from typing import Union
import numpy as np
import torch
__all__ = ["random_temporal_delete", "RandomTemporalDelete"]
[文档]
def random_temporal_delete(
x_seq: Union[torch.Tensor, np.ndarray], T_remain: int, batch_first: bool
):
r"""
**API Language:**
:ref:`中文 <random_temporal_delete-cn>` | :ref:`English <random_temporal_delete-en>`
----
.. _random_temporal_delete-cn:
* **中文**
在 `Deep Residual Learning in Spiking Neural Networks <https://arxiv.org/abs/2102.04159>`_ 中使用的随机时间删除数据增强。
:param x_seq: 一个序列, 其 `shape = [T, N, *]`, 其中 `T` 是序列长度, `N` 是批次大小
:type x_seq: Union[torch.Tensor, np.ndarray]
:param T_remain: 剩余的长度
:type T_remain: int
:param batch_first: 如果 `True`, `x_seq` 将被视为 `shape = [N, T, *]`
:type batch_first: bool
:return: 长度为 `T_remain` 的序列, 通过随机移除 `T - T_remain` 个切片获得
:rtype: Union[torch.Tensor, np.ndarray]
:raises ValueError: 当 ``T_remain`` 为负数, 或大于当前时间维长度时由 ``numpy.random.choice`` 抛出。
----
.. _random_temporal_delete-en:
* **English**
The random temporal delete data augmentation used in `Deep Residual Learning in Spiking Neural Networks <https://arxiv.org/abs/2102.04159>`_.
:param x_seq: a sequence with `shape = [T, N, *]`, where `T` is the sequence length and `N` is the batch size
:type x_seq: Union[torch.Tensor, np.ndarray]
:param T_remain: the remained length
:type T_remain: int
:param batch_first: if `True`, `x_seq` will be regarded as `shape = [N, T, *]`
:type batch_first: bool
:return: the sequence with length `T_remain`, which is obtained by randomly removing `T - T_remain` slices
:rtype: Union[torch.Tensor, np.ndarray]
:raises ValueError: raised by ``numpy.random.choice`` when ``T_remain`` is
negative or larger than the current time dimension length.
----
* **代码示例 | Example**
.. code-block:: python
import torch
from spikingjelly.datasets import random_temporal_delete
T = 8
T_remain = 5
N = 4
x_seq = torch.arange(0, N * T).view([N, T])
print("x_seq=\n", x_seq)
print(
"random_temporal_delete(x_seq)=\n",
random_temporal_delete(x_seq, T_remain, batch_first=True),
)
Outputs:
.. code-block:: shell
x_seq=
tensor([[ 0, 1, 2, 3, 4, 5, 6, 7],
[ 8, 9, 10, 11, 12, 13, 14, 15],
[16, 17, 18, 19, 20, 21, 22, 23],
[24, 25, 26, 27, 28, 29, 30, 31]])
random_temporal_delete(x_seq)=
tensor([[ 0, 1, 4, 6, 7],
[ 8, 9, 12, 14, 15],
[16, 17, 20, 22, 23],
[24, 25, 28, 30, 31]])
"""
if batch_first:
sec_list = np.random.choice(x_seq.shape[1], T_remain, replace=False)
else:
sec_list = np.random.choice(x_seq.shape[0], T_remain, replace=False)
sec_list.sort()
if batch_first:
return x_seq[:, sec_list]
else:
return x_seq[sec_list]
[文档]
class RandomTemporalDelete(torch.nn.Module):
r"""
**API Language:**
:ref:`中文 <RandomTemporalDelete-cn>` | :ref:`English <RandomTemporalDelete-en>`
----
.. _RandomTemporalDelete-cn:
* **中文**
:func:`random_temporal_delete` 的 ``torch.nn.Module`` 封装。前向传播时会使用构造时给定的
``T_remain`` 和 ``batch_first`` 调用 :func:`random_temporal_delete`。
----
.. _RandomTemporalDelete-en:
* **English**
A ``torch.nn.Module`` wrapper around :func:`random_temporal_delete`. During
``forward``, it calls :func:`random_temporal_delete` with the ``T_remain``
and ``batch_first`` values provided at construction time.
"""
def __init__(self, T_remain: int, batch_first: bool):
r"""
**API Language:**
:ref:`中文 <RandomTemporalDelete.__init__-cn>` | :ref:`English <RandomTemporalDelete.__init__-en>`
----
.. _RandomTemporalDelete.__init__-cn:
* **中文**
* **中文**
在 `Deep Residual Learning in Spiking Neural Networks <https://arxiv.org/abs/2102.04159>`_ 中使用的随机时间删除数据增强。
详见 :func:`random_temporal_delete`。
:param T_remain: 剩余的长度
:type T_remain: int
:param batch_first: 如果 `True`, `x_seq` 将被视为 `shape = [N, T, *]`
:type batch_first: bool
----
.. _RandomTemporalDelete.__init__-en:
* **English**
* **English**
The random temporal delete data augmentation used in `Deep Residual Learning in Spiking Neural Networks <https://arxiv.org/abs/2102.04159>`_.
Refer to :func:`random_temporal_delete` for more details.
:param T_remain: the remained length
:type T_remain: int
:param batch_first: if `True`, `x_seq` will be regarded as `shape = [N, T, *]`
:type batch_first: bool
:return: None
:rtype: None
"""
super().__init__()
self.T_remain = T_remain
self.batch_first = batch_first
[文档]
def forward(
self, x_seq: Union[torch.Tensor, np.ndarray]
) -> Union[torch.Tensor, np.ndarray]:
r"""
**API Language:**
:ref:`中文 <RandomTemporalDelete.forward-cn>` | :ref:`English <RandomTemporalDelete.forward-en>`
----
.. _RandomTemporalDelete.forward-cn:
* **中文**
使用当前模块保存的 ``T_remain`` 和 ``batch_first`` 配置, 对输入序列执行
:func:`random_temporal_delete`。
:param x_seq: 输入序列。其时间维布局由 ``batch_first`` 决定。
:type x_seq: Union[torch.Tensor, np.ndarray]
:return: 随机删除时间切片后的序列。
:rtype: Union[torch.Tensor, np.ndarray]
:raises ValueError: 当 ``self.T_remain`` 非法时, 由
:func:`random_temporal_delete` 内部的 ``numpy.random.choice`` 抛出
----
.. _RandomTemporalDelete.forward-en:
* **English**
Apply :func:`random_temporal_delete` to the input sequence with the
``T_remain`` and ``batch_first`` configuration stored in this module.
:param x_seq: input sequence. The time-dimension layout is determined
by ``batch_first``.
:type x_seq: Union[torch.Tensor, np.ndarray]
:return: sequence after random temporal deletion.
:rtype: Union[torch.Tensor, np.ndarray]
:raises ValueError: raised by ``numpy.random.choice`` inside
:func:`random_temporal_delete` when ``self.T_remain`` is invalid
"""
return random_temporal_delete(x_seq, self.T_remain, self.batch_first)
