[Add] feature alpha dropout (#6679)

Author: megemini

docs/api/paddle/nn/FeatureAlphaDropout_cn.rst

@@ -0,0 +1,27 @@
+.. _cn_api_paddle_nn_FeatureAlphaDropout:
+
+FeatureAlphaDropout
+-------------------------------
+
+.. py:function:: paddle.nn.FeatureAlphaDropout(p=0.5, name=None)
+
+一个通道是一个特征图， `FeatureAlphaDropout` 会随机屏蔽掉整个通道。 `AlphaDropout` 是一种具有自归一化性质的 `dropout` 。均值为 0 ，方差为 1 的输入，经过 `AlphaDropout` 计算之后，输出的均值和方差与输入保持一致。 `AlphaDropout` 通常与 SELU 激活函数组合使用。论文请参考：`Self-Normalizing Neural Networks <https://arxiv.org/abs/1706.02515>`_
+
+在动态图模式下，请使用模型的 `eval()` 方法切换至测试阶段。
+
+.. note::
+   对应的 `functional 方法` 请参考：:ref:`cn_api_paddle_nn_functional_feature_alpha_dropout` 。
+
+参数
+:::::::::
+ - **p** (float)：将输入节点置 0 的概率，即丢弃概率。默认：0.5。
+ - **name** (str，可选) - 具体用法请参见 :ref:`api_guide_Name`，一般无需设置，默认值为 None。
+
+返回
+:::::::::
+`Tensor` ，经过 FeatureAlphaDropout 之后的结果，与输入 x 形状相同的 `Tensor` 。
+
+代码示例
+:::::::::
+
+COPY-FROM: paddle.nn.FeatureAlphaDropout

docs/api/paddle/nn/Overview_cn.rst

@@ -231,6 +231,7 @@ Dropout 层
 
 
     " :ref:`paddle.nn.AlphaDropout <cn_api_paddle_nn_AlphaDropout>` ", "具有自归一化性质的 dropout"
+    " :ref:`paddle.nn.FeatureAlphaDropout <cn_api_paddle_nn_FeatureAlphaDropout>` ", "具有自归一化性质的 dropout ，屏蔽整个通道"
     " :ref:`paddle.nn.Dropout <cn_api_paddle_nn_Dropout>` ", "Dropout"
     " :ref:`paddle.nn.Dropout2D <cn_api_paddle_nn_Dropout2D>` ", "一维 Dropout"
     " :ref:`paddle.nn.Dropout3D <cn_api_paddle_nn_Dropout3D>` ", "二维 Dropout"
@@ -462,6 +463,7 @@ Dropout 方法
 
 
     " :ref:`paddle.nn.functional.alpha_dropout <cn_api_paddle_nn_functional_alpha_dropout>` ", "一种具有自归一化性质的 dropout"
+    " :ref:`paddle.nn.functional.feature_alpha_dropout <cn_api_paddle_nn_functional_feature_alpha_dropout>` ", "一种具有自归一化性质的 dropout ，屏蔽整个通道"
     " :ref:`paddle.nn.functional.dropout <cn_api_paddle_nn_functional_dropout>` ", "Dropout"
     " :ref:`paddle.nn.functional.dropout2d <cn_api_paddle_nn_functional_dropout2d>` ", "一维 Dropout"
     " :ref:`paddle.nn.functional.dropout3d <cn_api_paddle_nn_functional_dropout3d>` ", "二维 Dropout"

docs/api/paddle/nn/functional/feature_alpha_dropout_cn.rst

@@ -0,0 +1,24 @@
+.. _cn_api_paddle_nn_functional_feature_alpha_dropout:
+
+feature_alpha_dropout
+-------------------------------
+
+.. py:function:: paddle.nn.functional.feature_alpha_dropout(x, p=0.5, training=True, name=None)
+
+一个通道是一个特征图， `feature alpha dropout` 会随机屏蔽掉整个通道。 `alpha dropout` 是一种具有自归一化性质的 `dropout` 。均值为 0 ，方差为 1 的输入，经过 `alpha dropout` 计算之后，输出的均值和方差与输入保持一致。 `alpha dropout` 通常与 SELU 激活函数组合使用。
+
+参数
+:::::::::
+ - **x** (Tensor)：输入的多维 `Tensor`，数据类型为：bfloat16、float16、float32 或 float64。
+ - **p** (float)：将输入节点置 0 的概率，即丢弃概率。默认：0.5。
+ - **training** (bool)：标记是否为训练阶段。默认：True。
+ - **name** (str，可选) - 具体用法请参见 :ref:`api_guide_Name`，一般无需设置，默认值为 None。
+
+返回
+:::::::::
+`Tensor` ，经过 feature_alpha_dropout 之后的结果，与输入 x 形状相同的 `Tensor` 。
+
+代码示例
+:::::::::
+
+COPY-FROM: paddle.nn.functional.feature_alpha_dropout

docs/guides/model_convert/convert_from_pytorch/api_difference/functional/torch.nn.functional.feature_alpha_dropout.md

@@ -0,0 +1,24 @@
+## [ torch 参数更多 ]torch.nn.functional.feature_alpha_dropout
+
+### [torch.nn.functional.feature\_alpha\_dropout](https://pytorch.org/docs/stable/generated/torch.nn.functional.feature_alpha_dropout.html)
+
+```python
+torch.nn.functional.feature_alpha_dropout(input, p=0.5, training=False, inplace=False)
+```
+
+### [paddle.nn.functional.feature\_alpha\_dropout](https://www.paddlepaddle.org.cn/documentation/docs/zh/api/paddle/nn/functional/feature_alpha_dropout_cn.html#feature_alpha-dropout)
+
+```python
+paddle.nn.functional.feature_alpha_dropout(x, p=0.5, training=True, name=None)
+```
+
+其中 PyTorch 相比 Paddle 支持更多其他参数，具体如下：
+
+### 参数映射
+
+| PyTorch  | PaddlePaddle | 备注 |
+| -------- | ------------ | -- |
+| input    | x            | 输入的多维 Tensor，仅参数名不一致。 |
+| p        | p            | 将输入节点置 0 的概率。 |
+| training | training     | 标记是否为训练阶段，PyTorch 默认值为 False，paddle 默认值为 True。 |
+| inplace  | -            | 表示在不更改变量的内存地址的情况下，直接修改变量的值，Paddle 无此参数，一般对网络训练结果影响不大，可直接删除。 |

docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.FeatureAlphaDropout.md

@@ -0,0 +1,22 @@
+## [torch 参数更多]torch.nn.FeatureAlphaDropout
+
+### [torch.nn.FeatureAlphaDropout](https://pytorch.org/docs/stable/generated/torch.nn.FeatureAlphaDropout.html#torch.nn.FeatureAlphaDropout)
+
+```python
+torch.nn.FeatureAlphaDropout(p=0.5, inplace=False)
+```
+
+### [paddle.nn.FeatureAlphaDropout](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/api/paddle/nn/FeatureAlphaDropout_cn.html)
+
+```python
+paddle.nn.FeatureAlphaDropout(p=0.5, name=None)
+```
+
+PyTorch 相比 Paddle 支持更多其他参数，具体如下：
+
+### 参数映射
+
+| PyTorch | PaddlePaddle | 备注                                                                                                            |
+| ------- | ------------ | --------------------------------------------------------------------------------------------------------------- |
+| p       | p            | 将输入节点置 0 的概率，即丢弃概率。                                                                             |
+| inplace | -            | 表示在不更改变量的内存地址的情况下，直接修改变量的值，Paddle 无此参数，一般对网络训练结果影响不大，可直接删除。 |

docs/guides/model_convert/convert_from_pytorch/pytorch_api_mapping_cn.md

@@ -64,6 +64,7 @@
 | REFERENCE-MAPPING-ITEM(`torch.all`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/ops/torch.all.md) |
 | REFERENCE-MAPPING-ITEM(`torch.allclose`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/ops/torch.allclose.md) |
 | REFERENCE-MAPPING-ITEM(`torch.alpha_dropout`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/ops/torch.alpha_dropout.md) |
+| REFERENCE-MAPPING-ITEM(`torch.feature_alpha_dropout`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/ops/torch.feature_alpha_dropout.md) |
 | REFERENCE-MAPPING-ITEM(`torch.amax`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/ops/torch.amax.md) |
 | REFERENCE-MAPPING-ITEM(`torch.amin`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/ops/torch.amin.md) |
 | REFERENCE-MAPPING-ITEM(`torch.aminmax`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/ops/torch.aminmax.md) |
@@ -435,6 +436,7 @@
 | REFERENCE-MAPPING-ITEM(`torch.nn.AdaptiveMaxPool2d`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.AdaptiveMaxPool2d.md) |
 | REFERENCE-MAPPING-ITEM(`torch.nn.AdaptiveMaxPool3d`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.AdaptiveMaxPool3d.md) |
 | REFERENCE-MAPPING-ITEM(`torch.nn.AlphaDropout`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.AlphaDropout.md) |
+| REFERENCE-MAPPING-ITEM(`torch.nn.FeatureAlphaDropout`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.FeatureAlphaDropout.md) |
 | REFERENCE-MAPPING-ITEM(`torch.nn.AvgPool1d`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.AvgPool1d.md) |
 | REFERENCE-MAPPING-ITEM(`torch.nn.AvgPool2d`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.AvgPool2d.md) |
 | REFERENCE-MAPPING-ITEM(`torch.nn.AvgPool3d`, https://github.com/PaddlePaddle/docs/tree/develop/docs/guides/model_convert/convert_from_pytorch/api_difference/nn/torch.nn.AvgPool3d.md) |