def apply_filter_data(fname,raw=None,filter_method="mne",filter_type='bp',fcut1=1.0,fcut2=45.0,notch=None,notch_max=None,order=4,
remove_dcoffset = False,njobs=1,overwrite = False,do_run=True,verbose=False,save=True,picks=None,
fif_postfix=None, fif_extention='-raw.fif'):
'''
Applies the FIR FFT filter [bp,hp,lp,notches] to data array.
filter_method : mne => fft mne-filter
bw => fft butterwoth
ws => fft - windowed sinc
'''
from jumeg.filter import jumeg_filter
#--- define filter
jfilter = jumeg_filter(filter_method=filter_method,filter_type=filter_type,fcut1=fcut1,fcut2=fcut2,njobs=njobs,
remove_dcoffset=True,order=order)
jfilter.verbose = verbose
if do_run :
if raw is None:
if fname is None:
print"ERROR no file foumd!!\n"
return
raw = mne.io.Raw(fname,preload=True)
print"\n"
if picks is None :
picks = jumeg_base.pick_channels_nobads(raw)
#- apply filter for picks, exclude stim,resp,bads
jfilter.sampling_frequency = raw.info['sfreq']
#--- calc notch array 50,100,150 .. max
if notch :
jfilter.calc_notches(notch,notch_max)
jfilter.apply_filter(raw._data,picks=picks )
jfilter.update_info_filter_settings(raw)
#--- make output filename
name_raw = fname.split('-')[0]
fnfilt = name_raw + "," + jfilter.filter_name_postfix + fif_extention
raw.info['filename'] = fnfilt
if save :
fnfilt = jumeg_base.apply_save_mne_data(raw,fname=fnfilt)
else:
#--- calc notch array 50,100,150 .. max
if notch :
jfilter.calc_notches(notch,notch_max)
#--- make output filename
name_raw = fname.split('-')[0]
fnfilt = name_raw + "," + jfilter.filter_name_postfix + fif_extention
return (fnfilt, raw)
def noise_reducer_4raw_data(fname_raw,
raw=None,
signals=[],
noiseref=[],
detrending=None,
tmin=None,
tmax=None,
reflp=None,
refhp=None,
refnotch=None,
exclude_artifacts=True,
checkresults=True,
fif_extention="-raw.fif",
fif_postfix="nr",
reject={
'grad': 4000e-13,
'mag': 4e-12,
'eeg': 40e-6,
'eog': 250e-6
},
complementary_signal=False,
fnout=None,
verbose=False,
save=True):
"""Apply noise reduction to signal channels using reference channels.
!!! ONLY ONE RAW Obj Interface Version FB !!!
Parameters
----------
fname_raw : rawfile name
raw : fif raw object
signals : list of string
List of channels to compensate using noiseref.
If empty use the meg signal channels.
noiseref : list of string | str
List of channels to use as noise reference.
If empty use the magnetic reference channsls (default).
signals and noiseref may contain regexp, which are resolved
using mne.pick_channels_regexp(). All other channels are copied.
tmin : lower latency bound for weight-calc [start of trace]
tmax : upper latency bound for weight-calc [ end of trace]
Weights are calc'd for (tmin,tmax), but applied to entire data set
refhp : high-pass frequency for reference signal filter [None]
reflp : low-pass frequency for reference signal filter [None]
reflp < refhp: band-stop filter
reflp > refhp: band-pass filter
reflp is not None, refhp is None: low-pass filter
reflp is None, refhp is not None: high-pass filter
refnotch : (base) notch frequency for reference signal filter [None]
use raw(ref)-notched(ref) as reference signal
exclude_artifacts: filter signal-channels thru _is_good() [True]
(parameters are at present hard-coded!)
complementary_signal : replaced signal by traces that would be subtracted [False]
(can be useful for debugging)
checkresults : boolean to control internal checks and overall success [True]
reject = dict for rejection threshold
units:
grad: T / m (gradiometers)
mag: T (magnetometers)
eeg/eog: uV (EEG channels)
default=>{'grad':4000e-13,'mag':4e-12,'eeg':40e-6,'eog':250e-6}
save : save data to fif file
Outputfile:
-------
<wawa>,nr-raw.fif for input <wawa>-raw.fif
Returns
-------
TBD
Bugs
----
- artifact checking is incomplete (and with arb. window of tstep=0.2s)
- no accounting of channels used as signal/reference
- non existing input file handled ungracefully
"""
tc0 = time.clock()
tw0 = time.time()
if type(complementary_signal) != bool:
raise ValueError("Argument complementary_signal must be of type bool")
raw, fname_raw = jumeg_base.get_raw_obj(fname_raw, raw=raw)
if detrending:
raw = perform_detrending(None, raw=raw, save=False)
tc1 = time.clock()
tw1 = time.time()
if verbose:
print ">>> loading raw data took %.1f ms (%.2f s walltime)" % (
1000. * (tc1 - tc0), (tw1 - tw0))
# Time window selection
# weights are calc'd based on [tmin,tmax], but applied to the entire data set.
# tstep is used in artifact detection
# tmin,tmax variables must not be changed here!
if tmin is None:
itmin = 0
else:
itmin = int(floor(tmin * raw.info['sfreq']))
if tmax is None:
itmax = raw.last_samp
else:
itmax = int(ceil(tmax * raw.info['sfreq']))
if itmax - itmin < 2:
raise ValueError(
"Time-window for noise compensation empty or too short")
if verbose:
print ">>> Set time-range to [%7.3f,%7.3f]" % \
(raw.index_as_time(itmin)[0], raw.index_as_time(itmax)[0])
if signals is None or len(signals) == 0:
sigpick = jumeg_base.pick_meg_nobads(raw)
else:
sigpick = channel_indices_from_list(raw.info['ch_names'][:], signals,
raw.info.get('bads'))
nsig = len(sigpick)
if nsig == 0:
raise ValueError("No channel selected for noise compensation")
if noiseref is None or len(noiseref) == 0:
# References are not limited to 4D ref-chans, but can be anything,
# incl. ECG or powerline monitor.
if verbose:
print ">>> Using all refchans."
refexclude = "bads"
refpick = jumeg_base.pick_ref_nobads(raw)
else:
refpick = channel_indices_from_list(raw.info['ch_names'][:], noiseref,
raw.info.get('bads'))
nref = len(refpick)
if nref == 0:
raise ValueError("No channel selected as noise reference")
if verbose:
print ">>> sigpick: %3d chans, refpick: %3d chans" % (nsig, nref)
if reflp is None and refhp is None and refnotch is None:
use_reffilter = False
use_refantinotch = False
else:
use_reffilter = True
if verbose:
print "########## Filter reference channels:"
use_refantinotch = False
if refnotch is not None:
if reflp is None and reflp is None:
use_refantinotch = True
freqlast = np.min([5.01 * refnotch, 0.5 * raw.info['sfreq']])
if verbose:
print ">>> notches at freq %.1f and harmonics below %.1f" % (
refnotch, freqlast)
else:
raise ValueError("Cannot specify notch- and high-/low-pass"
"reference filter together")
else:
if verbose:
if reflp is not None:
print ">>> low-pass with cutoff-freq %.1f" % reflp
if refhp is not None:
print ">>> high-pass with cutoff-freq %.1f" % refhp
# Adapt followg drop-chans cmd to use 'all-but-refpick'
droplist = [
raw.info['ch_names'][k] for k in xrange(raw.info['nchan'])
if not k in refpick
]
tct = time.clock()
twt = time.time()
fltref = raw.drop_channels(droplist, copy=True)
if use_refantinotch:
rawref = raw.drop_channels(droplist, copy=True)
freqlast = np.min([5.01 * refnotch, 0.5 * raw.info['sfreq']])
fltref.notch_filter(np.arange(refnotch, freqlast, refnotch),
picks=np.array(xrange(nref)),
method='iir')
fltref._data = (rawref._data - fltref._data)
else:
fltref.filter(refhp,
reflp,
picks=np.array(xrange(nref)),
method='iir')
tc1 = time.clock()
tw1 = time.time()
if verbose:
print ">>> filtering ref-chans took %.1f ms (%.2f s walltime)" % (
1000. * (tc1 - tct), (tw1 - twt))
if verbose:
print "########## Calculating sig-ref/ref-ref-channel covariances:"
# Calculate sig-ref/ref-ref-channel covariance:
# (there is no need to calc inter-signal-chan cov,
# but there seems to be no appropriat fct available)
# Here we copy the idea from compute_raw_data_covariance()
# and truncate it as appropriate.
tct = time.clock()
twt = time.time()
# The following reject and infosig entries are only
# used in _is_good-calls.
# _is_good() from mne-0.9.git-py2.7.egg/mne/epochs.py seems to
# ignore ref-channels (not covered by dict) and checks individual
# data segments - artifacts across a buffer boundary are not found.
#--- !!! FB put to kwargs
#reject = dict(grad=4000e-13, # T / m (gradiometers)
# mag=4e-12, # T (magnetometers)
# eeg=40e-6, # uV (EEG channels)
# eog=250e-6) # uV (EOG channels)
infosig = copy.copy(raw.info)
infosig['chs'] = [raw.info['chs'][k] for k in sigpick]
infosig['ch_names'] = [raw.info['ch_names'][k] for k in sigpick]
infosig['nchan'] = len(sigpick)
idx_by_typesig = channel_indices_by_type(infosig)
# Read data in chunks:
tstep = 0.2
itstep = int(ceil(tstep * raw.info['sfreq']))
sigmean = 0
refmean = 0
sscovdata = 0
srcovdata = 0
rrcovdata = 0
n_samples = 0
for first in range(itmin, itmax, itstep):
last = first + itstep
if last >= itmax:
last = itmax
raw_segmentsig, times = raw[sigpick, first:last]
if use_reffilter:
raw_segmentref, times = fltref[:, first:last]
else:
raw_segmentref, times = raw[refpick, first:last]
if not exclude_artifacts or \
_is_good(raw_segmentsig, infosig['ch_names'], idx_by_typesig, reject, flat=None,
ignore_chs=raw.info['bads']):
sigmean += raw_segmentsig.sum(axis=1)
refmean += raw_segmentref.sum(axis=1)
sscovdata += (raw_segmentsig * raw_segmentsig).sum(axis=1)
srcovdata += np.dot(raw_segmentsig, raw_segmentref.T)
rrcovdata += np.dot(raw_segmentref, raw_segmentref.T)
n_samples += raw_segmentsig.shape[1]
else:
logger.info("Artefact detected in [%d, %d]" % (first, last))
if n_samples <= 1:
raise ValueError('Too few samples to calculate weights')
sigmean /= n_samples
refmean /= n_samples
sscovdata -= n_samples * sigmean[:] * sigmean[:]
sscovdata /= (n_samples - 1)
srcovdata -= n_samples * sigmean[:, None] * refmean[None, :]
srcovdata /= (n_samples - 1)
rrcovdata -= n_samples * refmean[:, None] * refmean[None, :]
rrcovdata /= (n_samples - 1)
sscovinit = np.copy(sscovdata)
if verbose:
print ">>> Normalize srcov..."
rrslope = copy.copy(rrcovdata)
for iref in xrange(nref):
dtmp = rrcovdata[iref, iref]
if dtmp > TINY:
srcovdata[:, iref] /= dtmp
rrslope[:, iref] /= dtmp
else:
srcovdata[:, iref] = 0.
rrslope[:, iref] = 0.
if verbose:
print ">>> Number of samples used : %d" % n_samples
tc1 = time.clock()
tw1 = time.time()
print ">>> sigrefchn covar-calc took %.1f ms (%.2f s walltime)" % (
1000. * (tc1 - tct), (tw1 - twt))
if checkresults:
if verbose:
print "########## Calculated initial signal channel covariance:"
# Calculate initial signal channel covariance:
# (only used as quality measure)
print ">>> initl rt(avg sig pwr) = %12.5e" % np.sqrt(
np.mean(sscovdata))
for i in xrange(5):
print ">>> initl signal-rms[%3d] = %12.5e" % (
i, np.sqrt(sscovdata.flatten()[i]))
print ">>>"
U, s, V = np.linalg.svd(rrslope, full_matrices=True)
if verbose:
print ">>> singular values:"
print s
print ">>> Applying cutoff for smallest SVs:"
dtmp = s.max() * SVD_RELCUTOFF
s *= (abs(s) >= dtmp)
sinv = [1. / s[k] if s[k] != 0. else 0. for k in xrange(nref)]
if verbose:
print ">>> singular values (after cutoff):"
print s
stat = np.allclose(rrslope, np.dot(U, np.dot(np.diag(s), V)))
if verbose:
print ">>> Testing svd-result: %s" % stat
if not stat:
print " (Maybe due to SV-cutoff?)"
# Solve for inverse coefficients:
# Set RRinv.tr=U diag(sinv) V
RRinv = np.transpose(np.dot(U, np.dot(np.diag(sinv), V)))
if checkresults:
stat = np.allclose(np.identity(nref), np.dot(RRinv, rrslope))
if stat:
if verbose:
print ">>> Testing RRinv-result (should be unit-matrix): ok"
else:
print ">>> Testing RRinv-result (should be unit-matrix): failed"
print np.transpose(np.dot(RRinv, rrslope))
print ">>>"
if verbose:
print "########## Calc weight matrix..."
# weights-matrix will be somewhat larger than necessary,
# (to simplify indexing in compensation loop):
weights = np.zeros((raw._data.shape[0], nref))
for isig in xrange(nsig):
for iref in xrange(nref):
weights[sigpick[isig], iref] = np.dot(srcovdata[isig, :],
RRinv[:, iref])
if verbose:
print "########## Compensating signal channels:"
if complementary_signal:
print ">>> Caveat: REPLACING signal by compensation signal"
tct = time.clock()
twt = time.time()
# Work on entire data stream:
for isl in xrange(raw._data.shape[1]):
slice = np.take(raw._data, [isl], axis=1)
if use_reffilter:
refslice = np.take(fltref._data, [isl], axis=1)
refarr = refslice[:].flatten() - refmean
# refarr = fltres[:,isl]-refmean
else:
refarr = slice[refpick].flatten() - refmean
subrefarr = np.dot(weights[:], refarr)
if not complementary_signal:
raw._data[:, isl] -= subrefarr
else:
raw._data[:, isl] = subrefarr
if (isl % 10000 == 0) and verbose:
print "\rProcessed slice %6d" % isl
if verbose:
print "\nDone."
tc1 = time.clock()
tw1 = time.time()
print ">>> compensation loop took %.1f ms (%.2f s walltime)" % (
1000. * (tc1 - tct), (tw1 - twt))
if checkresults:
if verbose:
print "########## Calculating final signal channel covariance:"
# Calculate final signal channel covariance:
# (only used as quality measure)
tct = time.clock()
twt = time.time()
sigmean = 0
sscovdata = 0
n_samples = 0
for first in range(itmin, itmax, itstep):
last = first + itstep
if last >= itmax:
last = itmax
raw_segmentsig, times = raw[sigpick, first:last]
# Artifacts found here will probably differ from pre-noisered artifacts!
if not exclude_artifacts or \
_is_good(raw_segmentsig, infosig['ch_names'], idx_by_typesig, reject,
flat=None, ignore_chs=raw.info['bads']):
sigmean += raw_segmentsig.sum(axis=1)
sscovdata += (raw_segmentsig * raw_segmentsig).sum(axis=1)
n_samples += raw_segmentsig.shape[1]
sigmean /= n_samples
sscovdata -= n_samples * sigmean[:] * sigmean[:]
sscovdata /= (n_samples - 1)
if verbose:
print ">>> no channel got worse: ", np.all(
np.less_equal(sscovdata, sscovinit))
print ">>> final rt(avg sig pwr) = %12.5e" % np.sqrt(
np.mean(sscovdata))
for i in xrange(5):
print ">>> final signal-rms[%3d] = %12.5e" % (
i, np.sqrt(sscovdata.flatten()[i]))
tc1 = time.clock()
tw1 = time.time()
print ">>> signal covar-calc took %.1f ms (%.2f s walltime)" % (
1000. * (tc1 - tct), (tw1 - twt))
print ">>>"
#--- fb update 21.07.2015
fname_out = jumeg_base.get_fif_name(raw=raw,
postfix=fif_postfix,
extention=fif_extention)
if save:
jumeg_base.apply_save_mne_data(raw, fname=fname_out, overwrite=True)
tc1 = time.clock()
tw1 = time.time()
if verbose:
print ">>> Total run took %.1f ms (%.2f s walltime)" % (1000. *
(tc1 - tc0),
(tw1 - tw0))
return raw, fname_out
def apply_noise_reducer_data(fname,
raw=None,
do_run=True,
verbose=False,
save=True,
plot=False,
reflp=None,
refhp=None,
refnotch=None,
fif_postfix="nr",
fif_extention="-raw.fif",
**kwargs):
'''
Applies the noise reducer to raw obj data or to fif-file.
the magic ee once
fb modified for raw obj support
imports jumeg_noise_reducer_4raw_data
'''
import os
#--- import noise_reducer and plot_power_spectrum function
from jumeg.jumeg_4raw_data_noise_reducer import noise_reducer_4raw_data, plot_denoising_4raw_data
fname_out = None
nr_done = False
if do_run:
raw, fname_raw = jumeg_base.get_raw_obj(fname, raw=raw)
fname_out = jumeg_base.get_fif_name(raw=raw,
postfix=fif_postfix,
extention=fif_extention,
update_raw_fname=False)
#--- apply noise reducer for 50 Hz (and harmonics)
if (reflp or refhp):
raw, fname_out = noise_reducer_4raw_data(fname,
raw=raw,
reflp=reflp,
refhp=refhp,
verbose=verbose,
save=False,
**kwargs['parameter'])
kwargs['parameter']['detrending'] = None
nr_done = True
if refnotch:
for refn in refnotch:
raw, fname_out = noise_reducer_4raw_data(None,
raw=raw,
refnotch=refn,
verbose=verbose,
save=False,
**kwargs['parameter'])
kwargs['parameter']['detrending'] = None
nr_done = True
# raw.info['filename'] = fname_out
if not nr_done:
return fname_raw, raw
raw.info['filename'] = fname_out
if save:
fname_out = jumeg_base.apply_save_mne_data(raw, fname=fname_out)
if plot:
print " --> noise reducer plot power spectrum"
from distutils.dir_util import mkpath
p, pdf = os.path.split(fname_raw)
plot_dir = p + '/plots/'
mkpath(plot_dir)
fn_power_spect = plot_dir + pdf[:pdf.rfind('-raw.fif'
)] + ',denoising'
plot_denoising_4raw_data([fname_raw, fname_out],
show=False,
fnout=fn_power_spect)
print "---> noise reducer plot :" + fn_power_spect
print "---> Done noise reducer: " + fname_out
return (fname_out, raw)
def apply_noise_reducer_data(fname,raw=None,do_run=True,verbose=False,save=True,plot=False,
reflp=None, refhp=None, refnotch=None,fif_postfix="nr",fif_extention="-raw.fif",**kwargs):
'''
Applies the noise reducer to raw obj data or to fif-file.
the magic ee once
fb modified for raw obj support
imports jumeg_noise_reducer_4raw_data
'''
import os
#--- import noise_reducer and plot_power_spectrum function
from jumeg.jumeg_4raw_data_noise_reducer import noise_reducer_4raw_data, plot_denoising_4raw_data
fname_out = None
nr_done = False
if do_run :
raw,fname_raw = jumeg_base.get_raw_obj(fname,raw=raw)
fname_out = jumeg_base.get_fif_name(raw=raw,postfix=fif_postfix,extention=fif_extention,update_raw_fname=False)
#--- apply noise reducer for 50 Hz (and harmonics)
if (reflp or refhp):
raw,fname_out = noise_reducer_4raw_data(fname,raw=raw,reflp=reflp,refhp=refhp,verbose=verbose,save=False,**kwargs['parameter'])
kwargs['parameter']['detrending'] = None
nr_done = True
if refnotch:
for refn in refnotch:
raw,fname_out = noise_reducer_4raw_data(None,raw=raw,refnotch=refn,verbose=verbose,save=False,**kwargs['parameter'])
kwargs['parameter']['detrending'] = None
nr_done = True
# raw.info['filename'] = fname_out
if not nr_done :
return fname_raw,raw
raw.info['filename'] = fname_out
if save:
fname_out = jumeg_base.apply_save_mne_data(raw,fname=fname_out)
if plot:
print " --> noise reducer plot power spectrum"
from distutils.dir_util import mkpath
p,pdf = os.path.split(fname_raw)
plot_dir = p+ '/plots/'
mkpath(plot_dir)
fn_power_spect = plot_dir + pdf[:pdf.rfind('-raw.fif') ]+ ',denoising'
plot_denoising_4raw_data([fname_raw,fname_out],show=False,fnout=fn_power_spect)
print"---> noise reducer plot :" + fn_power_spect
print "---> Done noise reducer: "+ fname_out
return (fname_out,raw)
def noise_reducer_4raw_data(fname_raw,raw=None,signals=[],noiseref=[],detrending=None,
tmin=None,tmax=None,reflp=None,refhp=None,refnotch=None,
exclude_artifacts=True,checkresults=True,
fif_extention="-raw.fif",fif_postfix="nr",
reject={'grad':4000e-13,'mag':4e-12,'eeg':40e-6,'eog':250e-6},
complementary_signal=False,fnout=None,verbose=False,save=True):
"""Apply noise reduction to signal channels using reference channels.
!!! ONLY ONE RAW Obj Interface Version FB !!!
Parameters
----------
fname_raw : rawfile name
raw : fif raw object
signals : list of string
List of channels to compensate using noiseref.
If empty use the meg signal channels.
noiseref : list of string | str
List of channels to use as noise reference.
If empty use the magnetic reference channsls (default).
signals and noiseref may contain regexp, which are resolved
using mne.pick_channels_regexp(). All other channels are copied.
tmin : lower latency bound for weight-calc [start of trace]
tmax : upper latency bound for weight-calc [ end of trace]
Weights are calc'd for (tmin,tmax), but applied to entire data set
refhp : high-pass frequency for reference signal filter [None]
reflp : low-pass frequency for reference signal filter [None]
reflp < refhp: band-stop filter
reflp > refhp: band-pass filter
reflp is not None, refhp is None: low-pass filter
reflp is None, refhp is not None: high-pass filter
refnotch : (base) notch frequency for reference signal filter [None]
use raw(ref)-notched(ref) as reference signal
exclude_artifacts: filter signal-channels thru _is_good() [True]
(parameters are at present hard-coded!)
complementary_signal : replaced signal by traces that would be subtracted [False]
(can be useful for debugging)
checkresults : boolean to control internal checks and overall success [True]
reject = dict for rejection threshold
units:
grad: T / m (gradiometers)
mag: T (magnetometers)
eeg/eog: uV (EEG channels)
default=>{'grad':4000e-13,'mag':4e-12,'eeg':40e-6,'eog':250e-6}
save : save data to fif file
Outputfile:
-------
<wawa>,nr-raw.fif for input <wawa>-raw.fif
Returns
-------
TBD
Bugs
----
- artifact checking is incomplete (and with arb. window of tstep=0.2s)
- no accounting of channels used as signal/reference
- non existing input file handled ungracefully
"""
tc0 = time.clock()
tw0 = time.time()
if type(complementary_signal) != bool:
raise ValueError("Argument complementary_signal must be of type bool")
raw,fname_raw = jumeg_base.get_raw_obj(fname_raw,raw=raw)
if detrending:
raw = perform_detrending(None,raw=raw, save=False)
tc1 = time.clock()
tw1 = time.time()
if verbose:
print ">>> loading raw data took %.1f ms (%.2f s walltime)" % (1000. * (tc1 - tc0), (tw1 - tw0))
# Time window selection
# weights are calc'd based on [tmin,tmax], but applied to the entire data set.
# tstep is used in artifact detection
# tmin,tmax variables must not be changed here!
if tmin is None:
itmin = 0
else:
itmin = int(floor(tmin * raw.info['sfreq']))
if tmax is None:
itmax = raw.last_samp
else:
itmax = int(ceil(tmax * raw.info['sfreq']))
if itmax - itmin < 2:
raise ValueError("Time-window for noise compensation empty or too short")
if verbose:
print ">>> Set time-range to [%7.3f,%7.3f]" % \
(raw.index_as_time(itmin)[0], raw.index_as_time(itmax)[0])
if signals is None or len(signals) == 0:
sigpick = jumeg_base.pick_meg_nobads(raw)
else:
sigpick = channel_indices_from_list(raw.info['ch_names'][:], signals,
raw.info.get('bads'))
nsig = len(sigpick)
if nsig == 0:
raise ValueError("No channel selected for noise compensation")
if noiseref is None or len(noiseref) == 0:
# References are not limited to 4D ref-chans, but can be anything,
# incl. ECG or powerline monitor.
if verbose:
print ">>> Using all refchans."
refexclude = "bads"
refpick = jumeg_base.pick_ref_nobads(raw)
else:
refpick = channel_indices_from_list(raw.info['ch_names'][:], noiseref,
raw.info.get('bads'))
nref = len(refpick)
if nref == 0:
raise ValueError("No channel selected as noise reference")
if verbose:
print ">>> sigpick: %3d chans, refpick: %3d chans" % (nsig, nref)
if reflp is None and refhp is None and refnotch is None:
use_reffilter = False
use_refantinotch = False
else:
use_reffilter = True
if verbose:
print "########## Filter reference channels:"
use_refantinotch = False
if refnotch is not None:
if reflp is None and reflp is None:
use_refantinotch = True
freqlast = np.min([5.01 * refnotch, 0.5 * raw.info['sfreq']])
if verbose:
print ">>> notches at freq %.1f and harmonics below %.1f" % (refnotch, freqlast)
else:
raise ValueError("Cannot specify notch- and high-/low-pass"
"reference filter together")
else:
if verbose:
if reflp is not None:
print ">>> low-pass with cutoff-freq %.1f" % reflp
if refhp is not None:
print ">>> high-pass with cutoff-freq %.1f" % refhp
# Adapt followg drop-chans cmd to use 'all-but-refpick'
droplist = [raw.info['ch_names'][k] for k in xrange(raw.info['nchan']) if not k in refpick]
tct = time.clock()
twt = time.time()
fltref = raw.drop_channels(droplist, copy=True)
if use_refantinotch:
rawref = raw.drop_channels(droplist, copy=True)
freqlast = np.min([5.01 * refnotch, 0.5 * raw.info['sfreq']])
fltref.notch_filter(np.arange(refnotch, freqlast, refnotch),
picks=np.array(xrange(nref)), method='iir')
fltref._data = (rawref._data - fltref._data)
else:
fltref.filter(refhp, reflp, picks=np.array(xrange(nref)), method='iir')
tc1 = time.clock()
tw1 = time.time()
if verbose:
print ">>> filtering ref-chans took %.1f ms (%.2f s walltime)" % (1000. * (tc1 - tct), (tw1 - twt))
if verbose:
print "########## Calculating sig-ref/ref-ref-channel covariances:"
# Calculate sig-ref/ref-ref-channel covariance:
# (there is no need to calc inter-signal-chan cov,
# but there seems to be no appropriat fct available)
# Here we copy the idea from compute_raw_data_covariance()
# and truncate it as appropriate.
tct = time.clock()
twt = time.time()
# The following reject and infosig entries are only
# used in _is_good-calls.
# _is_good() from mne-0.9.git-py2.7.egg/mne/epochs.py seems to
# ignore ref-channels (not covered by dict) and checks individual
# data segments - artifacts across a buffer boundary are not found.
#--- !!! FB put to kwargs
#reject = dict(grad=4000e-13, # T / m (gradiometers)
# mag=4e-12, # T (magnetometers)
# eeg=40e-6, # uV (EEG channels)
# eog=250e-6) # uV (EOG channels)
infosig = copy.copy(raw.info)
infosig['chs'] = [raw.info['chs'][k] for k in sigpick]
infosig['ch_names'] = [raw.info['ch_names'][k] for k in sigpick]
infosig['nchan'] = len(sigpick)
idx_by_typesig = channel_indices_by_type(infosig)
# Read data in chunks:
tstep = 0.2
itstep = int(ceil(tstep * raw.info['sfreq']))
sigmean = 0
refmean = 0
sscovdata = 0
srcovdata = 0
rrcovdata = 0
n_samples = 0
for first in range(itmin, itmax, itstep):
last = first + itstep
if last >= itmax:
last = itmax
raw_segmentsig, times = raw[sigpick, first:last]
if use_reffilter:
raw_segmentref, times = fltref[:, first:last]
else:
raw_segmentref, times = raw[refpick, first:last]
if not exclude_artifacts or \
_is_good(raw_segmentsig, infosig['ch_names'], idx_by_typesig, reject, flat=None,
ignore_chs=raw.info['bads']):
sigmean += raw_segmentsig.sum(axis=1)
refmean += raw_segmentref.sum(axis=1)
sscovdata += (raw_segmentsig * raw_segmentsig).sum(axis=1)
srcovdata += np.dot(raw_segmentsig, raw_segmentref.T)
rrcovdata += np.dot(raw_segmentref, raw_segmentref.T)
n_samples += raw_segmentsig.shape[1]
else:
logger.info("Artefact detected in [%d, %d]" % (first, last))
if n_samples <= 1:
raise ValueError('Too few samples to calculate weights')
sigmean /= n_samples
refmean /= n_samples
sscovdata -= n_samples * sigmean[:] * sigmean[:]
sscovdata /= (n_samples - 1)
srcovdata -= n_samples * sigmean[:, None] * refmean[None, :]
srcovdata /= (n_samples - 1)
rrcovdata -= n_samples * refmean[:, None] * refmean[None, :]
rrcovdata /= (n_samples - 1)
sscovinit = np.copy(sscovdata)
if verbose:
print ">>> Normalize srcov..."
rrslope = copy.copy(rrcovdata)
for iref in xrange(nref):
dtmp = rrcovdata[iref, iref]
if dtmp > TINY:
srcovdata[:, iref] /= dtmp
rrslope[:, iref] /= dtmp
else:
srcovdata[:, iref] = 0.
rrslope[:, iref] = 0.
if verbose:
print ">>> Number of samples used : %d" % n_samples
tc1 = time.clock()
tw1 = time.time()
print ">>> sigrefchn covar-calc took %.1f ms (%.2f s walltime)" % (1000. * (tc1 - tct), (tw1 - twt))
if checkresults:
if verbose:
print "########## Calculated initial signal channel covariance:"
# Calculate initial signal channel covariance:
# (only used as quality measure)
print ">>> initl rt(avg sig pwr) = %12.5e" % np.sqrt(np.mean(sscovdata))
for i in xrange(5):
print ">>> initl signal-rms[%3d] = %12.5e" % (i, np.sqrt(sscovdata.flatten()[i]))
print ">>>"
U, s, V = np.linalg.svd(rrslope, full_matrices=True)
if verbose:
print ">>> singular values:"
print s
print ">>> Applying cutoff for smallest SVs:"
dtmp = s.max() * SVD_RELCUTOFF
s *= (abs(s) >= dtmp)
sinv = [1. / s[k] if s[k] != 0. else 0. for k in xrange(nref)]
if verbose:
print ">>> singular values (after cutoff):"
print s
stat = np.allclose(rrslope, np.dot(U, np.dot(np.diag(s), V)))
if verbose:
print ">>> Testing svd-result: %s" % stat
if not stat:
print " (Maybe due to SV-cutoff?)"
# Solve for inverse coefficients:
# Set RRinv.tr=U diag(sinv) V
RRinv = np.transpose(np.dot(U, np.dot(np.diag(sinv), V)))
if checkresults:
stat = np.allclose(np.identity(nref), np.dot(RRinv, rrslope))
if stat:
if verbose:
print ">>> Testing RRinv-result (should be unit-matrix): ok"
else:
print ">>> Testing RRinv-result (should be unit-matrix): failed"
print np.transpose(np.dot(RRinv, rrslope))
print ">>>"
if verbose:
print "########## Calc weight matrix..."
# weights-matrix will be somewhat larger than necessary,
# (to simplify indexing in compensation loop):
weights = np.zeros((raw._data.shape[0], nref))
for isig in xrange(nsig):
for iref in xrange(nref):
weights[sigpick[isig],iref] = np.dot(srcovdata[isig,:], RRinv[:,iref])
if verbose:
print "########## Compensating signal channels:"
if complementary_signal:
print ">>> Caveat: REPLACING signal by compensation signal"
tct = time.clock()
twt = time.time()
# Work on entire data stream:
for isl in xrange(raw._data.shape[1]):
slice = np.take(raw._data, [isl], axis=1)
if use_reffilter:
refslice = np.take(fltref._data, [isl], axis=1)
refarr = refslice[:].flatten() - refmean
# refarr = fltres[:,isl]-refmean
else:
refarr = slice[refpick].flatten() - refmean
subrefarr = np.dot(weights[:], refarr)
if not complementary_signal:
raw._data[:, isl] -= subrefarr
else:
raw._data[:, isl] = subrefarr
if (isl % 10000 == 0) and verbose:
print "\rProcessed slice %6d" % isl
if verbose:
print "\nDone."
tc1 = time.clock()
tw1 = time.time()
print ">>> compensation loop took %.1f ms (%.2f s walltime)" % (1000. * (tc1 - tct), (tw1 - twt))
if checkresults:
if verbose:
print "########## Calculating final signal channel covariance:"
# Calculate final signal channel covariance:
# (only used as quality measure)
tct = time.clock()
twt = time.time()
sigmean = 0
sscovdata = 0
n_samples = 0
for first in range(itmin, itmax, itstep):
last = first + itstep
if last >= itmax:
last = itmax
raw_segmentsig, times = raw[sigpick, first:last]
# Artifacts found here will probably differ from pre-noisered artifacts!
if not exclude_artifacts or \
_is_good(raw_segmentsig, infosig['ch_names'], idx_by_typesig, reject,
flat=None, ignore_chs=raw.info['bads']):
sigmean += raw_segmentsig.sum(axis=1)
sscovdata += (raw_segmentsig * raw_segmentsig).sum(axis=1)
n_samples += raw_segmentsig.shape[1]
sigmean /= n_samples
sscovdata -= n_samples * sigmean[:] * sigmean[:]
sscovdata /= (n_samples - 1)
if verbose:
print ">>> no channel got worse: ", np.all(np.less_equal(sscovdata, sscovinit))
print ">>> final rt(avg sig pwr) = %12.5e" % np.sqrt(np.mean(sscovdata))
for i in xrange(5):
print ">>> final signal-rms[%3d] = %12.5e" % (i, np.sqrt(sscovdata.flatten()[i]))
tc1 = time.clock()
tw1 = time.time()
print ">>> signal covar-calc took %.1f ms (%.2f s walltime)" % (1000. * (tc1 - tct), (tw1 - twt))
print ">>>"
#--- fb update 21.07.2015
fname_out = jumeg_base.get_fif_name(raw=raw,postfix=fif_postfix,extention=fif_extention)
if save:
jumeg_base.apply_save_mne_data(raw,fname=fname_out,overwrite=True)
tc1 = time.clock()
tw1 = time.time()
if verbose:
print ">>> Total run took %.1f ms (%.2f s walltime)" % (1000. * (tc1 - tc0), (tw1 - tw0))
return raw,fname_out
