def get_dataset(subject, ins):
ch_fs_instances = []
ch_tags_instances = []
samples_trial = 7000 # samples per trial
no_ch = 14
sr = 128
instance_len = sr * 2 # to create sub instances of 2 sec
EEG_data = spio.loadmat(
'EID-M.mat',
squeeze_me=True)['eeg_close_ubicomp_8sub'].T # (15, 168000)
_labels = EEG_data[14, :] # len(168000)
s_index = consecutive_index(EEG_data[no_ch, :], subject)[0] # [0, 20999]
for s_instance in range(s_index[0] + 14000, s_index[1] + 1, samples_trial):
_instance = EEG_data[:no_ch, s_instance:s_instance +
samples_trial] # (14, 7000)
max_instances = _instance.shape[
1] / instance_len # this is not necessary, but I added it just FYI, 27
for _i in range(0, ins): # sub instances
if _i < max_instances:
index_start, index_end = instance_len * _i, instance_len * (
_i + 1)
sub_instance = _instance[:, index_start:index_end]
ch_fs_instances.append(get_features_emd(sub_instance, sr))
ch_tags_instances.append('subject_{0}'.format(subject))
return {"data": ch_fs_instances, "target": ch_tags_instances}
def main():
logging.info(" ***** EMD, CHANNELS:7, FEATURES: energy ***** \n")
INSTANCES = [10, 20, 40, 60]
sr = 200
ch = [1, 42, 46, 51, 52, 54, 55]
for ins in INSTANCES:
logging.info(" -------- Instance: {0} --------".format(ins))
ch_fs_instances = []
ch_tags_instances = []
for subject in range(1, 27): # 26
for session in range(1, 5): # 4
s_s_chs = get_subdataset(subject, session)
_index = [
i + 1 for i, d in enumerate(s_s_chs[:, -1]) if d == 1
]
instances = get_samples(_index, s_s_chs, sr)
for f_instance in range(0, ins):
if f_instance not in [15, 21, 23, 45]:
instance = preprocessing_P300(instances, f_instance,
sr, ch)
ch_fs_instances.append(get_features_emd(
instance, sr)) # CHANNELS: 14
ch_tags_instances.append('subject_{0}'.format(subject))
dataset = {"data": ch_fs_instances, "target": ch_tags_instances}
dataTraining = dataset['data']
targetTraining = dataset['target']
result = selector(dataTraining, targetTraining)
logging.info("Best classifier {0} with accuracy {1} \n".format(
result['classifier'], result['accuracy']))
# saving the model
model_name = 'P300_EMD_ch7_energy_ins%02d.sav' % ins
pickle.dump(result["model"], open(model_name, 'wb'))
def main():
logging.info(" ***** EMD, CHANNELS:7, FEATURES: energy + hht ***** ")
logging.info("--------------no preprocessing-------------- \n")
INSTANCES = [10, 20]
sr = 200
ch = [1, 42, 46, 51, 52, 54, 55]
# ch = ch = [0, 1, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 33, 34, 36, 36, 38, 40,
# 41, 42, 44, 46, 48, 50, 51, 52, 53, 54, 55] 32 channels
for ins in INSTANCES:
logging.info(" -------- Instance: {0} --------".format(ins))
ch_fs_instances = []
ch_tags_instances = []
for subject in range(1, 27): # 26
for session in range(1, 5): # 4
s_s_chs = get_subdataset(subject, session)
_index = [i + 1 for i, d in enumerate(s_s_chs[:, -1]) if d == 1]
instances = get_samples(_index, s_s_chs, sr)
for f_instance in range(0, ins):
if f_instance not in [15, 21, 23, 45]:
instance = np.array(instances[f_instance, :, 1:-1]).transpose()
ins7 = instance[ch, :]
ch_fs_instances.append(get_features_emd(ins7, sr)) # CHANNELS: 14
ch_tags_instances.append('subject_{0}'.format(subject))
dataset = {"data": ch_fs_instances, "target": ch_tags_instances}
dataTraining = dataset['data']
targetTraining = dataset['target']
result = selector(dataTraining, targetTraining)
logging.info("Best classifier {0} with accuracy {1} \n".format(result['classifier'], result['accuracy']))
# saving the model
model_name = 'P300_EMD_ch7_energy_hht_nopre_ins%02d.sav' % ins
pickle.dump(result["model"], open(model_name, 'wb'))
def main():
logging.info(" ***** Resting state, EMD, FEATURES: energy ***** \n")
ch_fs_instances = []
ch_tags_instances = []
samples_subject = 21000 # samples per subject
samples_trial = 7000 # samples per trial
no_subjects = 8
no_ch = 14
sr = 128
for ins in [10, 20]:
logging.info(" -------- Instance: {0} --------".format(ins))
instance_len = sr * 2 # to create sub instances of 2 sec
EEG_data = spio.loadmat(
'EID-M.mat',
squeeze_me=True)['eeg_close_ubicomp_8sub'].T # (15, 168000)
_labels = EEG_data[14, :] # len(168000)
for subject in range(1, no_subjects + 1):
s_index = consecutive_index(EEG_data[no_ch, :],
subject)[0] # [0, 20999],
for s_instance in range(s_index[0], s_index[0] + samples_trial + 1,
samples_trial):
_instance = EEG_data[:no_ch, s_instance:s_instance +
samples_trial] # (14, 7000)
max_instances = _instance.shape[
1] / instance_len # this is not necessary, but I added it just FYI, 27
for _i in range(0, ins): # sub instances
if _i < max_instances:
index_start, index_end = instance_len * _i, instance_len * (
_i + 1)
sub_instance = _instance[:, index_start:index_end]
sub_ins = preprocessing_resting(sub_instance)
ch_fs_instances.append(get_features_emd(sub_ins, sr))
ch_tags_instances.append('subject_{0}'.format(subject))
dataset = {"data": ch_fs_instances, "target": ch_tags_instances}
dataTraining = dataset['data']
targetTraining = dataset['target']
result = selector(dataTraining, targetTraining)
logging.info("Best classifier {0} with accuracy {1}".format(
result['classifier'], result['accuracy']))
# saving the model
model_name = 'EMD_resting_energy_ins%02d.sav' % ins
pickle.dump(result["model"], open(model_name, 'wb'))
def get_dataset(subject, ins):
sr = 200
ch = [1, 42, 46, 51, 52, 54, 55] # 7 channels
# ch = ch = [0, 1, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 33, 34, 36, 36, 38, 40,
# 41, 42, 44, 46, 48, 50, 51, 52, 53, 54, 55] 32 channels
session = 5
ch_fs_instances = []
ch_tags_instances = []
s_s_chs = get_subdataset(subject, session)
_index = [i + 1 for i, d in enumerate(s_s_chs[:, -1]) if d == 1]
instances = get_samples(_index, s_s_chs, sr)
for f_instance in range(0, ins):
if f_instance not in [15, 21, 23, 45]:
instance = preprocessing_P300(instances, f_instance, sr, ch)
ch_fs_instances.append(get_features_emd(instance,
sr)) # CHANNELS: 14
ch_tags_instances.append('subject_{0}'.format(subject))
return {"data": ch_fs_instances, "target": ch_tags_instances}
def main():
logging.info(" ***** Resting state, EMD , FEATURES: energy ***** ")
logging.info(" ---------- No preprocessing ---------- \n \n")
ch_fs_instances = []
ch_tags_instances = []
samples_subject = 21000 # samples per subject
samples_trial = 7000 # samples per trial
no_subjects = 8
no_ch = 14
sr = 128
for ins in [10, 20]:
logging.info(" -------- Instance: {0} --------".format(ins))
instance_len = sr * 2 # to create sub instances of 2 sec
EEG_data = spio.loadmat('EID-M.mat', squeeze_me=True)['eeg_close_ubicomp_8sub'].T # (15, 168000)
_labels = EEG_data[14, :] # len(168000)
for subject in range(1, no_subjects + 1):
s_index = consecutive_index(EEG_data[no_ch, :], subject)[0] # [0, 20999],
for s_instance in range(s_index[0], s_index[0] + samples_trial + 1, samples_trial):
_instance = EEG_data[:no_ch, s_instance:s_instance + samples_trial] # (14, 7000)
max_instances = _instance.shape[1] / instance_len # this is not necessary, but I added it just FYI, 27
for _i in range(0, ins): # sub instances
if _i < max_instances:
index_start, index_end = instance_len * _i, instance_len * (_i + 1)
sub_instance = _instance[:, index_start:index_end]
ch_fs_instances.append(get_features_emd(sub_instance, sr))
ch_tags_instances.append('subject_{0}'.format(subject))
dataset = {"data": ch_fs_instances, "target": ch_tags_instances}
dataTraining = dataset['data']
targetTraining = dataset['target']
print("targetTraining: ", targetTraining)
