def load_tfd_fold(fold=0):
"""
Return train, val, test data for the particular fold.
"""
# note that the training set used here is the 'unlabeled' set in TFD
x_train, _, _ = tfd.load_proper_fold(fold, 'unlabeled', scale=True)
x_val, _, _ = tfd.load_proper_fold(fold, 'val', scale=True)
x_test, _, _ = tfd.load_proper_fold(fold, 'test', scale=True)
imsz = np.prod(x_train.shape[1:])
return x_train.reshape(x_train.shape[0], imsz), \
x_val.reshape(x_val.shape[0], imsz), \
x_test.reshape(x_test.shape[0], imsz)
def load_tfd_fold(fold=0):
"""
Return train, val, test data for the particular fold.
"""
# note that the training set used here is the 'unlabeled' set in TFD
x_train, _, _ = tfd.load_proper_fold(fold, 'unlabeled', scale=True)
x_val, _, _ = tfd.load_proper_fold(fold, 'val', scale=True)
x_test, _, _ = tfd.load_proper_fold(fold, 'test', scale=True)
imsz = np.prod(x_train.shape[1:])
return x_train.reshape(x_train.shape[0], imsz), \
x_val.reshape(x_val.shape[0], imsz), \
x_test.reshape(x_test.shape[0], imsz)
def load_tfd_all_folds(set_name='val', n_folds=5):
x = []
for i_fold in range(n_folds):
#xx, _, _ = tfd.load_fold(i_fold, set_name, scale=True)
xx, _, _ = tfd.load_proper_fold(i_fold, set_name, scale=True)
x.append(xx.reshape(xx.shape[0], np.prod(xx.shape[1:])))
return x
def load_tfd_all_folds(set_name='val', n_folds=5):
x = []
for i_fold in range(n_folds):
#xx, _, _ = tfd.load_fold(i_fold, set_name, scale=True)
xx, _, _ = tfd.load_proper_fold(i_fold, set_name, scale=True)
x.append(xx.reshape(xx.shape[0], np.prod(xx.shape[1:])))
return x
def load_train_data(dataset='mnist'):
if dataset == 'mnist':
train_data, _, _ = mnistio.load_data()
elif dataset == 'tfd':
train_data, _, _ = tfd.load_proper_fold(0, 'unlabeled', scale=True)
train_data = train_data.reshape(train_data.shape[0], np.prod(train_data.shape[1:]))
return train_data
def load_train_data(dataset='mnist'):
if dataset == 'mnist':
train_data, _, _ = mnistio.load_data()
elif dataset == 'tfd':
train_data, _, _ = tfd.load_proper_fold(0, 'unlabeled', scale=True)
train_data = train_data.reshape(train_data.shape[0],
np.prod(train_data.shape[1:]))
return train_data
