def make_subjects_hdf5(**kwargs):
dataset = SSVEPExo()
data = dataset.get_data()
for subject in data:
session_data = data[subject]['session_0']
make_subject_hdf5(session_data,
output=f'data/{subject}.hdf5',
**kwargs)
import mne
import numpy as np
from mne.time_frequency import psd_welch
from numpy.lib import stride_tricks
from scipy.signal import filtfilt, lfilter
from moabb.datasets import SSVEPExo
dataset = SSVEPExo()
event_id = dataset.event_id
def compute_mega_raw(raw, filt_method, frequencies, order):
from timeflux_dsp.utils.filters import construct_iir_filter
if filt_method == 'lfilter':
filt_method = lfilter
elif filt_method == 'filtfilt':
filt_method = filtfilt
else:
raise ValueError(f'Unknown filt_method {filt_method}')
# read from info
ch_names = np.array(raw.info['ch_names'])
info_kind_dict = {2 : 'eeg', 3: 'stim'}
ch_types = np.array([info_kind_dict.get(info['kind']) for info in raw.info['chs']])
sfreq = raw.info['sfreq']
_stim_data = raw._data[ch_types=='stim', :]
_eeg_data = raw._data[ch_types=='eeg', :]
_eeg_ch_names = ch_names[ch_types=='eeg']
n_ch = len(_eeg_ch_names)
# loop over frrquencies, filter and concatenate channels
############################################################################### # Loading dataset # --------------- # # We will load the data from the first 2 subjects of the SSVEP_Exo dataset # and compare two algorithms on this set. One of the algorithm could only # process class associated with a stimulation frequency, we will thus drop # the resting class. As the resting class is the last defined class, picking # the first three classes (out of four) allows to focus only on the stimulation # frequency. n_subject = 2 # for i in range(n_subject): # SSVEPExo()._get_single_subject_data(i + 1) dataset = SSVEPExo() dataset.subject_list = dataset.subject_list[:n_subject] interval = dataset.interval paradigm = SSVEP(fmin=10, fmax=25, n_classes=3) paradigm_fb = FilterBankSSVEP(filters=None, n_classes=3) # Classes are defined by the frequency of the stimulation, here we use # the first two frequencies of the dataset, 13 and 17 Hz. # The evaluation function uses a LabelEncoder, transforming them # to 0 and 1 freqs = paradigm.used_events(dataset) ############################################################################## # Create pipelines # ----------------
moabb.set_log_level("info")
###############################################################################
# Loading dataset
# ---------------
#
# We will load the data from the first 2 subjects of the SSVEP_Exo dataset
# and compare two algorithms on this set. One of the algorithm could only
# process class associated with a stimulation frequency, we will thus drop
# the resting class. As the resting class is the last defined class, picking
# the first three classes (out of four) allows to focus only on the stimulation
# frequency.
n_subject = 2
for i in range(n_subject):
SSVEPExo()._get_single_subject_data(i + 1)
dataset = SSVEPExo()
dataset.subject_list = dataset.subject_list[:n_subject]
interval = dataset.interval
###############################################################################
# Choose paradigm
# ---------------
#
# We define the paradigms (SSVEP and FilterBankSSVEP) and use the dataset
# SSVEPExo. The SSVEP paradigm applied a bandpass filter (10-25 Hz) on
# the data while the FilterBankSSVEP paradigm uses as many bandpass filters as
# there are stimulation frequencies (here 2). For each stimulation frequency
# the EEG is filtered with a 1 Hz-wide bandpass filter centered on the
# frequency. This results in n_classes copies of the signal, filtered for each
# class, as used in filterbank motor imagery paradigms.
n_trials, n_freqs, n_channels, n_times = sig_ext.shape
X_sig = sig_ext.reshape((n_trials, n_channels * n_freqs, n_times))
X_sig = np.concatenate([X_sig[i, :, :] for i in range(X_sig.shape[0])],
axis=1)
cov = Covariances(estimator='lwf').transform(
X_sig.reshape((1, *(X_sig.shape))))
cov_list.append(cov)
return np.concatenate(cov_list, axis=0)
metric = 'logdet' #'riemann' #
##############################################################
# Load SSVEP database
paradigm = BaseSSVEP()
SSVEPExo().download(update_path=False, verbose=False)
datasets = SSVEPExo()
X, y, metadata = paradigm.get_data(dataset=datasets)
# with gzip.open('SSVEPExo.pkz', 'wb') as f:
# o = {'X':X, 'y':y, 'metadata':metadata}
# pickle.dump(o, f)
# with gzip.open('SSVEPExo.pkz', 'rb') as f:
# o = pickle.load(f)
# X, y, metadata = o['X'], o['y'], o['metadata']
n_subjects = len(np.unique(np.array(metadata['subject'])))
##############################################################
# Aggregate all signals for a class from subjects
# and compute covmats. Then, compute dist between cov and
def timeWindowInit(inlet):
sample, timestamp = inlet.pull_sample()
time_window = np.array(sample)
sample, timestamp = inlet.pull_sample()
sample = np.array(sample)
return np.column_stack((sample, time_window))
###############################################################################
## Option to online retraining
retrain = True
## FIles to offilne base
dataset = SSVEPExo(subjects=11)
interval = dataset.interval
paradigm_fb = FilterBankSSVEP(filters=None, n_classes=4)
filtered_db = paradigm_fb.get_data(dataset, return_epochs=False)
pipeline = make_pipeline(
ExtendedSSVEPSignal(),
Covariances(estimator="lwf"),
TangentSpace(),
LogisticRegression(solver="lbfgs", multi_class="auto"),
)
model = pipeline.fit()