[API 2.0] add pool2d3d API,test=develop #26331

LDOUBLEV · 2020-08-16T11:07:13Z

PR types

New features

PR changes

APIs

Describe

add API include:

paddle.nn.functional.avg_pool2d
paddle.nn.functional.max_pool2d
paddle.nn.functional.avg_pool3d
paddle.nn.functional.max_pool3d
paddle.nn.layers.AvgPool2d
paddle.nn.layers.MaxPool2d
paddle.nn.layers.AvgPool3d
paddle.nn.layers.MaxPool3d
paddle.nn.functional.avg_pool1d
paddle.nn.functional.max_pool1d
paddle.nn.functional.adaptive_avg_pool1d
paddle.nn.functional.adaptive_max_pool1d
paddle.nn.AvgPool1d
paddle.nn.MaxPool1d
paddle.nn.AdaptiveAvgPool1d
paddle.nn.AdaptiveMaxPool1d

中文文档PR：
https://github.com/PaddlePaddle/FluidDoc/pull/2434/files

paddle-bot-old · 2020-08-16T11:07:20Z

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

littletomatodonkey · 2020-08-17T02:20:49Z

python/paddle/nn/functional/pooling.py

+    return padding
+
+
+def max_pool2d(input,


littletomatodonkey · 2020-08-17T02:21:45Z

python/paddle/nn/functional/pooling.py

+    kernel_size = utils.convert_to_list(kernel_size, 2, 'pool_size')
+    stride = utils.convert_to_list(stride, 2, 'pool_stride')
+
+    data_format = "NCHW"


这块作为参数传进来吧？x可能试nhwc的

littletomatodonkey · 2020-08-17T02:22:16Z

python/paddle/nn/functional/pooling.py

+    return (pool_out, mask) if return_indices else pool_out
+
+
+def avg_pool2d(input,


littletomatodonkey · 2020-08-17T02:22:51Z

python/paddle/nn/functional/pooling.py

+            padding_algorithm = "SAME"
+            padding = [0, 0, 0]
+
+    data_format = "NCDHW"


littletomatodonkey · 2020-08-17T02:23:25Z

python/paddle/nn/functional/pooling.py

+    mask = helper.create_variable_for_type_inference(dtype)
+    outputs = {"Out": pool_out, "Mask": mask}
+
+    helper.append_op(


动态图模式用core.ops.xxx?

littletomatodonkey · 2020-08-17T02:23:57Z

python/paddle/nn/layer/common.py

+        self.name = name
+
+    def forward(self, input):
+        return F.max_pool3d(


littletomatodonkey · 2020-08-17T02:24:16Z

python/paddle/nn/layer/common.py

+                             None by default.
+
+    Returns:
+        Variable: The output tensor of pooling result. The data type is same as input tensor.


return: None

littletomatodonkey · 2020-08-17T02:24:27Z

python/paddle/nn/layer/common.py

+    H is the height of the feature,  D is the depth of the feature, and W is the width of the feature.
+
+    Args:
+        input (Variable): The input tensor of pooling operator, which is a 5-D tensor with


class里面没有input

littletomatodonkey · 2020-08-17T02:24:40Z

python/paddle/nn/layer/common.py

+        self.count_include_pad = count_include_pad
+        self.name = name
+
+    def forward(self, inputs):


WuHaobo · 2020-08-17T02:28:26Z

python/paddle/nn/functional/__init__.py

@@ -155,6 +157,10 @@
 from .pooling import pool3d  #DEFINE_ALIAS
 from .pooling import adaptive_pool2d  #DEFINE_ALIAS
 from .pooling import adaptive_pool3d  #DEFINE_ALIAS
+from .pooling import avg_pool2d


加上 #DEFINE_ALIAS 标示，做api映射

WuHaobo · 2020-08-17T02:28:53Z

python/paddle/nn/layer/__init__.py

@@ -42,6 +42,10 @@
 from .common import Linear  #DEFINE_ALIAS
 from .common import Flatten  #DEFINE_ALIAS
 from .common import UpSample  #DEFINE_ALIAS
+from .common import AvgPool2d
+from .common import MaxPool2d
+from .common import AvgPool3d


#DEFINE_ALIAS

WuHaobo · 2020-08-17T03:55:24Z

python/paddle/nn/layer/common.py

+
+          # max pool2d
+          input = paddle.to_variable(np.random.uniform(-1, 1, [1, 3, 32, 32]).astype(np.float32))
+          output = F.avg_pool2d(input,


这个是functional的example吧

WuHaobo · 2020-08-17T04:50:36Z

python/paddle/nn/layer/common.py

+
+    def __init__(self,
+                 kernel_size=-1,
+                 stride=1,


stride 的默认值应该是None ，不是1；
以及注意None时候的行为

WuHaobo · 2020-08-17T04:50:57Z

python/paddle/nn/layer/common.py

+
+    def __init__(self,
+                 kernel_size,
+                 stride=1,


stride 的默认值应该是None ，不是1；
以及注意None时候的行为

WuHaobo · 2020-08-17T04:58:19Z

python/paddle/nn/layer/common.py

+                 stride=1,
+                 padding=0,
+                 ceil_mode=False,
+                 count_include_pad=True,


divisor_override，这个没对齐

WuHaobo · 2020-08-17T05:05:00Z

python/paddle/nn/layer/common.py

+
+          # max pool2d
+          input = paddle.to_variable(np.random.uniform(-1, 1, [1, 3, 32, 32]).astype(np.float32))
+          output = F.max_pool2d(input,


构造Module的样例

WuHaobo · 2020-08-17T05:11:36Z

python/paddle/nn/layer/common.py

+                 stride=1,
+                 padding=0,
+                 ceil_mode=False,
+                 return_indices=False,


ceil_mode return_indices的传参的顺序

WuHaobo · 2020-08-18T08:01:27Z

python/paddle/nn/layer/common.py

                                stride=2, padding=0)
+          output = Poo2d(input)


fix the typo "Poo2d"

WuHaobo · 2020-08-18T08:04:03Z

python/paddle/nn/layer/common.py

-                 name=None):
+                 ceil_mode=False,
+                 name=None,
+                 data_format="NCHW"):


参数顺序，name 一般是最后一个

WuHaobo · 2020-08-18T08:04:39Z

python/paddle/nn/layer/common.py

-                 name=None):
+                 ceil_mode=False,
+                 name=None,
+                 data_format="NCDHW"):


data_format 和 name的顺序

WuHaobo · 2020-08-18T08:05:05Z

python/paddle/nn/layer/common.py

        super(AvgPool3d, self).__init__()
        self.ksize = kernel_size
        self.stride = stride
        self.padding = padding
        self.ceil_mode = ceil_mode
        self.count_include_pad = count_include_pad
        self.name = name
+        self.data_format = data_format
+        self.divisor = divisor_override


WuHaobo

LGTM

jzhang533 · 2020-08-21T08:41:27Z

python/paddle/nn/functional/pooling.py

-    'pool2d', 'pool3d', 'adaptive_pool2d', 'adaptive_pool3d',
-    'adaptive_avg_pool2d', 'adaptive_avg_pool3d'
+    'pool2d',
+    'pool3d',


pool2d, pool3d 在2.0的API体系里不再需要了。对应的fluid下的api定义上加上deprecated

收到，这个改动和另外两条意见能否新提一个PR修改，functional/common.py 修改较多，很容易和其他PR冲突。解决冲突后还得重新过CI

jzhang533 · 2020-08-21T08:44:29Z

python/paddle/nn/functional/pooling.py

+    'adaptive_avg_pool2d',
+    'adaptive_avg_pool3d',
+    'adaptive_pool2d',
+    'adaptive_pool3d',


adaptive_pool2d, adaptive_pool3d应该是也不需要了，直接对fluid下的api定义上加上deprecated吧。
好像还缺少adaptive_max_pool2d, adaptive_max_pool3d ?

这两个API分给了一帆，他提了PR

jzhang533 · 2020-08-21T08:46:10Z

python/paddle/nn/functional/pooling.py

+    The output value of the layer with input size (N, C, L),
+    output (N, C, L_{out}) and kernel_size k can be precisely described as
+    For average pool1d:
+


这部分可以直接加个reference link 到class 版本的API里。
正在请 @TCChenlong 确认具体的写法。

收到，我线下找 chenlong 确认下。

jzhang533 · 2020-08-21T08:48:02Z

python/paddle/nn/layer/common.py

+    in NCHW format, where N is batch size, C is the number of channels,
+    H is the height of the feature, and W is the width of the feature.
+
+    Example:


输入输出的形状规范成torch文档里的Shape的写法。

收到，这个改动和另外两条意见能否新提一个PR修改，functional/common.py 修改较多，很容易和其他PR冲突。解决冲突后还得重新过CI

jzhang533

will merge
some followup:

split pooling implementation to separate file
fix api deprecation and document issues discussed in this pr.

jzhang533

lgtm(resolved conflicts)

raindrops2sea

LGTM

TCChenlong · 2020-08-23T02:02:56Z

python/paddle/nn/functional/pooling.py

+    Args:
+        x (Tensor): The input tensor of pooling operator which is a 3-D tensor with
+                          shape [N, C, L]. where `N` is batch size, `C` is the number of channels,
+                          `L` is the length of the feature. The data type if float32 or float64.


TCChenlong · 2020-08-23T02:29:17Z

python/paddle/nn/functional/pooling.py

+        ShapeError: If the output's shape calculated is not greater than 0.
+    Examples:
+        .. code-block:: python
+          import paddle


这里应该有一行间隔不然会预览bug

TCChenlong · 2020-08-23T02:29:40Z

python/paddle/nn/functional/pooling.py

+    Args:
+        x (Tensor): The input tensor of pooling operator which is a 3-D tensor with
+                          shape [N, C, L], where `N` is batch size, `C` is the number of channels,
+                          `L` is the length of the feature. The data type if float32 or float64.


TCChenlong · 2020-08-23T02:33:01Z

python/paddle/nn/functional/pooling.py

+            when `data_format` is `"NHWC"`, `pool_padding` can be in the form
+            `[[0,0], [pad_height_top, pad_height_bottom], [pad_width_left, pad_width_right], [0,0]]`.
+            Otherwise, the pool padding size will be a square of an int.
+        ceil_mode (bool): when True, will use `ceil` instead of `floor` to compute the output shape


少了'Default is False。'

TCChenlong · 2020-08-23T02:33:48Z

python/paddle/nn/functional/pooling.py

+        ShapeError: If the output's shape calculated is not greater than 0.
+    Examples:
+        .. code-block:: python
+          import paddle


同理应该加一个空行

add pool2d3d API,test=develop

02edddb

LDOUBLEV added 2 commits August 16, 2020 11:18

add api unitest,test=develop

94614cb

fix unittest, test=develop

086225d

littletomatodonkey reviewed Aug 17, 2020

View reviewed changes

WuHaobo reviewed Aug 17, 2020

View reviewed changes

LDOUBLEV added 2 commits August 17, 2020 06:22

fix reviews, test=develop

b5389ec

return one element when return indices is true, test=develop

4244206

LDOUBLEV changed the title ~~add pool2d3d API,test=develop~~ [API 2.0] add pool2d3d API,test=develop Aug 17, 2020

fix low converage; to_variable to to_tensor, test=develop

54039d9

WuHaobo reviewed Aug 18, 2020

View reviewed changes

python/paddle/nn/layer/common.py Outdated

stride=2, padding=0)

output = Poo2d(input)

Copy link

Contributor

WuHaobo Aug 18, 2020 •

edited

Loading

Choose a reason for hiding this comment

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

fix the typo "Poo2d"

WuHaobo reviewed Aug 18, 2020

View reviewed changes

LDOUBLEV added 2 commits August 18, 2020 13:20

sort API params, test=develop

d54ba32

fix conflicts, test=develop

550c260

WuHaobo previously approved these changes Aug 20, 2020

View reviewed changes

fix en doc, merge PR#26108 to here, test=develop

a89c65f

LDOUBLEV dismissed WuHaobo’s stale review via a89c65f August 21, 2020 02:27

LDOUBLEV mentioned this pull request Aug 21, 2020

[API 2.0] add functional pool1d API #26108

Closed

LDOUBLEV added 2 commits August 21, 2020 03:21

fix en doc, test=develop

a6d7806

fix conflicts

149b466

jzhang533 reviewed Aug 21, 2020

View reviewed changes

jzhang533 previously approved these changes Aug 21, 2020

View reviewed changes

dyning previously approved these changes Aug 21, 2020

View reviewed changes

fix conflicts, test=develop

c581fc8

LDOUBLEV dismissed stale reviews from dyning and jzhang533 via c581fc8 August 24, 2020 05:12

dyning approved these changes Aug 24, 2020

View reviewed changes

jzhang533 approved these changes Aug 24, 2020

View reviewed changes

raindrops2sea approved these changes Aug 24, 2020

View reviewed changes

jzhang533 merged commit c8e1836 into PaddlePaddle:develop Aug 24, 2020

TCChenlong reviewed Aug 24, 2020

View reviewed changes

shippingwang mentioned this pull request Aug 25, 2020

[API2.0]Unify pooling function and add adaptive max pooling function #26483

Merged

		return (pool_out, mask) if return_indices else pool_out


		def avg_pool2d(input,

[API 2.0] add pool2d3d API,test=develop #26331

[API 2.0] add pool2d3d API,test=develop #26331

Conversation

LDOUBLEV commented Aug 16, 2020 • edited Loading

PR types

PR changes

Describe

paddle-bot-old bot commented Aug 16, 2020

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

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

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

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

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

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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WuHaobo Aug 17, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WuHaobo Aug 18, 2020 • edited Loading

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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WuHaobo left a 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

Choose a reason for hiding this comment

jzhang533 Aug 21, 2020 • edited Loading

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

jzhang533 left a comment

Choose a reason for hiding this comment

jzhang533 left a comment

Choose a reason for hiding this comment

raindrops2sea left a 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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

LDOUBLEV commented Aug 16, 2020 •

edited

Loading

WuHaobo Aug 17, 2020 •

edited

Loading

WuHaobo Aug 18, 2020 •

edited

Loading

jzhang533 Aug 21, 2020 •

edited

Loading