"SonyAIBORobotSurface2", "StarLightCurves", "Strawberry", "SwedishLeaf", "Symbols", "SyntheticControl",
"ToeSegmentation1", "ToeSegmentation2", "Trace", "TwoLeadECG", "TwoPatterns", "UWaveGestureLibraryAll",
"UWaveGestureLibraryX", "UWaveGestureLibraryY", "UWaveGestureLibraryZ", "Wafer", "Wine", "WordSynonyms",
"Worms", "WormsTwoClass", "Yoga", "ACSF1", "AllGestureWiimoteX", "AllGestureWiimoteY", "AllGestureWiimoteZ",
"BME", "Chinatown", "Crop", "DodgerLoopDay", "DodgerLoopGame", "DodgerLoopWeekend", "EOGHorizontalSignal",
"EOGVerticalSignal", "EthanolLevel", "FreezerRegularTrain", "FreezerSmallTrain", "Fungi", "GestureMidAirD1",
"GestureMidAirD2", "GestureMidAirD3", "GesturePebbleZ1", "GesturePebbleZ2", "GunPointAgeSpan",
"GunPointMaleVersusFemale", "GunPointOldVersusYoung", "HouseTwenty", "InsectEPGRegularTrain", "InsectEPGSmallTrain",
"MelbournePedestrian", "MixedShapesRegularTrain", "MixedShapesSmallTrain", "PickupGestureWiimoteZ",
"PigAirwayPressure", "PigArtPressure", "PigCVP", "PLAID", "PowerCons","Rock","SemgHandGenderCh2",
"SemgHandMovementCh2","SemgHandSubjectCh2","ShakeGestureWiimoteZ","SmoothSubspace","UMD"]
total = 0
for i, dataset in enumerate(datasets):
args.dataset = dataset
nb_class = ds.nb_classes(args.dataset)
nb_dims = ds.nb_dims(args.dataset)
# Load data
x_train, y_train, x_test, y_test = get_datasets(args)
nb_timesteps = int(x_train.shape[1] / nb_dims)
input_shape = (nb_timesteps , nb_dims)
# Process data
x_test = x_test.reshape((-1, input_shape[0], input_shape[1]))
x_train = x_train.reshape((-1, input_shape[0], input_shape[1]))
y_test = to_categorical(ds.class_offset(y_test, args.dataset), nb_class)
y_train = to_categorical(ds.class_offset(y_train, args.dataset), nb_class)
# Augment data
def load_ucr2018(dataset_path, dataset_name):
##################
# load raw data
##################
nb_class = ds.nb_classes(dataset_name)
nb_dims = ds.nb_dims(dataset_name)
if dataset_name in ['MFPT', 'XJTU']:
x = np.load("{}/{}/{}_data.npy".format(dataset_path, dataset_name,
dataset_name))
y = np.load("{}/{}/{}_label.npy".format(dataset_path, dataset_name,
dataset_name))
(x_train, x_test) = (x[:100], x[100:])
(y_train, y_test) = (y[:100], y[100:])
elif dataset_name in ['EpilepticSeizure']:
data_x, data_y = get_EpilepticSeizure(dataset_path, dataset_name)
(x_train, x_test) = (data_x[:int(0.5 * data_x.shape[0])],
data_x[int(0.5 * data_x.shape[0]):])
(y_train, y_test) = (data_y[:int(0.5 * data_x.shape[0])],
data_y[int(0.5 * data_x.shape[0]):])
else:
x_train, y_train, x_test, y_test = TSC_data_loader(
dataset_path, dataset_name)
nb_timesteps = int(x_train.shape[1] / nb_dims)
input_shape = (nb_timesteps, nb_dims)
############################################
# Combine all train and test data for resample
############################################
x_all = np.concatenate((x_train, x_test), axis=0)
y_all = np.concatenate((y_train, y_test), axis=0)
ts_idx = list(range(x_all.shape[0]))
np.random.shuffle(ts_idx)
x_all = x_all[ts_idx]
y_all = y_all[ts_idx]
label_idxs = np.unique(y_all)
class_stat_all = {}
for idx in label_idxs:
class_stat_all[idx] = len(np.where(y_all == idx)[0])
print("[Stat] All class: {}".format(class_stat_all))
test_idx = []
val_idx = []
train_idx = []
for idx in label_idxs:
target = list(np.where(y_all == idx)[0])
nb_samp = int(len(target))
test_idx += target[:int(nb_samp * 0.2)]
val_idx += target[int(nb_samp * 0.2):int(nb_samp * 0.4)]
train_idx += target[int(nb_samp * 0.4):]
x_test = x_all[test_idx]
y_test = y_all[test_idx]
x_val = x_all[val_idx]
y_val = y_all[val_idx]
x_train = x_all[train_idx]
y_train = y_all[train_idx]
label_idxs = np.unique(y_train)
class_stat = {}
for idx in label_idxs:
class_stat[idx] = len(np.where(y_train == idx)[0])
# print("[Stat] Train class: {}".format(class_stat))
print("[Stat] Train class: mean={}, std={}".format(
np.mean(list(class_stat.values())), np.std(list(class_stat.values()))))
label_idxs = np.unique(y_val)
class_stat = {}
for idx in label_idxs:
class_stat[idx] = len(np.where(y_val == idx)[0])
# print("[Stat] Test class: {}".format(class_stat))
print("[Stat] Val class: mean={}, std={}".format(
np.mean(list(class_stat.values())), np.std(list(class_stat.values()))))
label_idxs = np.unique(y_test)
class_stat = {}
for idx in label_idxs:
class_stat[idx] = len(np.where(y_test == idx)[0])
# print("[Stat] Test class: {}".format(class_stat))
print("[Stat] Test class: mean={}, std={}".format(
np.mean(list(class_stat.values())), np.std(list(class_stat.values()))))
########################################
# Data Split End
########################################
# Process data
x_test = x_test.reshape((-1, input_shape[0], input_shape[1]))
x_val = x_val.reshape((-1, input_shape[0], input_shape[1]))
x_train = x_train.reshape((-1, input_shape[0], input_shape[1]))
print("Train:{}, Test:{}, Class:{}".format(x_train.shape, x_test.shape,
nb_class))
# Normalize
x_train_max = np.max(x_train)
x_train_min = np.min(x_train)
x_train = 2. * (x_train - x_train_min) / (x_train_max - x_train_min) - 1.
# Test is secret
x_val = 2. * (x_val - x_train_min) / (x_train_max - x_train_min) - 1.
x_test = 2. * (x_test - x_train_min) / (x_train_max - x_train_min) - 1.
return x_train, y_train, x_val, y_val, x_test, y_test, nb_class, class_stat_all