飞桨API 阅读记录.ipynb · Neal/百度飞桨--AI入门之深度学习笔记

加入 Gitee

与超过 1200万开发者一起发现、参与优秀开源项目，私有仓库也完全免费：）

克隆/下载

飞桨API 阅读记录.ipynb 11.92 KB

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# fluid  API框架名\n",
    "\n",
    "fluid.backward  ： 反向传播\n",
    "    \n",
    "fluid.clip ： 修剪数据\n",
    "\n",
    "dataset\n",
    "\n",
    "fluid.dataset\n",
    "\n",
    "fluid.dygraph\n",
    "\n",
    "fluid.executor\n",
    "\n",
    "fluid.initializer\n",
    "\n",
    "fluid.io\n",
    "\n",
    "fluid.layers\n",
    "\n",
    "fluid.metrics\n",
    "\n",
    "fluid.nets\n",
    "\n",
    "fluid.optimizer\n",
    "\n",
    "fluid.profiler\n",
    "\n",
    "fluid.regularizer\n",
    "\n",
    "fluid.transpiler\n",
    "\n",
    "fluid.unique_name\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##  fluid.backward  ： 反向传播\n",
    "\n",
    ">以下两个API ：声明式编程(静态图)专用API\n",
    " \n",
    " \n",
    "#### 1. append_backward： 根据用户的前向** 自动反向传播。此函数由优化器（Optimizer）的 minimize 函数自动调用，一般无需手动调用\n",
    "\n",
    "函数：\n",
    "\n",
    "**paddle.fluid.backward.append_backward(loss, parameter_list=None, no_grad_set=None, callbacks=None)**\n",
    "\n",
    "参数说明：\n",
    "> loss ( Variable ) - 网络的损失变量\n",
    "\n",
    "> parameter_list （list [Variable|str]，可选）- 指定优化器需要更新的参数或参数名称\b列表。如果为 None ，则将更新所有参数。默认值为 None。\n",
    "\n",
    ">no_grad_set （set [Variable|str]，可选）- 在 block0 ( Block ) 中要忽略梯度的 Variable 的名字的集合。所有的 Block 中带有 stop_gradient = True 的所有 Variable 的名字都会被自动添加到此集合中。如果该参数不为 None，则会将该参数集合的内容添加到默认的集合中。默认值为 None。\n",
    "\n",
    ">callbacks （list [callable object]，可选）- 回调函数列表。用于在反向传播构建中执行一些自定义作业。每次将新的梯度OP添加到程序中时，将调用其中的所有可调用对象。可调用对象必须有两个输入参数： Block 和 context 。 Block 是将被添加到新梯度算子的块。 context 是一个映射，其键是梯度 Variable 名，值是对应的原始 Variable 。除此之外， context 还有另一个特殊的键值对：键是字符串 __ current_op_desc__ ，值是刚刚触发可调用对象的梯度OP的 op_desc 。默认值为 None。\n",
    "    \n",
    "返回值：\n",
    ">参数及其梯度 Variable 的元组的列表。元组的第一个值为参数，第二个值为该参数的梯度 Variable 。\n",
    "\n",
    "返回类型\n",
    ">list[( Variable , Variable )]\n",
    "\n",
    "抛出异常\n",
    ">AssertionError - 如果 loss 不是 Variable 的实例。\n",
    "\n",
    "API位置：\n",
    "\n",
    "https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/backward_cn/append_backward_cn.html\n",
    "    \n",
    "\n",
    "#### 2. gradients： 将目标梯度反向传播到输入\n",
    "\n",
    "\n",
    "函数：\n",
    "\n",
    "**paddle.fluid.backward.gradients(targets, inputs, target_gradients=None, no_grad_set=None)**\n",
    "\n",
    "参数：\n",
    ">targets (Variable|list[Variable]) – 目标变量\n",
    "\n",
    ">inputs (Variable|list[Variable]) – 输入变量\n",
    "\n",
    ">target_gradients (Variable|list[Variable]，可选) – 目标的梯度变量，应与目标变量形状相同；如果设置为None，则以1初始化所有梯度变量\n",
    "\n",
    ">no_grad_set (set[Variable|str]，可选) – 在 block0 ( Block ) 中要忽略梯度的 Variable 的名字的集合。所有的 Block 中带有 stop_gradient = True 的所有 Variable 的名字都会被自动添加到此集合中。如果该参数不为 None，则会将该参数集合的内容添加到默认的集合中。默认值为 None。\n",
    "\n",
    "返回值：\n",
    "\n",
    ">数组，包含与输入对应的梯度。如果一个输入不影响目标函数，则对应的梯度变量为None\n",
    "\n",
    "返回值类型:\n",
    "\n",
    ">(list[Variable])\n",
    "\n",
    "\n",
    "API位置：\n",
    "https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/backward_cn/gradients_cn.html\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## fluid.clip ： 修剪\n",
    "\n",
    ">专用对于数据的修剪功能的API\n",
    "\n",
    "\n",
    "#### 1. ErrorClipByValue : 给予超出边界的张量值设定，最大和最小的值\n",
    "    \n",
    "\n",
    "函数：\n",
    "\n",
    "**class paddle.fluid.clip.ErrorClipByValue(max, min=None)**\n",
    "            \n",
    "参数\n",
    ">max:任何大于max（最大值）的值都被设置为 max\n",
    "\n",
    ">min:任何小于min（最小值）的值都被设置为 min\n",
    "            \n",
    "\n",
    "返回值：无\n",
    "\n",
    "API位置：\n",
    "\n",
    "https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/clip_cn/ErrorClipByValue_cn.html\n",
    "\n",
    "    \n",
    "####  2. GradientClipByGlobalNorm：将一个 张量列表 t_list 中所有Tensor的L2范数之和，限定在 clip_norm 参数范围内\n",
    "    \n",
    "> 使用限制\n",
    "\n",
    "\n",
    ">* 该类需要在初始化 optimizer 时进行设置后才能生效。\n",
    "\n",
    ">* L2范数定义为向量所有元素的平方和的开平方\n",
    "\n",
    "\n",
    "函数：\n",
    "\n",
    "**class paddle.fluid.clip.GradientClipByGlobalNorm(clip_norm, group_name='default_group', need_clip=None)**\n",
    "\n",
    "参数\n",
    ">* clip_norm: 所允许的范数最大值\n",
    "\n",
    ">* group_name='default_group' :剪切的组名\n",
    "\n",
    ">* need_clip=None:默认为None，此时会裁剪网络中全部参数。 \n",
    "    用于指定需要梯度裁剪的参数，该函数接收一个 Parameter ，返回一个布尔值 (True表示需要裁剪，False不需要裁剪)。\n",
    "\n",
    "返回值：无\n",
    "\n",
    "API位置：\n",
    "https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/clip_cn/GradientClipByGlobalNorm_cn.html\n",
    "\n",
    "    \n",
    "####  3. GradientClipByNorm\n",
    "    \n",
    "    \n",
    "####  4. GradientClipByValue\n",
    "    \n",
    "    \n",
    "####  5.  set_gradient_clip\n",
    "    \n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## dataset\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "----"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## fluid.dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## fluid.dygraph\n",
    "BackwardStrategy\n",
    "\n",
    "BatchNorm\n",
    "\n",
    "BCELoss\n",
    "\n",
    "BilinearTensorProduct\n",
    "\n",
    "\n",
    "#### Conv2D :卷积算子 构建一个二维卷积层\n",
    "> 其将在神经网络中构建一个二维卷积层 , 用于构建 Conv2D 类的一个可调用对象\n",
    "\n",
    "#### 1. ErrorClipByValue : 给予超出边界的张量值设定，最大和最小的值\n",
    "    \n",
    "\n",
    "函数：\n",
    "\n",
    "**class paddle.fluid.dygraph.Conv2D(num_channels, num_filters, filter_size, stride=1, padding=0, dilation=1, groups=None, param_attr=None, bias_attr=None, use_cudnn=True, act=None, dtype='float32')**\n",
    "            \n",
    "参数\n",
    "    常用的参数\n",
    ">  - num_channels (int) - 输入图像的通道数。\n",
    "\n",
    ">  - num_fliters (int) - 卷积核的个数，和输出特征图通道数相同\n",
    "\n",
    ">  - filter_size(int|tuple) - 卷积核大小，可以是整数，比如3，表示卷积核的高和宽均为3 ；或者是两个整数的list，例如[3,2]，表示卷积核的高为3，宽为2。\n",
    "\n",
    ">  - stride(int|tuple) - 步幅，可以是整数，默认值为1，表示垂直和水平滑动步幅均为1；或者是两个整数的list，例如[3,2]，表示垂直滑动步幅为3，水平滑动步幅为2。\n",
    "\n",
    "\n",
    ">  - padding(int|tuple) - 填充大小，可以是整数，比如1，表示竖直和水平边界填充大小均为1；或者是两个整数的list，例如[2,1]，表示竖直边界填充大小为2，水平边界填充大小为1。\n",
    "\n",
    "\n",
    ">  - act（str）-  应用于输出上的激活函数，如Tanh、Softmax、Sigmoid，Relu等，默认值为None。\n",
    "\n",
    ">  - dilation (int|tuple, 可选) - 膨胀系数大小。如果 dialation 为元组，则必须包含两个整型数，分别表示垂直和水平膨胀系数。否则，表示垂直和水平膨胀系数均为 dialation 。默认值：1\n",
    "\n",
    ">  - groups (int, 可选) - 二维卷积层的组数。根据Alex Krizhevsky的深度卷积神经网络（CNN）论文中的分组卷积：当group=2，滤波器的前一半仅和输入特征图的前一半连接。滤波器的后一半仅和输入特征图的后一半连接。默认值：1\n",
    "\n",
    ">  - param_attr (ParamAttr, 可选) - 指定权重参数属性的对象。默认值为None，表示使用默认的权重参数属性。具体用法请参见 ParamAttr \n",
    "\n",
    ">  - bias_attr (ParamAttr|bool, 可选) - 指定偏置参数属性的对象。默认值为None，表示使用默认的偏置参数属性。具体用法请参见 ParamAttr \n",
    "\n",
    ">  - use_cudnn (bool, 可选) - 是否用cudnn核，只有已安装cudnn库时才有效。默认值：True。\n",
    "\n",
    ">  - dtype (str, 可选) - 数据类型，可以为\"float32\"或\"float64\"。默认值：\"float32\"\n",
    "   \n",
    "\n",
    "返回值：无\n",
    "\n",
    "API位置：\n",
    "\n",
    "https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/Conv2D_cn.html\n",
    "\n",
    "----\n",
    "\n",
    "\n",
    "Conv2DTranspose\n",
    "\n",
    "Conv3D\n",
    "\n",
    "Conv3DTranspose\n",
    "\n",
    "CosineDecay\n",
    "\n",
    "DataParallel\n",
    "\n",
    "disable_dygraph\n",
    "\n",
    "disable_imperative\n",
    "\n",
    "Dropout\n",
    "\n",
    "declarative\n",
    "\n",
    "Embedding\n",
    "\n",
    "enable_dygraph\n",
    "\n",
    "enable_imperative\n",
    "\n",
    "enabled\n",
    "\n",
    "ExponentialDecay\n",
    "\n",
    "grad\n",
    "\n",
    "GroupNorm\n",
    "\n",
    "GRUUnit\n",
    "\n",
    "guard\n",
    "\n",
    "InstanceNorm\n",
    "\n",
    "InverseTimeDecay\n",
    "\n",
    "L1Loss\n",
    "\n",
    "Layer\n",
    "\n",
    "LayerList\n",
    "\n",
    "LayerNorm\n",
    "\n",
    "Linear\n",
    "\n",
    "load_dygraph\n",
    "\n",
    "MSELoss\n",
    "\n",
    "NaturalExpDecay\n",
    "\n",
    "NCE\n",
    "\n",
    "NLLLoss\n",
    "\n",
    "NoamDecay\n",
    "\n",
    "ParallelEnv\n",
    "\n",
    "ParameterList\n",
    "\n",
    "no_grad\n",
    "\n",
    "PiecewiseDecay\n",
    "\n",
    "PolynomialDecay\n",
    "\n",
    "Pool2D\n",
    "\n",
    "PRelu\n",
    "\n",
    "prepare_context\n",
    "\n",
    "ProgramTranslator\n",
    "\n",
    "save_dygraph\n",
    "\n",
    "Sequential\n",
    "\n",
    "SpectralNorm\n",
    "\n",
    "to_variable\n",
    "\n",
    "TracedLayer\n",
    "\n",
    "TreeConv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}

一键复制编辑原始数据按行查看历史

提交于 2020-08-22 19:46 . 计算机视觉基础概念更新

马建仓 AI 助手

尝试更多

代码解读

代码找茬

代码优化

Python

https://gitee.com/neal_in_real/PaddlePaddle.git

git@gitee.com:neal_in_real/PaddlePaddle.git

neal_in_real

PaddlePaddle

百度飞桨--AI入门之深度学习笔记

master

Neal/百度飞桨--AI入门之深度学习笔记

简介

发行版

贡献者

近期动态

Neal/百度飞桨--AI入门之深度学习笔记 .gitee-modal { width: 500px !important; }

简介

发行版

贡献者

近期动态

搜索帮助

Neal/百度飞桨--AI入门之深度学习笔记