Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

【Hackathon 5th No.24】Add SubsetRandomSampler #648

Merged
merged 2 commits into from
Oct 18, 2023
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
138 changes: 138 additions & 0 deletions rfcs/APIs/20230925_api_design_for_SubsetRandomSampler.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,138 @@
# paddle.utils.data.SubsetRandomSampler API 增强设计文档

| API名称 | paddle.utils.data.SubsetRandomSampler|
| ------------ | -------------------------------------- |
| 提交作者 | Asthestarsfalll |
| 提交时间 | 2023-09-25 |
| 版本号 | V1.0 |
| 依赖飞桨版本 | develop |
| 文件名 | 20230925_api_design_for_SubsetRandomSampler.md |


# 一、概述
## 1、相关背景

`SubsetRandomSampler`子集随机采样器,支持从数据集的指定子集中随机选择样本,可以用于将数据集分成训练集和验证集等子集。

## 2、功能目标

增加`paddle.utils.data.SubsetRandomSampler`,实现对给定子集的随机采样。

## 3、意义

飞桨支持`SubsetRandomSampler`。

# 二、飞桨现状

目前paddle缺少相关功能实现,但是有类似功能的API,只需要继承`Sampler` 基类,并重写`__iter__`和`__iter__`方法实现相关功能即可。


# 三、业内方案调研

## PyTorch

Pytorch中有API`torch.utils.data.SubsetRandomSampler(indices, generator)`.在pytorch中,介绍为:

```
Samples elements randomly from a given list of indices, without replacement.
```

### 实现方法

实现方法较为简单,其子集通过给定的`indices`确定,采样时只需要从`indices`中采样便可达到相应的效果。

```python
class SubsetRandomSampler(Sampler[int]):
r"""Samples elements randomly from a given list of indices, without replacement.

Args:
indices (sequence): a sequence of indices
generator (Generator): Generator used in sampling.
"""
indices: Sequence[int]

def __init__(self, indices: Sequence[int], generator=None) -> None:
self.indices = indices
self.generator = generator

def __iter__(self) -> Iterator[int]:
for i in torch.randperm(len(self.indices), generator=self.generator):
yield self.indices[i]

def __len__(self) -> int:
return len(self.indices)

```

## MindSpore

MindSpore 中`Sampler`的整体设计与`Paddle`并不相同

``` python
class SubsetRandomSampler(SubsetSampler):
"""
Samples the elements randomly from a sequence of indices.

Args:
indices (Iterable): A sequence of indices (Any iterable Python object but string).
num_samples (int, optional): Number of elements to sample. Default: ``None`` , which means sample all elements.

Raises:
TypeError: If elements of `indices` are not of type number.
TypeError: If `num_samples` is not of type int.
ValueError: If `num_samples` is a negative value.

Examples:
>>> import mindspore.dataset as ds
>>> indices = [0, 1, 2, 3, 7, 88, 119]
>>>
>>> # create a SubsetRandomSampler, will sample from the provided indices
>>> sampler = ds.SubsetRandomSampler(indices)
>>> data = ds.ImageFolderDataset(image_folder_dataset_dir, num_parallel_workers=8, sampler=sampler)
"""

def parse(self):
""" Parse the sampler."""
num_samples = self.num_samples if self.num_samples is not None else 0
c_sampler = cde.SubsetRandomSamplerObj(self.indices, num_samples)
c_child_sampler = self.parse_child()
c_sampler.add_child(c_child_sampler)
return c_sampler

def is_shuffled(self):
return True

def parse_for_minddataset(self):
"""Parse the sampler for MindRecord."""
c_sampler = cde.MindrecordSubsetSampler(self.indices, ds.config.get_seed())
c_child_sampler = self.parse_child_for_minddataset()
c_sampler.add_child(c_child_sampler)
c_sampler.set_num_samples(self.get_num_samples())
return c_sampler
```

## API实现方案

pytorch的sampler整体设计与paddle类似,因此考虑参考pytorch的方案实现
在 python\paddle\io\sampler.py 中添加对应类。但是由于paddle并没有完全支持`generator`,因此将该参数移除。

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

subsetrandomsampler中__iter__方法会通过randperm函数将indices打散,并支持设置随机数生成器generator,请确认一下paddle中提供的randperm函数的用法,并且对比一下配置generator和未配置generator的区别

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

paddle 中 randperm 参数如下:

paddle.randperm(n, dtype='int64', name=None)

并不支持 generator。不仅是 randperm,randn、randint 等也不支持。

相关issue:


# 六、测试和验收的考量

测试考虑的 case 如下:
- 确保结果符合预期:一次遍历中遍历所有的 `index` 一次且仅有一次,确保不重复不遗漏.

# 七、可行性分析和排期规划

方案实施难度可控,工期上可以满足在当前版本周期内开发完成。

# 八、影响面

为已有 API 的增强,对其他模块没有影响

# 名词解释


# 附件及参考资料