refine paddle.stack #25305

zhiqiu · 2020-07-01T08:06:11Z

PR types

Function optimization

PR changes

APIs

Describe

Refine paddle.stack API

Variable -> Tensor
use core.ops.stack in dygraph mode
refine document
fluid.layers.stack(x, axis=0) -> fluid.layers.stack(x, axis=0, name=None), which is compatible upgrade.
reuse code of fluid.layers.stack

lanxianghit · 2020-07-28T15:50:49Z

python/paddle/tensor/manipulation.py

@@ -243,65 +248,39 @@ def stack(x, axis=0, out=None, name=None):
                          [5.0, 6.0] ] ]

    Args:
-        x (Variable|list(Variable)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
+        x (Tensor|list(Tensor)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.


Suggested change

x (Tensor|list(Tensor)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.

x (Tensor|list(Tensor)): Input :code:`x` can be a single Tensor, or a :code:`list` of Tensors.

Done, thanks.

lanxianghit · 2020-07-28T15:52:16Z

python/paddle/tensor/manipulation.py

@@ -243,65 +248,39 @@ def stack(x, axis=0, out=None, name=None):
                          [5.0, 6.0] ] ]

    Args:
-        x (Variable|list(Variable)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
+        x (Tensor|list(Tensor)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
                                     If :code:`x` is a :code:`list`, the shapes of all these Tensors


为啥这个文档里都是“:code:x”这样的写法，其他文档都是“ x ”的写法？

两种语法，网页展示的结果应该是一样的。

lanxianghit · 2020-07-28T15:53:48Z

python/paddle/tensor/manipulation.py



-    This OP stacks all the inputs :code:`x` along axis.
+    This OP stacks all the input tensors :code:`x` along :code:`axis` dimemsion. 
+    All tensors should to be of the same shape. 


dtype是否要求相同？代码里有相关检查和处理吗？

要求相同的dtype，c++代码中会做check和报错。

InvalidArgumentError: The DataType of stack Op's duplicable Variable X must be consistent. The current variable type is (double), but the previous variable type is (float). at (/paddle/paddle/fluid/framework/operator.cc:1298)

lanxianghit · 2020-07-28T15:54:38Z

python/paddle/tensor/manipulation.py

@@ -243,65 +248,39 @@ def stack(x, axis=0, out=None, name=None):
                          [5.0, 6.0] ] ]

    Args:
-        x (Variable|list(Variable)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
+        x (Tensor|list(Tensor)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
                                     If :code:`x` is a :code:`list`, the shapes of all these Tensors
                                     must be the same. Supposing input is N dims
                                     Tensors :math:`[d_0, d_1, ..., d_{n-1}]`, the output is N+1 dims


没有文档预览图，看起来太费劲了

lanxianghit · 2020-07-28T15:55:55Z

python/paddle/tensor/manipulation.py

@@ -243,65 +248,39 @@ def stack(x, axis=0, out=None, name=None):
                          [5.0, 6.0] ] ]

    Args:
-        x (Variable|list(Variable)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
+        x (Tensor|list(Tensor)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
                                     If :code:`x` is a :code:`list`, the shapes of all these Tensors
                                     must be the same. Supposing input is N dims
                                     Tensors :math:`[d_0, d_1, ..., d_{n-1}]`, the output is N+1 dims


语法有问题

lanxianghit · 2020-07-28T15:57:53Z

python/paddle/tensor/manipulation.py

                                     If :code:`x` is a :code:`list`, the shapes of all these Tensors
                                     must be the same. Supposing input is N dims
                                     Tensors :math:`[d_0, d_1, ..., d_{n-1}]`, the output is N+1 dims
                                     Tensor :math:`[d_0, d_1, d_{axis-1}, len(x), d_{axis}, ..., d_{n-1}]`.
                                     Support data types: float32, float64, int32, int64.
        axis (int, optional): The axis along which all inputs are stacked. ``axis`` range is :math:`[-(R+1), R+1)`.
-                              R is the first tensor of inputs. If ``axis`` < 0, :math:`axis=axis+rank(x[0])+1`.
+                              R is the first tensor of inputs. If ``axis`` < 0, :math:`axis = axis+ndim(x[0])+1`.


R的定义没有说清楚

已修改， R is the number of dimensions of the first input tensor :code:x[0].

lanxianghit · 2020-07-28T16:00:22Z

python/paddle/tensor/manipulation.py

+            # result shape: [3, 1, 2]
+            # result value: [[[1.0, 2.0]],
+            #                [[3.0, 4.0]],
+            #                [[5.0, 6.0]]]


缺少axis为负数时的例子，并应该详细说明负数情况下是如何起作用的

case 2中axis是负数。

lanxianghit · 2020-07-28T16:00:32Z

python/paddle/tensor/manipulation.py

+            #                [[3.0, 4.0]],
+            #                [[5.0, 6.0]]]
+    """
+    return layers.stack(x, axis, name)


报错信息怎么处理的？

paddle.stack复用了layer.stack的报错信息

zhiqiu · 2020-08-05T13:21:45Z

lanxianghit

LGTM

jzhang533 · 2020-08-06T04:37:20Z

python/paddle/tensor/manipulation.py

+            paddle.enable_imperative()
+            x1 = fluid.dygraph.to_variable(data1)
+            x2 = fluid.dygraph.to_variable(data2)
+            x3 = fluid.dygraph.to_variable(data3)


use paddle.imperative.to_variable

done, thanks

jzhang533

lgtm

zhiqiu added 4 commits July 24, 2020 10:25

refine stack in dygraph mode, test=develop

966af16

update stack, test=develop

481756b

rename Variable to Tensor, test=develop

f008d08

add name args to fluid.layers.stack, test=develop

03a0dbd

zhiqiu force-pushed the dev/refine_dygraph_stack branch from 056378f to 03a0dbd Compare July 24, 2020 03:03

zhiqiu changed the title ~~refine stack in dygraph mode, test=develop~~ refine stack Jul 28, 2020

lanxianghit reviewed Jul 28, 2020

View reviewed changes

zhiqiu added 3 commits August 5, 2020 17:47

follow comments

edd0070

update doc

1311645

update doc

feec883

lanxianghit previously approved these changes Aug 6, 2020

View reviewed changes

jzhang533 reviewed Aug 6, 2020

View reviewed changes

zhiqiu dismissed lanxianghit’s stale review via a615ca6 August 6, 2020 05:37

follow comments

3d7d800

zhiqiu force-pushed the dev/refine_dygraph_stack branch from a615ca6 to 3d7d800 Compare August 6, 2020 06:14

zhiqiu changed the title ~~refine stack~~ refine paddle.stack Aug 6, 2020

jzhang533 approved these changes Aug 6, 2020

View reviewed changes

lanxianghit approved these changes Aug 6, 2020

View reviewed changes

zhiqiu merged commit 6e7f0bb into PaddlePaddle:develop Aug 6, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

refine paddle.stack #25305

refine paddle.stack #25305

zhiqiu commented Jul 1, 2020 •

edited

lanxianghit Jul 28, 2020

zhiqiu Aug 5, 2020

lanxianghit Jul 28, 2020

zhiqiu Aug 5, 2020

lanxianghit Jul 28, 2020

zhiqiu Aug 5, 2020 •

edited

zhiqiu Aug 5, 2020

lanxianghit Jul 28, 2020

lanxianghit Jul 28, 2020

lanxianghit Jul 28, 2020

zhiqiu Aug 5, 2020 •

edited

lanxianghit Jul 28, 2020

zhiqiu Aug 5, 2020

lanxianghit Jul 28, 2020

zhiqiu Aug 5, 2020

zhiqiu commented Aug 5, 2020

lanxianghit left a comment

jzhang533 Aug 6, 2020

zhiqiu Aug 6, 2020

jzhang533 left a comment

	x (Tensor\|list(Tensor)): Input :code:`x` can be a single Tensor, a :code:`list` of Tensors.
	x (Tensor\|list(Tensor)): Input :code:`x` can be a single Tensor, or a :code:`list` of Tensors.

refine paddle.stack #25305

refine paddle.stack #25305

Conversation

zhiqiu commented Jul 1, 2020 • edited

PR types

PR changes

Describe

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zhiqiu Aug 5, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zhiqiu Aug 5, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

zhiqiu commented Aug 5, 2020

lanxianghit left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jzhang533 left a comment

Choose a reason for hiding this comment

zhiqiu commented Jul 1, 2020 •

edited

zhiqiu Aug 5, 2020 •

edited

zhiqiu Aug 5, 2020 •

edited