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.21】为 Paddle 新增 LinearLR #647

Merged
merged 2 commits into from
Oct 7, 2023
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
98 changes: 98 additions & 0 deletions rfcs/APIs/20230925_api_design_for_LinearLR.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,98 @@

# paddle.optimizer.lr.LinearLR设计文档

| API名称 | paddle.optimizer.lr.LinearLR |
| ------------------------------------------------------------ | ----------------------------------- |
| 提交作者<input type="checkbox" class="rowselector hidden"> | Asthestarsfalll |
| 提交时间<input type="checkbox" class="rowselector hidden"> | 2023-09-25 |
| 版本号 | V1.0 |
| 依赖飞桨版本<input type="checkbox" class="rowselector hidden"> | develop |
| 文件名 | 20230925_api_design_for_LinearLR.md<br> |

# 一、概述

## 1、相关背景


LinearLR 学习率调度器在训练开始时通过乘法因子降低学习率, 但是它会在一定数量的训练步骤中线性地改变学习率,直到它达到最终设定的学习率。

## 2、功能目标

在 Paddle 框架中,新增 LinearLR 优化调度器,调用路径为:paddle.optimizer.lr.LinearLR。

## 3、意义
飞桨支持LinearLR优化调度器

# 二、飞桨现状

目前paddle缺少相关功能实现。

# 三、业内方案调研

## Pytorch

Pytorch中有API``,以及对应的`torch.optim.lr_scheduler.``LinearLR(optimizer, start_factor, end_factor, total_iters, last_epoch=-1, verbose=False)`.

在pytorch中,介绍为:

```
Decays the learning rate of each parameter group by linearly changing small multiplicative factor until the number of epoch reaches a pre-defined milestone: total_iters. Notice that such decay can happen simultaneously with other changes to the learning rate from outside this scheduler. When last_epoch=-1, sets initial lr as lr.

```

### 实现方法

在实现方法上, Pytorch是直接通过python实现的,[代码位置](https://github.com/pytorch/pytorch/blob/main/torch/optim/lr_scheduler.py#L551)。

核心代码为:

```python
def get_lr(self):
if not self._get_lr_called_within_step:
warnings.warn("To get the last learning rate computed by the scheduler, "
"please use `get_last_lr()`.", UserWarning)

if self.last_epoch == 0:
return [group['lr'] * self.start_factor for group in self.optimizer.param_groups]

if self.last_epoch > self.total_iters:
return [group['lr'] for group in self.optimizer.param_groups]

return [group['lr'] * (1. + (self.end_factor - self.start_factor) /
(self.total_iters * self.start_factor + (self.last_epoch - 1) * (self.end_factor - self.start_factor)))
for group in self.optimizer.param_groups]
```


# 四、方案设计

## 命名与参数设计

API设计为`paddle.optimizer.lr.LinearLR(base_learning_rate,total_steps,start_factor,end_factor,last_epoch=-1,verbose=False)`

将默认参数 `total_iters` 改为非默认参数 `total_steps`, 名称上与其他学习率调度器保持一致。


## 底层OP设计

直接使用python实现,不再单独设计OP。

## API实现方案

参考Pytorch进行实现,实现位置为`paddle/optimizer/lr.py`。

# 六、测试和验收的考量

测试考虑的case如下:

- 动态图,静态图,与numpy的结果保持一致;
- 错误检查:`start_factor`和 `end_factor`数值不正确时时能正确抛出错误;
- 错误检查:`total_steps` 小于等于0时抛出错误;

# 七、可行性分析及规划排期

方案仅使用python实现,工期上可以满足在当前版本周期内开发完成。

# 八、影响面

为独立新增API,对其他模块没有影响