def __init__(self, sf, npts, nbins=18, pha_f=[2, 4], pha_meth='hilbert',
pha_cycle=3, amp_f=[60, 200], amp_meth='hilbert', amp_cycle=6,
window=None, width=None, step=None, time=None,
**kwargs):
# Check pha and amp methods:
_checkref('pha_meth', pha_meth, ['hilbert', 'hilbert1', 'hilbert2'])
_checkref('amp_meth', amp_meth, ['hilbert', 'hilbert1', 'hilbert2'])
# Check the type of f:
if (len(pha_f) == 4) and isinstance(pha_f[0], (int, float)):
pha_f = binarize(
pha_f[0], pha_f[1], pha_f[2], pha_f[3], kind='list')
if (len(amp_f) == 4) and isinstance(amp_f[0], (int, float)):
amp_f = binarize(
amp_f[0], amp_f[1], amp_f[2], amp_f[3], kind='list')
self.xvec = []
# Binarize phase vector :
self._binsize = 360 / nbins
self._phabin = np.arange(0, 360, self._binsize)
self.phabin = np.concatenate((self._phabin[:, np.newaxis],
self._phabin[:, np.newaxis]+self._binsize), axis=1)
# Initialize coupling:
_coupling.__init__(self, pha_f, 'phase', pha_meth, pha_cycle,
amp_f, 'amplitude', amp_meth, amp_cycle,
sf, npts, window, width, step, time, **kwargs)
self._nbins = nbins
def __init__(self, sf, npts, f=(2, 200, 10, 5), baseline=(1, 2), norm=0,
method='hilbert1', window=None, width=None, step=None,
time=None, **kwargs):
_checkref('method', method, ['hilbert', 'hilbert1', 'hilbert2',
'wavelet'])
_spectral.__init__(self, sf, npts, 'power', f, baseline, norm,
method, window, width, step, None, time,
True, **kwargs)
def __init__(self, sf, npts, f=[60, 200], baseline=(1, 2), norm=0,
method='hilbert1', window=None, width=None, step=None,
split=None, time=None, **kwargs):
_checkref('method', method, ['hilbert', 'hilbert1', 'hilbert2',
'wavelet'])
_spectral.__init__(self, sf, npts, 'amplitude', f, baseline, norm,
method, window, width, step, split, time,
False, **kwargs)
def __init__(self,
sf,
npts,
f=[60, 200],
method='hilbert',
window=None,
width=None,
step=None,
time=None,
**kwargs):
_checkref('method', method, ['hilbert', 'hilbert1', 'hilbert2'])
_spectral.__init__(self, sf, npts, 'phase', f, None, 0, method, window,
width, step, None, time, False, **kwargs)
def __init__(self,
sf,
npts,
f=(2, 200, 10, 5),
baseline=(1, 2),
norm=0,
method='hilbert1',
window=None,
width=None,
step=None,
time=None,
**kwargs):
_checkref('method', method,
['hilbert', 'hilbert1', 'hilbert2', 'wavelet'])
_spectral.__init__(self, sf, npts, 'power', f, baseline, norm, method,
window, width, step, None, time, True, **kwargs)
def __init__(self,
sf,
npts,
f=[60, 200],
baseline=None,
norm=None,
method='hilbert1',
window=None,
width=None,
step=None,
split=None,
time=None,
**kwargs):
_checkref('method', method,
['hilbert', 'hilbert1', 'hilbert2', 'wavelet'])
_spectral.__init__(self, sf, npts, 'power', f, baseline, norm, method,
window, width, step, split, time, False, **kwargs)
def __init__(self, sf, npts, pha_f=[2, 4], pha_meth='hilbert',
pha_cycle=3, amp_f=[60, 200], amp_meth='hilbert', amp_cycle=6,
window=None, step=None, width=None, time=None, **kwargs):
# Check pha and amp methods:
_checkref('pha_meth', pha_meth, ['hilbert', 'hilbert1', 'hilbert2'])
_checkref('amp_meth', amp_meth, ['hilbert', 'hilbert1', 'hilbert2'])
# Check the type of f:
if (len(pha_f) == 4) and isinstance(pha_f[0], (int, float)):
pha_f = binarize(
pha_f[0], pha_f[1], pha_f[2], pha_f[3], kind='list')
if (len(amp_f) == 4) and isinstance(amp_f[0], (int, float)):
amp_f = binarize(
amp_f[0], amp_f[1], amp_f[2], amp_f[3], kind='list')
# Initialize cfc :
_coupling.__init__(self, pha_f, 'phase', pha_meth, pha_cycle,
amp_f, 'amplitude', amp_meth, amp_cycle,
sf, npts, window, width, step, time, **kwargs)
def __init__(self, sf, npts, Id='113', pha_f=[2, 4], pha_meth='hilbert',
pha_cycle=3, amp_f=[60, 200], amp_meth='hilbert', amp_cycle=6,
nbins=18, window=None, width=None, step=None, time=None,
**kwargs):
# Check pha and amp methods:
_checkref('pha_meth', pha_meth, ['hilbert', 'hilbert1', 'hilbert2'])
_checkref('amp_meth', amp_meth, ['hilbert', 'hilbert1', 'hilbert2'])
# Check the type of f:
if (len(pha_f) == 4) and isinstance(pha_f[0], (int, float)):
pha_f = binarize(
pha_f[0], pha_f[1], pha_f[2], pha_f[3], kind='list')
if (len(amp_f) == 4) and isinstance(amp_f[0], (int, float)):
amp_f = binarize(
amp_f[0], amp_f[1], amp_f[2], amp_f[3], kind='list')
self.xvec = []
# Initalize pac object :
self.Id = Id
me = Id[0]
# Manage settings :
# 1 - Choose if we extract phase or amplitude :
# - Methods using phase // amplitude :
if me in ['1', '2', '3', '5', '6']:
pha_kind, amp_kind = 'phase', 'amplitude'
# - Methods using phase // phase :
elif me in ['4']:
pha_kind, amp_kind = 'phase', 'amplitude'
# 2 - Specific case of Ozkurt :
if me == '5':
Id = '500'
# Initialize cfc :
_coupling.__init__(self, pha_f, pha_kind, pha_meth, pha_cycle,
amp_f, amp_kind, amp_meth, amp_cycle,
sf, npts, window, width, step, time, **kwargs)
# Get pac model :
_, _, _, ModelStr, SurStr, NormStr = CfcSettings(Id, nbins)
self.model = ['Method : '+ModelStr, 'Surrogates : '+SurStr,
'Normalization : '+NormStr]
self._nbins = nbins
def __init__(self, sf, npts, f=[2, 4], method='hilbert', cycle=3,
sample=None, time=None, **kwargs):
# Check pha and amp methods:
_checkref('pha_meth', method, ['hilbert', 'hilbert1', 'hilbert2'])
# Check the type of f:
if (len(f) == 4) and isinstance(f[0], (int, float)):
f = binarize(f[0], f[1], f[2], f[3], kind='list')
# Initialize PLV :
_coupling.__init__(self, f, 'phase', method, cycle,
f, 'phase', method, cycle,
sf, npts, None, None, None, time, **kwargs)
if time is None:
time = np.arange(npts)
else:
time = time
if sample is None:
sample = slice(npts)
self._sample = sample
self.time = time[sample]
del self.amp
def __init__(self, sf, npts, f=[60, 200], method='hilbert', window=None,
width=None, step=None, time=None, **kwargs):
_checkref('method', method, ['hilbert', 'hilbert1', 'hilbert2'])
_spectral.__init__(self, sf, npts, 'phase', f, None, 0, method,
window, width, step, None, time, False, **kwargs)
def get(self, x, statmeth=None, tail=2, maxstat=-1, metric='m_center',
n_perm=200, n_jobs=-1):
"""Get the spectral informations of the signal x.
Args:
x: array
Data. x should have a shape of
(n_electrodes x n_pts x n_trials)
Kargs:
n_perm: integer, optional, [def: 200]
Number of permutations for assessing statistical significiancy.
statmeth: string, optional, [def: None]
Method to evaluate the statistical significiancy. To get p-values,
the program will compare real values with a defined baseline. As a
consequence, the 'norm' parameter should not be equal to zero.
- None: no statistical evaluation
- 'permutation': randomly shuffle real data with baseline.Control the number of permutations with the n_perm parameter. For example, if n_perm = 1000, this mean that minimum p-valueswill be 0.001.
- 'wilcoxon': Wilcoxon signed-rank test
- 'kruskal': Kruskal-Wallis H-test
tail: int, optional, [def: 1]
For the permutation method, get p-values from one or two tails of
the distribution.
maxtstat: integer, optional, [def -1]
Correct p-values with maximum statistique. maxstat correspond to
the dimension of perm for correction. Use -1 to correct through all
dimensions. Otherwise, use d1, d2, ... or dn to correct through a
specific dimension.
n_perm: integer, optional, [def: 200]
Number of permutations for assessing statistical significiancy.
Return:
xF: array
The un/normalized feature of x, with a shape of
(n_frequency x n_electrodes x n_window x n_trials)
"""
# Get variables :
self._statmeth = statmeth
self._n_perm = n_perm
self._2t = tail
self._mxst = maxstat
self._metric = metric
# Check input size :
if len(x.shape) == 2:
x = x[np.newaxis, ...]
if x.shape[1] != self._npts:
raise ValueError('The second dimension must be '+str(self._npts))
nfeat = x.shape[0]
# Check statistical method :
_checkref('statmeth', statmeth, ['permutation', 'wilcoxon', 'kruskal'])
# run feature computation:
data = Parallel(n_jobs=n_jobs)(
delayed(_get)(x[k, ...], self) for k in range(nfeat))
# xF, pvalues = zip(*data)
# Re-organize data :
xF = np.swapaxes(np.array(data), 0, 1)
# Remove last dimension (for TF):
if self._meanT:
xF = xF[..., 0]
return xF # , np.swapaxes(np.array(pvalues), 0, 1)
def get(self,
x,
statmeth=None,
tail=2,
maxstat=-1,
metric='m_center',
n_perm=200,
n_jobs=-1):
"""Get the spectral informations of the signal x.
Args:
x: array
Data. x should have a shape of
(n_electrodes x n_pts x n_trials)
Kargs:
n_perm: integer, optional, [def: 200]
Number of permutations for assessing statistical significiancy.
statmeth: string, optional, [def: None]
Method to evaluate the statistical significiancy. To get p-values,
the program will compare real values with a defined baseline. As a
consequence, the 'norm' parameter should not be equal to zero.
- None: no statistical evaluation
- 'permutation': randomly shuffle real data with baseline.Control the number of permutations with the n_perm parameter. For example, if n_perm = 1000, this mean that minimum p-valueswill be 0.001.
- 'wilcoxon': Wilcoxon signed-rank test
- 'kruskal': Kruskal-Wallis H-test
tail: int, optional, [def: 1]
For the permutation method, get p-values from one or two tails of
the distribution.
maxtstat: integer, optional, [def -1]
Correct p-values with maximum statistique. maxstat correspond to
the dimension of perm for correction. Use -1 to correct through all
dimensions. Otherwise, use d1, d2, ... or dn to correct through a
specific dimension.
n_perm: integer, optional, [def: 200]
Number of permutations for assessing statistical significiancy.
Return:
xF: array
The un/normalized feature of x, with a shape of
(n_frequency x n_electrodes x n_window x n_trials)
"""
# Get variables :
self._statmeth = statmeth
self._n_perm = n_perm
self._2t = tail
self._mxst = maxstat
self._metric = metric
# Check input size :
if len(x.shape) == 2:
x = x[np.newaxis, ...]
if x.shape[1] != self._npts:
raise ValueError('The second dimension must be ' + str(self._npts))
nfeat = x.shape[0]
# Check statistical method :
_checkref('statmeth', statmeth, ['permutation', 'wilcoxon', 'kruskal'])
# run feature computation:
data = Parallel(n_jobs=n_jobs)(delayed(_get)(x[k, ...], self)
for k in range(nfeat))
# xF, pvalues = zip(*data)
# Re-organize data :
xF = np.swapaxes(np.array(data), 0, 1)
# Remove last dimension (for TF):
if self._meanT:
xF = xF[..., 0]
return xF # , np.swapaxes(np.array(pvalues), 0, 1)
def get(self,
x,
statmeth=None,
tail=2,
n_perm=200,
metric='m_center',
maxstat=False,
n_jobs=-1):
"""Get the spectral feature of the signal x.
Args:
x: array
Data with a shape of (n_electrodes x n_pts x n_trials)
Kargs:
statmeth: string, optional, [def: None]
Method to evaluate the statistical significiancy. To get p-values,
the program will compare real values with a defined baseline. As a
consequence, the 'norm' and 'baseline' parameter should not be None.
- 'permutation': randomly shuffle real data with baseline.Control the number of permutations with the n_perm parameter. For example, if n_perm = 1000, this mean that minimum p-valueswill be 0.001.
- 'wilcoxon': Wilcoxon signed-rank test
- 'kruskal': Kruskal-Wallis H-test
n_perm: integer, optional, [def: 200]
Number of permutations for assessing statistical significiancy.
tail: int, optional, [def: 2]
For the permutation method, get p-values from one or two tails of
the distribution. Use -1 for testing A<B, 1 for A>B and 2 for A~=B.
metric: string/function type, optional, [def: 'm_center']
Use diffrent metrics to normalize data and permutations by the
defined baseline. Use:
- None: compare directly values without transformation
- 'm_center': (A-B)/mean(B) transformation
- 'm_zscore': (A-B)/std(B) transformation
- 'm_minus': (A-B) transformation
- function: user defined function [def myfcn(A, B): return array_like]
maxstat: bool, optional, [def: False]
Correct p-values with maximum statistique. If maxstat is True,
the correction will be applied only trhough frequencies.
n_jobs: integer, optional, [def: -1]
Control the number of jobs to extract features. If
n_jobs = -1, all the jobs are used.
Return:
xF: array
The un/normalized feature of x, with a shape of
(n_frequency x n_electrodes x n_window x n_trials)
pvalues: array
p-values with a shape of (n_frequency x n_electrodes x n_window)
"""
# Get variables :
self._statmeth = statmeth
self._n_perm = n_perm
self._2t = tail
self._mxst = maxstat
self._metric = metric
# Check input size :
if len(x.shape) == 2:
x = x[np.newaxis, ...]
if x.shape[1] != self._npts:
raise ValueError('The second dimension must be ' + str(self._npts))
nfeat = x.shape[0]
warnmsg = 'You define a normalization but no baseline has been' + \
' specified. Normalization will be ignore'
if (self._norm is not None) and (self._baseline is None):
warn(warnmsg)
self._norm = None
# Check statistical method :
if statmeth is not None:
_checkref('statmeth', statmeth,
['permutation', 'wilcoxon', 'kruskal'])
# run feature computation:
data = Parallel(n_jobs=n_jobs)(delayed(_get)(x[k, ...], self)
for k in range(nfeat))
xF, pvalues = zip(*data)
# Re-organize data :
xF = np.swapaxes(np.array(xF), 0, 1)
if pvalues[0] is not None:
pvalues = np.swapaxes(np.array(pvalues), 0, 1)
else:
pvalues = None
# Remove last dimension (for TF):
if self._meanT:
xF = xF[..., 0]
return xF, pvalues
def get(self, x, statmeth=None, tail=2, n_perm=200, metric='m_center',
maxstat=False, n_jobs=-1):
"""Get the spectral feature of the signal x.
Args:
x: array
Data with a shape of (n_electrodes x n_pts x n_trials)
Kargs:
statmeth: string, optional, [def: None]
Method to evaluate the statistical significiancy. To get p-values,
the program will compare real values with a defined baseline. As a
consequence, the 'norm' and 'baseline' parameter should not be None.
- 'permutation': randomly shuffle real data with baseline.Control the number of permutations with the n_perm parameter. For example, if n_perm = 1000, this mean that minimum p-valueswill be 0.001.
- 'wilcoxon': Wilcoxon signed-rank test
- 'kruskal': Kruskal-Wallis H-test
n_perm: integer, optional, [def: 200]
Number of permutations for assessing statistical significiancy.
tail: int, optional, [def: 2]
For the permutation method, get p-values from one or two tails of
the distribution. Use -1 for testing A<B, 1 for A>B and 2 for A~=B.
metric: string/function type, optional, [def: 'm_center']
Use diffrent metrics to normalize data and permutations by the
defined baseline. Use:
- None: compare directly values without transformation
- 'm_center': (A-B)/mean(B) transformation
- 'm_zscore': (A-B)/std(B) transformation
- 'm_minus': (A-B) transformation
- function: user defined function [def myfcn(A, B): return array_like]
maxstat: bool, optional, [def: False]
Correct p-values with maximum statistique. If maxstat is True,
the correction will be applied only trhough frequencies.
n_jobs: integer, optional, [def: -1]
Control the number of jobs to extract features. If
n_jobs = -1, all the jobs are used.
Return:
xF: array
The un/normalized feature of x, with a shape of
(n_frequency x n_electrodes x n_window x n_trials)
pvalues: array
p-values with a shape of (n_frequency x n_electrodes x n_window)
"""
# Get variables :
self._statmeth = statmeth
self._n_perm = n_perm
self._2t = tail
self._mxst = maxstat
self._metric = metric
# Check input size :
if len(x.shape) == 2:
x = x[np.newaxis, ...]
if x.shape[1] != self._npts:
raise ValueError('The second dimension must be '+str(self._npts))
nfeat = x.shape[0]
warnmsg = 'You define a normalization but no baseline has been' + \
' specified. Normalization will be ignore'
if (self._norm is not None) and (self._baseline is None):
warn(warnmsg)
self._norm = None
# Check statistical method :
if statmeth is not None:
_checkref('statmeth', statmeth, ['permutation', 'wilcoxon', 'kruskal'])
# run feature computation:
data = Parallel(n_jobs=n_jobs)(
delayed(_get)(x[k, ...], self) for k in range(nfeat))
xF, pvalues = zip(*data)
# Re-organize data :
xF = np.swapaxes(np.array(xF), 0, 1)
if pvalues[0] is not None:
pvalues = np.swapaxes(np.array(pvalues), 0, 1)
else:
pvalues = None
# Remove last dimension (for TF):
if self._meanT:
xF = xF[..., 0]
return xF, pvalues
