcut2=None,remove_dcoffset=remove_dcoffset,njobs=njobs)
jfi_bw.sampling_frequency = self.ica_raw.info['sfreq']
epocher_condition_list = self.ctps_update_hdf_condition_list(condition_list)
for condi in epocher_condition_list:
self.hdf_obj_reset_key('/ctps/'+ condi)
#--- get fresh IC's data & filter inplace
print " ---> get ica sources ...\n"
ica_orig = self.ica_raw.get_sources(self.raw)
#---for filter bands
for idx_freq in range( self.ctps_freq_bands.shape[0] ):
print " ---> START CTPS Filter Band ==> %d / %d\n" % (idx_freq+1, self.ctps_freq_bands.shape[0]+1 )
print self.ctps_freq_bands[idx_freq]
#--- get fresh IC's data & filter inplace
print " ---> copy ica sources ...\n"
ica = ica_orig.copy() # self.ica_raw.get_sources(self.raw)
print " ---> apply filter ...\n"
jfi_bw.fcut1 = self.ctps_freq_bands[idx_freq][0]
jfi_bw.fcut2 = self.ctps_freq_bands[idx_freq][1]
jfi_bw.verbose = self.verbose
jfi_bw.apply_filter(ica._data)
#--- for epocher condition
for condi in epocher_condition_list:
print " ---> START condition : " + condi + " CTPS Filter Band ==> %d / %d \n" % (idx_freq+1, self.ctps_freq_bands.shape[0] )
ctps_key = '/ctps/' + condi
#stim,ep_param,info_param = self.ctps_update_condition_parameter(condi,artifact_events)
#self.ctps_update_ctps_hdf_parameter_time(ctps_parameter=ctps_parameter,ep_param=ep_param)
if not( ctps_key in self.HDFobj.keys() ):
print"---> NEW HDF key: " + ctps_key
stim,ep_param,info_param = self.ctps_update_condition_parameter(condi,artifact_events)
self.ctps_update_ctps_hdf_parameter_time(ctps_parameter=ctps_parameter,ep_param=ep_param)
self.HDFobj[ctps_key] = pd.DataFrame( self.ctps_freq_bands ).astype(np.int16)
Hstorer = self.HDFobj.get_storer(ctps_key)
Hstorer.attrs['ctps_hdf_parameter'] = self.ctps_hdf_parameter
self.HDFobj[ctps_key+'/events'] = pd.Series( stim['events'][:,0] ).astype(np.int32)
# d=np.zeros( [ len(self.ctps_freq_bands_list ),248,1000 ] ).astype(np.int16)
self.HDFobj.flush()
print"--->done update storer: " + ctps_key
else:
ev = self.HDFobj.get(ctps_key+'/events')
Hstorer = self.HDFobj.get_storer(ctps_key)
self.ctps_hdf_parameter = Hstorer.attrs.ctps_hdf_parameter
stim['events'] = np.zeros(( ev.size, 3), dtype=np.float64)
stim['events'][:,0]= ev
stim['events'][:,2]= self.idx_hit
#pk_dynamics_key = ctps_key +'/pk_dynamics'
pk_dynamics_key = ctps_key +'/pk_dynamics/'+ self.ctps_freq_bands_list[idx_freq]
#--- make epochs
# print self.ctps_hdf_parameter
ica_epochs = mne.Epochs(ica,events=stim['events'],picks=self.ica_picks,
event_id=self.ctps_hdf_parameter['event_id'],
tmin=self.ctps_hdf_parameter['time_pre'],
tmax=self.ctps_hdf_parameter['time_post'],
baseline=self.ctps_hdf_parameter['baseline'],verbose=self.verbose,proj=proj)
#--- compute CTPS
#-------
#--- ks_dynamics : ndarray, shape (n_sources, n_times)
# The kuiper statistics.
#--- pk_dynamics : ndarray, shape (n_sources, n_times)
# The normalized kuiper index for ICA sources and
# time slices.
#--- phase_angles : ndarray, (n_epochs, n_sources, n_times) | None
# The phase values for epochs, sources and time slices. If ``assume_raw``
# is False, None is returned.
print " ---> apply compute_ctps ...\n"
if save_phase_angles :
phase_angles_key = ctps_key +'/phase_angle/'+ self.ctps_freq_bands_list[idx_freq]
_,pk_dynamics_f64,phase_angles_f64 = ctps( ica_epochs.get_data() )
#_,pk_dynamics_f64,phase_angles_f64 = ctps.compute_ctps( ica_epochs.get_data() )
self.HDFobj[ phase_angles_key ]= pd.Panel( (phase_angles_f64 * self.ctps_hdf_parameter['scale_factor'])).astype( np.int16 )
else :
_,pk_dynamics_f64,_ = ctps( ica_epochs.get_data() )
#_,pk_dynamics_f64,_ = ctps.compute_ctps( ica_epochs.get_data() )
self.HDFobj[pk_dynamics_key] = pd.DataFrame( (pk_dynamics_f64 * self.ctps_hdf_parameter['scale_factor']) ).astype( np.int16 )
self.HDFobj.flush()
# print self.HDFobj[pk_dynamics_key].transpose().max()
fhdr = self.HDFobj.filename
self.HDFobj.close()
return fhdr
def ctps_ica_steady_state_artifacts_update(self,fname,raw=None,fname_ica=None,ica_raw=None,template_name=None,condition_list=None,
filter_method="bw",remove_dcoffset=False,jobs=4,
freq_ctps=None,proj=False,njobs=None,
ctps_parameter = {'time_pre':-1.0,'time_post':1.0,'baseline':None},
save_phase_angles=False,verbose=False):
self.ctps_init_brain_response_data(fname,raw=raw,fname_ica=fname_ica,ica_raw=ica_raw,template_name=template_name)
if not freq_ctps:
self.ctps_init_freq_bands(freq_ctps=self.steady_state_artifact_bands)
else:
self.ctps_init_freq_bands(freq_ctps=freq_ctps)
#--- init/define bw-bp-filter-obj for ctps to filter ica_raw data within freq bands
jfi_bw = jumeg_filter(filter_method=filter_method,filter_type='bp',fcut1=None,fcut2=None,remove_dcoffset=remove_dcoffset,njobs=njobs)
jfi_bw.sampling_frequency = self.ica_raw.info['sfreq']
#--- init ctps_hdf_parameter
ctps_parameter['dt'] = 4 # ~every 4 sec
ctps_parameter['t0'] = 1
ctps_parameter['t1'] = int ( self.raw.index_as_time( self.raw.n_times)[0] )
self.ctps_update_ctps_hdf_parameter_time(ctps_parameter=ctps_parameter)
#--- make evs
ev_id = 2048
tpoints = np.linspace(ctps_parameter['t0'],ctps_parameter['t1'],int(ctps_parameter['t1']/ctps_parameter['dt']),endpoint=False)
# dtp= tp[1:]-tp[0:-1]
ev = np.zeros(( tpoints.size, 3), dtype=np.float64)
ev[:,0]= self.raw.time_as_index(tpoints)
ev[:,2]= ev_id
# dtsls=tsls[1:]-tsls[0:-1]
#--- make HDF node for steady-state artifacsts
storer_attrs = {'ctps_parameter': self.ctps_hdf_parameter}
stst_key='/artifacts/steady-state'
self.hdf_obj_update_dataframe(pd.DataFrame( self.ctps_freq_bands ).astype(np.int16),key=stst_key,**storer_attrs )
#--- get fresh IC's data & filter inplace
print " ---> get ica sources ...\n"
ica_orig = self.ica_raw.get_sources(self.raw)
#---for filter bands
for idx_freq in range( self.ctps_freq_bands.shape[0] ):
print " ---> START CTPS Steady-State Artifact Detection Filter Band ==> %d / %d\n" % (idx_freq+1, self.ctps_freq_bands.shape[0]+1 )
print self.ctps_freq_bands[idx_freq]
#--- get fresh IC's data & filter inplace
print " ---> copy ica sources ...\n"
ica = ica_orig.copy() # self.ica_raw.get_sources(self.raw)
print " ---> apply filter ...\n"
jfi_bw.fcut1 = self.ctps_freq_bands[idx_freq][0]
jfi_bw.fcut2 = self.ctps_freq_bands[idx_freq][1]
jfi_bw.verbose = self.verbose
jfi_bw.apply_filter(ica._data)
pk_dynamics_key = stst_key +'/pk_dynamics/'+ self.ctps_freq_bands_list[idx_freq]
ica_epochs = mne.Epochs(ica,events=ev,event_id=ev_id,
tmin=self.ctps_hdf_parameter['time_pre'],
tmax=self.ctps_hdf_parameter['time_post'],
baseline=self.ctps_hdf_parameter['baseline'],
verbose=self.verbose,proj=proj)
print " ---> Steady-State Artifact -> apply compute_ ctps ...\n"
#--- compute CTPS
#-------
#--- ks_dynamics : ndarray, shape (n_sources, n_times)
# The kuiper statistics.
#--- pk_dynamics : ndarray, shape (n_sources, n_times)
# The normalized kuiper index for ICA sources and
# time slices.
#--- phase_angles : ndarray, (n_epochs, n_sources, n_times) | None
# The phase values for epochs, sources and time slices. If ``assume_raw``
# is False, None is returned.
if save_phase_angles :
phase_angles_key = stst_key +'/phase_angle/'+ self.ctps_freq_bands_list[idx_freq]
_,pk_dynamics_f64,phase_angles_f64 = ctps( ica_epochs.get_data() )
self.HDFobj[ phase_angles_key ]= pd.Panel( (phase_angles_f64 * self.ctps_hdf_parameter['scale_factor'])).astype( np.int16 )
else :
_,pk_dynamics_f64,_ = ctps( ica_epochs.get_data() )
self.HDFobj[pk_dynamics_key] = pd.DataFrame( (pk_dynamics_f64 * self.ctps_hdf_parameter['scale_factor']) ).astype( np.int16 )
print " ---> done Steady-State Artifact -> "+ pk_dynamics_key
print "Max : %f" % ( pk_dynamics_f64.max() )
print "\n"
self.HDFobj.flush()
fhdr = self.HDFobj.filename
self.HDFobj.close()
return fhdr
def get_ics_cardiac(meg_raw,
ica,
flow=8,
fhigh=25,
tmin=-0.4,
tmax=0.4,
name_ecg='ECG 001',
use_CTPS=True,
event_id=999,
score_func='pearsonr',
thresh=0.25):
'''
Identify components with cardiac artefacts
'''
from mne.preprocessing import find_ecg_events
idx_ecg = []
if name_ecg in meg_raw.ch_names:
# get and filter ICA signals
ica_raw = ica.get_sources(meg_raw)
ica_raw.filter(l_freq=flow, h_freq=fhigh, n_jobs=2, method='fft')
# get ECG events
events_ecg, _, _ = find_ecg_events(meg_raw,
ch_name=name_ecg,
event_id=event_id,
l_freq=flow,
h_freq=fhigh,
verbose=False)
# CTPS
if use_CTPS:
# create epochs
picks = np.arange(ica.n_components_)
ica_epochs = mne.Epochs(ica_raw,
events=events_ecg,
event_id=event_id,
tmin=tmin,
tmax=tmax,
baseline=None,
proj=False,
picks=picks,
verbose=False)
# compute CTPS
_, pk, _ = ctps.ctps(ica_epochs.get_data())
pk_max = np.max(pk, axis=1)
scores_ecg = pk_max
ic_ecg = np.where(pk_max >= thresh)[0]
else:
# use correlation
idx_ecg = [meg_raw.ch_names.index(name_ecg)]
ecg_filtered = mne.filter.filter_data(meg_raw[idx_ecg, :][0],
meg_raw.info['sfreq'],
l_freq=flow,
h_freq=fhigh)
scores_ecg = ica.score_sources(meg_raw,
target=ecg_filtered,
score_func=score_func)
ic_ecg = np.where(np.abs(scores_ecg) >= thresh)[0]
else:
logger.warning(">>>> Warning: Could not find ECG channel %s" %
name_ecg)
events_ecg = []
if len(ic_ecg) == 0:
ic_ecg = np.array([-1])
scores_ecg = np.zeros(
ica.n_components) #scores_ecg = np.array([-1]) ???
events_ecg = np.array([-1])
else:
events_ecg[:,
0] -= meg_raw.first_samp # make sure event samples start from 0
return [ic_ecg, scores_ecg, events_ecg]
