[Typing][A-48,A-49,A-50][debug] Add type annotations for 3 optimizers (RAdam, RMSProp, Rprop)

[DrRyanHuang]

PR Category

User Experience

PR Types

Improvements

Description

目前阻塞在🚧 #65076

类型标注：

• paddle/optimizer/radam.py
• paddle/optimizer/rmsprop.py
• paddle/optimizer/rprop.py

Related links

• [Type Hints] 为公开 API 标注类型提示信息 #65008

@SigureMo @megemini

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[SigureMo]

这个 PR 建议在 #65076 之后再推进，另外，最好参考 https://github.com/cattidea/paddlepaddle-stubs 标注更详细些～比如泛型不要省略参数，dict 用更细致的 TypedDict 等

[SigureMo]

#65076 已合，这个 PR 可以推进下了～

[megemini]

@DrRyanHuang 关于这几个 Optimizer 的 ParameterConfig ，可以参考一下 #65273 (comment) ～ 🤟

[SigureMo]

标题还是不要用 xxx/*，这样会认为这个 PR 将 optimizer 全完成了

[SigureMo]

#65273 已合，麻烦一起改一下～

[SigureMo]

@megemini 开了 debug 好像看到了一些奇怪的报错

[megemini]

@megemini 开了 debug 好像看到了一些奇怪的报错

2024-06-21 14:50:55 Traceback (most recent call last):
2024-06-21 14:50:55   File "/paddle/tools/type_checking.py", line 325, in <module>
2024-06-21 14:50:55     run_type_checker(args, mypy_checker)
2024-06-21 14:50:55   File "/paddle/tools/type_checking.py", line 304, in run_type_checker
2024-06-21 14:50:55     test_results = get_test_results(type_checker, docstrings_to_test)
2024-06-21 14:50:55   File "/paddle/tools/type_checking.py", line 272, in get_test_results
2024-06-21 14:50:55     with multiprocessing.Pool(initializer=init_worker) as pool:
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/context.py", line 119, in Pool
2024-06-21 14:50:55     return Pool(processes, initializer, initargs, maxtasksperchild,
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/pool.py", line 215, in __init__
2024-06-21 14:50:55     self._repopulate_pool()
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/pool.py", line 306, in _repopulate_pool
2024-06-21 14:50:55     return self._repopulate_pool_static(self._ctx, self.Process,
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/pool.py", line 329, in _repopulate_pool_static
2024-06-21 14:50:55     w.start()
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/process.py", line 121, in start
2024-06-21 14:50:55     self._popen = self._Popen(self)
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/context.py", line 281, in _Popen
2024-06-21 14:50:55     return Popen(process_obj)
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/popen_fork.py", line 19, in __init__
2024-06-21 14:50:55     self._launch(process_obj)
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/popen_fork.py", line 71, in _launch
2024-06-21 14:50:55     code = process_obj._bootstrap(parent_sentinel=child_r)
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _bootstrap
2024-06-21 14:50:55     self.run()
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/process.py", line 108, in run
2024-06-21 14:50:55     self._target(*self._args, **self._kwargs)
2024-06-21 14:50:55   File "/usr/lib/python3.10/multiprocessing/pool.py", line 125, in worker
2024-06-21 14:50:55     result = (True, func(*args, **kwds))
2024-06-21 14:50:55   File "mypy/semanal.py", line 6714, in accept
2024-06-21 14:50:55   File "mypy/nodes.py", line 787, in accept
2024-06-21 14:50:55   File "mypy/semanal.py", line 835, in visit_func_def
2024-06-21 14:50:55   File "mypy/semanal.py", line 870, in analyze_func_def
2024-06-21 14:50:55   File "mypy/semanal.py", line 6386, in defer
2024-06-21 14:50:55 AssertionError: Must not defer during final iteration
2024-06-21 14:50:55 --------------------

应该还是之前 Optimizer 的那个 config 的 defer 顺序导致的 ～ 我看已经 merge 了，再跑一下看看吧 ～

defer 错误，但不是之前 Optimizer 的那个，还得定位一下 ～

[megemini]

@SigureMo @DrRyanHuang

线下做了一些实验，mypy 有问题，pyright 没问题。

定位的地方为 python/paddle/optimizer/optimizer.py：

if TYPE_CHECKING:
    from typing_extensions import NotRequired, TypedDict

    # from paddle import Tensor # 不能用这个
    from paddle.nn.clip import GradientClipBase

    from ..base.framework import Operator, Program
    from ..tensor.tensor import Tensor # 使用这个代替

    class _ParameterConfig(TypedDict):
        params: Sequence[Tensor]
        weight_decay: NotRequired[float | WeightDecayRegularizer | None]
        learning_rate: NotRequired[float | Tensor | LRScheduler | None]

这里需要用 from ..tensor.tensor import Tensor 代替 from paddle import Tensor ～

这里可能：

• mypy 解析顺序有问题，paddle 其他模块标注的没问题，pyright 没问题
• mypy 没问题，paddle 其他模块的标注有问题，pyright 也存在 bug

现在还不能下定论，也没搜到相关 issue ～

二位是否可以实验一下？@DrRyanHuang 如果修改 python/paddle/optimizer/optimizer.py，需要同时修改示例程序，如：

                >>> rmsprop = paddle.optimizer.RMSProp(
                ...     learning_rate=0.1,
                ...     parameters=[{ # type: ignore
                ...         'params': linear_1.parameters()
                ...     }, {
                ...         'params': linear_2.parameters(),
                ...         'weight_decay': 0.001,
                ...         'learning_rate': 0.1
                ...     }],
                ...     weight_decay=0.01
                ... )

这里的 parameters 要加个 # type: ignore ～

[SigureMo]

LGTMeow