def __init__(self, eeg_training, eeg_test, n_components=2):
"""
Parameters
----------
n_components : int
The number of components to use as relevant features in the CSP algorithm
eeg_training : Eeg
The Eeg object that contains the raw training eeg data separated by class (left and right)
eeg_test : Eeg
The Eeg object that contains the raw test eeg data separated by class (left and right)
"""
self.n_components = n_components
self.__bands = range(QUANTITY_BANDS)
self.__csp_by_band = [
CSP(average_trial_covariance=False, n_components=self.n_components)
for _ in self.__bands
]
self.training_labels = eeg_training.labels
self.test_labels = eeg_test.labels
self.training_features = self.extract_features(eeg_training)
self.test_features = self.extract_features(eeg_test)
self.n_features = self.training_features.shape[1]
if self.n_features != self.test_features.shape[1]:
raise Exception(
"The number of features extracted from the training and test dataset's should be equal"
)
right_data_file = f"{DATA_FOLDER}/{subject}-right.csv"
data = read_eeg_file(left_data_file, right_data_file, TIME_LENGTH,
TIME_WINDOW, EPOCH_SIZE)
# Pre-processing
print("Pre-processing ...")
print("Applying 5º order Butterworth bandpass filter (7-30 Hz)")
b, a = signal.butter(5, [7, 30], btype="bandpass", fs=FS)
data.left_data = signal.filtfilt(b, a, data.left_data, axis=1)
data.right_data = signal.filtfilt(b, a, data.right_data, axis=1)
print("Spatial-filtering ...")
data.X = np.concatenate((data.left_data, data.right_data))
csp = CSP(average_trial_covariance=True, n_components=CSP_COMPONENTS)
csp.fit(data.left_data, data.right_data)
data.Z = np.array([csp.project(x) for x in data.X])
# print(data.Z.shape) #(399,700,2)
# 첫번째 trial의 채널 1
# plot.specgram(pd.DataFrame(data.Z[0]).iloc[:,0], NFFT=64, Fs=250, noverlap=32) # 기존
# plot.specgram(pd.DataFrame(data.Z[0]).iloc[:,0], NFFT=32, Fs=250, noverlap=16)
# plot.show()
# Feature extraction
print("Extracting features ...")
data.F = np.zeros((data.X.shape[0], 2, CSP_COMPONENTS))
for n_epoch in range(0, data.X.shape[0]):
epoch = data.Z[n_epoch]
print(data.left_data.shape)
print(data.right_data.shape)
train_left_index = [index for index in train_index if index < trials]
train_right_index = [index - trials for index in train_index if index >= trials]
X_left_train, X_right_train = data.left_data[train_left_index], data.right_data[train_right_index]
test_left_index = [index for index in test_index if index < trials]
test_right_index = [index - trials for index in test_index if index >= trials]
X_left_test, X_right_test = data.left_data[test_left_index], data.right_data[test_right_index]
y_train, y_test = data.labels[train_index], data.labels[test_index]
# Feature extraction
N_CSP_COMPONENTS = 2
csp_by_band = [CSP(average_trial_covariance=False, n_components=N_CSP_COMPONENTS)
for _ in bands]
left_bands_training = filter_bank(X_left_train)
right_bands_training = filter_bank(X_right_train)
left_bands_test = filter_bank(X_left_test)
right_bands_test = filter_bank(X_right_test)
features_train = None
features_test = None
for n_band in range(quantity_bands):
left_band_training = left_bands_training[n_band]
right_band_training = right_bands_training[n_band]
left_band_test = left_bands_test[n_band]
right_band_test = right_bands_test[n_band]
data = read_eeg_file(left_data_file, right_data_file, TIME_LENGTH,
TIME_WINDOW, EPOCH_SIZE)
# Pre-processing
print("Pre-processing ...")
print("Applying 5º order Butterworth bandpass filter (7-30 Hz)")
b, a = signal.butter(5, [7, 30], btype="bandpass", fs=FS)
# left_data2, right_data2 전달된 상태
data.left_data = signal.filtfilt(b, a, data.left_data, axis=1)
data.right_data = signal.filtfilt(b, a, data.right_data, axis=1)
print("Spatial-filtering ...")
data.X = np.concatenate((data.left_data, data.right_data))
csp = CSP(average_trial_covariance=True, n_components=CSP_COMPONENTS)
csp.fit(data.left_data, data.right_data)
# print(csp.W.shape) #(3,2)
# print(data.left_data.shape) #(159, 500, 3)
# print(data.right_data.shape)# (165, 500, 3)
# print(data.X.shape)
# counts=0
counts = 303
imgtype = 'left'
print("왼쪽 데이터 이미지 생성 시작")
for i in data.X:
print("i_check{}".format(i))
counts += 1
outlines="skirt",
show_names=True)
# Load training data
print("Loading data ...")
left_data_file = f"{DATA_FOLDER}/left-hand-subject-{subject}.csv"
right_data_file = f"{DATA_FOLDER}/right-hand-subject-{subject}.csv"
eeg = read_eeg_file(left_data_file, right_data_file, TIME_LENGTH, TIME_WINDOW,
EPOCH_SIZE)
apply_bandpass_filter(eeg)
n_epoch = 25
left_epoch = eeg.left_data[n_epoch, :, :]
right_epoch = eeg.right_data[n_epoch, :, :]
ch_names = ["FC1", "FC2", "C1", "C2", "C3", "C4", "CP1", "CP2"]
#
# plot_topomap(ch_names, left_epoch)
# plot_topomap(ch_names, right_epoch)
csp = CSP(average_trial_covariance=True, n_components=8)
csp.fit(eeg.left_data, eeg.right_data)
left_epoch_projected = csp.project(left_epoch)
right_epoch_projected = csp.project(right_epoch)
plot_topomap(ch_names, left_epoch_projected)
plot_topomap(ch_names, right_epoch_projected)