def __getitem__(self, index):
if isinstance(index, int):
raise ValueError(
'The index of a `CombinationMetaDataset` must be '
'a tuple of integers, and not an integer. For example, call '
'`dataset[({0})]` to get a task with classes from 0 to {1} '
'(got `{2}`).'.format(
', '.join([
str(idx) for idx in range(self.num_classes_per_task)
]), self.num_classes_per_task - 1, index))
assert len(index) == self.num_classes_per_task
datasets = [self.dataset[i] for i in index]
# Use deepcopy on `Categorical` target transforms, to avoid any side
# effect across tasks.
task = ConcatTask(datasets,
self.num_classes_per_task,
target_transform=wrap_transform(
self.target_transform,
self._copy_categorical,
transform_type=Categorical))
if self.dataset_transform is not None:
task = self.dataset_transform(task)
return task
def __init__(self,
dataset,
batch_size=1,
shuffle=True,
sampler=None,
batch_sampler=None,
num_workers=0,
collate_fn=None,
pin_memory=False,
drop_last=False,
timeout=0,
worker_init_fn=None):
if collate_fn is None:
collate_fn = no_collate
if isinstance(dataset, CombinationMetaDataset) and (sampler is None):
if shuffle:
sampler = CombinationRandomSampler(dataset)
else:
sampler = CombinationSequentialSampler(dataset)
shuffle = False
elif isinstance(dataset, (list, tuple)):
if shuffle:
sampler = MultiCombinationRandomSampler(dataset)
dataset = ConcatTask(
dataset,
dataset[0].num_classes,
target_transform=dataset[0].target_transform)
else:
raise NotImplementedError()
shuffle = False
# sample combinations within each dataset
# assumes indices are concatenated
# RandomSampler
super(MetaDataLoader, self).__init__(dataset,
batch_size=batch_size,
shuffle=shuffle,
sampler=sampler,
batch_sampler=batch_sampler,
num_workers=num_workers,
collate_fn=collate_fn,
pin_memory=pin_memory,
drop_last=drop_last,
timeout=timeout,
worker_init_fn=worker_init_fn)