def test_cubic_errors(self):
# Check error ouputs for mis-settings of the parameters
# Empty signal
self.assertRaises(
ValueError, cubic.cubic,
neo.AnalogSignalArray([] * pq.dimensionless,
sampling_period=10 * pq.ms))
# Multidimensional array
self.assertRaises(
ValueError, cubic.cubic,
neo.AnalogSignalArray([[1, 2, 3], [1, 2, 3]] * pq.dimensionless,
sampling_period=10 * pq.ms))
self.assertRaises(ValueError, cubic.cubic,
numpy.array([[1, 2, 3], [1, 2, 3]]))
# Negative alpha
self.assertRaises(ValueError, cubic.cubic, self.data_array, alpha=-0.1)
# Negative number of iterations ximax
self.assertRaises(ValueError, cubic.cubic, self.data_array, ximax=-100)
# Checking case in which the second cumulant of the signal is smaller
# than the first cumulant (analitycal constrain of the method)
self.assertRaises(ValueError,
cubic.cubic,
neo.AnalogSignalArray(numpy.array(
[1] * 1000).reshape(1000, 1),
units=pq.dimensionless,
sampling_period=10 * pq.ms),
alpha=self.alpha)
def setUp(self):
# Generate test data of a harmonic function over a long time
time = np.arange(0, 1000, 0.1) * pq.ms
freq = 10 * pq.Hz
self.amplitude = np.array([
np.linspace(1, 10, len(time)),
np.linspace(1, 10, len(time)),
np.ones((len(time))),
np.ones((len(time))) * 10.
]).T
self.phase = np.array([
(time * freq).simplified.magnitude * 2. * np.pi,
(time * freq).simplified.magnitude * 2. * np.pi + np.pi / 2,
(time * freq).simplified.magnitude * 2. * np.pi + np.pi,
(time * freq).simplified.magnitude * 2. * 2. * np.pi
]).T
self.phase = np.mod(self.phase + np.pi, 2. * np.pi) - np.pi
# rising amplitude cosine, random ampl. sine, flat inverse cosine,
# flat cosine at double frequency
sigs = np.vstack([
self.amplitude[:, 0] * np.cos(self.phase[:, 0]),
self.amplitude[:, 1] * np.cos(self.phase[:, 1]),
self.amplitude[:, 2] * np.cos(self.phase[:, 2]),
self.amplitude[:, 3] * np.cos(self.phase[:, 3])
])
self.long_signals = neo.AnalogSignalArray(
sigs.T,
units='mV',
t_start=0. * pq.ms,
sampling_rate=(len(time) / (time[-1] - time[0])).rescale(pq.Hz),
dtype=float)
# Generate test data covering a single oscillation cycle in 1s only
phases = np.arange(0, 2 * np.pi, np.pi / 256)
sigs = np.vstack([
np.sin(phases),
np.cos(phases),
np.sin(2 * phases),
np.cos(2 * phases)
])
self.one_period = neo.AnalogSignalArray(sigs.T,
units=pq.mV,
sampling_rate=len(phases) *
pq.Hz)
def test_butter_filter_function(self):
# generate white noise AnalogSignalArray
noise = neo.AnalogSignalArray(np.random.normal(size=5000),
sampling_rate=1000 * pq.Hz,
units='mV')
# test if the filter performance is as well with filftunc=lfilter as
# with filtfunc=filtfilt (i.e. default option)
kwds = {
'signal': noise,
'highpass_freq': 250.0 * pq.Hz,
'lowpass_freq': None,
'filter_function': 'filtfilt'
}
filtered_noise = elephant.signal_processing.butter(**kwds)
_, psd_filtfilt = spsig.welch(filtered_noise,
nperseg=1024,
fs=1000.0,
detrend=lambda x: x)
kwds['filter_function'] = 'lfilter'
filtered_noise = elephant.signal_processing.butter(**kwds)
_, psd_lfilter = spsig.welch(filtered_noise,
nperseg=1024,
fs=1000.0,
detrend=lambda x: x)
self.assertAlmostEqual(psd_filtfilt[0], psd_lfilter[0])
def test_butter_input_types(self):
# generate white noise data of different types
noise_np = np.random.normal(size=5000)
noise_pq = noise_np * pq.mV
noise = neo.AnalogSignalArray(noise_pq, sampling_rate=1000.0 * pq.Hz)
# check input as NumPy ndarray
filtered_noise_np = elephant.signal_processing.butter(noise_np,
400.0,
100.0,
fs=1000.0)
self.assertTrue(isinstance(filtered_noise_np, np.ndarray))
self.assertFalse(isinstance(filtered_noise_np, pq.quantity.Quantity))
self.assertFalse(isinstance(filtered_noise_np, neo.AnalogSignalArray))
self.assertEqual(filtered_noise_np.shape, noise_np.shape)
# check input as Quantity array
filtered_noise_pq = elephant.signal_processing.butter(noise_pq,
400.0 * pq.Hz,
100.0 * pq.Hz,
fs=1000.0)
self.assertTrue(isinstance(filtered_noise_pq, pq.quantity.Quantity))
self.assertFalse(isinstance(filtered_noise_pq, neo.AnalogSignalArray))
self.assertEqual(filtered_noise_pq.shape, noise_pq.shape)
# check input as neo AnalogSignalArray
filtered_noise = elephant.signal_processing.butter(
noise, 400.0 * pq.Hz, 100.0 * pq.Hz)
self.assertTrue(isinstance(filtered_noise, neo.AnalogSignalArray))
self.assertEqual(filtered_noise.shape, noise.shape)
# check if the results from different input types are identical
self.assertTrue(
np.all(filtered_noise_pq.magnitude == filtered_noise_np))
self.assertTrue(np.all(filtered_noise.magnitude == filtered_noise_np))
def get_weights(self):
return neo.AnalogSignalArray(self._weights,
units='nA',
sampling_period=self.interval * ms,
channel_index=numpy.arange(
len(self._weights[0])),
name="weight")
def _get_current_segment(self,
filter_ids=None,
variables='all',
clear=False):
segment = neo.Segment(
name="segment%03d" % self._simulator.state.segment_counter,
description=self.population.describe(),
rec_datetime=datetime.now()
) # would be nice to get the time at the start of the recording, not the end
variables_to_include = set(self.recorded.keys())
if variables is not 'all':
variables_to_include = variables_to_include.intersection(
set(variables))
for variable in variables_to_include:
if variable == 'spikes':
t_stop = self._simulator.state.t * pq.ms # must run on all MPI nodes
segment.spiketrains = [
neo.SpikeTrain(
self._get_spiketimes(id),
t_start=self._recording_start_time,
t_stop=t_stop,
units='ms',
source_population=self.population.label,
source_id=int(id),
source_index=self.population.id_to_index(id))
for id in sorted(self.filter_recorded(
'spikes', filter_ids))
]
else:
ids = sorted(self.filter_recorded(variable, filter_ids))
signal_array = self._get_all_signals(variable,
ids,
clear=clear)
t_start = self._recording_start_time
sampling_period = self.sampling_interval * pq.ms
current_time = self._simulator.state.t * pq.ms
mpi_node = self._simulator.state.mpi_rank # for debugging
if signal_array.size > 0: # may be empty if none of the recorded cells are on this MPI node
channel_indices = numpy.array(
[self.population.id_to_index(id) for id in ids])
units = self.population.find_units(variable)
source_ids = numpy.fromiter(ids, dtype=int)
segment.analogsignalarrays.append(
neo.AnalogSignalArray(
signal_array,
units=units,
t_start=t_start,
sampling_period=sampling_period,
name=variable,
source_population=self.population.label,
channel_index=channel_indices,
source_ids=source_ids))
logger.debug("%d **** ids=%s, channels=%s", mpi_node,
source_ids, channel_indices)
assert segment.analogsignalarrays[
0].t_stop - current_time < 2 * sampling_period
# need to add `Unit` and `RecordingChannelGroup` objects
return segment
def test_butter_multidim_input(self):
noise_pq = np.random.normal(size=(4, 5000)) * pq.mV
noise_neo = neo.AnalogSignalArray(noise_pq.T,
sampling_rate=1000.0 * pq.Hz)
noise_neo1d = neo.AnalogSignalArray(noise_pq[0],
sampling_rate=1000.0 * pq.Hz)
filtered_noise_pq = elephant.signal_processing.butter(noise_pq,
250.0,
fs=1000.0)
filtered_noise_neo = elephant.signal_processing.butter(
noise_neo, 250.0)
filtered_noise_neo1d = elephant.signal_processing.butter(
noise_neo1d, 250.0)
self.assertTrue(
np.all(
filtered_noise_pq.magnitude == filtered_noise_neo.T.magnitude))
self.assertTrue(
np.all(filtered_noise_neo1d.magnitude ==
filtered_noise_neo.T.magnitude[0]))
def test_zscore_list_inplace(self):
'''
Test zscore on a list of AnalogSignalArray objects, asking for an
inplace operation.
'''
signal1 = neo.AnalogSignalArray(np.transpose(
np.vstack([self.test_seq1, self.test_seq1])),
units='mV',
t_start=0. * pq.ms,
sampling_rate=1000. * pq.Hz,
dtype=float)
signal2 = neo.AnalogSignalArray(np.transpose(
np.vstack([self.test_seq1, self.test_seq2])),
units='mV',
t_start=0. * pq.ms,
sampling_rate=1000. * pq.Hz,
dtype=float)
signal_list = [signal1, signal2]
m = np.mean(np.hstack([self.test_seq1, self.test_seq1]))
s = np.std(np.hstack([self.test_seq1, self.test_seq1]))
target11 = (self.test_seq1 - m) / s
target21 = (self.test_seq1 - m) / s
m = np.mean(np.hstack([self.test_seq1, self.test_seq2]))
s = np.std(np.hstack([self.test_seq1, self.test_seq2]))
target12 = (self.test_seq1 - m) / s
target22 = (self.test_seq2 - m) / s
# Call elephant function
result = elephant.signal_processing.zscore(signal_list, inplace=True)
assert_array_almost_equal(result[0].magnitude,
np.transpose(np.vstack([target11,
target12])),
decimal=9)
assert_array_almost_equal(result[1].magnitude,
np.transpose(np.vstack([target21,
target22])),
decimal=9)
# Assert original signal is overwritten
self.assertEqual(signal1[0, 0].magnitude, target11[0])
self.assertEqual(signal2[0, 0].magnitude, target21[0])
def complexity_pdf(spiketrains, binsize):
"""
Complexity Distribution [1] of a list of :attr:`neo.SpikeTrain` objects.
Probability density computed from the complexity histogram which is the
histogram of the entries of the population histogram of clipped (binary)
spike trains computed with a bin width of binsize.
It provides for each complexity (== number of active neurons per bin) the
number of occurrences. The normalization of that histogram to 1 is the
probability density.
Parameters
----------
spiketrains : List of neo.SpikeTrain objects
Spiketrains with a common time axis (same `t_start` and `t_stop`)
binsize : quantities.Quantity
Width of the histogram's time bins.
Returns
-------
time_hist : neo.AnalogSignalArray
A neo.AnalogSignalArray object containing the histogram values.
`AnalogSignal[j]` is the histogram computed between .
See also
--------
elephant.conversion.BinnedSpikeTrain
References
----------
[1]Gruen, S., Abeles, M., & Diesmann, M. (2008). Impact of higher-order
correlations on coincidence distributions of massively parallel data.
In Dynamic Brain-from Neural Spikes to Behaviors (pp. 96-114).
Springer Berlin Heidelberg.
"""
# Computing the population histogram with parameter binary=True to clip the
# spike trains before summing
pophist = time_histogram(spiketrains, binsize, binary=True)
# Computing the histogram of the entries of pophist (=Complexity histogram)
complexity_hist = np.histogram(pophist.magnitude,
bins=range(0,
len(spiketrains) + 2))[0]
# Normalization of the Complexity Histogram to 1 (probabilty distribution)
complexity_hist = complexity_hist / complexity_hist.sum()
# Convert the Complexity pdf to an neo.AnalogSignalArray
complexity_distribution = neo.AnalogSignalArray(
np.array(complexity_hist).reshape(len(complexity_hist), 1) *
pq.dimensionless,
t_start=0 * pq.dimensionless,
sampling_period=1 * pq.dimensionless)
return complexity_distribution
def setUp(self):
n2 = 300
n0 = 100000 - n2
self.xi = 10
self.data_signal = neo.AnalogSignalArray(
numpy.array([self.xi] * n2 + [0] * n0).reshape(n0 + n2, 1) *
pq.dimensionless,
sampling_period=1 * pq.s)
self.data_array = numpy.array([self.xi] * n2 + [0] * n0)
self.alpha = 0.05
self.ximax = 10
def test_butter_missing_cutoff_freqs(self):
# generate a dummy AnalogSignalArray
anasig_dummy = neo.AnalogSignalArray(np.zeros(5000),
sampling_rate=1000 * pq.Hz,
units='mV')
# test a case where no cut-off frequencies are given
kwds = {
'signal': anasig_dummy,
'highpass_freq': None,
'lowpass_freq': None
}
self.assertRaises(ValueError, elephant.signal_processing.butter,
**kwds)
def test_butter_invalid_filter_function(self):
# generate a dummy AnalogSignalArray
anasig_dummy = neo.AnalogSignalArray(np.zeros(5000),
sampling_rate=1000 * pq.Hz,
units='mV')
# test exception upon invalid filtfunc string
kwds = {
'signal': anasig_dummy,
'highpass_freq': 250.0 * pq.Hz,
'filter_function': 'invalid_filter'
}
self.assertRaises(ValueError, elephant.signal_processing.butter,
**kwds)
def segment_from_recording_device(devices,
variables_to_include,
id_lists,
t_stop,
name="segment00"):
"""
Extract data from a NEST recording device and return it as a Neo Segment object.
"""
def get_data(device, variable, id_list):
events = nest.GetStatus(device, 'events')[0]
ids = events['senders']
values = events[variable]
data = {}
for id in id_list:
data[id] = values[ids == id]
assert len(data) > 0
return data
segment = neo.Segment(name=name, rec_datetime=datetime.now())
for device, variable, id_list in zip(devices, variables_to_include,
id_lists):
print(name, device, variable)
data = get_data(device, variable, id_list)
if variable == 'times':
print(" adding spiketrain")
id0 = min(id_list)
segment.spiketrains = [
neo.SpikeTrain(spiketrain,
t_start=0.0,
t_stop=t_stop,
units='ms',
source_id=id,
source_index=id - id0)
for id, spiketrain in data.items()
]
else:
print(" adding signal")
source_ids = np.array(id_list)
channel_indices = source_ids - source_ids.min()
signal_array = np.vstack(data.values()).T
segment.analogsignalarrays = [
neo.AnalogSignalArray(signal_array,
units='mV',
t_start=0.0 * ms,
sampling_period=0.1 * ms,
name=variable,
channel_index=channel_indices,
source_ids=source_ids)
]
return segment
def test_butter_filter_type(self):
"""
Test if correct type of filtering is performed according to how cut-off
frequencies are given
"""
# generate white noise AnalogSignalArray
noise = neo.AnalogSignalArray(np.random.normal(size=5000),
sampling_rate=1000 * pq.Hz,
units='mV')
# test high-pass filtering: power at the lowest frequency
# should be almost zero
# Note: the default detrend function of scipy.signal.welch() seems to
# cause artificial finite power at the lowest frequencies. Here I avoid
# this by using an identity function for detrending
filtered_noise = elephant.signal_processing.butter(
noise, 250.0 * pq.Hz, None)
_, psd = spsig.welch(filtered_noise,
nperseg=1024,
fs=1000.0,
detrend=lambda x: x)
self.assertAlmostEqual(psd[0], 0)
# test low-pass filtering: power at the highest frequency
# should be almost zero
filtered_noise = elephant.signal_processing.butter(
noise, None, 250.0 * pq.Hz)
_, psd = spsig.welch(filtered_noise, nperseg=1024, fs=1000.0)
self.assertAlmostEqual(psd[-1], 0)
# test band-pass filtering: power at the lowest and highest frequencies
# should be almost zero
filtered_noise = elephant.signal_processing.butter(
noise, 200.0 * pq.Hz, 300.0 * pq.Hz)
_, psd = spsig.welch(filtered_noise,
nperseg=1024,
fs=1000.0,
detrend=lambda x: x)
self.assertAlmostEqual(psd[0], 0)
self.assertAlmostEqual(psd[-1], 0)
# test band-stop filtering: power at the intermediate frequency
# should be almost zero
filtered_noise = elephant.signal_processing.butter(
noise, 400.0 * pq.Hz, 100.0 * pq.Hz)
_, psd = spsig.welch(filtered_noise, nperseg=1024, fs=1000.0)
self.assertAlmostEqual(psd[256], 0)
def test_zscore_single_inplace_int(self):
'''
Test if the z-score is correctly calculated even if the input is an
AnalogSignalArray of type int, asking for an inplace operation.
'''
signal = neo.AnalogSignalArray(self.test_seq1,
units='mV',
t_start=0. * pq.ms,
sampling_rate=1000. * pq.Hz,
dtype=int)
m = np.mean(self.test_seq1)
s = np.std(self.test_seq1)
target = (self.test_seq1 - m) / s
assert_array_almost_equal(elephant.signal_processing.zscore(
signal, inplace=True).magnitude,
target.astype(int),
decimal=9)
# Assert original signal is overwritten
self.assertEqual(signal[0].magnitude, target.astype(int)[0])
def test_zscore_single_dup_int(self):
'''
Test if the z-score is correctly calculated even if the input is an
AnalogSignalArray of type int, asking for a duplicate (duplicate should
be of type float).
'''
signal = neo.AnalogSignalArray(self.test_seq1,
units='mV',
t_start=0. * pq.ms,
sampling_rate=1000. * pq.Hz,
dtype=int)
m = np.mean(self.test_seq1)
s = np.std(self.test_seq1)
target = (self.test_seq1 - m) / s
assert_array_almost_equal(elephant.signal_processing.zscore(
signal, inplace=False).magnitude,
target,
decimal=9)
# Assert original signal is untouched
self.assertEqual(signal.magnitude[0], self.test_seq1[0])
def test_zscore_single_dup(self):
'''
Test z-score on a single AnalogSignalArray, asking to return a
duplicate.
'''
signal = neo.AnalogSignalArray(self.test_seq1,
units='mV',
t_start=0. * pq.ms,
sampling_rate=1000. * pq.Hz,
dtype=float)
m = np.mean(self.test_seq1)
s = np.std(self.test_seq1)
target = (self.test_seq1 - m) / s
assert_array_equal(target, scipy.stats.zscore(self.test_seq1))
result = elephant.signal_processing.zscore(signal, inplace=False)
assert_array_almost_equal(result.magnitude, target, decimal=9)
self.assertEqual(result.units, pq.Quantity(1. * pq.dimensionless))
# Assert original signal is untouched
self.assertEqual(signal[0].magnitude, self.test_seq1[0])
def test_zscore_single_multidim_dup(self):
'''
Test z-score on a single AnalogSignal with multiple dimensions, asking
to return a duplicate.
'''
signal = neo.AnalogSignalArray(np.transpose(
np.vstack([self.test_seq1, self.test_seq2])),
units='mV',
t_start=0. * pq.ms,
sampling_rate=1000. * pq.Hz,
dtype=float)
m = np.mean(signal.magnitude, axis=0, keepdims=True)
s = np.std(signal.magnitude, axis=0, keepdims=True)
target = (signal.magnitude - m) / s
assert_array_almost_equal(elephant.signal_processing.zscore(
signal, inplace=False).magnitude,
target,
decimal=9)
# Assert original signal is untouched
self.assertEqual(signal[0, 0].magnitude, self.test_seq1[0])
def test_zscore_single_inplace(self):
'''
Test z-score on a single AnalogSignalArray, asking for an inplace
operation.
'''
signal = neo.AnalogSignalArray(self.test_seq1,
units='mV',
t_start=0. * pq.ms,
sampling_rate=1000. * pq.Hz,
dtype=float)
m = np.mean(self.test_seq1)
s = np.std(self.test_seq1)
target = (self.test_seq1 - m) / s
result = elephant.signal_processing.zscore(signal, inplace=True)
assert_array_almost_equal(result.magnitude, target, decimal=9)
self.assertEqual(result.units, pq.Quantity(1. * pq.dimensionless))
# Assert original signal is overwritten
self.assertEqual(signal[0].magnitude, target[0])
def _cch_memory(st_1, st_2, win, mode, norm, border_corr, binary, kern):
# Check that the spike trains are binned with the saem temporal
# resolution
if not st1.matrix_rows == 1:
raise AssertionError("Spike train must be one dimensional")
if not st2.matrix_rows == 1:
raise AssertionError("Spike train must be one dimensional")
if not st1.binsize == st2.binsize:
raise AssertionError("Bin sizes must be equal")
# Retrieve unclipped matrix
st1_spmat = st_1.to_sparse_array()
st2_spmat = st_2.to_sparse_array()
binsize = st_1.binsize
max_num_bins = max(st_1.num_bins, st_2.num_bins)
# Set the time window in which is computed the cch
if win is not None:
# Window parameter given in number of bins (integer)
if isinstance(win[0], int) and isinstance(win[1], int):
# Check the window parameter values
if win[0] >= win[1] or win[0] <= -max_num_bins \
or win[1] >= max_num_bins:
raise ValueError(
"The window exceeds the length of the spike trains")
# Assign left and right edges of the cch
l, r = win[0], win[1]
# Window parameter given in time units
else:
# Check the window parameter values
if win[0].rescale(binsize.units).magnitude % \
binsize.magnitude != 0 or win[1].rescale(
binsize.units).magnitude % binsize.magnitude != 0:
raise ValueError(
"The window has to be a multiple of the binsize")
if win[0] >= win[1] or win[0] <= -max_num_bins * binsize \
or win[1] >= max_num_bins * binsize:
raise ValueError("The window exceeds the length of the"
" spike trains")
# Assign left and right edges of the cch
l, r = int(win[0].rescale(binsize.units) / binsize), int(
win[1].rescale(binsize.units) / binsize)
# Case without explicit window parameter
else:
# cch computed for all the possible entries
if mode == 'full':
# Assign left and right edges of the cch
r = st_2.num_bins - 1
l = -st_1.num_bins + 1
# cch compute only for the entries that completely overlap
elif mode == 'valid':
# Assign left and right edges of the cch
r = max(st_2.num_bins - st_1.num_bins, 0)
l = min(st_2.num_bins - st_1.num_bins, 0)
# Check the mode parameter
else:
raise KeyError("The possible entries for mode parameter are" +
"'full' and 'valid'")
# For each row, extract the nonzero column indices
# and the corresponding # data in the matrix (for performance reasons)
st1_bin_idx_unique = st1_spmat.nonzero()[1]
st2_bin_idx_unique = st2_spmat.nonzero()[1]
# Case with binary entries
if binary:
st1_bin_counts_unique = np.array(st1_spmat.data > 0, dtype=int)
st2_bin_counts_unique = np.array(st2_spmat.data > 0, dtype=int)
# Case with all values
else:
st1_bin_counts_unique = st1_spmat.data
st2_bin_counts_unique = st2_spmat.data
# Initialize the counts to an array of zeroes,
# and the bin IDs to integers
# spanning the time axis
counts = np.zeros(np.abs(l) + np.abs(r) + 1)
bin_ids = np.arange(l, r + 1)
# Compute the CCH at lags in l,...,r only
for idx, i in enumerate(st1_bin_idx_unique):
timediff = st2_bin_idx_unique - i
timediff_in_range = np.all([timediff >= l, timediff <= r], axis=0)
timediff = (timediff[timediff_in_range]).reshape((-1, ))
counts[timediff + np.abs(l)] += st1_bin_counts_unique[idx] * \
st2_bin_counts_unique[timediff_in_range]
# Correct the values taking into account lacking contributes
# at the edges
if border_corr is True:
correction = float(max_num_bins + 1) / np.array(
max_num_bins + 1 - abs(np.arange(l, r + 1)), float)
counts = counts * correction
# Define the kern for smoothing as an ndarray
if hasattr(kern, '__iter__'):
if len(kern) > np.abs(l) + np.abs(r) + 1:
raise ValueError(
'The length of the kernel cannot be larger than the '
'length %d of the resulting CCH.' %
(np.abs(l) + np.abs(r) + 1))
kern = np.array(kern, dtype=float)
kern = 1. * kern / sum(kern)
# Check kern parameter
elif kern is not None:
raise ValueError('Invalid smoothing kernel.')
# Smooth the cross-correlation histogram with the kern
if kern is not None:
counts = np.convolve(counts, kern, mode='same')
# Rescale the histogram so that the central bin has height 1,
# if requested
if norm and l <= 0 <= r:
if counts[np.abs(l)] != 0:
counts = counts / counts[np.abs(l)]
else:
warnings.warn('CCH not normalized because no value for 0 lag')
# Transform the array count into an AnalogSignalArray
cch_result = neo.AnalogSignalArray(
signal=counts.reshape(counts.size, 1),
units=pq.dimensionless,
t_start=(bin_ids[0] - 0.5) * st_1.binsize,
sampling_period=st_1.binsize)
# Return only the hist_bins bins and counts before and after the
# central one
return cch_result, bin_ids
def instantaneous_rate(spiketrain,
sampling_period,
kernel='auto',
cutoff=5.0,
t_start=None,
t_stop=None,
trim=False):
"""
Estimates instantaneous firing rate by kernel convolution.
Parameters
-----------
spiketrain : 'neo.SpikeTrain'
Neo object that contains spike times, the unit of the time stamps
and t_start and t_stop of the spike train.
sampling_period : Time Quantity
Time stamp resolution of the spike times. The same resolution will
be assumed for the kernel
kernel : string 'auto' or callable object of :class:`Kernel` from module
'kernels.py'. Currently implemented kernel forms are rectangular,
triangular, epanechnikovlike, gaussian, laplacian, exponential,
and alpha function.
Example: kernel = kernels.RectangularKernel(sigma=10*ms, invert=False)
The kernel is used for convolution with the spike train and its
standard deviation determines the time resolution of the instantaneous
rate estimation.
Default: 'auto'. In this case, the optimized kernel width for the
rate estimation is calculated according to [1] and with this width
a gaussian kernel is constructed. Automatized calculation of the
kernel width is not available for other than gaussian kernel shapes.
cutoff : float
This factor determines the cutoff of the probability distribution of
the kernel, i.e., the considered width of the kernel in terms of
multiples of the standard deviation sigma.
Default: 5.0
t_start : Time Quantity (optional)
Start time of the interval used to compute the firing rate. If None
assumed equal to spiketrain.t_start
Default: None
t_stop : Time Quantity (optional)
End time of the interval used to compute the firing rate (included).
If None assumed equal to spiketrain.t_stop
Default: None
trim : bool
if False, the output of the Fast Fourier Transformation being a longer
vector than the input vector by the size of the kernel is reduced back
to the original size of the considered time interval of the spiketrain
using the median of the kernel.
if True, only the region of the convolved signal is returned, where
there is complete overlap between kernel and spike train. This is
achieved by reducing the length of the output of the Fast Fourier
Transformation by a total of two times the size of the kernel, and
t_start and t_stop are adjusted.
Default: False
Returns
-------
rate : neo.AnalogSignalArray
Contains the rate estimation in unit hertz (Hz).
Has a property 'rate.times' which contains the time axis of the rate
estimate. The unit of this property is the same as the resolution that
is given via the argument 'sampling_period' to the function.
Raises
------
TypeError:
If `spiketrain` is not an instance of :class:`SpikeTrain` of Neo.
If `sampling_period` is not a time quantity.
If `kernel` is neither instance of :class:`Kernel` or string 'auto'.
If `cutoff` is neither float nor int.
If `t_start` and `t_stop` are neither None nor a time quantity.
If `trim` is not bool.
ValueError:
If `sampling_period` is smaller than zero.
Example
--------
kernel = kernels.AlphaKernel(sigma = 0.05*s, invert = True)
rate = instantaneous_rate(spiketrain, sampling_period = 2*ms, kernel)
References
----------
..[1] H. Shimazaki, S. Shinomoto, J Comput Neurosci (2010) 29:171–182.
"""
# Checks of input variables:
if not isinstance(spiketrain, SpikeTrain):
raise TypeError(
"spiketrain must be instance of :class:`SpikeTrain` of Neo!\n"
" Found: %s, value %s" % (type(spiketrain), str(spiketrain)))
if not (isinstance(sampling_period, pq.Quantity)
and sampling_period.dimensionality.simplified == pq.Quantity(
1, "s").dimensionality):
raise TypeError("The sampling period must be a time quantity!\n"
" Found: %s, value %s" %
(type(sampling_period), str(sampling_period)))
if sampling_period.magnitude < 0:
raise ValueError("The sampling period must be larger than zero.")
if kernel == 'auto':
kernel_width = sskernel(spiketrain.magnitude, tin=None,
bootstrap=True)['optw']
unit = spiketrain.units
sigma = 1 / (2.0 * 2.7) * kernel_width * unit
# factor 2.0 connects kernel width with its half width,
# factor 2.7 connects half width of Gaussian distribution with
# 99% probability mass with its standard deviation.
kernel = kernels.GaussianKernel(sigma)
elif not isinstance(kernel, kernels.Kernel):
raise TypeError("kernel must be either instance of :class:`Kernel` "
"or the string 'auto'!\n"
" Found: %s, value %s" %
(type(kernel), str(kernel)))
if not (isinstance(cutoff, float) or isinstance(cutoff, int)):
raise TypeError("cutoff must be float or integer!")
if not (t_start is None or (isinstance(t_start, pq.Quantity)
and t_start.dimensionality.simplified
== pq.Quantity(1, "s").dimensionality)):
raise TypeError("t_start must be a time quantity!")
if not (t_stop is None or (isinstance(t_stop, pq.Quantity)
and t_stop.dimensionality.simplified
== pq.Quantity(1, "s").dimensionality)):
raise TypeError("t_stop must be a time quantity!")
if not (isinstance(trim, bool)):
raise TypeError("trim must be bool!")
# main function:
units = pq.CompoundUnit("%s*s" %
str(sampling_period.rescale('s').magnitude))
spiketrain = spiketrain.rescale(units)
if t_start is None:
t_start = spiketrain.t_start
else:
t_start = t_start.rescale(spiketrain.units)
if t_stop is None:
t_stop = spiketrain.t_stop
else:
t_stop = t_stop.rescale(spiketrain.units)
time_vector = np.zeros(int((t_stop - t_start)) + 1)
spikes_slice = spiketrain.time_slice(t_start, t_stop) \
if len(spiketrain) else np.array([])
for spike in spikes_slice:
index = int((spike - t_start))
time_vector[index] += 1
if cutoff < kernel.min_cutoff:
cutoff = kernel.min_cutoff
warnings.warn("The width of the kernel was adjusted to a minimally "
"allowed width.")
t_arr = np.arange(
-cutoff * kernel.sigma.rescale(units).magnitude,
cutoff * kernel.sigma.rescale(units).magnitude +
sampling_period.rescale(units).magnitude,
sampling_period.rescale(units).magnitude) * units
r = scipy.signal.fftconvolve(time_vector,
kernel(t_arr).rescale(pq.Hz).magnitude,
'full')
if np.any(r < 0):
warnings.warn("Instantaneous firing rate approximation contains "
"negative values, possibly caused due to machine "
"precision errors.")
if not trim:
r = r[kernel.median_index(t_arr):-(kernel(t_arr).size -
kernel.median_index(t_arr))]
elif trim:
r = r[2 * kernel.median_index(t_arr):-2 *
(kernel(t_arr).size - kernel.median_index(t_arr))]
t_start += kernel.median_index(t_arr) * spiketrain.units
t_stop -= (kernel(t_arr).size -
kernel.median_index(t_arr)) * spiketrain.units
rate = neo.AnalogSignalArray(signal=r.reshape(r.size, 1),
sampling_period=sampling_period,
units=pq.Hz,
t_start=t_start,
t_stop=t_stop)
return rate
def time_histogram(spiketrains, binsize, t_start=None, t_stop=None,
output='counts', binary=False):
"""
Time Histogram of a list of :attr:`neo.SpikeTrain` objects.
Parameters
----------
spiketrains : List of neo.SpikeTrain objects
Spiketrains with a common time axis (same `t_start` and `t_stop`)
binsize : quantities.Quantity
Width of the histogram's time bins.
t_start, t_stop : Quantity (optional)
Start and stop time of the histogram. Only events in the input
`spiketrains` falling between `t_start` and `t_stop` (both included)
are considered in the histogram. If `t_start` and/or `t_stop` are not
specified, the maximum `t_start` of all :attr:spiketrains is used as
`t_start`, and the minimum `t_stop` is used as `t_stop`.
Default: t_start = t_stop = None
output : str (optional)
Normalization of the histogram. Can be one of:
* `counts`'`: spike counts at each bin (as integer numbers)
* `mean`: mean spike counts per spike train
* `rate`: mean spike rate per spike train. Like 'mean', but the
counts are additionally normalized by the bin width.
binary : bool (optional)
If **True**, indicates whether all spiketrain objects should first
binned to a binary representation (using the `BinnedSpikeTrain` class
in the `conversion` module) and the calculation of the histogram is
based on this representation.
Note that the output is not binary, but a histogram of the converted,
binary representation.
Default: False
Returns
-------
time_hist : neo.AnalogSignalArray
A neo.AnalogSignalArray object containing the histogram values.
`AnalogSignal[j]` is the histogram computed between
`t_start + j * binsize` and `t_start + (j + 1) * binsize`.
See also
--------
elephant.conversion.BinnedSpikeTrain
"""
min_tstop = 0
if t_start is None:
# Find the internal range for t_start, where all spike trains are
# defined; cut all spike trains taking that time range only
max_tstart, min_tstop = _get_start_stop_from_input(spiketrains)
t_start = max_tstart
if not all([max_tstart == t.t_start for t in spiketrains]):
warnings.warn(
"Spiketrains have different t_start values -- "
"using maximum t_start as t_start.")
if t_stop is None:
# Find the internal range for t_stop
if min_tstop:
t_stop = min_tstop
if not all([min_tstop == t.t_stop for t in spiketrains]):
warnings.warn(
"Spiketrains have different t_stop values -- "
"using minimum t_stop as t_stop.")
else:
min_tstop = _get_start_stop_from_input(spiketrains)[1]
t_stop = min_tstop
if not all([min_tstop == t.t_stop for t in spiketrains]):
warnings.warn(
"Spiketrains have different t_stop values -- "
"using minimum t_stop as t_stop.")
sts_cut = [st.time_slice(t_start=t_start, t_stop=t_stop) for st in
spiketrains]
# Bin the spike trains and sum across columns
bs = BinnedSpikeTrain(sts_cut, t_start=t_start, t_stop=t_stop,
binsize=binsize)
if binary:
bin_hist = bs.to_sparse_bool_array().sum(axis=0)
else:
bin_hist = bs.to_sparse_array().sum(axis=0)
# Flatten array
bin_hist = np.ravel(bin_hist)
# Renormalise the histogram
if output == 'counts':
# Raw
bin_hist = bin_hist * pq.dimensionless
elif output == 'mean':
# Divide by number of input spike trains
bin_hist = bin_hist * 1. / len(spiketrains) * pq.dimensionless
elif output == 'rate':
# Divide by number of input spike trains and bin width
bin_hist = bin_hist * 1. / len(spiketrains) / binsize
else:
raise ValueError('Parameter output is not valid.')
return neo.AnalogSignalArray(signal=bin_hist.reshape(bin_hist.size, 1),
sampling_period=binsize, units=bin_hist.units,
t_start=t_start)
def read_in_signal(self, segment, block, signal_array, data_indexes,
view_indexes, variable, recording_start_time,
sampling_interval, units, label):
""" Reads in a data item that's not spikes (likely v, gsyn e, gsyn i)\
and saves this data to the segment.
:param segment: Segment to add data to
:type segment: neo.Segment
:param block: neo block
:type block: neo.Block
:param signal_array: the raw signal data
:type signal_array: nparray
:param data_indexes: The indexes for the recorded data
:type data_indexes: list(int)
:param view_indexes: The indexes for which data should be returned.\
If None all data (view_index = data_indexes)
:type view_indexes: list(int)
:param variable: the variable name
:param recording_start_time: when recording started
:param sampling_interval: how often a neuron is recorded
:param units: the units of the recorded value
:param label: human readable label
"""
# pylint: disable=too-many-arguments
t_start = recording_start_time * quantities.ms
sampling_period = sampling_interval * quantities.ms
if view_indexes is None:
indexes = numpy.array(data_indexes)
elif view_indexes == data_indexes:
indexes = numpy.array(data_indexes)
else:
# keep just the view indexes in the data
indexes = [i for i in view_indexes if i in data_indexes]
# keep just data columns in the view
map_indexes = [data_indexes.index(i) for i in indexes]
signal_array = signal_array[:, map_indexes]
ids = list(map(self._population.index_to_id, indexes))
if pynn8_syntax:
data_array = neo.AnalogSignalArray(signal_array,
units=units,
t_start=t_start,
sampling_period=sampling_period,
name=variable,
source_population=label,
channel_index=indexes,
source_ids=ids)
data_array.shape = (data_array.shape[0], data_array.shape[1])
segment.analogsignalarrays.append(data_array)
else:
data_array = neo.AnalogSignal(signal_array,
units=units,
t_start=t_start,
sampling_period=sampling_period,
name=variable,
source_population=label,
source_ids=ids)
channel_index = _get_channel_index(indexes, block)
data_array.channel_index = channel_index
data_array.shape = (data_array.shape[0], data_array.shape[1])
segment.analogsignals.append(data_array)
channel_index.analogsignals.append(data_array)
def _cch_fast(st_1, st_2, win, mode, norm, border_corr, binary, kern):
# Check that the spike trains are binned with the same temporal
# resolution
if not st1.matrix_rows == 1:
raise AssertionError("Spike train must be one dimensional")
if not st2.matrix_rows == 1:
raise AssertionError("Spike train must be one dimensional")
if not st1.binsize == st2.binsize:
raise AssertionError("Bin sizes must be equal")
# Retrieve the array of the binne spik train
st1_arr = st1.to_array()[0, :]
st2_arr = st2.to_array()[0, :]
binsize = st1.binsize
# Convert the to binary version
if binary:
st1_arr = np.array(st1_arr > 0, dtype=int)
st2_arr = np.array(st2_arr > 0, dtype=int)
max_num_bins = max(len(st1_arr), len(st2_arr))
# Cross correlate the spiketrains
# Case explicit temporal window
if win is not None:
# Window parameter given in number of bins (integer)
if isinstance(win[0], int) and isinstance(win[1], int):
# Check the window parameter values
if win[0] >= win[1] or win[0] <= -max_num_bins \
or win[1] >= max_num_bins:
raise ValueError(
"The window exceed the length of the spike trains")
# Assign left and right edges of the cch
l, r = win[0], win[1]
# Window parameter given in time units
else:
# Check the window parameter values
if win[0].rescale(binsize.units).magnitude % \
binsize.magnitude != 0 or win[1].rescale(
binsize.units).magnitude % binsize.magnitude != 0:
raise ValueError(
"The window has to be a multiple of the binsize")
if win[0] >= win[1] or win[0] <= -max_num_bins * binsize \
or win[1] >= max_num_bins * binsize:
raise ValueError("The window exceed the length of the"
" spike trains")
# Assign left and right edges of the cch
l, r = int(win[0].rescale(binsize.units) / binsize), int(
win[1].rescale(binsize.units) / binsize)
# Cross correlate the spike trains
corr = np.correlate(st2_arr, st1_arr, mode='full')
counts = corr[len(st1_arr) + l + 1:len(st1_arr) + 1 + r + 1]
# Case generic
else:
# Cross correlate the spike trains
counts = np.correlate(st2_arr, st1_arr, mode=mode)
# Assign the edges of the cch for the different mode parameters
if mode == 'full':
# Assign left and right edges of the cch
r = st_2.num_bins - 1
l = -st_1.num_bins + 1
# cch compute only for the entries that completely overlap
elif mode == 'valid':
# Assign left and right edges of the cch
r = max(st_2.num_bins - st_1.num_bins, 0)
l = min(st_2.num_bins - st_1.num_bins, 0)
bin_ids = np.r_[l:r + 1]
# Correct the values taking into account lacking contributes
# at the edges
if border_corr is True:
correction = float(max_num_bins + 1) / np.array(
max_num_bins + 1 - abs(np.arange(l, r + 1)), float)
counts = counts * correction
# Define the kern for smoothing as an ndarray
if hasattr(kern, '__iter__'):
if len(kern) > np.abs(l) + np.abs(r) + 1:
raise ValueError(
'The length of the kernel cannot be larger than the '
'length %d of the resulting CCH.' %
(np.abs(l) + np.abs(r) + 1))
kern = np.array(kern, dtype=float)
kern = 1. * kern / sum(kern)
# Check kern parameter
elif kern is not None:
raise ValueError('Invalid smoothing kernel.')
# Smooth the cross-correlation histogram with the kern
if kern is not None:
counts = np.convolve(counts, kern, mode='same')
# Rescale the histogram so that the central bin has height 1,
# if requested
if norm and l <= 0 <= r:
if counts[np.abs(l)] != 0:
counts = counts / counts[np.abs(l)]
else:
warnings.warn('CCH not normalized because no value for 0 lag')
# Transform the array count into an AnalogSignalArray
cch_result = neo.AnalogSignalArray(
signal=counts.reshape(counts.size, 1),
units=pq.dimensionless,
t_start=(bin_ids[0] - 0.5) * st_1.binsize,
sampling_period=st_1.binsize)
# Return only the hist_bins bins and counts before and after the
# central one
return cch_result, bin_ids
def estimate_csd(lfp,
coord_electrode,
sigma,
method='standard',
diam=None,
h=None,
sigma_top=None,
tol=1E-6,
num_steps=200,
f_type='identity',
f_order=None):
"""
Estimates current source density (CSD) from local field potential (LFP)
recordings from multiple depths of the cortex.
Parameters
----------
lfp : neo.AnalogSignalArray
LFP signals from which CSD is estimated.
coord_electrode : Quantity array
Depth of evenly spaced electrode contact points.
sigma : Quantity float
Conductivity of tissue.
method : string
CSD estimation method, either of 'standard': the standard
double-derivative method, 'delta': delta-iCSD method, 'step':
step-iCSD method, 'spline': spline-iCSD method. Default is 'standard'
diam : Quantity float
Diamater of the assumed circular planar current sources centered at
each contact, required by iCSD methods (= 'delta', 'step',
'spline'). Default is `None`.
h : float or np.ndarray * quantity.Quantity
assumed thickness of the source cylinders at all or each contact
sigma_top : Quantity float
Conductivity on top of tissue. When set to `None`, the same value as
sigma: is used. Default is `None`.
tol : float
Tolerance of numerical integration, required by step- and
spline-iCSD methods. Default is 1E-6.
num_steps : int
Number of data points for the spatially upsampled LFP/CSD data,
required by spline-iCSD method. Default is 200.
f_type : string
Type of spatial filter used for smoothing of the result, either of
'boxcar' (uses `scipy.signal.baxcar()`), 'hamming' (
`scipy.signal.hamming()`), 'triangular' (`scipy.signal.tri()`),
'gaussian' (`scipy.signal.gaussian`), 'identity' (no smoothing is
applied). Default is 'identity'.
f_order : float tuple
Parameters to be passed to the scipy.signal function associated with
the specified filter type.
Returns
-------
tuple : (csd, csd_filtered)
csd : neo.AnalogSignalArray
Estimated CSD
csd_filtered : neo.AnalogSignalArray
Estimated CSD, spatially filtered
Example
-------
import numpy as np
import matplotlib.pyplot as plt
from scipy import io
import quantities as pq
import neo
import icsd
#loading test data
test_data = io.loadmat('test_data.mat')
#prepare lfp data for use, by changing the units to SI and append
#quantities, along with electrode geometry and conductivities
lfp_data = test_data['pot1'] * 1E-3 * pq.V # [mV] -> [V]
z_data = np.linspace(100E-6, 2300E-6, 23) * pq.m # [m]
diam = 500E-6 * pq.m # [m]
sigma = 0.3 * pq.S / pq.m # [S/m] or [1/(ohm*m)]
sigma_top = 0. * pq.S / pq.m # [S/m] or [1/(ohm*m)]
lfp = neo.AnalogSignalArray(lfp_data.T, sampling_rate=2.0*pq.kHz)
# Input dictionaries for each method
params = {}
params['delta'] = {
'method': 'delta',
'lfp' : lfp,
'coord_electrode' : z_data,
'diam' : diam, # source diameter
'sigma' : sigma, # extracellular conductivity
'sigma_top' : sigma, # conductivity on top of cortex
}
params['step'] = {
'method': 'step',
'lfp' : lfp,
'coord_electrode' : z_data,
'diam' : diam,
'sigma' : sigma,
'sigma_top' : sigma,
'tol' : 1E-12, # Tolerance in numerical integration
}
params['spline'] = {
'method': 'spline',
'lfp' : lfp,
'coord_electrode' : z_data,
'diam' : diam,
'sigma' : sigma,
'sigma_top' : sigma,
'num_steps' : 201, # Spatial CSD upsampling to N steps
'tol' : 1E-12,
}
params['standard'] = {
'method': 'standard',
'lfp' : lfp,
'coord_electrode' : z_data,
'sigma' : sigma,
}
#plot LFP signal
fig, axes = plt.subplots(len(params)+1, 1, figsize=(6, 8))
ax = axes[0]
im = ax.imshow(lfp.magnitude.T, origin='upper', vmin=-abs(lfp).max(),
vmax=abs(lfp).max(), cmap='jet_r', interpolation='nearest')
ax.axis(ax.axis('tight'))
cb = plt.colorbar(im, ax=ax)
cb.set_label('LFP (%s)' % lfp_data.dimensionality.string)
ax.set_xticklabels([])
ax.set_title('LFP')
ax.set_ylabel('ch #')
i_ax = 1
for method, param in params.items():
ax = axes[i_ax]
i_ax += 1
csd = icsd.estimate_csd(**param)
im = ax.imshow(csd.magnitude.T, origin='upper', vmin=-abs(csd).max(),
vmax=abs(csd).max(), cmap='jet_r',
interpolation='nearest')
ax.axis(ax.axis('tight'))
ax.set_title(method)
cb = plt.colorbar(im, ax=ax)
cb.set_label('CSD (%s)' % csd.dimensionality.string)
ax.set_xticklabels([])
ax.set_ylabel('ch #')
plt.show()
"""
supported_methods = ('standard', 'delta', 'step', 'spline')
icsd_methods = ('delta', 'step', 'spline')
if method not in supported_methods:
print("Pamareter `method` must be either of {}".format(
", ".join(supported_methods)))
raise ValueError
elif method in icsd_methods and diam is None:
print("Parameter `diam` must be specified for iCSD methods: {}".format(
", ".join(icsd_methods)))
raise ValueError
if not isinstance(lfp, neo.AnalogSignalArray):
print('Parameter `lfp` must be neo.AnalogSignalArray')
raise TypeError
if f_type is not 'identity' and f_order is None:
print("The order of {} filter must be specified".format(f_type))
raise ValueError
lfp_pqarr = lfp.magnitude.T * lfp.units
if sigma_top is None:
sigma_top = sigma
arg_dict = {
'lfp': lfp_pqarr,
'coord_electrode': coord_electrode,
'sigma': sigma,
'f_type': f_type,
'f_order': f_order,
}
if method == 'standard':
csd_estimator = StandardCSD(**arg_dict)
else:
arg_dict['diam'] = diam
arg_dict['sigma_top'] = sigma_top
if method == 'delta':
csd_estimator = DeltaiCSD(**arg_dict)
else:
arg_dict['tol'] = tol
if method == 'step':
arg_dict['h'] = h
csd_estimator = StepiCSD(**arg_dict)
else:
arg_dict['num_steps'] = num_steps
csd_estimator = SplineiCSD(**arg_dict)
csd_pqarr = csd_estimator.get_csd()
csd_pqarr_filtered = csd_estimator.filter_csd(csd_pqarr)
csd = neo.AnalogSignalArray(csd_pqarr.T,
t_start=lfp.t_start,
sampling_rate=lfp.sampling_rate)
csd_filtered = neo.AnalogSignalArray(csd_pqarr_filtered.T,
t_start=lfp.t_start,
sampling_rate=lfp.sampling_rate)
return csd, csd_filtered
def _cch_memory(binned_st1, binned_st2, win, border_corr, binary, kern):
# Retrieve unclipped matrix
st1_spmat = binned_st1.to_sparse_array()
st2_spmat = binned_st2.to_sparse_array()
binsize = binned_st1.binsize
max_num_bins = max(binned_st1.num_bins, binned_st2.num_bins)
# Set the time window in which is computed the cch
if not isinstance(win, str):
# Window parameter given in number of bins (integer)
if isinstance(win[0], int) and isinstance(win[1], int):
# Check the window parameter values
if win[0] >= win[1] or win[0] <= -max_num_bins \
or win[1] >= max_num_bins:
raise ValueError(
"The window exceeds the length of the spike trains")
# Assign left and right edges of the cch
l, r = win[0], win[1]
# Window parameter given in time units
else:
# Check the window parameter values
if win[0].rescale(binsize.units).magnitude % \
binsize.magnitude != 0 or win[1].rescale(
binsize.units).magnitude % binsize.magnitude != 0:
raise ValueError(
"The window has to be a multiple of the binsize")
if win[0] >= win[1] or win[0] <= -max_num_bins * binsize \
or win[1] >= max_num_bins * binsize:
raise ValueError("The window exceeds the length of the"
" spike trains")
# Assign left and right edges of the cch
l, r = int(win[0].rescale(binsize.units) / binsize), int(
win[1].rescale(binsize.units) / binsize)
# Case without explicit window parameter
elif window == 'full':
# cch computed for all the possible entries
# Assign left and right edges of the cch
r = binned_st2.num_bins - 1
l = -binned_st1.num_bins + 1
# cch compute only for the entries that completely overlap
elif window == 'valid':
# cch computed only for valid entries
# Assign left and right edges of the cch
r = max(binned_st2.num_bins - binned_st1.num_bins, 0)
l = min(binned_st2.num_bins - binned_st1.num_bins, 0)
# Check the mode parameter
else:
raise KeyError("Invalid window parameter")
# For each row, extract the nonzero column indices
# and the corresponding # data in the matrix (for performance reasons)
st1_bin_idx_unique = st1_spmat.nonzero()[1]
st2_bin_idx_unique = st2_spmat.nonzero()[1]
# Case with binary entries
if binary:
st1_bin_counts_unique = np.array(st1_spmat.data > 0, dtype=int)
st2_bin_counts_unique = np.array(st2_spmat.data > 0, dtype=int)
# Case with all values
else:
st1_bin_counts_unique = st1_spmat.data
st2_bin_counts_unique = st2_spmat.data
# Initialize the counts to an array of zeroes,
# and the bin IDs to integers
# spanning the time axis
counts = np.zeros(np.abs(l) + np.abs(r) + 1)
bin_ids = np.arange(l, r + 1)
# Compute the CCH at lags in l,...,r only
for idx, i in enumerate(st1_bin_idx_unique):
il = np.searchsorted(st2_bin_idx_unique, l + i)
ir = np.searchsorted(st2_bin_idx_unique, r + i, side='right')
timediff = st2_bin_idx_unique[il:ir] - i
assert ((timediff >= l) & (timediff <= r)).all(), 'Not all the '
'entries of cch lie in the window'
counts[timediff + np.abs(l)] += (st1_bin_counts_unique[idx] *
st2_bin_counts_unique[il:ir])
st2_bin_idx_unique = st2_bin_idx_unique[il:]
st2_bin_counts_unique = st2_bin_counts_unique[il:]
# Border correction
if border_corr is True:
counts = _border_correction(counts, max_num_bins, l, r)
if kern is not None:
# Smoothing
counts = _kernel_smoothing(counts, kern, l, r)
# Transform the array count into an AnalogSignalArray
cch_result = neo.AnalogSignalArray(
signal=counts.reshape(counts.size, 1),
units=pq.dimensionless,
t_start=(bin_ids[0] - 0.5) * binned_st1.binsize,
sampling_period=binned_st1.binsize)
# Return only the hist_bins bins and counts before and after the
# central one
return cch_result, bin_ids
def instantaneous_rate(spiketrain, sampling_period, form,
sigma='auto', t_start=None, t_stop=None,
acausal=True, trim=False):
"""
Estimate instantaneous firing rate by kernel convolution.
Parameters
-----------
spiketrain: 'neo.SpikeTrain'
Neo object that contains spike times, the unit of the time stamps
and t_start and t_stop of the spike train.
sampling_period : Quantity
time stamp resolution of the spike times. the same resolution will
be assumed for the kernel
form : {'BOX', 'TRI', 'GAU', 'EPA', 'EXP', 'ALP'}
Kernel form. Currently implemented forms are BOX (boxcar),
TRI (triangle), GAU (gaussian), EPA (epanechnikov), EXP (exponential),
ALP (alpha function). EXP and ALP are asymmetric kernel forms and
assume optional parameter `direction`.
sigma : string or Quantity
Standard deviation of the distribution associated with kernel shape.
This parameter defines the time resolution of the kernel estimate
and makes different kernels comparable (cf. [1] for symmetric kernels).
This is used here as an alternative definition to the cut-off
frequency of the associated linear filter.
Default value is 'auto'. In this case, the optimized kernel width for
the rate estimation is calculated according to [1].
t_start : Quantity (Optional)
start time of the interval used to compute the firing rate, if None
assumed equal to spiketrain.t_start
Default:None
t_stop : Qunatity
End time of the interval used to compute the firing rate (included).
If none assumed equal to spiketrain.t_stop
Default:None
acausal : bool
if True, acausal filtering is used, i.e., the gravity center of the
filter function is aligned with the spike to convolve
Default:None
m_idx : int
index of the value in the kernel function vector that corresponds
to its gravity center. this parameter is not mandatory for
symmetrical kernels but it is required when asymmetrical kernels
are to be aligned at their gravity center with the event times if None
is assumed to be the median value of the kernel support
Default : None
trim : bool
if True, only the 'valid' region of the convolved
signal are returned, i.e., the points where there
isn't complete overlap between kernel and spike train
are discarded
NOTE: if True and an asymmetrical kernel is provided
the output will not be aligned with [t_start, t_stop]
Returns
-------
rate : neo.AnalogSignalArray
Contains the rate estimation in unit hertz (Hz).
Has a property 'rate.times' which contains the time axis of the rate
estimate. The unit of this property is the same as the resolution that
is given as an argument to the function.
Raises
------
TypeError:
If argument value for the parameter `sigma` is not a quantity object
or string 'auto'.
See also
--------
elephant.statistics.make_kernel
References
----------
..[1] H. Shimazaki, S. Shinomoto, J Comput Neurosci (2010) 29:171–182.
"""
if sigma == 'auto':
unit = spiketrain.units
kernel_width = sskernel(spiketrain.magnitude, tin=None,
bootstrap=True)['optw']
sigma = kernel_width*unit
elif not isinstance(sigma, pq.Quantity):
raise TypeError('sigma must be either a quantities object or "auto".'
' Found: %s, value %s' %(type(sigma), str(sigma)))
kernel, norm, m_idx = make_kernel(form=form, sigma=sigma,
sampling_period=sampling_period)
units = pq.CompoundUnit("%s*s" % str(sampling_period.rescale('s').magnitude))
spiketrain = spiketrain.rescale(units)
if t_start is None:
t_start = spiketrain.t_start
else:
t_start = t_start.rescale(spiketrain.units)
if t_stop is None:
t_stop = spiketrain.t_stop
else:
t_stop = t_stop.rescale(spiketrain.units)
time_vector = np.zeros(int((t_stop - t_start)) + 1)
spikes_slice = spiketrain.time_slice(t_start, t_stop) \
if len(spiketrain) else np.array([])
for spike in spikes_slice:
index = int((spike - t_start))
time_vector[index] += 1
r = norm * scipy.signal.fftconvolve(time_vector, kernel, 'full')
if np.any(r < 0):
warnings.warn('Instantaneous firing rate approximation contains '
'negative values, possibly caused due to machine '
'precision errors')
if acausal:
if not trim:
r = r[m_idx:-(kernel.size - m_idx)]
elif trim:
r = r[2 * m_idx:-2*(kernel.size - m_idx)]
t_start = t_start + m_idx * spiketrain.units
t_stop = t_stop - ((kernel.size) - m_idx) * spiketrain.units
else:
if not trim:
r = r[m_idx:-(kernel.size - m_idx)]
elif trim:
r = r[2 * m_idx:-2*(kernel.size - m_idx)]
t_start = t_start + m_idx * spiketrain.units
t_stop = t_stop - ((kernel.size) - m_idx) * spiketrain.units
rate = neo.AnalogSignalArray(signal=r.reshape(r.size, 1),
sampling_period=sampling_period,
units=pq.Hz, t_start=t_start)
return rate
def estimate_csd(lfp,
coords=None,
method=None,
process_estimate=True,
**kwargs):
"""
Fuction call to compute the current source density (CSD) from extracellular
potential recordings(local-field potentials - LFP) using laminar electrodes
or multi-contact electrodes with 2D or 3D geometries.
Parameters
----------
lfp : list(neo.AnalogSignal type objects)
positions of electrodes can be added as neo.RecordingChannel
coordinate or sent externally as a func argument (See coords)
coords : [Optional] corresponding spatial coordinates of the electrodes
Defaults to None
Otherwise looks for RecordingChannels coordinate
method : string
Pick a method corresonding to the setup, in this implementation
For Laminar probe style (1D), use 'KCSD1D' or 'StandardCSD',
or 'DeltaiCSD' or 'StepiCSD' or 'SplineiCSD'
For MEA probe style (2D), use 'KCSD2D', or 'MoIKCSD'
For array of laminar probes (3D), use 'KCSD3D'
Defaults to None
process_estimate : bool
In the py_iCSD_toolbox this corresponds to the filter_csd -
the parameters are passed as kwargs here ie., f_type and f_order
In the kcsd methods this corresponds to cross_validate -
the parameters are passed as kwargs here ie., lambdas and Rs
Defaults to True
kwargs : parameters to each method
The parameters corresponding to the method chosen
See the documentation of the individual method
Default is {} - picks the best parameters,
Returns
-------
Estimated CSD
neo.AnalogSignalArray Object
annotated with the spatial coordinates
Raises
------
AttributeError
No units specified for electrode spatial coordinates
ValueError
Invalid function arguments, wrong method name, or
mismatching coordinates
TypeError
Invalid cv_param argument passed
"""
if not isinstance(lfp[0], neo.AnalogSignal):
raise TypeError('Parameter `lfp` must be a list(neo.AnalogSignal \
type objects')
if coords is None:
coords = []
for ii in lfp:
coords.append(ii.recordingchannel.coordinate.rescale(pq.mm))
else:
scaled_coords = []
for coord in coords:
try:
scaled_coords.append(coord.rescale(pq.mm))
except AttributeError:
raise AttributeError('No units given for electrode spatial \
coordinates')
coords = scaled_coords
if method is None:
raise ValueError('Must specify a method of CSD implementation')
if len(coords) != len(lfp):
raise ValueError('Number of signals and coords is not same')
for ii in coords: # CHECK for Dimensionality of electrodes
if len(ii) > 3:
raise ValueError('Invalid number of coordinate positions')
dim = len(coords[0]) # TODO : Generic co-ordinates!
if dim == 1 and (method not in available_1d):
raise ValueError('Invalid method, Available options are:',
available_1d)
if dim == 2 and (method not in available_2d):
raise ValueError('Invalid method, Available options are:',
available_2d)
if dim == 3 and (method not in available_3d):
raise ValueError('Invalid method, Available options are:',
available_3d)
if method in kernel_methods:
input_array = np.zeros((len(lfp), lfp[0].magnitude.shape[0]))
for ii, jj in enumerate(lfp):
input_array[ii, :] = jj.rescale(pq.mV).magnitude
kernel_method = getattr(KCSD, method) # fetch the class 'KCSD1D'
lambdas = kwargs.pop('lambdas', None)
Rs = kwargs.pop('Rs', None)
k = kernel_method(np.array(coords), input_array, **kwargs)
if process_estimate:
k.cross_validate(lambdas, Rs)
estm_csd = k.values()
estm_csd = np.rollaxis(estm_csd, -1, 0)
output = neo.AnalogSignalArray(estm_csd * pq.uA / pq.mm**3,
t_start=lfp[0].t_start,
sampling_rate=lfp[0].sampling_rate)
if dim == 1:
output.annotate(x_coords=k.estm_x)
elif dim == 2:
output.annotate(x_coords=k.estm_x, y_coords=k.estm_y)
elif dim == 3:
output.annotate(x_coords=k.estm_x,
y_coords=k.estm_y,
z_coords=k.estm_z)
elif method in py_iCSD_toolbox:
coords = np.array(coords) * coords[0].units
if method in icsd_methods:
try:
coords = coords.rescale(kwargs['diam'].units)
except KeyError: # Then why specify as a default in icsd?
# All iCSD methods explicitly assume a source
# diameter in contrast to the stdCSD that
# implicitly assume infinite source radius
raise ValueError("Parameter diam must be specified for iCSD \
methods: {}".format(", ".join(icsd_methods)))
if 'f_type' in kwargs:
if (kwargs['f_type'] is not 'identity') and \
(kwargs['f_order'] is None):
raise ValueError("The order of {} filter must be \
specified".format(kwargs['f_type']))
lfp = neo.AnalogSignalArray(np.asarray(lfp).T,
units=lfp[0].units,
sampling_rate=lfp[0].sampling_rate)
csd_method = getattr(icsd, method) # fetch class from icsd.py file
csd_estimator = csd_method(lfp=lfp.magnitude.T * lfp.units,
coord_electrode=coords.flatten(),
**kwargs)
csd_pqarr = csd_estimator.get_csd()
if process_estimate:
csd_pqarr_filtered = csd_estimator.filter_csd(csd_pqarr)
output = neo.AnalogSignalArray(csd_pqarr_filtered.T,
t_start=lfp.t_start,
sampling_rate=lfp.sampling_rate)
else:
output = neo.AnalogSignalArray(csd_pqarr.T,
t_start=lfp.t_start,
sampling_rate=lfp.sampling_rate)
output.annotate(x_coords=coords)
return output
seg.spikes.append(sp)
for ind2 in range(3):
irr = neo.IrregularlySampledSignal(name='IrregularlySampled' + str(ind2), times=np.random.rand(10) * qu.s,
signal=np.random.rand(10) * qu.mV)
irr.segment = seg
seg.irregularlysampledsignals.append(irr)
for ind2 in range(3):
an = neo.AnalogSignal(name='AnalogSignal' + str(ind2), signal=np.random.rand(10) * qu.mV,
sampling_rate=10 * qu.Hz)
an.segment = seg
seg.analogsignals.append(an)
for ind2 in range(3):
an = neo.AnalogSignalArray(name='AnalogSignalArray' + str(ind2), signal=np.random.rand(10, 10) * qu.mV,
sampling_rate=10 * qu.Hz)
sp.segment = seg
seg.analogsignalarrays.append(an)
for ind2 in range(3):
ev = neo.Event(name='Event' + str(ind2), time=np.random.rand() * qu.s, label='h')
ev.segment = seg
seg.events.append(ev)
for ind2 in range(3):
eva = neo.EventArray(name='EventArray' + str(ind2), times=np.random.rand(10) * qu.s, label=['h'] * 10)
eva.segment = seg
seg.eventarrays.append(eva)
for ind2 in range(3):
ep = neo.Epoch(name='Epoch' + str(ind2), time=np.random.rand() * qu.s, duration=np.random.rand() * qu.s,
def _cch_speed(binned_st1, binned_st2, win, border_corr, binary, kern):
# Retrieve the array of the binne spik train
st1_arr = binned_st1.to_array()[0, :]
st2_arr = binned_st2.to_array()[0, :]
binsize = binned_st1.binsize
# Convert the to binary version
if binary:
st1_arr = np.array(st1_arr > 0, dtype=int)
st2_arr = np.array(st2_arr > 0, dtype=int)
max_num_bins = max(len(st1_arr), len(st2_arr))
# Cross correlate the spiketrains
# Case explicit temporal window
if not isinstance(win, str):
# Window parameter given in number of bins (integer)
if isinstance(win[0], int) and isinstance(win[1], int):
# Check the window parameter values
if win[0] >= win[1] or win[0] <= -max_num_bins \
or win[1] >= max_num_bins:
raise ValueError(
"The window exceed the length of the spike trains")
# Assign left and right edges of the cch
l, r = win
# Window parameter given in time units
else:
# Check the window parameter values
if win[0].rescale(binsize.units).magnitude % \
binsize.magnitude != 0 or win[1].rescale(
binsize.units).magnitude % binsize.magnitude != 0:
raise ValueError(
"The window has to be a multiple of the binsize")
if win[0] >= win[1] or win[0] <= -max_num_bins * binsize \
or win[1] >= max_num_bins * binsize:
raise ValueError("The window exceed the length of the"
" spike trains")
# Assign left and right edges of the cch
l, r = int(win[0].rescale(binsize.units) / binsize), int(
win[1].rescale(binsize.units) / binsize)
# Zero padding
st1_arr = np.pad(st1_arr,
(int(np.abs(np.min([l, 0]))), np.max([r, 0])),
mode='constant')
cch_mode = 'valid'
else:
# Assign the edges of the cch for the different mode parameters
if win == 'full':
# Assign left and right edges of the cch
r = binned_st2.num_bins - 1
l = -binned_st1.num_bins + 1
# cch compute only for the entries that completely overlap
elif win == 'valid':
# Assign left and right edges of the cch
r = max(binned_st2.num_bins - binned_st1.num_bins, 0)
l = min(binned_st2.num_bins - binned_st1.num_bins, 0)
cch_mode = win
# Cross correlate the spike trains
counts = np.correlate(st2_arr, st1_arr, mode=cch_mode)
bin_ids = np.r_[l:r + 1]
# Border correction
if border_corr is True:
counts = _border_correction(counts, max_num_bins, l, r)
if kern is not None:
# Smoothing
counts = _kernel_smoothing(counts, kern, l, r)
# Transform the array count into an AnalogSignalArray
cch_result = neo.AnalogSignalArray(
signal=counts.reshape(counts.size, 1),
units=pq.dimensionless,
t_start=(bin_ids[0] - 0.5) * binned_st1.binsize,
sampling_period=binned_st1.binsize)
# Return only the hist_bins bins and counts before and after the
# central one
return cch_result, bin_ids