def __init__(self, fn, channels, to_freq, montage_channels, montage, idx=1, csp_time=[0, 0.3], a_time=0.5):
self.data = sp.signalParser(fn)
self.tags = self.data.get_p300_tags(idx=idx, samples=False)
signal = self.data.prep_signal(to_freq, channels, montage_channels, montage)
signal2 = np.zeros(signal.shape)
self.fs = to_freq
signal2 = np.zeros(signal.shape)
self.wrong_tags = self.data.get_p300_tags(idx=idx, rest=True, samples=False)
artifacts_data = np.zeros([len(channels), self.fs * a_time, len(self.tags)])
for i, p in enumerate(self.tags):
artifacts_data[..., i] = signal[
:, p * self.data.sampling_frequency : (p + a_time) * self.data.sampling_frequency
]
self.a_features, self.bands = artifactsCalibration(artifacts_data, self.data.sampling_frequency)
signal2 = np.zeros(signal.shape)
self.fs = to_freq
self.channels = channels
self.idx = idx
b, a = ss.butter(3, 2 * 1.0 / self.fs, btype="high")
b_l, a_l = ss.butter(3, 20.0 * 2 / self.fs, btype="low")
for e in xrange(len(self.channels)):
tmp = filtfilt(b, a, signal[e, :])
signal2[e, :] = filtfilt(b_l, a_l, tmp)
self.signal_original = signal2
self.t1, self.t2 = self.show_mean(csp_time, "Cz", dont_plot=False)
P, vals = self.train_csp(signal2, [self.t1, self.t2])
self.P = P
self.signal = np.dot(P[:, 0], signal2)
def analyse(self, blink, data):
"""Fired as often as defined in hashtable configuration:
# Define from which moment in time (ago) we want to get samples (in seconds)
'ANALYSIS_BUFFER_FROM':
# Define how many samples we wish to analyse every tick (in seconds)
'ANALYSIS_BUFFER_COUNT':
# Define a tick duration (in seconds).
'ANALYSIS_BUFFER_EVERY':
# To SUMP UP - above default values (0.5, 0.4, 0.25) define that
# every 0.25s we will get buffer of length 0.4s starting from a sample
# that we got 0.5s ago.
# Some more typical example would be for values (0.5, 0.5 0.25).
# In that case, every 0.25 we would get buffer of samples from 0.5s ago till now.
data format is determined by another hashtable configuration:
# possible values are: 'PROTOBUF_SAMPLES', 'NUMPY_CHANNELS'
# it indicates format of buffered data returned to analysis
# NUMPY_CHANNELS is a numpy 2D array with data divided by channels
# PROTOBUF_SAMPLES is a list of protobuf Sample() objects
'ANALYSIS_BUFFER_RET_FORMAT'
"""
self.logger.debug("Got data to analyse... after: " +
str(time.time() - self.last_time))
self.logger.debug("first and last value: " + str(data[0][0]) + " - " +
str(data[0][-1]))
self.logger.debug("DATA SIZE: " + str(data.shape))
self.logger.debug("BLINK index / ts / real_ts: " + str(blink.index) +
" / " + str(blink.timestamp) + " / " +
str(time.time()))
self.last_time = time.time()
#Wszystko dalej powinno się robić dla każdego nowego sygnału
signal = np.dot(self.montage_matrix.T, data)
self.logger.debug("AFTER MONTAGE SIGNAL SIZE: " + str(signal.shape))
tmp_sig = np.zeros(signal.shape)
for e in xrange(len(self.montage_matrix.T)):
tmp = filtfilt(self.b, self.a, signal[e, :])
tmp_sig[e, :] = filtfilt(self.b_l, self.a_l, tmp)
if artifactsClasifier(tmp_sig, self.a_features, self.bands, self.fs):
#2 Montujemy CSP
#sig = np.dot(self.q.P[:, 0], tmp_sig)
#3 Klasyfikacja: indeks pola albo -1, gdy nie ma detekcji
ix = self.analyze.analyze(tmp_sig, blink.index, tr=self.treshold)
if ix >= 0:
self.send_func(ix)
else:
self.logger.info("Got -1 ind- no decision")
else:
self.logger.info("Got -1 ind- no decision")
def analyse(self, blink, data):
"""Fired as often as defined in hashtable configuration:
# Define from which moment in time (ago) we want to get samples (in seconds)
'ANALYSIS_BUFFER_FROM':
# Define how many samples we wish to analyse every tick (in seconds)
'ANALYSIS_BUFFER_COUNT':
# Define a tick duration (in seconds).
'ANALYSIS_BUFFER_EVERY':
# To SUMP UP - above default values (0.5, 0.4, 0.25) define that
# every 0.25s we will get buffer of length 0.4s starting from a sample
# that we got 0.5s ago.
# Some more typical example would be for values (0.5, 0.5 0.25).
# In that case, every 0.25 we would get buffer of samples from 0.5s ago till now.
data format is determined by another hashtable configuration:
# possible values are: 'PROTOBUF_SAMPLES', 'NUMPY_CHANNELS'
# it indicates format of buffered data returned to analysis
# NUMPY_CHANNELS is a numpy 2D array with data divided by channels
# PROTOBUF_SAMPLES is a list of protobuf Sample() objects
'ANALYSIS_BUFFER_RET_FORMAT'
"""
self.logger.debug("Got data to analyse... after: "+str(time.time()-self.last_time))
self.logger.debug("first and last value: "+str(data[0][0])+" - "+str(data[0][-1]))
self.logger.debug("DATA SIZE: "+str(data.shape))
self.logger.debug("BLINK index / ts / real_ts: "+str(blink.index)+" / "+str(blink.timestamp)+" / "+str(time.time()))
self.last_time = time.time()
#Wszystko dalej powinno się robić dla każdego nowego sygnału
signal = np.dot(self.montage_matrix.T, data)
self.logger.debug("AFTER MONTAGE SIGNAL SIZE: "+str(signal.shape))
tmp_sig = np.zeros(signal.shape)
for e in xrange(len(self.montage_matrix.T)):
tmp = filtfilt(self.b,self.a, signal[e, :])
tmp_sig[e, :] = filtfilt(self.b_l, self.a_l, tmp)
if artifactsClasifier(tmp_sig, self.a_features, self.bands, self.fs):
#2 Montujemy CSP
#sig = np.dot(self.q.P[:, 0], tmp_sig)
#3 Klasyfikacja: indeks pola albo -1, gdy nie ma detekcji
ix = self.analyze.analyze(tmp_sig, blink.index, tr=self.treshold)
if ix >= 0:
self.send_func(ix)
else:
self.logger.info("Got -1 ind- no decision")
else:
self.logger.info("Got -1 ind- no decision")
def __init__(self,
fn,
channels,
to_freq,
montage_channels,
montage,
idx=1,
csp_time=[0, 0.3],
a_time=0.5):
self.data = sp.signalParser(fn)
self.tags = self.data.get_p300_tags(idx=idx, samples=False)
signal = self.data.prep_signal(to_freq, channels, montage_channels,
montage)
signal2 = np.zeros(signal.shape)
self.fs = to_freq
signal2 = np.zeros(signal.shape)
self.wrong_tags = self.data.get_p300_tags(idx=idx,
rest=True,
samples=False)
artifacts_data = np.zeros(
[len(channels), self.fs * a_time,
len(self.tags)])
for i, p in enumerate(self.tags):
artifacts_data[...,
i] = signal[:, p *
self.data.sampling_frequency:(p +
a_time) *
self.data.sampling_frequency]
self.a_features, self.bands = artifactsCalibration(
artifacts_data, self.data.sampling_frequency)
signal2 = np.zeros(signal.shape)
self.fs = to_freq
self.channels = channels
self.idx = idx
b, a = ss.butter(3, 2 * 1.0 / self.fs, btype='high')
b_l, a_l = ss.butter(3, 20.0 * 2 / self.fs, btype='low')
for e in xrange(len(self.channels)):
tmp = filtfilt(b, a, signal[e, :])
signal2[e, :] = filtfilt(b_l, a_l, tmp)
self.signal_original = signal2
self.t1, self.t2 = self.show_mean(csp_time, 'Cz', dont_plot=False)
P, vals = self.train_csp(signal2, [self.t1, self.t2])
self.P = P
self.signal = np.dot(P[:, 0], signal2)
