from typing import Iterator, List, Optional, Union
from collections import Counter
import logging
from operator import itemgetter
from random import choices, sample
import numpy as np
import torch
from torch.utils.data import DistributedSampler
from torch.utils.data.sampler import BatchSampler, Sampler
from catalyst.contrib.utils.misc import find_value_ids
from catalyst.data.dataset.torch import DatasetFromSampler
[docs]class BalanceClassSampler(Sampler):
"""Allows you to create stratified sample on unbalanced classes.
Args:
labels: list of class label for each elem in the dataset
mode: Strategy to balance classes.
Must be one of [downsampling, upsampling]
"""
[docs] def __init__(
self, labels: List[int], mode: Union[str, int] = "downsampling"
):
"""Sampler initialisation."""
super().__init__(labels)
labels = np.array(labels)
samples_per_class = {
label: (labels == label).sum() for label in set(labels)
}
self.lbl2idx = {
label: np.arange(len(labels))[labels == label].tolist()
for label in set(labels)
}
if isinstance(mode, str):
assert mode in ["downsampling", "upsampling"]
if isinstance(mode, int) or mode == "upsampling":
samples_per_class = (
mode
if isinstance(mode, int)
else max(samples_per_class.values())
)
else:
samples_per_class = min(samples_per_class.values())
self.labels = labels
self.samples_per_class = samples_per_class
self.length = self.samples_per_class * len(set(labels))
[docs] def __iter__(self) -> Iterator[int]:
"""
Yields:
indices of stratified sample
"""
indices = []
for key in sorted(self.lbl2idx):
replace_flag = self.samples_per_class > len(self.lbl2idx[key])
indices += np.random.choice(
self.lbl2idx[key], self.samples_per_class, replace=replace_flag
).tolist()
assert len(indices) == self.length
np.random.shuffle(indices)
return iter(indices)
[docs] def __len__(self) -> int:
"""
Returns:
length of result sample
"""
return self.length
[docs]class BalanceBatchSampler(Sampler):
"""
This kind of sampler can be used for both metric learning and
classification task.
Sampler with the given strategy for the C unique classes dataset:
- Selection P of C classes for the 1st batch
- Selection K instances for each class for the 1st batch
- Selection P of C - P remaining classes for 2nd batch
- Selection K instances for each class for the 2nd batch
- ...
The epoch ends when there are no classes left.
So, the batch sise is P * K except the last one.
Thus, in each epoch, all the classes will be selected once, but this
does not mean that all the instances will be selected during the epoch.
One of the purposes of this sampler is to be used for
forming triplets and pos/neg pairs inside the batch.
To guarante existance of these pairs in the batch,
P and K should be > 1. (1)
Behavior in corner cases:
- If a class does not contain K instances,
a choice will be made with repetition.
- If C % P == 1 then one of the classes should be dropped
otherwise statement (1) will not be met.
This type of sampling can be found in the classical paper of Person Re-Id,
where P equals 32 and K equals 4:
`In Defense of the Triplet Loss for Person Re-Identification`_.
Args:
labels: list of classes labeles for each elem in the dataset
p: number of classes in a batch, should be > 1
k: number of instances of each class in a batch, should be > 1
.. _In Defense of the Triplet Loss for Person Re-Identification:
https://arxiv.org/abs/1703.07737
"""
[docs] def __init__(self, labels: Union[List[int], np.ndarray], p: int, k: int):
"""Sampler initialisation."""
super().__init__(self)
classes = set(labels)
assert isinstance(p, int) and isinstance(k, int)
assert (1 < p <= len(classes)) and (1 < k)
assert all(
n > 1 for n in Counter(labels).values()
), "Each class shoud contain at least 2 instances to fit (1)"
self._labels = labels
self._p = p
self._k = k
self._batch_size = self._p * self._k
self._classes = classes
# to satisfy statement (1)
num_classes = len(self._classes)
if num_classes % self._p == 1:
self._num_epoch_classes = num_classes - 1
else:
self._num_epoch_classes = num_classes
@property
def batch_size(self) -> int:
"""
Returns:
this value should be used in DataLoader as batch size
"""
return self._batch_size
@property
def batches_in_epoch(self) -> int:
"""
Returns:
number of batches in an epoch
"""
return int(np.ceil(self._num_epoch_classes / self._p))
[docs] def __len__(self) -> int:
"""
Returns:
number of samples in an epoch
"""
return self._num_epoch_classes * self._k
[docs] def __iter__(self) -> Iterator[int]:
"""
Returns:
indeces for sampling dataset elems during an epoch
"""
inds = []
for cls_id in sample(self._classes, self._num_epoch_classes):
all_cls_inds = find_value_ids(self._labels, cls_id)
# we've checked in __init__ that this value must be > 1
num_samples_exists = len(all_cls_inds)
if num_samples_exists < self._k:
selected_inds = sample(
all_cls_inds, k=num_samples_exists
) + choices(all_cls_inds, k=self._k - num_samples_exists)
else:
selected_inds = sample(all_cls_inds, k=self._k)
inds.extend(selected_inds)
return iter(inds)
[docs]class DynamicBalanceClassSampler(Sampler):
"""
This kind of sampler can be used for classification tasks with significant
class imbalance.
The idea of this sampler that we start with the original class distribution
and gradually move to uniform class distribution like with downsampling.
Let's define :math: D_i = #C_i/ #C_min where :math: #C_i is a size of class
i and :math: #C_min is a size of the rarest class, so :math: D_i define
class distribution. Also define :math: g(n_epoch) is a exponential
scheduler. On each epoch current :math: D_i calculated as
:math: current D_i = D_i ^ g(n_epoch),
after this data samples according this distribution.
Notes:
In the end of the training, epochs will contain only
min_size_class * n_classes examples. So, possible it will not
necessary to do validation on each epoch. For this reason use
ControlFlowCallback.
Examples:
>>> import torch
>>> import numpy as np
>>> from catalyst.data import DynamicBalanceClassSampler
>>> from torch.utils import data
>>> features = torch.Tensor(np.random.random((200, 100)))
>>> labels = np.random.randint(0, 4, size=(200,))
>>> sampler = DynamicBalanceClassSampler(labels)
>>> labels = torch.LongTensor(labels)
>>> dataset = data.TensorDataset(features, labels)
>>> loader = data.dataloader.DataLoader(dataset, batch_size=8)
>>> for batch in loader:
>>> b_features, b_labels = batch
Sampler was inspired by https://arxiv.org/abs/1901.06783
"""
[docs] def __init__(
self,
labels: List[Union[int, str]],
exp_lambda: float = 0.9,
start_epoch: int = 0,
max_d: Optional[int] = None,
mode: Union[str, int] = "downsampling",
ignore_warning: bool = False,
):
"""
Args:
labels: list of labels for each elem in the dataset
exp_lambda: exponent figure for schedule
start_epoch: start epoch number, can be useful for multi-stage
experiments
max_d: if not None, limit on the difference between the most
frequent and the rarest classes, heuristic
mode: number of samples per class in the end of training. Must be
"downsampling" or number. Before change it, make sure that you
understand how does it work
ignore_warning: ignore warning about min class size
"""
assert isinstance(start_epoch, int)
assert 0 < exp_lambda < 1, "exp_lambda must be in (0, 1)"
super().__init__(labels)
self.exp_lambda = exp_lambda
if max_d is None:
max_d = np.inf
self.max_d = max_d
self.epoch = start_epoch
labels = np.array(labels)
samples_per_class = Counter(labels)
self.min_class_size = min(samples_per_class.values())
if self.min_class_size < 100 and not ignore_warning:
logger = logging.getLogger(__name__)
logger.warning(
f"the smallest class contains only"
f" {self.min_class_size} examples. At the end of"
f" training, epochs will contain only"
f" {self.min_class_size * len(samples_per_class)}"
f" examples"
)
self.original_d = {
key: value / self.min_class_size
for key, value in samples_per_class.items()
}
self.label2idxes = {
label: np.arange(len(labels))[labels == label].tolist()
for label in set(labels)
}
if isinstance(mode, int):
self.min_class_size = mode
else:
assert mode == "downsampling"
self.labels = labels
self._update()
def _update(self) -> None:
"""
Update d coefficients
Returns: None
"""
current_d = {
key: min(value ** self._exp_scheduler(), self.max_d)
for key, value in self.original_d.items()
}
samples_per_classes = {
key: int(value * self.min_class_size)
for key, value in current_d.items()
}
self.samples_per_classes = samples_per_classes
self.length = np.sum(list(samples_per_classes.values()))
self.epoch += 1
def _exp_scheduler(self) -> float:
return self.exp_lambda ** self.epoch
[docs] def __iter__(self) -> Iterator[int]:
"""
Yields:
indices of stratified sample
"""
indices = []
for key in sorted(self.label2idxes):
samples_per_class = self.samples_per_classes[key]
replace_flag = samples_per_class > len(self.label2idxes[key])
indices += np.random.choice(
self.label2idxes[key], samples_per_class, replace=replace_flag
).tolist()
assert len(indices) == self.length
np.random.shuffle(indices)
self._update()
return iter(indices)
[docs] def __len__(self) -> int:
"""
Returns:
length of result sample
"""
return self.length
[docs]class MiniEpochSampler(Sampler):
"""
Sampler iterates mini epochs from the dataset used by ``mini_epoch_len``.
Args:
data_len: Size of the dataset
mini_epoch_len: Num samples from the dataset used in one
mini epoch.
drop_last: If ``True``, sampler will drop the last batches
if its size would be less than ``batches_per_epoch``
shuffle: one of ``"always"``, ``"real_epoch"``, or `None``.
The sampler will shuffle indices
> "per_mini_epoch" - every mini epoch (every ``__iter__`` call)
> "per_epoch" -- every real epoch
> None -- don't shuffle
Example:
>>> MiniEpochSampler(len(dataset), mini_epoch_len=100)
>>> MiniEpochSampler(len(dataset), mini_epoch_len=100, drop_last=True)
>>> MiniEpochSampler(len(dataset), mini_epoch_len=100,
>>> shuffle="per_epoch")
"""
[docs] def __init__(
self,
data_len: int,
mini_epoch_len: int,
drop_last: bool = False,
shuffle: str = None,
):
"""Sampler initialisation."""
super().__init__(None)
self.data_len = int(data_len)
self.mini_epoch_len = int(mini_epoch_len)
self.steps = int(data_len / self.mini_epoch_len)
self.state_i = 0
has_reminder = data_len - self.steps * mini_epoch_len > 0
if self.steps == 0:
self.divider = 1
elif has_reminder and not drop_last:
self.divider = self.steps + 1
else:
self.divider = self.steps
self._indices = np.arange(self.data_len)
self.indices = self._indices
self.end_pointer = max(self.data_len, self.mini_epoch_len)
if not (shuffle is None or shuffle in ["per_mini_epoch", "per_epoch"]):
raise ValueError(
"Shuffle must be one of ['per_mini_epoch', 'per_epoch']. "
+ f"Got {shuffle}"
)
self.shuffle_type = shuffle
[docs] def shuffle(self) -> None:
"""Shuffle sampler indices."""
if self.shuffle_type == "per_mini_epoch" or (
self.shuffle_type == "per_epoch" and self.state_i == 0
):
if self.data_len >= self.mini_epoch_len:
self.indices = self._indices
np.random.shuffle(self.indices)
else:
self.indices = np.random.choice(
self._indices, self.mini_epoch_len, replace=True
)
[docs] def __iter__(self) -> Iterator[int]:
"""Iterate over sampler.
Returns:
python iterator
"""
self.state_i = self.state_i % self.divider
self.shuffle()
start = self.state_i * self.mini_epoch_len
stop = (
self.end_pointer
if (self.state_i == self.steps)
else (self.state_i + 1) * self.mini_epoch_len
)
indices = self.indices[start:stop].tolist()
self.state_i += 1
return iter(indices)
[docs] def __len__(self) -> int:
"""
Returns:
int: length of the mini-epoch
"""
return self.mini_epoch_len
[docs]class DynamicLenBatchSampler(BatchSampler):
"""
A dynamic batch length data sampler.
Should be used with `catalyst.utils.trim_tensors`.
Adapted from `Dynamic minibatch trimming to improve BERT training speed`_.
Args:
sampler: Base sampler.
batch_size: Size of minibatch.
drop_last: If ``True``, the sampler will drop the last batch
if its size would be less than ``batch_size``.
Usage example:
>>> from torch.utils import data
>>> from catalyst.data import DynamicLenBatchSampler
>>> from catalyst import utils
>>> dataset = data.TensorDataset(
>>> input_ids, input_mask, segment_ids, labels
>>> )
>>> sampler_ = data.RandomSampler(dataset)
>>> sampler = DynamicLenBatchSampler(
>>> sampler_, batch_size=16, drop_last=False
>>> )
>>> loader = data.DataLoader(dataset, batch_sampler=sampler)
>>> for batch in loader:
>>> tensors = utils.trim_tensors(batch)
>>> b_input_ids, b_input_mask, b_segment_ids, b_labels = \
>>> tuple(t.to(device) for t in tensors)
.. _`Dynamic minibatch trimming to improve BERT training speed`:
https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/discussion/94779
"""
[docs] def __iter__(self):
"""
Iteration over BatchSampler.
"""
buckets = [[]] * 100
yielded = 0
for idx in self.sampler:
count_zeros = torch.sum(self.sampler.data_source[idx][0] == 0)
count_zeros = int(count_zeros / 64)
if len(buckets[count_zeros]) == 0:
buckets[count_zeros] = []
buckets[count_zeros].append(idx)
if len(buckets[count_zeros]) == self.batch_size:
batch = list(buckets[count_zeros])
yield batch
yielded += 1
buckets[count_zeros] = []
batch = []
leftover = [idx2 for bucket in buckets for idx2 in bucket]
for idx3 in leftover:
batch.append(idx3)
if len(batch) == self.batch_size:
yielded += 1
yield batch
batch = []
if len(batch) > 0 and not self.drop_last:
yielded += 1
yield batch
assert len(self) == yielded, (
"produced an inccorect number of batches. "
+ "expected %i, but yielded %i" % (len(self), yielded)
)
[docs]class DistributedSamplerWrapper(DistributedSampler):
"""
Wrapper over `Sampler` for distributed training.
Allows you to use any sampler in distributed mode.
It is especially useful in conjunction with
`torch.nn.parallel.DistributedDataParallel`. In such case, each
process can pass a DistributedSamplerWrapper instance as a DataLoader
sampler, and load a subset of subsampled data of the original dataset
that is exclusive to it.
.. note::
Sampler is assumed to be of constant size.
"""
[docs] def __init__(
self,
sampler,
num_replicas: Optional[int] = None,
rank: Optional[int] = None,
shuffle: bool = True,
):
"""
Args:
sampler: Sampler used for subsampling
num_replicas (int, optional): Number of processes participating in
distributed training
rank (int, optional): Rank of the current process
within ``num_replicas``
shuffle (bool, optional): If true (default),
sampler will shuffle the indices
"""
super(DistributedSamplerWrapper, self).__init__(
DatasetFromSampler(sampler),
num_replicas=num_replicas,
rank=rank,
shuffle=shuffle,
)
self.sampler = sampler
[docs] def __iter__(self):
"""@TODO: Docs. Contribution is welcome."""
self.dataset = DatasetFromSampler(self.sampler)
indexes_of_indexes = super().__iter__()
subsampler_indexes = self.dataset
return iter(itemgetter(*indexes_of_indexes)(subsampler_indexes))
__all__ = [
"BalanceClassSampler",
"BalanceBatchSampler",
"MiniEpochSampler",
"DistributedSamplerWrapper",
"DynamicBalanceClassSampler",
"DynamicLenBatchSampler",
]