def load_data(_path, _ftype, coords, joints, cycles=3):
file_feature = os.path.join(_path, 'features' + _ftype + '.h5')
ff = h5py.File(file_feature, 'r')
file_label = os.path.join(_path, 'labels' + _ftype + '.h5')
fl = h5py.File(file_label, 'r')
data_list = []
num_samples = len(ff.keys())
time_steps = 0
labels = np.empty(num_samples)
for si in range(num_samples):
ff_group_key = list(ff.keys())[si]
data_list.append(list(ff[ff_group_key])) # Get the data
time_steps_curr = len(ff[ff_group_key])
if time_steps_curr > time_steps:
time_steps = time_steps_curr
labels[si] = fl[list(fl.keys())[si]][()]
data = np.empty((num_samples, time_steps * cycles, joints * coords))
for si in range(num_samples):
data_list_curr = np.tile(
data_list[si], (int(np.ceil(time_steps / len(data_list[si]))), 1))
for ci in range(cycles):
data[si, time_steps * ci:time_steps *
(ci + 1), :] = data_list_curr[0:time_steps]
data = common.get_affective_features(
np.reshape(data,
(data.shape[0], data.shape[1], joints, coords)))[:, :, :48]
data_train, data_test, labels_train, labels_test = train_test_split(
data, labels, test_size=0.1)
return data, labels, data_train, labels_train, data_test, labels_test
def load_data(_path, _ftype_real, _ftype_synth, coords, joints, cycles=3):
file_feature_real = os.path.join(_path, 'features' + _ftype_real + '.h5')
ffr = h5py.File(file_feature_real, 'r')
file_label_real = os.path.join(_path, 'labels' + _ftype_real + '.h5')
flr = h5py.File(file_label_real, 'r')
file_feature_synth = os.path.join(_path, 'features' + _ftype_synth + '.h5')
ffs = h5py.File(file_feature_synth, 'r')
file_label_synth = os.path.join(_path, 'labels' + _ftype_synth + '.h5')
fls = h5py.File(file_label_synth, 'r')
data_list = []
num_samples_real = len(ffr.keys())
num_samples_synth = len(ffs.keys())
num_samples = num_samples_real + num_samples_synth
time_steps = 0
labels_real = np.empty(num_samples_real)
labels_synth = np.empty(num_samples_synth)
for si in range(num_samples_real):
ffr_group_key = list(ffr.keys())[si]
data_list.append(list(ffr[ffr_group_key])) # Get the data
time_steps_curr = len(ffr[ffr_group_key])
if time_steps_curr > time_steps:
time_steps = time_steps_curr
labels_real[si] = flr[list(flr.keys())[si]][()]
for si in range(num_samples_synth):
ffs_group_key = list(ffs.keys())[si]
data_list.append(list(ffs[ffs_group_key])) # Get the data
time_steps_curr = len(ffs[ffs_group_key])
if time_steps_curr > time_steps:
time_steps = time_steps_curr
labels_synth[si] = fls[list(fls.keys())[si]][()]
labels = np.concatenate((labels_real, labels_synth), axis=0)
data = np.empty((num_samples, time_steps*cycles, joints*coords))
for si in range(num_samples):
data_list_curr = np.tile(data_list[si], (int(np.ceil(time_steps / len(data_list[si]))), 1))
for ci in range(cycles):
data[si, time_steps * ci:time_steps * (ci + 1), :] = data_list_curr[0:time_steps]
data = common.get_affective_features(np.reshape(data, (data.shape[0], data.shape[1], joints, coords)))[:, :, :48]
data_train, data_test, labels_train, labels_test = train_test_split(data, labels, test_size=0.1)
return data, labels, data_train, labels_train, data_test, labels_test