def read_data(path=SAMPLES_PATH, windows=4, overlapping=3, raw_data=False):
# datas 和 labels 都是 list. datas 中的每一项是 shape=(32, 128*windows) 的数组
datas, labels = get_samples_data(path, windows, overlapping)
datas_result = []
for data in datas:
data_list = []
if not raw_data:
# 数据预处理，提取特征(每一个样本处理后的结果是 numpy.narray 且 shape=(features, seq_length))
for window in range(windows): # 1S 为一个单位提取特征
features_list = []
for i in range(32): # 依次处理 32 通道的 EEG 信号
X = data[i, 128 * (window):128 * ((window + 1))]
theta, alpha, beta, gamma = data_filter(
X, params) # 获取各个频率段的数据
features_list.append(differential_entropy(theta))
features_list.append(differential_entropy(alpha))
features_list.append(differential_entropy(beta))
features_list.append(differential_entropy(gamma))
_max = max(features_list)
_min = min(features_list)
data_list.append(
(np.array(features_list).reshape(-1, 1) - _min) /
(_max - _min)) # 0-1化处理
datas_result.append(
np.c_[tuple(data_list)]) # shape=(features, seq_length)
if (raw_data):
datas_result = datas
del datas # 释放内存
assert len(datas_result) == len(labels)
if not raw_data:
np.save("./data_set/datas_features", datas_result)
np.save("./data_set/label_features", labels)
return (datas_result, labels)
def read_data(path=SAMPLES_PATH, train=True, test=False, seed=42, raw_data=False):
'''默认读取训练集数据（880个）'''
# datas 和 labels 都是 list. datas 中的每一项是 shape=(40, 7680) 的数组
datas, labels = get_samples_data(path, train=train, test=test, seed=42)
# 移除前 3S 的数据，利用后 60S 的数据
datas_result = []
time_window = 4 # 时间窗口大小是 2s
for data in datas:
data_list = []
if not raw_data:
# 数据预处理，提取特征(每一个样本处理后的结果是 numpy.narray 且 shape=(features, seq_length))
for window in range(60 // time_window):
features_list = []
for i in range(32): # 依次处理 32 通道的 EEG 信号
X = data[i, 128*(window*time_window):128*(time_window*(window+1))]
theta, alpha, beta, gamma = data_filter(X, params) # 获取各个频率段的数据
features_list.append(differential_entropy(theta))
features_list.append(differential_entropy(alpha))
features_list.append(differential_entropy(beta))
features_list.append(differential_entropy(gamma))
_max = max(features_list)
_min = min(features_list)
data_list.append((np.array(features_list).reshape(-1, 1) - _min)/(_max - _min)) # 0-1化处理
datas_result.append(np.c_[tuple(data_list)]) # shape=(features, seq_length)
else:
datas_result.append(data[:32,:])
datas.clear() # 释放内存
if train:
np.save("./data_set/datas_train", datas_result)
np.save("./data_set/label_train", labels)
if test:
np.save("./data_set/datas_test", datas_result)
np.save("./data_set/label_test", labels)
return (datas_result, labels)
def read_data(path=SAMPLES_PATH,
train=True,
test=False,
seed=42,
raw_data=False):
# datas 和 labels 都是 list. datas 中的每一项是 shape=(40, 8064) 的数组
datas, labels = get_samples_data(path, train=train, test=test, seed=42)
# 移除前 3S 的数据，利用后 60S 的数据
datas_result = []
time_window = 4 # 时间窗口大小是 2s
for data in datas:
data_list = []
if not raw_data:
# 数据预处理，提取特征(每一个样本处理后的结果是 numpy.narray 且 shape=(features, seq_length))
for window in range(60 // time_window):
features_list = []
for i in range(32): # 依次处理 32 通道的 EEG 信号
X = data[i, 128 * (3 + window * time_window):128 *
(3 + time_window * (window + 1))]
theta, alpha, beta, gamma = data_filter(
X, params) # 获取各个频率段的数据
features_list.append(differential_entropy(theta))
features_list.append(differential_entropy(alpha))
features_list.append(differential_entropy(beta))
features_list.append(differential_entropy(gamma))
data_list.append(np.array(features_list).reshape(-1, 1))
datas_result.append(
np.c_[tuple(data_list)]) # shape=(features, seq_length)
else:
datas_result.append(data[:32, 128 * 3:])
datas.clear() # 释放内存
return (datas_result, labels)