class NodeTaskUSPS(NodeTaskDictCluster):
"""Basic USPS task where the digits are considered "domains"
The digits (0, 1, ..., 9) are regarded as domains (to be separated by unsupervised clustering). Based on NodeTaskMNISTColor10 from DomainLab.
"""
@property
def list_str_y(self):
"""
USPS task has no labels (digits are considered domains)
"""
return mk_dummy_label_list_str("dummy", 10)
@property
def isize(self):
"""
:return: image size object storing image channels, height, width.
"""
return ImSize(1, 16, 16)
[docs] def get_list_domains(self):
"""
Get list of domain names
:return: list of domain names
"""
return mk_dummy_label_list_str("digit", 10)
[docs] def get_dset_by_domain(self, args, na_domain, split=False):
"""Get a dataset by digit
:param args: command line arguments
:param na_domain: domain name
:param split: whether a training/validation split is performed (the
training split portion will be determined by args.split); for test
set, no need to split; args.split: by default, split is set to be
zero which in python can be evaluated in if statement, in which case,
no separate validation set will be created. Otherwise, this argument
is the percentage of the data to be used as training set, while the
rest will be used as validation set.
:return: training dataset, validation dataset
"""
ratio_split = float(args.split) if split else False
# by default, split is set to be zero which in python can
# be evaluated in if statement, in which case, no validation
# set will be created. Otherwise, this argument is
# the split ratio
trans = [transforms.Resize((16, 16)), transforms.ToTensor()]
ind_global = self.get_list_domains().index(na_domain)
dset = DsetUSPS(digit=ind_global, args=args, list_transforms=trans)
# split dset into training and validation sets
if ratio_split > 0:
train_len = int(len(dset) * ratio_split)
val_len = len(dset) - train_len
train_set, val_set = random_split(dset, [train_len, val_len])
else:
train_set = dset
val_set = dset
return train_set, val_set