# ******************************************************************************
# Copyright 2017-2019 Intel Corporation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ******************************************************************************
import torch
from typing import List
[docs]class ParallelDataset(torch.utils.data.Dataset):
def __init__(self, *datasets):
self.datasets = datasets
def __getitem__(self, i):
return tuple(d[i] for d in self.datasets)
def __len__(self):
return min(len(d) for d in self.datasets)
[docs]class ConcatTensorDataset(torch.utils.data.Dataset):
r"""Dataset as a concatenation of multiple TensorDataset datasets with same number of tensors.
Each sample will be retrieved by indexing tensors along the first dimension.
Arguments:
dataset (TensorDataset): dataset to which rest datasets will be concatinated.
datasets (List[TensorDataset]): datasets to concat to the dataset.
"""
def __init__(
self,
dataset: torch.utils.data.TensorDataset,
datasets: List[torch.utils.data.TensorDataset],
):
tensors = dataset.tensors
for ds in datasets:
concat_tensors = []
for i, t in enumerate(ds.tensors):
concat_tensors.append(torch.cat((tensors[i], t), 0))
tensors = concat_tensors
assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors)
self.tensors = tensors
def __getitem__(self, index):
return tuple(tensor[index] for tensor in self.tensors)
def __len__(self):
return self.tensors[0].size(0)
[docs]class CombinedTensorDataset(torch.utils.data.Dataset):
r"""Dataset as a concatenation of tensor datasets with different number of
tensors (labeled dataset/ unlabeled dataset). Labels of unlabeled dataset will
be represented as a tensor of zeros.
Each sample will be retrieved by indexing tensors along the first dimension.
Arguments:
datasets (List[TensorDataset]): datasets to concat.
"""
def __init__(self, datasets: List[torch.utils.data.TensorDataset]):
max_ds_len = max([len(ds.tensors) for ds in datasets])
tensors = ()
# match tensors count
for ds in datasets:
if len(ds.tensors) < max_ds_len: # no labels
ds.tensors += (torch.tensor(torch.zeros(ds.tensors[0].shape), dtype=int),)
# concat
for i in range(max_ds_len):
tensors += (torch.cat([ds.tensors[i] for ds in datasets], dim=0),)
assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors)
self.tensors = tensors
def __getitem__(self, index):
return tuple(tensor[index] for tensor in self.tensors)
def __len__(self):
return self.tensors[0].size(0)