def reg_loader_subset(self, indices):
"""Returns the torch dataloader over the regularization set (unsupervised examples only). """
# transform the unsupervised set the same way as the training set:
mini_batch_size = self.mini_batch_size()
return torch.utils.data.DataLoader(self.unsupset, batch_size=mini_batch_size, shuffle=False,
sampler=ProtectedSubsetRandomSampler(indices),
num_workers=2)
def loader_for_dataset(self, dataset):
mini_batch_size = self.mini_batch_size()
return torch.utils.data.DataLoader(
dataset,
batch_size=mini_batch_size,
shuffle=False,
sampler=ProtectedSubsetRandomSampler(range(0, len(dataset))))
def train_loader_subset(self, indices):
"""Returns the torch dataloader over the training set, shuffled,
but limited to the example range start-end."""
mini_batch_size = self.mini_batch_size()
trainloader = torch.utils.data.DataLoader(self.trainset, batch_size=mini_batch_size, shuffle=False,
sampler=ProtectedSubsetRandomSampler(indices),
num_workers=2)
return trainloader
def test_loader_subset(self, indices):
"""Returns the torch dataloader over the test set. """
mini_batch_size = self.mini_batch_size()
return torch.utils.data.DataLoader(
self.testset,
sampler=ProtectedSubsetRandomSampler(indices),
batch_size=mini_batch_size,
shuffle=False,
num_workers=2)
def loader_for_dataset(self, dataset):
mini_batch_size = self.mini_batch_size()
return torch.utils.data.DataLoader(
dataset,
batch_size=mini_batch_size,
shuffle=False,
sampler=ProtectedSubsetRandomSampler(range(0, len(dataset))),
collate_fn=stl10_collate,
num_workers=self.num_workers)
def test_loader_subset(self, indices):
"""Returns the torch dataloader over the test set, limiting to the examples
identified by the indices. """
mini_batch_size = self.mini_batch_size()
return torch.utils.data.DataLoader(
self._testset,
collate_fn=stl10_collate,
sampler=ProtectedSubsetRandomSampler(indices),
batch_size=mini_batch_size,
shuffle=False,
num_workers=self.num_workers)