def __init__(self, opt, type_of_data, mydir=None, noisiness=-1):
MNISTDataset.__init__(self, opt, type_of_data, mydir)
if noisiness == -1 or noisiness >= self.num_classes:
raise Exception(
'Noisiness -1 is not allowed, please provide a vaid arguement')
self.noisiness = noisiness
self.allign_data_for_func()
def __init__(self, opt, type_of_data, mydir = None,\ additional_constraint_on_data = lambda a,b: True, \ relational_func = lambda a,b: a and b,\ domain_constraints = lambda a: True, y_classes = 2): MNISTDataset.__init__(self, opt, type_of_data, mydir) self.additional_constraint_on_data = additional_constraint_on_data self.relational_func = relational_func self.domain_constraints = domain_constraints self.y_classes = y_classes self.allign_data_for_func(additional_constraint_on_data, relational_func, domain_constraints)
def __init__(self,
opt,
type_of_data,
mydir=None,
additional_constraint_on_data=True,
relational_func=lambda a, b: a and b):
MNISTDataset.__init__(self, opt, type_of_data, mydir)
self.additional_constraint_on_data = additional_constraint_on_data
self.relational_func = relational_func
self.allign_data_for_func(additional_constraint_on_data,
relational_func)
def __init__(self, opt, type_of_data, mydir=None):
MNISTDataset.__init__(self, opt, type_of_data, mydir)
self.allign_data_for_func()
def create_transfered(opt):
type_of_data = 'test'
mnist_train = MNISTDataset(opt,
'train',
trim_data=False,
image_size=(28, 28))
mnist_valid = MNISTDataset(opt,
'valid',
trim_data=False,
image_size=(28, 28))
mnist_input_data = np.concatenate((mnist_train.inputs, mnist_valid.inputs),
axis=0)
mnist_target_data = np.concatenate(
(mnist_train.targets, mnist_valid.targets), axis=0)
print(mnist_input_data.shape, mnist_target_data.shape)
usps_train = USPSDataset(opt,
'train',
trim_data=False,
image_size=(28, 28))
usps_valid = USPSDataset(opt,
'valid',
trim_data=False,
image_size=(28, 28))
usps_input_data = np.concatenate((usps_train.inputs, usps_valid.inputs),
axis=0)
usps_target_data = np.concatenate((usps_train.targets, usps_valid.targets),
axis=0)
print(usps_input_data.shape, usps_input_data.shape)
return
possible_idxs_larger = list(range(len(larger_data_set)))
possible_idxs_smaller = list(range(len(smaller_data_set)))
inputs = []
targets = []
labels = []
for i in range(len(smaller_data_set)):
smaller_idx_idx = random.choice(range(len(possible_idxs_smaller)))
smaller_idx = possible_idxs_smaller[smaller_idx_idx]
smaller_label = smaller_data_set.targets[smaller_idx]
# print(possible_idxs_smaller, smaller_idx, smaller_label)
del possible_idxs_smaller[smaller_idx_idx]
for larger_idx_idx, larger_idx in enumerate(possible_idxs_larger):
larger_label = larger_data_set.targets[larger_idx]
if smaller_label == larger_label:
break
if smaller_label != larger_label:
continue
del possible_idxs_larger[larger_idx_idx]
inputs.append(smaller_data_set.inputs[smaller_idx])
targets.append(larger_data_set.inputs[larger_idx])
labels.append(smaller_label)
usps = np.array(inputs)
mnist = np.array(targets)
labels = np.array(labels)
print('smaller: {}, larger: {}'.format(smaller_data_set, larger_data_set))
print('usps ', inputs.shape, ' mnist ', mnist.shape, ' label ',
labels.shape)