def adding_list(start,end,env,Data,filtOBj,ch,ListObj,cutoff,info,exclusion):
for s, e in zip(start, end):
print '.',
sys.stdout.flush()
index = np.arange(s,e)
if exclusion is not None:
ex_s = exclusion.__getlist__('start_sec')*Data.sample_rate
ex_e = exclusion.__getlist__('end_sec')*Data.sample_rate
aux = np.array([])
for es, ee in zip(ex_s, ex_e):
aux = np.append(aux,np.arange(int(es),int(ee)))
aux2 = list(set(aux).intersection(index))
if len(aux2) > 0:
print 'excluded',
continue
HFOwaveform = env[index]
tstamp_points = s + np.argmax(HFOwaveform)
tstamp = Data.time_vec[tstamp_points]
if tstamp_points-int(Data.sample_rate/2) > 0 and tstamp_points+int(Data.sample_rate/2)+1 < Data.data.shape[0]:
Lindex = np.arange(tstamp_points-int(Data.sample_rate/2),tstamp_points+int(Data.sample_rate/2)+1)
elif tstamp_points-int(Data.sample_rate/2) < 0:
Lindex = np.arange(0,Data.sample_rate+1)
elif tstamp_points+int(Data.sample_rate/2)+1 > Data.data.shape[0]:
Lindex = np.arange(Data.data.shape[0]-Data.sample_rate-1,Data.data.shape[0])
tstamp_idx = np.nonzero(Lindex==tstamp_points)[0][0]
waveform = np.empty((Lindex.shape[0],3))
waveform[:] = np.NAN
if ch == 'common':
waveform[:,0] = Data.common_ref[Lindex]
waveform[:,1] = filtOBj.common_ref[Lindex]
waveform[:,2] = env[Lindex]
else:
waveform[:,0] = Data.data[Lindex,ch]
waveform[:,1] = filtOBj.data[Lindex,ch]
waveform[:,2] = env[Lindex]
try:
start_idx = np.nonzero(Lindex==s)[0][0]
end_idx = np.nonzero(Lindex==e)[0][0]
hfo = hfoObj(ch,tstamp,tstamp_idx, waveform,start_idx,end_idx,ths_value,Data.sample_rate,cutoff,info)
ListObj.__addEvent__(hfo)
except IndexError:
continue
def findHFO_filtHilbert(Data,
low_cut,
high_cut=None,
order=None,
window=('kaiser', 0.5),
ths=5,
ths_method='STD',
min_dur=3,
min_separation=2,
energy=False):
'''
Find HFO by Filter-Hilbert method.
Parameters
----------
Data: DataObj
Data object to filt/find HFO
low_cut: int
Low cut frequency.
high_cut: int
High cut frequency. If None, high_cut = nyrqst
order: int, optional
None (default) - Order of the filter calculated as 1/10 of sample rate
window : string or tuple of string and parameter values
Desired window to use. See `scipy.signal.get_window` for a list
of windows and required parameters.
ths : int, optional
5 (default) - times value of threshold (5*STD for example)
ths_method: str, optional
'STD' - Standard desviation above the mean
'Tukey' - Interquartil interval above percentile 75
min_dur: int, optional
3 (default) - minimal number of cicle that event should last. Calculeted
the number of points that event should last by formula ceil(min_dur*sample_rate/high_cut)
min_separation: int, optional
2 (defalt) - minimal number of cicle that separete events. Calculetad
the number of points that separete events by formula ceil(min_separation*sample_rate/low_cut)
'''
if low_cut == None and high_cut == None:
raise Exception('You should determine the cutting frequencies')
sample_rate = Data.sample_rate
# if no high cut, =nyrqst
if high_cut == None:
high_cut = sample_rate / 2
cutoff = [low_cut, high_cut]
# Transform min_dur from cicles to poinst - minimal duration of HFO (Default is 3 cicles)
min_dur = math.ceil(min_dur * sample_rate / high_cut)
# Transform min_separation from cicles to points - minimal separation between events
min_separation = math.ceil(min_separation * sample_rate / low_cut)
# filtering
filtOBj = eegfilt(Data, low_cut, high_cut, order, window)
nch = filtOBj.n_channels
if order == None:
order = int(sample_rate / 10)
info = str(low_cut) + '-' + str(high_cut) + ' Hz filtering; order: ' + str(
order) + ', window: ' + str(window) + ' ; ' + str(
ths) + '*' + ths_method + '; min_dur = ' + str(
min_dur) + '; min_separation = ' + str(min_separation)
HFOs = EventList(Data.ch_labels, (Data.time_vec[0], Data.time_vec[-1]))
if nch == 1:
print 'Finding in channel'
filt = filtOBj.data
env = np.abs(sig.hilbert(filt))
if ths_method == 'STD':
ths_value = np.mean(env) + ths * np.std(env)
elif ths_method == 'Tukey':
ths_value = np.percentile(
env,
75) + ths * (np.percentile(env, 75) - np.percentile(env, 25))
start, end = findStartEnd(filt, env, ths_value, min_dur,
min_separation)
for s, e in zip(start, end):
index = np.arange(s, e)
HFOwaveform = env[index]
tstamp_points = s + np.argmax(HFOwaveform)
tstamp = Data.time_vec[tstamp_points]
Lindex = np.arange(tstamp_points - int(sample_rate / 2),
tstamp_points + int(sample_rate / 2) + 1)
tstamp_idx = np.nonzero(Lindex == tstamp_points)[0][0]
waveform = np.empty((Lindex.shape[0], 2))
waveform[:] = np.NAN
waveform[:, 0] = Data.data[Lindex]
waveform[:, 1] = filtOBj.data[Lindex]
start_idx = np.nonzero(Lindex == s)[0][0]
end_idx = np.nonzero(Lindex == e)[0][0]
hfo = hfoObj(0, tstamp, tstamp_idx, waveform, start_idx, end_idx,
ths_value, sample_rate, cutoff, info)
HFOs.__addEvent__(hfo)
else:
for ch in range(nch):
if ch not in filtOBj.bad_channels:
print 'Finding in channel ' + filtOBj.ch_labels[ch]
filt = filtOBj.data[:, ch]
if energy:
env = np.abs(sig.hilbert(filt))**2
else:
env = np.abs(sig.hilbert(filt))
if ths_method == 'STD':
ths_value = np.mean(env) + ths * np.std(env)
elif ths_method == 'Tukey':
ths_value = np.percentile(env, 75) + ths * (
np.percentile(env, 75) - np.percentile(env, 25))
start, end = findStartEnd(filt, env, ths_value, min_dur,
min_separation)
for s, e in zip(start, end):
index = np.arange(s, e)
HFOwaveform = env[index]
tstamp_points = s + np.argmax(HFOwaveform)
tstamp = Data.time_vec[tstamp_points]
Lindex = np.arange(
tstamp_points - int(sample_rate / 2),
tstamp_points + int(sample_rate / 2) + 1)
tstamp_idx = np.nonzero(Lindex == tstamp_points)[0][0]
waveform = np.empty((Lindex.shape[0], 2))
waveform[:] = np.NAN
waveform[:, 0] = Data.data[Lindex, ch]
waveform[:, 1] = filtOBj.data[Lindex, ch]
start_idx = np.nonzero(Lindex == s)[0][0]
end_idx = np.nonzero(Lindex == e)[0][0]
hfo = hfoObj(ch, tstamp, tstamp_idx, waveform, start_idx,
end_idx, ths_value, sample_rate, cutoff, info)
HFOs.__addEvent__(hfo)
return HFOs
def open_dataset(file_name,dataset_name,htype = 'auto'):
'''
open a dataset in a specific file_name
Parameters
----------
file_name: str
Name of the HDF5 (.h5) file
dataset_name: str
Name of dataset to open
htype: str, optional
auto (the default) - read htype from HDF file
Data - DataObj type
Spike - SpikeObj type
hfo - hfoObj type
'''
# reading h5 file
h5 = h5py.File(file_name,'r+')
# loading dataset
dataset = h5[dataset_name]
# getting htype
if htype == 'auto':
htype = dataset.attrs['htype']
if htype == 'Data':
# Sample Rate attribute
sample_rate = dataset.attrs['SampleRate[Hz]']
n_points = dataset.shape[0]
end_time = n_points/sample_rate
# Amplitude Unit
if 'amp_unit' in dataset.attrs:
amp_unit = dataset.attrs['amp_unit']
else:
amp_unit = 'AU'
# Time vector
if 'Time_vec_edge' in dataset.attrs:
edge = dataset.attrs['Time_vec_edge']
if edge[0] == edge[1]:
time_vec = np.linspace(0,end_time,n_points,endpoint=False)
else:
time_vec = np.linspace(edge[0],edge[1],n_points,endpoint=False)
else:
time_vec = np.linspace(0,end_time,n_points,endpoint=False)
# Check if has 'Bad_channels' attribute, if not, create one empty
if len([x for x in dataset.attrs.keys() if x == 'Bad_channels']) == 0:
dataset.attrs.create("Bad_channels",[],dtype=int)
# Load bad channels
bad_channels = dataset.attrs["Bad_channels"]
# Creating dictionary
Data = DataObj(dataset[:],sample_rate,amp_unit,dataset.attrs['Channel_Labels'],time_vec,bad_channels,file_name,dataset_name)
elif htype == 'list':
# Time vector
keys = dataset.keys()
ch_labels = dataset.attrs['ch_labels']
time_edge = dataset.attrs['time_edge']
Data = EventList(ch_labels,time_edge,file_name,dataset_name)
for k in keys:
waveform = dataset[k][:]
tstamp = dataset[k].attrs['tstamp']
evhtype = dataset[k].attrs['htype']
channel = dataset[k].attrs['channel']
if evhtype == 'Spike':
clus = dataset[k].attrs['cluster']
feat = dataset[k].attrs['features']
time_edge = dataset[k].attrs['time_edge']
spk = SpikeObj(channel,waveform,tstamp,clus,feat,time_edge)
Data.__addEvent__(spk)
elif evhtype == 'HFO':
tstamp_idx = dataset[k].attrs['tstamp_idx']
start_idx = dataset[k].attrs['start_idx']
end_idx = dataset[k].attrs['end_idx']
ths_value = dataset[k].attrs['ths_value']
sample_rate = dataset[k].attrs['sample_rate']
cutoff = dataset[k].attrs['cutoff']
info = dataset[k].attrs['info']
hfo = hfoObj(channel,tstamp,tstamp_idx, waveform,start_idx,end_idx,ths_value,sample_rate,cutoff,info)
Data.__addEvent__(hfo)
h5.close()
return Data
def findHFO_filtbank(Data,low_cut = 50,high_cut= None, ths = 5, max_ths = 10,par = False, save = None, replace = False, exclude = [],rc = None ,dview = None):
'''
Find HFO by Filter-bank method.
by Anderson Brito da Silva - 29/jul/2015
Parameters
----------
Data: DataObj
Data object to filt/find HFO
low_cut: int
50 (default) - Low cut frequency in Hz.
high_cut: int
High cut frequency in Hz. If None, high_cut = nyrqst
ths : int, optional
3 (default) - threshold for z-score
max_ths : int, optional
20 (default) - max threshold for z-score
'''
import sys
import time
def clear_cache(rc,dview):
rc.purge_results('all')
rc.results.clear()
rc.metadata.clear()
dview.results.clear()
assert not rc.outstanding
rc.history = []
dview.history = []
def create_wavelet(f,time):
import numpy as np
numcycles = 13
std = numcycles/(2*np.pi*f)
wavelet = np.exp(2*1j*np.pi*f*time)*np.exp(-(time**2)/(2*(std**2)))
wavelet /= max(wavelet)
return wavelet
def filt_wavelet(data_ch,wavelet):
import scipy.signal as sig
x = sig.fftconvolve(data_ch,wavelet,'same')
return x
def bin_filt(filt):
import numpy as np
#bin_x = np.array([1 if y < 5 or y > 10 else 0 for y in abs((filt-np.mean(filt[200:-200]))/np.std(filt[200:-200]))])
#bin_x = np.zeros(filt.shape)
q75, q25 = np.percentile(np.abs(filt[200:-200]), [75 ,25])
iqr = q75 - q25
#bin_x[np.nonzero(filt.real<(q75+3*iqr))] = 1
bin_x = np.array([1 if y < (q75+3*iqr) else 0 for y in np.abs(filt)])
return bin_x
def find_min_duration(f,sample_rate):
import math
# Transform min_dur from cicles to poinst - minimal duration of HFO (Default is 3 cicles)
min_dur = math.ceil(3*sample_rate/f)
return min_dur
def find_min_separation(f,sample_rate):
import math
# Transform min_separation from cicles to points - minimal separation between events
min_separation = math.ceil(2*sample_rate/f)
return min_separation
def find_start_end(x, bin_x,min_dur,min_separation,max_local=True):
import numpy as np
subthsIX = bin_x.nonzero()[0] # subthreshold index
subthsInterval = np.diff(subthsIX) # interval between subthreshold
sIX = subthsInterval > min_dur # index of subthsIX bigger then minimal duration
start_ix = subthsIX[sIX] + 1 # start index of events
end_ix = start_ix + subthsInterval[sIX]-1 # end index of events
to_remove = np.array(np.nonzero(start_ix[1:]-end_ix[0:-1] < min_separation)[0]) # find index of events separeted by less the minimal interval
start_ix = np.delete(start_ix, to_remove+1) # removing
end_ix = np.delete(end_ix, to_remove) #removing
if max_local:
if start_ix.shape[0] != 0:
locs = np.diff(np.sign(np.diff(x))).nonzero()[0] + 1 # local min+max
to_remove = []
for ii in range(start_ix.shape[0]):
if np.nonzero((locs > start_ix[ii]) & (locs < end_ix[ii]))[0].shape[0] < 6:
to_remove.append(ii)
start_ix = np.delete(start_ix, to_remove) # removing
end_ix = np.delete(end_ix, to_remove) #removing
return start_ix, end_ix
def se_to_array(arrlen,se):
import numpy as np
z = np.zeros((arrlen,1))
for ii in range(se[0].shape[0]):
z[se[0][ii]:se[1][ii],0] = 1
return z
print 'Finding HFO by Wavelet Filter Bank'
sys.stdout.flush()
if low_cut == None and high_cut == None:
raise Exception('You should determine the cutting frequencies')
sample_rate = Data.sample_rate
# if no high cut, =nyrqst
if high_cut == None:
high_cut = sample_rate/2
cutoff = [low_cut,high_cut] # define cutoff
scales = np.logspace(np.log(cutoff[0]),np.log(cutoff[1]), base=np.e,num = 30, endpoint=False)
noffilters = len(scales) # number of filters
seg_len = 400. # milisecond
npoints = seg_len*sample_rate/1000 # number of points of wavelet
time_vec = np.linspace(-seg_len/2000,seg_len/2000,npoints) #time_vec
if save is not None:
file_name = save[0]
obj_name = save[1]
save_opt = True
HFOs = EventList(Data.ch_labels,(Data.time_vec[0],Data.time_vec[-1]),file_name = file_name, dataset_name = obj_name)
# open or creating file
h5 = h5py.File(file_name,'a')
#deleting previous dataset
if obj_name in h5:
if replace:
del h5[obj_name]
group = h5.create_group(obj_name)
group.attrs.create('htype',HFOs.htype)
group.attrs.create('time_edge',[HFOs.time_edge[0],HFOs.time_edge[-1]])
group.attrs.create('ch_labels', HFOs.ch_labels[:])
ev_count = 0
print 'Created Dataset ' + file_name + ' ' + obj_name
else:
group = h5[obj_name]
ev_count = len(group.items())
print 'Open Dataset ' + file_name + ' ' + obj_name
else:
group = h5.create_group(obj_name)
group.attrs.create('htype',HFOs.htype)
group.attrs.create('time_edge',[HFOs.time_edge[0],HFOs.time_edge[-1]])
group.attrs.create('ch_labels', HFOs.ch_labels[:])
ev_count = 0
print 'Created Dataset ' + file_name + ' ' + obj_name
else:
save_opt = False
HFOs = EventList(Data.ch_labels,(Data.time_vec[0],Data.time_vec[-1]))
info = str(low_cut) + '-' + str(high_cut) + ' Hz Wavelet Filter Bank'
if par:
print 'Using Parallel processing',
print str(len(rc.ids)) + ' cores'
min_durs = map(find_min_duration,scales,itertools.repeat(sample_rate,noffilters))
print 'Durations',
min_seps = map(find_min_separation,scales,itertools.repeat(sample_rate,noffilters))
print '/ Separations',
sys.stdout.flush()
wavelets = map(create_wavelet,scales,itertools.repeat(time_vec,noffilters))
print '/ Wavelets'
sys.stdout.flush()
nch = Data.n_channels
for ch in [x for x in range(nch) if not x in exclude]:
if ch not in Data.bad_channels:
btime = time.time()
if save_opt:
del HFOs
h5.close()
h5 = h5py.File(file_name,'a')
group = h5[obj_name]
ev_count = len(group.items())
print group, ev_count
HFOs = EventList(Data.ch_labels,(Data.time_vec[0],Data.time_vec[-1]),file_name = file_name, dataset_name = obj_name)
print 'Finding in channel ' + Data.ch_labels[ch]
sys.stdout.flush()
arrlen = Data.data[:,ch].shape[0]
zsc = np.zeros((arrlen,noffilters))
spect = np.zeros((arrlen,noffilters), dtype='complex' )
if par:
sys.stdout.flush()
filt_waves = dview.map_sync(filt_wavelet,itertools.repeat(Data.data[:,ch],noffilters),wavelets)
clear_cache(rc,dview)
print 'Convolved',
sys.stdout.flush()
spect = np.array(filt_waves)
bin_xs= dview.map_sync(bin_filt,filt_waves)
clear_cache(rc,dview)
print '/ Binarised',
sys.stdout.flush()
se = dview.map_sync(find_start_end,filt_waves,bin_xs,min_durs,min_seps)
clear_cache(rc,dview)
#print 'here'
#_vars = sys.modules[__name__]
#delattr(_vars, filt_waves)
#delattr(_vars, bin_xs)
print '/ Found',
sys.stdout.flush()
zsc = np.squeeze(dview.map_sync(se_to_array,itertools.repeat(arrlen,noffilters),se))
clear_cache(rc,dview)
upIX = np.unique(np.nonzero(zsc==1)[1])
other = np.ones(Data.data[:,ch].shape)
other[upIX] = 0
print '/ Start-End'
sys.stdout.flush()
start_ix, end_ix = find_start_end([],other,find_min_duration(cutoff[1],sample_rate),find_min_separation(cutoff[0],sample_rate),max_local=False)
else:
wavelets = map(create_wavelet,scales,itertools.repeat(time_vec,noffilters))
print 'Creating list',
sys.stdout.flush()
for s, e in zip(start_ix, end_ix):
HFOwaveform = np.argmax(np.mean(spect[np.unique(zsc[:,np.arange(s,e)].nonzero()[0]),:][:,np.arange(s,e)],0))
tstamp_points = s + HFOwaveform
if HFOwaveform > int(sample_rate/2):
continue
if tstamp_points+int(sample_rate/2)+1 < e:
continue
if tstamp_points-int(sample_rate/2) < 0 or tstamp_points+int(sample_rate/2)+1 > zsc.shape[1]:
continue
waveform = np.empty((np.arange(tstamp_points-int(sample_rate/2),tstamp_points+int(sample_rate/2)+1).shape[0],2))
waveform[:] = np.NAN
waveform[:,0] = Data.data[np.arange(tstamp_points-int(sample_rate/2),tstamp_points+int(sample_rate/2)+1),ch]
waveform[:,1] = np.mean(spect[np.unique(zsc[:,np.arange(tstamp_points-int(sample_rate/2),tstamp_points+int(sample_rate/2)+1)].nonzero()[0]),:][:,np.arange(tstamp_points-int(sample_rate/2),tstamp_points+int(sample_rate/2)+1)],0)
start_idx = np.nonzero(np.arange(tstamp_points-int(sample_rate/2),tstamp_points+int(sample_rate/2)+1)==s)[0][0]
end_idx = np.nonzero(np.arange(tstamp_points-int(sample_rate/2),tstamp_points+int(sample_rate/2)+1)==e)[0][0]
hfo = hfoObj(ch,Data.time_vec[tstamp_points],np.nonzero(np.arange(tstamp_points-int(sample_rate/2),tstamp_points+int(sample_rate/2)+1)==tstamp_points)[0][0], waveform,start_idx,end_idx,ths,sample_rate,cutoff,info)
#if hfo.spectrum.peak_freq > low_cut or hfo.spectrum.peak_freq < high_cut:
HFOs.__addEvent__(hfo)
print '.',
sys.stdout.flush()
print '\n'
print HFOs
if save_opt:
print '... Saving ...'
sys.stdout.flush()
for idx, ev in enumerate(HFOs.event):
name = ev.htype + '_' + str(idx+ev_count)
dataset = group.create_dataset(name,data=ev.waveform)
dataset.attrs.create('htype', ev.htype)
dataset.attrs.create('tstamp', ev.tstamp)
dataset.attrs.create('channel', ev.channel)
dataset.attrs.create('tstamp_idx', ev.tstamp_idx)
dataset.attrs.create('start_idx', ev.start_idx)
dataset.attrs.create('end_idx', ev.end_idx)
dataset.attrs.create('ths_value', ev.ths_value)
dataset.attrs.create('sample_rate', ev.sample_rate)
dataset.attrs.create('cutoff', ev.cutoff)
dataset.attrs.create('info', ev.info)
print 'Elapsed: %s' % (time.time() - btime)
h5.close()
def findHFO_filtHilbert(
Data,
low_cut,
high_cut=None,
order=None,
window=("kaiser", 0.5),
ths=5,
ths_method="STD",
min_dur=3,
min_separation=2,
energy=False,
):
"""
Find HFO by Filter-Hilbert method.
Parameters
----------
Data: DataObj
Data object to filt/find HFO
low_cut: int
Low cut frequency.
high_cut: int
High cut frequency. If None, high_cut = nyrqst
order: int, optional
None (default) - Order of the filter calculated as 1/10 of sample rate
window : string or tuple of string and parameter values
Desired window to use. See `scipy.signal.get_window` for a list
of windows and required parameters.
ths : int, optional
5 (default) - times value of threshold (5*STD for example)
ths_method: str, optional
'STD' - Standard desviation above the mean
'Tukey' - Interquartil interval above percentile 75
min_dur: int, optional
3 (default) - minimal number of cicle that event should last. Calculeted
the number of points that event should last by formula ceil(min_dur*sample_rate/high_cut)
min_separation: int, optional
2 (defalt) - minimal number of cicle that separete events. Calculetad
the number of points that separete events by formula ceil(min_separation*sample_rate/low_cut)
"""
if low_cut == None and high_cut == None:
raise Exception("You should determine the cutting frequencies")
sample_rate = Data.sample_rate
# if no high cut, =nyrqst
if high_cut == None:
high_cut = sample_rate / 2
cutoff = [low_cut, high_cut]
# Transform min_dur from cicles to poinst - minimal duration of HFO (Default is 3 cicles)
min_dur = math.ceil(min_dur * sample_rate / high_cut)
# Transform min_separation from cicles to points - minimal separation between events
min_separation = math.ceil(min_separation * sample_rate / low_cut)
# filtering
filtOBj = eegfilt(Data, low_cut, high_cut, order, window)
nch = filtOBj.n_channels
if order == None:
order = int(sample_rate / 10)
info = (
str(low_cut)
+ "-"
+ str(high_cut)
+ " Hz filtering; order: "
+ str(order)
+ ", window: "
+ str(window)
+ " ; "
+ str(ths)
+ "*"
+ ths_method
+ "; min_dur = "
+ str(min_dur)
+ "; min_separation = "
+ str(min_separation)
)
HFOs = EventList(Data.ch_labels, (Data.time_vec[0], Data.time_vec[-1]))
if nch == 1:
print "Finding in channel"
filt = filtOBj.data
env = np.abs(sig.hilbert(filt))
if ths_method == "STD":
ths_value = np.mean(env) + ths * np.std(env)
elif ths_method == "Tukey":
ths_value = np.percentile(env, 75) + ths * (np.percentile(env, 75) - np.percentile(env, 25))
start, end = findStartEnd(filt, env, ths_value, min_dur, min_separation)
for s, e in zip(start, end):
index = np.arange(s, e)
HFOwaveform = env[index]
tstamp_points = s + np.argmax(HFOwaveform)
tstamp = Data.time_vec[tstamp_points]
Lindex = np.arange(tstamp_points - int(sample_rate / 2), tstamp_points + int(sample_rate / 2) + 1)
tstamp_idx = np.nonzero(Lindex == tstamp_points)[0][0]
waveform = np.empty((Lindex.shape[0], 2))
waveform[:] = np.NAN
waveform[:, 0] = Data.data[Lindex]
waveform[:, 1] = filtOBj.data[Lindex]
start_idx = np.nonzero(Lindex == s)[0][0]
end_idx = np.nonzero(Lindex == e)[0][0]
hfo = hfoObj(0, tstamp, tstamp_idx, waveform, start_idx, end_idx, ths_value, sample_rate, cutoff, info)
HFOs.__addEvent__(hfo)
else:
for ch in range(nch):
if ch not in filtOBj.bad_channels:
print "Finding in channel " + filtOBj.ch_labels[ch]
filt = filtOBj.data[:, ch]
if energy:
env = np.abs(sig.hilbert(filt)) ** 2
else:
env = np.abs(sig.hilbert(filt))
if ths_method == "STD":
ths_value = np.mean(env) + ths * np.std(env)
elif ths_method == "Tukey":
ths_value = np.percentile(env, 75) + ths * (np.percentile(env, 75) - np.percentile(env, 25))
start, end = findStartEnd(filt, env, ths_value, min_dur, min_separation)
for s, e in zip(start, end):
index = np.arange(s, e)
HFOwaveform = env[index]
tstamp_points = s + np.argmax(HFOwaveform)
tstamp = Data.time_vec[tstamp_points]
Lindex = np.arange(tstamp_points - int(sample_rate / 2), tstamp_points + int(sample_rate / 2) + 1)
tstamp_idx = np.nonzero(Lindex == tstamp_points)[0][0]
waveform = np.empty((Lindex.shape[0], 2))
waveform[:] = np.NAN
waveform[:, 0] = Data.data[Lindex, ch]
waveform[:, 1] = filtOBj.data[Lindex, ch]
start_idx = np.nonzero(Lindex == s)[0][0]
end_idx = np.nonzero(Lindex == e)[0][0]
hfo = hfoObj(
ch, tstamp, tstamp_idx, waveform, start_idx, end_idx, ths_value, sample_rate, cutoff, info
)
HFOs.__addEvent__(hfo)
return HFOs
def open_dataset(file_name,dataset_name,htype = 'auto'):
'''
open a dataset in a specific file_name
Parameters
----------
file_name: str
Name of the HDF5 (.h5) file
dataset_name: str
Name of dataset to open
htype: str, optional
auto (the default) - read htype from HDF file
Data - DataObj type
Spike - SpikeObj type
hfo - hfoObj type
'''
# reading h5 file
h5 = h5py.File(file_name,'r+')
# loading dataset
dataset = h5[dataset_name]
# getting htype
if htype == 'auto':
htype = dataset.attrs['htype']
if htype == 'Data':
# Sample Rate attribute
sample_rate = dataset.attrs['SampleRate[Hz]']
n_points = dataset.shape[0]
end_time = n_points/sample_rate
# Amplitude Unit
if 'amp_unit' in dataset.attrs:
amp_unit = dataset.attrs['amp_unit']
else:
amp_unit = 'AU'
# Time vector
if 'Time_vec_edge' in dataset.attrs:
edge = dataset.attrs['Time_vec_edge']
time_vec = np.linspace(edge[0],edge[1],n_points,endpoint=False)
else:
time_vec = np.linspace(0,end_time,n_points,endpoint=False)
# Check if has 'Bad_channels' attribute, if not, create one empty
if len([x for x in dataset.attrs.keys() if x == 'Bad_channels']) == 0:
dataset.attrs.create("Bad_channels",[],dtype=int)
# Load bad channels
bad_channels = dataset.attrs["Bad_channels"]
# Creating dictionary
Data = DataObj(dataset[:],sample_rate,amp_unit,dataset.attrs['Channel_Labels'],time_vec,bad_channels,file_name,dataset_name)
elif htype == 'list':
# Time vector
keys = dataset.keys()
ch_labels = dataset.attrs['ch_labels']
time_edge = dataset.attrs['time_edge']
Data = EventList(ch_labels,time_edge,file_name,dataset_name)
for k in keys:
waveform = dataset[k][:]
tstamp = dataset[k].attrs['tstamp']
evhtype = dataset[k].attrs['htype']
if evhtype == 'Spike':
clus = dataset[k].attrs['cluster']
feat = dataset[k].attrs['features']
spk = SpikeObj(waveform,tstamp,clus,feat)
Data.__addEvent__(spk)
elif evhtype == 'HFO':
channel = dataset[k].attrs['channel']
tstamp_idx = dataset[k].attrs['tstamp_idx']
start_idx = dataset[k].attrs['start_idx']
end_idx = dataset[k].attrs['end_idx']
ths_value = dataset[k].attrs['ths_value']
sample_rate = dataset[k].attrs['sample_rate']
cutoff = dataset[k].attrs['cutoff']
info = dataset[k].attrs['info']
hfo = hfoObj(channel,tstamp,tstamp_idx, waveform,start_idx,end_idx,ths_value,sample_rate,cutoff,info)
Data.__addEvent__(hfo)
h5.close()
return Data
