idx_x = np.arange(0,368)
FileIO.save_mat('0_dataset/AB_156_to_186_walking.mat', is_walking = True)
#%%
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_st_AB_mat(data_path = 'data/AB_dataset/AB_', is_resize = True,
leave_one_num = 1)
#%% UCI DSADS datase
# read data:[label,subjects,segments, time, sensors]
x_mat, y_mat = FileIO.read_UCI_DSADS()
FileIO.save_UCI_DSADS(x_mat, y_mat, file_path = 'data/1_dataset_UCI_DSADS/Raw/')
#%% extract features and output data
x_mat = utils.extract_UCI_features(x_mat)
FileIO.save_UCI_DSADS(x_mat, y_mat, file_path = 'data/1_dataset_UCI_DSADS/Features/')
#%% load UCI data
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_UCI_mat(data_path = 'data/1_dataset_UCI_DSADS/Features/',
feature_length = 6*45, is_resize = True, leave_one_num = 1)
#%%
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_UCI_mat(data_path = 'data/1_dataset_UCI_DSADS/Raw/',
feature_length = 125*45, is_resize = True, leave_one_num = 1)
def dataset_read(source,
target,
batch_size,
is_resize=False,
leave_one_num=-1,
dataset='NW',
sensor_num=0):
S_train = {}
S_val = {}
S_test = {}
T_train = {}
T_val = {}
T_test = {}
if 'NW' == dataset:
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_st_AB_mat(data_path = 'data/AB_dataset/AB_', X_dim = 4,
is_resize = is_resize,
leave_one_num = leave_one_num,
sensor_num = sensor_num)
elif 'UCI' == dataset:
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_UCI_mat(data_path = 'data/1_dataset_UCI_DSADS/Features/',
feature_length = 6*45, X_dim = 4,
is_resize = is_resize, leave_one_num = leave_one_num,
sensor_num = sensor_num)
S_train['imgs'] = x_s_train
S_train['labels'] = y_s_train
T_train['imgs'] = x_t_train
T_train['labels'] = y_t_train
# input target samples for both
S_val['imgs'] = x_s_val
S_val['labels'] = y_s_val
T_val['imgs'] = x_t_val
T_val['labels'] = y_t_val
S_test['imgs'] = x_s_test
S_test['labels'] = y_s_test
T_test['imgs'] = x_t_test
T_test['labels'] = y_t_test
train_loader = UnalignedDataLoader()
train_loader.initialize(S_train, T_train, batch_size, batch_size)
# train_loader.initialize(T_train, S_train, batch_size, batch_size)
data_train = train_loader.load_data()
test_loader = UnalignedDataLoader()
test_loader.initialize(S_val, T_val, batch_size, batch_size)
# test_loader.initialize(T_val, S_val, batch_size, batch_size)
data_val = test_loader.load_data()
final_test_loader = UnalignedDataLoader()
final_test_loader.initialize(S_test, T_test, batch_size, batch_size)
# final_test_loader.initialize(T_test, S_test, batch_size, batch_size)
data_test = final_test_loader.load_data()
print('Target test shape: {}'.format(T_test['labels'].shape))
return data_train, data_val, data_test
def traditional_har(dataset='UCI'):
if 'UCI' == dataset:
sub_num = 8
class_num = 19
feature_length = 6
sensor_num = 45
elif 'NW' == dataset:
sub_num = 10
class_num = 7
acc_s_LDA = np.zeros(sub_num)
acc_t_LDA = np.zeros(sub_num)
acc_s_SVM = np.zeros(sub_num)
acc_t_SVM = np.zeros(sub_num)
acc_s_ANN = np.zeros(sub_num)
acc_t_ANN = np.zeros(sub_num)
for i in range(sub_num):
# load UCI dataset
if 'UCI' == dataset:
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_UCI_mat(data_path = 'data/1_dataset_UCI_DSADS/Features/',
feature_length = feature_length*45, X_dim = 2,
leave_one_num = i)
x_s_train = x_s_train[:, 0:feature_length * sensor_num]
x_s_val = x_s_val[:, 0:feature_length * sensor_num]
x_s_test = x_s_test[:, 0:feature_length * sensor_num]
x_t_train = x_t_train[:, 0:feature_length * sensor_num]
x_t_val = x_t_val[:, 0:feature_length * sensor_num]
x_t_test = x_t_test[:, 0:feature_length * sensor_num]
# load NW dataset
elif 'NW' == dataset:
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_st_AB_mat(data_path = 'data/AB_dataset/AB_', X_dim = 2,
leave_one_num = i)
# print(y_s_train.shape[0] + y_s_val.shape[0] + y_s_test.shape[0],
# y_t_train.shape[0] + y_t_val.shape[0] + y_t_test.shape[0])
# LDA, no domain adaptation
clf = LDA()
clf.fit(x_s_train, y_s_train)
y_s_test_pred = clf.predict(x_s_test)
start = time.clock()
for i in range(8):
out_prediction = clf.predict(x_s_test[[i]])
end = time.clock()
print('LDA: forward time for each segment:%.30f' %
((end - start) / 8.))
acc = accuracy_score(y_s_test, y_s_test_pred)
print("LDA: source domain accuracy: %.2f%%" % acc)
acc_s_LDA[i] = acc
y_t_test_pred = clf.predict(x_t_test)
acc = accuracy_score(y_t_test, y_t_test_pred)
print("LDA: target domain accuracy: %.2f%%" % (acc))
acc_t_LDA[i] = acc
# SVM, no domain adaptation
clf = svm.LinearSVC(max_iter=5000)
clf.fit(x_s_train, y_s_train)
y_s_test_pred = clf.predict(x_s_test)
start = time.clock()
for i in range(8):
out_prediction = clf.predict(x_s_test[[i]])
end = time.clock()
print('SVM: forward time for each segment:%.30f' %
((end - start) / 8.))
acc = accuracy_score(y_s_test, y_s_test_pred)
print("SVM: source domain accuracy: %.2f%%" % acc)
acc_s_SVM[i] = acc
y_t_test_pred = clf.predict(x_t_test)
acc = accuracy_score(y_t_test, y_t_test_pred)
print("SVM: target domain accuracy: %.2f%%" % (acc))
acc_t_SVM[i] = acc
#%% ANN, no domain adaptation
# load UCI dataset
if 'UCI' == dataset:
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_UCI_mat(data_path = 'data/1_dataset_UCI_DSADS/Features/',
is_one_hot = True, is_normalized = True,
feature_length = feature_length*45,
X_dim = 2, leave_one_num = i)
# load NW dataset
if 'NW' == dataset:
x_s_train, y_s_train, x_s_val, y_s_val, x_s_test, y_s_test, \
x_t_train, y_t_train, x_t_val, y_t_val, x_t_test, y_t_test = \
FileIO.load_st_AB_mat(data_path = 'data/AB_dataset/AB_', X_dim = 2,
is_one_hot = True, is_normalized = True,
leave_one_num = i)
clf = MLPClassifier(solver='sgd',
activation='tanh',
learning_rate='adaptive',
learning_rate_init=0.1,
hidden_layer_sizes=(10, class_num),
max_iter=2000)
clf.fit(x_s_train, y_s_train)
y_s_test_pred = clf.predict(x_s_test)
acc = accuracy_score(y_s_test, y_s_test_pred)
start = time.clock()
for i in range(8):
out_prediction = clf.predict(x_s_test[[i]])
end = time.clock()
print('ANN: forward time for each segment:%.30f' %
((end - start) / 8.))
print("ANN: source domain accuracy: %.2f%%" % acc)
acc_s_ANN[i] = acc
y_t_test_pred = clf.predict(x_t_test)
acc = accuracy_score(y_t_test, y_t_test_pred)
print("ANN: target domain accuracy: %.2f%%" % (acc))
acc_t_ANN[i] = acc
print('LDA: mean of test acc in the source domain:', np.mean(acc_s_LDA))
print('LDA: mean of test acc in the target domain:', np.mean(acc_t_LDA))
print('SVM: mean of test acc in the source domain:', np.mean(acc_s_SVM))
print('SVM: mean of test acc in the target domain:', np.mean(acc_t_SVM))
print('ANN: mean of test acc in the source domain:', np.mean(acc_s_ANN))
print('ANN: mean of test acc in the target domain:', np.mean(acc_t_ANN))
return np.transpose(np.c_[acc_s_LDA, acc_t_LDA, acc_s_SVM, acc_t_SVM,
acc_s_ANN, acc_t_ANN])
