Source code for olympus.datasets.sampling.group

"""BSD 3-Clause License

Copyright (c) Soumith Chintala 2016,
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

* Redistributions of source code must retain the above copyright notice, this
  list of conditions and the following disclaimer.

* Redistributions in binary form must reproduce the above copyright notice,
  this list of conditions and the following disclaimer in the documentation
  and/or other materials provided with the distribution.

* Neither the name of the copyright holder nor the names of its
  contributors may be used to endorse or promote products derived from
  this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
"""
from collections import defaultdict
from torch.utils.data.sampler import BatchSampler, Sampler


[docs]class GroupedBatchSampler(BatchSampler): """Wraps another sampler to yield a mini-batch of indices. It enforces that the batch only contain elements from the same group. It also tries to provide mini-batches which follows an ordering which is as close as possible to the ordering from the original sampler. Parameters ---------- sampler: Sampler Base sampler. group_ids: list[int] If the sampler produces indices in range [0, N), `group_ids` must be a list of `N` ints which contains the group id of each sample. The group ids must be a continuous set of integers starting from 0, i.e. they must be in the range [0, num_groups). batch_size: int Size of mini-batch. """ def __init__(self, sampler, group_ids, batch_size): if not isinstance(sampler, Sampler): raise ValueError( "sampler should be an instance of " "torch.utils.data.Sampler, but got sampler={}".format(sampler) ) self.sampler = sampler self.group_ids = group_ids self.batch_size = batch_size def __iter__(self): buffer_per_group = defaultdict(list) samples_per_group = defaultdict(list) num_batches = 0 for idx in self.sampler: group_id = self.group_ids[idx] buffer_per_group[group_id].append(idx) samples_per_group[group_id].append(idx) if len(buffer_per_group[group_id]) == self.batch_size: yield buffer_per_group[group_id] num_batches += 1 del buffer_per_group[group_id] assert len(buffer_per_group[group_id]) < self.batch_size # now we have run out of elements that satisfy # the group criteria, let's return the remaining # elements so that the size of the sampler is # deterministic expected_num_batches = len(self) num_remaining = expected_num_batches - num_batches if num_remaining > 0: # for the remaining batches, take first the buffers with largest number # of elements for group_id, _ in sorted(buffer_per_group.items(), key=lambda x: len(x[1]), reverse=True): remaining = self.batch_size - len(buffer_per_group[group_id]) buffer_per_group[group_id].extend( samples_per_group[group_id][:remaining]) assert len(buffer_per_group[group_id]) == self.batch_size yield buffer_per_group[group_id] num_remaining -= 1 if num_remaining == 0: break assert num_remaining == 0 def __len__(self): return len(self.sampler) // self.batch_size