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.

Sign up for GitHub

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

Jump to bottom

[xdoctest][task 337] reformat example code with google style in `python/paddle/base/framework.py` #57151

Merged

luotao1 merged 6 commits into PaddlePaddle:develop from ooooo-create:ooooo/xdoctest337

Sep 11, 2023

Contributor

ooooo-create commented Sep 10, 2023

PR types

Others

PR changes

Others

Description

修改如下文件的示例代码，使其通过 xdoctest 检查：

python/paddle/base/framework.py

预览：

Related links

@sunzhongkai588 @SigureMo @megemini


          [Doctest]fix No.337, test=docs_preview

0cba64c

paddle-bot bot commented Sep 10, 2023

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.

paddle-bot bot added the contributor label

SigureMo reviewed

View reviewed changes

Member

SigureMo left a comment

CI 里看起来还有别的失败的，也需要看下～

fix

Contributor Author

ooooo-create commented Sep 10, 2023

from paddle.base.dygraph import Linear,这个有Linear吗

Member

SigureMo commented Sep 10, 2023

CI 里看起来还有别的失败的，也需要看下～

咦，之前 review 的消息丢了

from paddle.base.dygraph import Linear,这个有Linear吗

paddle.nn.Linear

Contributor Author

ooooo-create commented Sep 10, 2023

咦，之前 review 的消息丢了
是丢了诶~

megemini reviewed

View reviewed changes

python/paddle/base/framework.py Outdated

Comment on lines 1004 to 1034

    
                          >>> import paddle

                          >>> paddle.enable_static()

                          >>> with paddle.static.name_scope("s1"):

                          ...     a = paddle.static.data(name='data', shape=[None, 1], dtype='int32')

                          ...     b = a + 1

                          >>> with paddle.static.name_scope("s2"):

                          ...     c = b * 1

                          >>> with paddle.static.name_scope("s3"):

                          ...     d = c / 1

                          >>> with paddle.static.name_scope("s1"):

                          ...     f = paddle.tensor.pow(d, 2.0)

                          >>> with paddle.static.name_scope("s4"):

                          ...     g = f - 1

                          >>> # Op are created in the default main program.

                          >>> for op in paddle.static.default_main_program().block(0).ops:

                          ...     # elementwise_add is created in /s1/

                          ...     if op.type == 'elementwise_add':

                          ...         assert op.desc.attr("op_namescope") == '/s1/'

                          ...     # elementwise_mul is created in '/s1/s2'

                          ...     elif op.type == 'elementwise_mul':

                          ...         assert op.desc.attr("op_namescope") == '/s1/s2/'

                          ...     # elementwise_div is created in '/s1/s3'

                          ...     elif op.type == 'elementwise_div':

                          ...         assert op.desc.attr("op_namescope") == '/s1/s3/'

                          ...     # elementwise_sum is created in '/s4'

                          ...     elif op.type == 'elementwise_sub':

                          ...         assert op.desc.attr("op_namescope") == '/s4/'

                          ...     # pow is created in /s1_1/

                          ...     elif op.type == 'pow':

                          ...         assert op.desc.attr("op_namescope") == '/s1_1/'

Contributor

megemini Sep 10, 2023

这段代码运行没什么问题，但实际上，很多 elif 都没跑到～

如果把 op 打印出来的话（print('>>>', op, op.type)），是这样：

>>> {Out=['tmp_0']} = scale(inputs={ScaleTensor=[], X=['data']}, bias = 1.0, bias_after_scale = True, op_device = , op_namescope = /s1/, op_role = 0, op_role_var = [], scale = 1.0, with_quant_attr = False) scale
>>> {Out=['tmp_1']} = scale(inputs={ScaleTensor=[], X=['tmp_0']}, bias = 0.0, bias_after_scale = True, op_device = , op_namescope = /s2/, op_role = 0, op_role_var = [], scale = 1.0, with_quant_attr = False) scale
>>> {Out=['tmp_2']} = cast(inputs={X=['tmp_1']}, in_dtype = 2, op_device = , op_namescope = /s3/, op_role = 0, op_role_var = [], out_dtype = 5, use_mkldnn = False, with_quant_attr = False) cast
>>> {Out=['tmp_3']} = scale(inputs={ScaleTensor=[], X=['tmp_2']}, bias = 0.0, bias_after_scale = True, op_device = , op_namescope = /s3/, op_role = 0, op_role_var = [], scale = 1.0, with_quant_attr = False) scale
>>> {Out=['pow_0.tmp_0']} = pow(inputs={FactorTensor=[], X=['tmp_3']}, factor = 2.0, op_device = , op_namescope = /s1_1/, op_role = 0, op_role_var = [], with_quant_attr = False) pow
>>> {Out=['tmp_4']} = scale(inputs={ScaleTensor=[], X=['pow_0.tmp_0']}, bias = -1.0, bias_after_scale = True, op_device = , op_namescope = /s4/, op_role = 0, op_role_var = [], scale = 1.0, with_quant_attr = False) scale

@SigureMo paddle 的这个流程是不是改了？这里算是遗留代码？

Member

SigureMo Sep 10, 2023

首先有些地方缩进层级少了，需要改一下 @ooooo-create

@SigureMo paddle 的这个流程是不是改了？这里算是遗留代码？

Paddle/python/paddle/base/layers/math_op_patch.py

Lines 464 to 494 in 3c19d80

    
           if isinstance(other_var, float): 
        
               # in all cases(+, -, *, /, **, //, %), we need cast tensor.dtype to float 
        
               if self.dtype in _supported_int_dtype_: 
        
                   self = astype(self, 'float32') 
        
               # here use `scale` replace `elementwise` to get better performance 
        
               # but only +, -, *, / can use this method 
        
               if scalar_method is not None: 
        
                   return scalar_method(self, other_var) 
        
           elif isinstance(other_var, int): 
        
               # in all cases(+, -, *, /, **, //, %), we can cast it to float 
        
               # because the output tensor.dtype depend on the type of input tensor 
        
               other_var = float(other_var) 
        
               # division is a special case 
        
               # NOTE(chenweihang): because we cast tensor to float32 instead float64, 
        
               # the division result can only guarantee the numerical accuracy of 6 digits 
        
               # after the decimal point. The result of numpy calculation is of float64 type, 
        
               # so the calculation result here and the calculation result of numpy are 
        
               # different after 6 decimal point. If necessary, we can also use float64 here. 
        
               # torch's behavior here is consistent with ours 
        
               if ( 
        
                   op_type == 'elementwise_div' 
        
                   and self.dtype in _supported_int_dtype_ 
        
               ): 
        
                   self = astype(self, 'float32') 
        
               # here use `scale` replace `elementwise` to get better performance 
        
               # but only +, -, *, / can use this method 
        
               if scalar_method is not None: 
        
                   return scalar_method(self, other_var) 
        
           else: 
        
               # do nothing 
        
               pass

我看了下这部分代码，在 rhs 为 int、float 时就会变成 scale op，不过这已经是 3 年前的代码了

但这段示例代码貌似历史更加悠久，貌似是 4 年前的……

根据 rhs 只有 scalar 才会变成 scale op，我们可以改成 b = a + paddle.to_tensor(1)，就可以确保是 elementwise_add op 了

python/paddle/base/framework.py Outdated

Comment on lines 1630 to 1631

		... loss2 = paddle.sum(ret2)
		... loss2.backward()

Contributor

megemini Sep 10, 2023

加一个 retain_grads() 吧～不然这 print 没啥意义 ... ...

loss2 = paddle.sum(ret2)
loss2.retain_grads()
loss2.backward()

python/paddle/base/framework.py Outdated

Comment on lines 1678 to 1682

    
                              ...     loss2 = paddle.sum(ret2)

                              ...     loss2.backward()

                              ...     print(loss2.gradient())

                              ...     loss2.clear_gradient()

                              ...     print("After clear {}".format(loss2.gradient()))

Contributor

megemini Sep 10, 2023

同上，加 retrain_grads() ～

另外，这个 print 的输出可以写出来～


          fix loss

1f57ebf

SigureMo reviewed

View reviewed changes

python/paddle/base/framework.py Outdated

@@ @@ -1664,7 +1665,7 @@ def clear_gradient(self): @@
                           .. code-block:: python
                               >>> import paddle
-                              >>> import paddle.base as base
+                              >>> import paddle.fluid as base

Member

SigureMo Sep 10, 2023

fluid？

python/paddle/base/framework.py Outdated

@@ @@ -1006,10 +1006,10 @@ def name_scope(prefix=None): @@
                           >>> with paddle.static.name_scope("s1"):
                           ...     a = paddle.static.data(name='data', shape=[None, 1], dtype='int32')
                           ...     b = a + 1

Member

SigureMo Sep 10, 2023

这里仍然使用 scalar 的话，基本不会走下面的分支的


          fix type

6c22ad6

SigureMo reviewed

View reviewed changes

python/paddle/base/framework.py Outdated

@@ @@ -1005,7 +1005,7 @@ def name_scope(prefix=None): @@
                           >>> paddle.enable_static()
                           >>> with paddle.static.name_scope("s1"):
                           ...     a = paddle.static.data(name='data', shape=[None, 1], dtype='int32')
-                          ...     b = a + 1
+                          ...     b = a + paddle.to_tensor(1)
                           ...     with paddle.static.name_scope("s2"):
                           ...         c = b * 1

Member

SigureMo Sep 11, 2023

下面这几个也需要改哦

python/paddle/base/framework.py Outdated

Comment on lines 7540 to 7543

+                          >>> #print the number of blocks in the program, 1 in this case
+                          >>> print(paddle.static.default_main_program().num_blocks) # 1
+                          >>> #print the default_main_program
+                          >>> print(paddle.static.default_main_program())

Member

SigureMo Sep 11, 2023

Suggested change

      
                        >>> #print the number of blocks in the program, 1 in this case
          
                        >>> print(paddle.static.default_main_program().num_blocks) # 1
          
                        >>> #print the default_main_program
          
                        >>> print(paddle.static.default_main_program())
          
                        >>> # print the number of blocks in the program, 1 in this case
          
                        >>> print(paddle.static.default_main_program().num_blocks) # 1
          
                        >>> # print the default_main_program
          
                        >>> print(paddle.static.default_main_program())


          fix type

9e1530a

SigureMo reviewed

View reviewed changes

python/paddle/base/framework.py Outdated

                           ...     with paddle.static.name_scope("s3"):
-                          ...         d = c / 1
+                          ...         d = c / paddle.to_tensor(1)
                           >>> with paddle.static.name_scope("s1"):
                           ...     f = paddle.tensor.pow(d, 2.0)

Member

SigureMo Sep 11, 2023

这个也要 😂

Contributor Author

ooooo-create Sep 11, 2023

😊是的，新的改好啦~


          fix type

b9fefd6

SigureMo approved these changes

View reviewed changes

Member

SigureMo left a comment

LGTMeow 🐾

luotao1 added the HappyOpenSource Pro label

luotao1 closed this

luotao1 reopened this

luotao1 merged commit 8d84b72 into PaddlePaddle:develop

luotao1 mentioned this pull request

[xdoctest] 分批次修改已有代码的示例 #55629

Closed

ooooo-create deleted the ooooo/xdoctest337 branch

September 23, 2023 05:27

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

contributor HappyOpenSource Pro