24.0.1 PyTorch Basic.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# PyTorch基础"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 前言"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "PyTorch是2018年GitHub增长最快的项目之一。\n",
    "\n",
    "深度学习神经网络的框架如果说之前是TensorFlow, Keras, Theano, Caffe, CNTK, PyTorch, MXNET等百家争鸣，现在已经发展到Google的TensorFlow (Keras被TensorFlow收入囊中)与Facebook的PyTorch双雄争霸的局面。\n",
    "\n",
    "网友表示，这就像现实版的 “Mac 和 PC 之争”（PyTorch 是 Mac）。\n",
    "\n",
    "有关TensorFlow与PyTorch之争的文章，参见：[TensorFlow 2.0和PyTorch谁更好？大牛们争了好几天](https://blog.csdn.net/yH0VLDe8VG8ep9VGe/article/details/82976191)\n",
    "\n",
    "PyTorch现在已经发布了1.0正式版：\n",
    "<img src='./image/pytorch.jpeg' />\n",
    "\n",
    "PyTorch的官方文档：[PyTorch中文文档](https://pytorch-cn.readthedocs.io/zh/latest/)\n",
    "\n",
    "PyTorch的[官方网站](https://pytorch.org/)\n",
    "\n",
    "<img src='./image/pytorchinstall.png' />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "本文是基于GitHub上的[PyTorch Book](https://github.com/chenyuntc/pytorch-book)的第二章：快速入门编写的，因为这篇文档的jupyter笔记文件，作者在文件名加入了冒号(:)，以至于无法在Windows平台正常下载。（Mac下可正常下载与使用），此文的目的是弥补这个遗憾。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## PyTorch第一步"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "PyTorch的简洁设计使得它入门很简单，在深入介绍PyTorch之前，本节将先介绍一些PyTorch的基础知识，使得读者能够对PyTorch有一个大致的了解，并能够用PyTorch搭建一个简单的神经网络。部分内容读者可能暂时不太理解，可先不予以深究。\n",
    "\n",
    "既然都已经开始神经网络应用了，那么假定读者已经掌握了Numpy的基础应用。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Tensor"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tensor是PyTorch中重要的数据结构，可认为是一个高维数组。它可以是一个数（标量）、一维数组（向量）、二维数组（矩阵）以及更高维的数组。\n",
    "\n",
    "Tensor和Numpy的ndarrays类似，但Tensor可以使用GPU进行加速。\n",
    "\n",
    "Tensor的使用和Numpy及Matlab的接口十分相似，下面通过几个例子来看看Tensor的基本使用。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'1.0.1'"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import torch as t\n",
    "t.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "构建5x3矩阵，只是分配空间，未初始化"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[0.0000e+00, 0.0000e+00, 0.0000e+00],\n",
      "        [0.0000e+00, 0.0000e+00, 0.0000e+00],\n",
      "        [0.0000e+00, 0.0000e+00, 0.0000e+00],\n",
      "        [0.0000e+00, 1.8890e-42, 0.0000e+00],\n",
      "        [0.0000e+00, 1.7704e+31, 0.0000e+00]])\n"
     ]
    }
   ],
   "source": [
    "x = t.Tensor(5, 3)\n",
    "print(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下面，我们用数字数组构建一个2x2的矩阵："
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[1., 2.],\n",
      "        [3., 4.]])\n"
     ]
    }
   ],
   "source": [
    "x = t.Tensor([[1, 2], [3, 4]])\n",
    "print(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "使用```[0,1]```均匀分布随机初始化二维数组"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[0.6869, 0.2421, 0.5610],\n",
      "        [0.5947, 0.4914, 0.1517],\n",
      "        [0.9055, 0.2603, 0.9848],\n",
      "        [0.5783, 0.0239, 0.4172],\n",
      "        [0.5441, 0.5163, 0.6383]])\n"
     ]
    }
   ],
   "source": [
    "x = t.rand(5, 3)\n",
    "print(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "查看x的形状\n",
      "torch.Size([5, 3])\n",
      "查看列的个数，两种写法等价\n",
      "3 3\n"
     ]
    }
   ],
   "source": [
    "print('查看x的形状')\n",
    "print(x.size())\n",
    "print('查看列的个数，两种写法等价')\n",
    "print(x.size()[1], x.size(1))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "torch.Size是tuple对象的子类，因此其支持tuple的所有操作，如x.size()[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "矩阵加法"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x\n",
      "tensor([[0.6869, 0.2421, 0.5610],\n",
      "        [0.5947, 0.4914, 0.1517],\n",
      "        [0.9055, 0.2603, 0.9848],\n",
      "        [0.5783, 0.0239, 0.4172],\n",
      "        [0.5441, 0.5163, 0.6383]])\n",
      "y\n",
      "tensor([[0.5530, 0.8533, 0.2583],\n",
      "        [0.6763, 0.3558, 0.0465],\n",
      "        [0.7478, 0.5108, 0.7239],\n",
      "        [0.6035, 0.8471, 0.8104],\n",
      "        [0.1674, 0.6108, 0.3192]])\n",
      "x+y\n",
      "tensor([[1.2399, 1.0954, 0.8193],\n",
      "        [1.2710, 0.8471, 0.1982],\n",
      "        [1.6533, 0.7710, 1.7087],\n",
      "        [1.1819, 0.8709, 1.2276],\n",
      "        [0.7115, 1.1271, 0.9575]])\n"
     ]
    }
   ],
   "source": [
    "y = t.rand(5, 3)\n",
    "print('x')\n",
    "print(x)\n",
    "print('y')\n",
    "print(y)\n",
    "print('x+y')\n",
    "print(x + y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "随机数是深度学习中很关键的基础，为什么？自行百度吧。\n",
    "\n",
    "设置pytorch的Tensor种子数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100\n"
     ]
    }
   ],
   "source": [
    "t.manual_seed(100)\n",
    "print(t.initial_seed())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下面通过不同的种子数，分别获取x与y，并确保每次执行，x与y的值不会发生变化"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x\n",
      "tensor([[0.1117, 0.8158, 0.2626],\n",
      "        [0.4839, 0.6765, 0.7539],\n",
      "        [0.2627, 0.0428, 0.2080],\n",
      "        [0.1180, 0.1217, 0.7356],\n",
      "        [0.7118, 0.7876, 0.4183]])\n",
      "y\n",
      "tensor([[0.5737, 0.1411, 0.8161],\n",
      "        [0.8392, 0.9081, 0.9018],\n",
      "        [0.0450, 0.4321, 0.4437],\n",
      "        [0.2886, 0.5823, 0.0332],\n",
      "        [0.4048, 0.3298, 0.4143]])\n",
      "x+y\n",
      "tensor([[0.6854, 0.9569, 1.0787],\n",
      "        [1.3231, 1.5846, 1.6557],\n",
      "        [0.3077, 0.4749, 0.6518],\n",
      "        [0.4066, 0.7040, 0.7688],\n",
      "        [1.1166, 1.1174, 0.8326]])\n"
     ]
    }
   ],
   "source": [
    "t.manual_seed(100)\n",
    "x= t.rand(5, 3)\n",
    "print('x')\n",
    "print(x)\n",
    "t.manual_seed(95)\n",
    "y = t.rand(5, 3)\n",
    "print('y')\n",
    "print(y)\n",
    "print('x+y')\n",
    "print(x + y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "加法的第二种写法"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[0.6854, 0.9569, 1.0787],\n",
      "        [1.3231, 1.5846, 1.6557],\n",
      "        [0.3077, 0.4749, 0.6518],\n",
      "        [0.4066, 0.7040, 0.7688],\n",
      "        [1.1166, 1.1174, 0.8326]])\n"
     ]
    }
   ],
   "source": [
    "print(t.add(x, y))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "加法的第三种写法：指定加法结果的输出目标为result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[0.6854, 0.9569, 1.0787],\n",
      "        [1.3231, 1.5846, 1.6557],\n",
      "        [0.3077, 0.4749, 0.6518],\n",
      "        [0.4066, 0.7040, 0.7688],\n",
      "        [1.1166, 1.1174, 0.8326]])\n"
     ]
    }
   ],
   "source": [
    "result = t.Tensor(5, 3) # 预先分配空间\n",
    "t.add(x, y, out=result) # 输入到result\n",
    "print(result)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "基于tensor变量做加法\n",
    "\n",
    ">注意，函数名后面带下划线_ 的函数会修改Tensor本身。<br>例如，x.add_(y)和x.t_()会改变 x，但x.add(y)和x.t()返回一个新的Tensor， 而x不变。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "最初y\n",
      "tensor([[0.5737, 0.1411, 0.8161],\n",
      "        [0.8392, 0.9081, 0.9018],\n",
      "        [0.0450, 0.4321, 0.4437],\n",
      "        [0.2886, 0.5823, 0.0332],\n",
      "        [0.4048, 0.3298, 0.4143]])\n",
      "第一种加法，y的结果\n",
      "tensor([[0.6854, 0.9569, 1.0787],\n",
      "        [1.3231, 1.5846, 1.6557],\n",
      "        [0.3077, 0.4749, 0.6518],\n",
      "        [0.4066, 0.7040, 0.7688],\n",
      "        [1.1166, 1.1174, 0.8326]])\n",
      "y\n",
      "tensor([[0.5737, 0.1411, 0.8161],\n",
      "        [0.8392, 0.9081, 0.9018],\n",
      "        [0.0450, 0.4321, 0.4437],\n",
      "        [0.2886, 0.5823, 0.0332],\n",
      "        [0.4048, 0.3298, 0.4143]])\n",
      "第二种加法，y的结果\n",
      "tensor([[0.6854, 0.9569, 1.0787],\n",
      "        [1.3231, 1.5846, 1.6557],\n",
      "        [0.3077, 0.4749, 0.6518],\n",
      "        [0.4066, 0.7040, 0.7688],\n",
      "        [1.1166, 1.1174, 0.8326]])\n",
      "y\n",
      "tensor([[0.6854, 0.9569, 1.0787],\n",
      "        [1.3231, 1.5846, 1.6557],\n",
      "        [0.3077, 0.4749, 0.6518],\n",
      "        [0.4066, 0.7040, 0.7688],\n",
      "        [1.1166, 1.1174, 0.8326]])\n"
     ]
    }
   ],
   "source": [
    "print('最初y')\n",
    "t.manual_seed(95)\n",
    "y = t.rand(5, 3)\n",
    "print(y)\n",
    "\n",
    "print('第一种加法，y的结果')\n",
    "print(y.add(x)) # 普通加法，不改变y的内容\n",
    "print('y')\n",
    "print(y)\n",
    "\n",
    "print('第二种加法，y的结果')\n",
    "print(y.add_(x)) # inplace 加法，y变了\n",
    "print('y')\n",
    "print(y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tensor的选取操作与Numpy类似"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x\n",
      "tensor([[0.1117, 0.8158, 0.2626],\n",
      "        [0.4839, 0.6765, 0.7539],\n",
      "        [0.2627, 0.0428, 0.2080],\n",
      "        [0.1180, 0.1217, 0.7356],\n",
      "        [0.7118, 0.7876, 0.4183]])\n",
      "获取前两行：x[:2]\n",
      "tensor([[0.1117, 0.8158, 0.2626],\n",
      "        [0.4839, 0.6765, 0.7539]])\n",
      "x[:,2]\n",
      "获取索引为2的列，即第三列：x[:, 2]\n",
      "tensor([0.2626, 0.7539, 0.2080, 0.7356, 0.4183])\n",
      "获取第一行第一列成员\n",
      "tensor(0.1117)\n",
      "获取第一行第一列成员的浮点值\n",
      "0.1116642951965332\n"
     ]
    }
   ],
   "source": [
    "print('x')\n",
    "print(x)\n",
    "print('获取前两行：x[:2]')\n",
    "print(x[:2])\n",
    "print('x[:,2]')\n",
    "print('获取索引为2的列，即第三列：x[:, 2]')\n",
    "print(x[:, 2])\n",
    "print('获取第一行第一列成员')\n",
    "print(x[0][0])\n",
    "print('获取第一行第一列成员的浮点值')\n",
    "print(x[0][0].item())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tensor还支持很多操作，包括数学运算、线性代数、选择、切片等等，其接口设计与Numpy极为相似。\n",
    "\n",
    "Tensor和Numpy的数组之间的互操作非常容易且快速。对于Tensor不支持的操作，可以先转为Numpy数组处理，之后再转回Tensor。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "新建一个全1的Tensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([1., 1., 1., 1., 1.])\n"
     ]
    }
   ],
   "source": [
    "a = t.ones(5)\n",
    "print(a)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "将Tensor转为numpy对象"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1. 1. 1. 1. 1.]\n"
     ]
    }
   ],
   "source": [
    "b = a.numpy()\n",
    "print(b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Numpy对象转为Tensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1. 1. 1. 1. 1.]\n",
      "tensor([1., 1., 1., 1., 1.], dtype=torch.float64)\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "a = np.ones(5)\n",
    "b = t.from_numpy(a)\n",
    "print(a)\n",
    "print(b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tensor和numpy对象共享内存，所以他们之间的转换很快，而且几乎不会消耗什么资源。但这也意味着，如果其中一个变了，另外一个也会随之改变。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[2. 2. 2. 2. 2.]\n",
      "tensor([2., 2., 2., 2., 2.], dtype=torch.float64)\n"
     ]
    }
   ],
   "source": [
    "b.add_(1)\n",
    "print(a)\n",
    "print(b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "如果你想获取某一个元素的值，可以使用scalar.item。 Scalar，即标量\n",
    "\n",
    "直接tensor[idx]得到的还是一个tensor: 一个0-dim 的tensor，一般称为scalar."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(2., dtype=torch.float64)\n"
     ]
    }
   ],
   "source": [
    "scalar = b[0]\n",
    "print(scalar)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([])\n"
     ]
    }
   ],
   "source": [
    "print(scalar.size())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.0\n"
     ]
    }
   ],
   "source": [
    "print(scalar.item())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor\n",
      "tensor([2])\n",
      "scalar\n",
      "tensor(2., dtype=torch.float64)\n"
     ]
    }
   ],
   "source": [
    "tensor = t.tensor([2])\n",
    "print('tensor')\n",
    "print(tensor)\n",
    "print('scalar')\n",
    "print(scalar)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor size\n",
      "torch.Size([1])\n",
      "scalar size\n",
      "torch.Size([])\n"
     ]
    }
   ],
   "source": [
    "print('tensor size')\n",
    "print(tensor.size())\n",
    "print('scalar size')\n",
    "print(scalar.size())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "只有一个元素的tensor也可以调用```tensor.item()```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2\n"
     ]
    }
   ],
   "source": [
    "print(tensor.item())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "此外在pytorch中还有一个和np.array 很类似的接口: torch.tensor, 二者的使用十分类似。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([3, 4])\n"
     ]
    }
   ],
   "source": [
    "tensor = t.tensor([3, 4]) # 新建一个包含3, 4两个元素的tensor\n",
    "print(tensor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(3)\n"
     ]
    }
   ],
   "source": [
    "scalar = t.tensor(3)\n",
    "print(scalar)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "old tensor: tensor([3, 4])\n",
      "new tensor: tensor([1111,    4])\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\python3.6\\lib\\site-packages\\ipykernel_launcher.py:2: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  \n"
     ]
    }
   ],
   "source": [
    "old_tensor = tensor\n",
    "new_tensor = t.tensor(old_tensor)\n",
    "new_tensor[0] = 1111\n",
    "print('old tensor: {0}'.format(old_tensor))\n",
    "print('new tensor: {0}'.format(new_tensor))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "需要注意的是，t.tensor()总是会进行数据拷贝，新tensor和原来的数据不再共享内存。\n",
    "\n",
    "所以如果你想共享内存的话，建议使用torch.from_numpy()或者tensor.detach()来新建一个tensor, 二者共享内存。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "old tensor: tensor([1111,    4])\n",
      "new tensor: tensor([1111,    4])\n"
     ]
    }
   ],
   "source": [
    "new_tensor = old_tensor.detach()\n",
    "new_tensor[0] = 1111\n",
    "print('old tensor: {0}'.format(old_tensor))\n",
    "print('new tensor: {0}'.format(new_tensor))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tensor可通过.cuda 方法转为GPU的Tensor，从而享受GPU带来的加速运算。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cuda:0\n",
      "tensor([[0.7970, 1.7728, 1.3413],\n",
      "        [1.8070, 2.2611, 2.4096],\n",
      "        [0.5704, 0.5178, 0.8598],\n",
      "        [0.5246, 0.8257, 1.5044],\n",
      "        [1.8284, 1.9050, 1.2509]], device='cuda:0')\n"
     ]
    }
   ],
   "source": [
    "# 在不支持CUDA的机器下，下一步还是在CPU上运行\n",
    "device = t.device(\"cuda:0\" if t.cuda.is_available() else \"cpu\")\n",
    "print(device)\n",
    "x = x.to(device)\n",
    "y = y.to(device)\n",
    "z = x+y\n",
    "print(z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "此外，还可以使用tensor.cuda() 的方式将tensor拷贝到gpu上，但是这种方式不太推荐。\n",
    "\n",
    "此处可能发现GPU运算的速度并未提升太多，这是因为x和y太小且运算也较为简单，而且将数据从内存转移到显存还需要花费额外的开销。\n",
    "\n",
    "GPU的优势需在大规模数据和复杂运算下才能体现出来。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### autograd: 自动微分"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "深度学习的算法本质上是通过反向传播求导数，而PyTorch的autograd模块则实现了此功能。\n",
    "\n",
    "在Tensor上的所有操作，autograd都能为它们自动提供微分，避免了手动计算导数的复杂过程。\n",
    "\n",
    "从0.4起, Variable 正式合并入Tensor。\n",
    "\n",
    "Variable 本来实现的自动微分功能，Tensor就能支持。\n",
    "\n",
    "读者还是可以使用Variable(tensor), 但是这个操作其实什么都没做。建议读者以后直接使用tensor.\n",
    "\n",
    "要想使得Tensor使用autograd功能，只需要设置tensor.requries_grad=True."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[1., 1.],\n",
      "        [1., 1.]], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "# 为tensor设置 requires_grad 标识，代表着需要求导数\n",
    "# pytorch 会自动调用autograd 记录操作\n",
    "x = t.ones(2, 2, requires_grad=True)\n",
    "print(x)\n",
    "# 上一步等价于\n",
    "# x = t.ones(2,2)\n",
    "# x.requires_grad = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(4., grad_fn=<SumBackward0>)\n"
     ]
    }
   ],
   "source": [
    "y = x.sum()\n",
    "print(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<SumBackward0 object at 0x0000015208F9C128>\n"
     ]
    }
   ],
   "source": [
    "print(y.grad_fn)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "第1次反向传播，梯度为：tensor([[1., 1.],\n",
      "        [1., 1.]])\n",
      "第2次反向传播，梯度为：tensor([[2., 2.],\n",
      "        [2., 2.]])\n",
      "第3次反向传播，梯度为：tensor([[3., 3.],\n",
      "        [3., 3.]])\n",
      "第4次反向传播，梯度为：tensor([[4., 4.],\n",
      "        [4., 4.]])\n",
      "第5次反向传播，梯度为：tensor([[5., 5.],\n",
      "        [5., 5.]])\n"
     ]
    }
   ],
   "source": [
    "for i in range(5):\n",
    "    y.backward() # 反向传播，计算梯度\n",
    "    print('第{0}次反向传播，梯度为：{1}'.format(i + 1, x.grad))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[1., 1.],\n",
      "        [1., 1.]], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "print(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "注意：grad在反向传播过程中是累加的(accumulated)。\n",
    "\n",
    "这意味着每一次运行反向传播，梯度都会累加之前的梯度，所以反向传播之前需把梯度清零。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[0., 0.],\n",
      "        [0., 0.]])\n"
     ]
    }
   ],
   "source": [
    "# 清零梯度\n",
    "print(x.grad.data.zero_())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "之后，再反向传播，即可重新计算梯度"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[1., 1.],\n",
      "        [1., 1.]])\n"
     ]
    }
   ],
   "source": [
    "y.backward()\n",
    "print(x.grad)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 神经网络"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Autograd实现了反向传播功能，但是直接用来写深度学习的代码在很多情况下还是稍显复杂。\n",
    "\n",
    "torch.nn是专门为神经网络设计的模块化接口。\n",
    "\n",
    "nn构建于 Autograd之上，可用来定义和运行神经网络。\n",
    "\n",
    "nn.Module是nn中最重要的类，可把它看成是一个网络的封装，包含网络各层定义以及forward方法，调用forward(input)方法，可返回前向传播的结果。\n",
    "\n",
    "下面就以最早的卷积神经网络：LeNet为例，来看看如何用nn.Module实现。LeNet的网络结构如图所示：\n",
    "\n",
    "此卷积神经网络，我们在[以Keras为例的神经网络模型](./24.%20Neural%20Network%20Introduction.ipynb)已经做了非常详细的阐述，这里我们主要看PyTorch的实现方式"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src='./image/lenet5.png' />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "LeNet每一层特征具现化："
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src='./image/lenet57_1.png' />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "这是一个基础的前向传播(feed-forward)网络: 接收输入，经过层层传递运算，得到输出。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 定义网络"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "定义网络时，需要继承nn.Module，并实现它的forward方法，把网络中具有可学习参数的层放在构造函数__init__中。\n",
    "\n",
    "如果某一层(如ReLU)不具有可学习的参数，则既可以放在构造函数中，也可以不放，但建议不放在其中，而在forward中使用nn.functional代替。\n",
    "\n",
    "下面的代码，我们只使用了两个卷积层，三个全连接层，最后一层nn.Linear(84, 10)表示10分类"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn as nn\n",
    "import torch.nn.functional as F"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "nn.Conv2d的参数：\n",
    "```\n",
    " |  Args:\n",
    " |      in_channels (int): Number of channels in the input image\n",
    " |      out_channels (int): Number of channels produced by the convolution\n",
    " |      kernel_size (int or tuple): Size of the convolving kernel\n",
    " |      stride (int or tuple, optional): Stride of the convolution. Default: 1\n",
    " |      padding (int or tuple, optional): Zero-padding added to both sides of the input. Default: 0\n",
    " |      dilation (int or tuple, optional): Spacing between kernel elements. Default: 1\n",
    " |      groups (int, optional): Number of blocked connections from input channels to output channels. Default: 1\n",
    " |      bias (bool, optional): If ``True``, adds a learnable bias to the output. Default: ``True``\n",
    " ```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下文的代码示例，没有使用torch.nn.Sequential，虽然下面的代码看起来粗糙一些，但是与原始算法的构建更贴切"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Net(\n",
      "  (conv1): Conv2d(1, 6, kernel_size=(5, 5), stride=(1, 1))\n",
      "  (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))\n",
      "  (fc1): Linear(in_features=400, out_features=120, bias=True)\n",
      "  (fc2): Linear(in_features=120, out_features=84, bias=True)\n",
      "  (fc3): Linear(in_features=84, out_features=10, bias=True)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "class Net(nn.Module):\n",
    "    def __init__(self):\n",
    "        # nn.Module子类的函数必须在构造函数中执行父类的构造函数\n",
    "        # 下式等价于nn.Module.__init__(self)\n",
    "        super(Net, self).__init__()\n",
    "        \n",
    "        # 卷积层 '1'表示输入图片为单通道, '6'表示输出通道数，'5'表示卷积核为5*5\n",
    "        self.conv1 = nn.Conv2d(1, 6, 5) \n",
    "        # 卷积层\n",
    "        self.conv2 = nn.Conv2d(6, 16, 5) \n",
    "        # 仿射层/全连接层，y = Wx + b\n",
    "        self.fc1   = nn.Linear(16*5*5, 120) \n",
    "        self.fc2   = nn.Linear(120, 84)\n",
    "        self.fc3   = nn.Linear(84, 10)\n",
    "\n",
    "    def forward(self, x): \n",
    "        # 卷积 -> 激活 -> 池化 \n",
    "        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))\n",
    "        x = F.max_pool2d(F.relu(self.conv2(x)), 2) \n",
    "        # reshape，‘-1’表示自适应\n",
    "        x = x.view(x.size()[0], -1) \n",
    "        x = F.relu(self.fc1(x))\n",
    "        x = F.relu(self.fc2(x))\n",
    "        x = self.fc3(x)        \n",
    "        return x\n",
    "\n",
    "net = Net()\n",
    "print(net)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<b><font color='red'>只要在nn.Module的子类中定义了forward函数，backward函数就会自动被实现(利用autograd)。</font></b>\n",
    "\n",
    "在forward 函数中可使用任何tensor支持的函数，还可以使用if、for循环、print、log等Python语法，写法和标准的Python写法一致。\n",
    "\n",
    "网络的可学习参数通过net.parameters()返回，net.named_parameters可同时返回可学习的参数及名称。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10\n"
     ]
    }
   ],
   "source": [
    "params = list(net.parameters())\n",
    "print(len(params))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "conv1.weight : torch.Size([6, 1, 5, 5])\n",
      "conv1.bias : torch.Size([6])\n",
      "conv2.weight : torch.Size([16, 6, 5, 5])\n",
      "conv2.bias : torch.Size([16])\n",
      "fc1.weight : torch.Size([120, 400])\n",
      "fc1.bias : torch.Size([120])\n",
      "fc2.weight : torch.Size([84, 120])\n",
      "fc2.bias : torch.Size([84])\n",
      "fc3.weight : torch.Size([10, 84])\n",
      "fc3.bias : torch.Size([10])\n"
     ]
    }
   ],
   "source": [
    "for name, parameters in net.named_parameters():\n",
    "    print(name,':',parameters.size())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "forward函数的输入和输出都是Tensor。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input: tensor([[[[ 0.1268,  1.3564,  0.5632,  ..., -1.7736, -0.2205,  2.7467],\n",
      "          [-1.7599,  1.0230, -0.1107,  ..., -0.8043, -0.6244, -0.5882],\n",
      "          [ 1.6700, -0.9275, -0.9759,  ...,  0.2521,  0.6541,  2.1024],\n",
      "          ...,\n",
      "          [-0.6924, -0.4866, -0.1484,  ..., -0.3634, -0.4904, -1.2284],\n",
      "          [ 0.0528, -0.6355, -0.3881,  ...,  1.4351, -0.9121,  0.7485],\n",
      "          [-1.3633,  0.6923,  0.0092,  ...,  0.5204, -0.9783, -0.3428]]]])\n",
      "input size: torch.Size([1, 1, 32, 32])\n",
      "out size: torch.Size([1, 10])\n"
     ]
    }
   ],
   "source": [
    "# Samples x Channels x Height x Width\n",
    "t.manual_seed(100)\n",
    "input = t.randn(1, 1, 32, 32)\n",
    "print('input: {0}'.format(input))\n",
    "print('input size: {0}'.format(input.size()))\n",
    "out = net(input)\n",
    "print('out size: {0}'.format(out.size()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "net.zero_grad() # 所有参数的梯度清零\n",
    "out.backward(t.ones(1,10)) # 反向传播"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "需要注意的是，torch.nn只支持mini-batches，不支持一次只输入一个样本，即一次必须是一个batch。\n",
    "\n",
    "但如果只想输入一个样本，则用 input.unsqueeze(0)将batch_size设为１。\n",
    "\n",
    "例如 nn.Conv2d 输入必须是4维的，形如$nSamples \\times nChannels \\times Height \\times Width$。可将nSample设为1，即$1 \\times nChannels \\times Height \\times Width$。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 损失函数"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "nn实现了神经网络中大多数的损失函数。\n",
    "\n",
    "例如nn.MSELoss用来计算均方误差。\n",
    "\n",
    "nn.CrossEntropyLoss用来计算交叉熵损失。\n",
    "\n",
    "还记得在之前说的么：\n",
    "\n",
    "采用交叉熵为损失函数，因为其是凸函数，只有凸函数才有可能得到全局最优解。 \n",
    "\n",
    "如果采用Sigmoid作为激活函数，则MSE（均方差）为非凸函数，易陷入局部最优。 \n",
    "\n",
    "此外，MSE比交叉熵容易出现梯度弥散，根据计算MSE的梯度一般是交叉熵的1/4或者更小\n",
    "\n",
    "还有更重要的总结：\n",
    "<br><b><font color='red'>\n",
    "PyTorch的MSE用于做回归的，因为目标参数，即yhat只能为浮点型: torch.float\n",
    "<br>\n",
    "PyTorch的交叉熵用于做分类的，因为目标参数，即yhat只能为长整型: torch.long\n",
    "</font></b>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### MSE作为损失函数"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "MSE的target必须与output的分类数目一致, 否则会报如下错误：\n",
    "<br>The size of tensor a (10) must match the size of tensor b (5) at non-singleton dimension 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "output size: torch.Size([1, 10])\n",
      "tensor([9., 1., 7., 5., 6., 7., 5., 7., 5., 0.])\n",
      "tensor(33.9477, grad_fn=<MseLossBackward>) 33.947723388671875\n"
     ]
    }
   ],
   "source": [
    "output = net(input)\n",
    "print('output size: {0}'.format(output.size()))\n",
    "# 注意t.arange(0., 10.)的写法，如果写为t.arange(0, 10)，则会报如下错误：\n",
    "# Expected object of scalar type Float but got scalar type Long for argument #2 'target'\n",
    "# print(t.arange(0., 10.))\n",
    "# target = t.arange(0.,10.).view(1,10) \n",
    "target = t.empty(10, dtype=t.float).random_(10)\n",
    "print(target)\n",
    "criterion = nn.MSELoss()\n",
    "loss = criterion(output, target)\n",
    "print(loss, loss.item()) # loss是个scalar"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "如果对loss进行反向传播溯源(使用gradfn属性)，可看到它的计算图如下：\n",
    "```\n",
    "input -> conv2d -> relu -> maxpool2d -> conv2d -> relu -> maxpool2d  \n",
    "      -> view -> linear -> relu -> linear -> relu -> linear \n",
    "      -> MSELoss\n",
    "      -> loss\n",
    "```\n",
    "当调用loss.backward()时，该图会动态生成并自动微分，也即会自动计算图中参数(Parameter)的导数。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "反向传播之前 conv1.bias的梯度\n",
      "tensor([0., 0., 0., 0., 0., 0.])\n",
      "反向传播之后 conv1.bias的梯度\n",
      "tensor([ 0.0211, -0.0390, -0.1102,  0.0096,  0.0166, -0.0149])\n"
     ]
    }
   ],
   "source": [
    "# 运行.backward，观察调用之前和调用之后的grad\n",
    "net.zero_grad() # 把net中所有可学习参数的梯度清零\n",
    "print('反向传播之前 conv1.bias的梯度')\n",
    "print(net.conv1.bias.grad)\n",
    "loss.backward()\n",
    "print('反向传播之后 conv1.bias的梯度')\n",
    "print(net.conv1.bias.grad)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### CrossEntropy(交叉熵)作为损失函数"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- PyTorch中的交叉熵使用，其target参数的size必须与batch-size大小一致，否则会报类似异常：\n",
    "```\n",
    "ValueError: Expected input batch_size (5) to match target batch_size (4).\n",
    "```\n",
    "- target设置的值，如果是10分类，必须在0~9之间，否则会报如下异常：\n",
    "```\n",
    "RuntimeError: Assertion `cur_target >= 0 && cur_target < n_classes' failed.  at d:\\build\\pytorch\\pytorch-1.0\\aten\\src\\thnn\\generic\\classnllcriterion.c:93\n",
    "```\n",
    "- 交叉熵的target参数，必须为t.long类型，否则会报错"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input size: torch.Size([5, 1, 32, 32])\n",
      "output size: torch.Size([5, 10])\n",
      "target: tensor([4, 8, 6, 2, 1]), size: torch.Size([5])\n",
      "tensor(2.3091, grad_fn=<NllLossBackward>) 2.309054374694824\n"
     ]
    }
   ],
   "source": [
    "# batch-size x Channels x Height x Width\n",
    "t.manual_seed(88)\n",
    "inputbatchsize5 = t.randn(5, 1, 32, 32)\n",
    "print('input size: {0}'.format(inputbatchsize5.size()))\n",
    "netforcross = Net()\n",
    "outputforcross = netforcross(inputbatchsize5)\n",
    "print('output size: {0}'.format(outputforcross.size()))\n",
    "targetforcross = t.empty(5, dtype=t.long).random_(10)\n",
    "print('target: {0}, size: {1}'.format(targetforcross, targetforcross.size()))\n",
    "criterionforcross = nn.CrossEntropyLoss()\n",
    "lossforcross = criterionforcross(outputforcross, targetforcross)\n",
    "print(lossforcross, lossforcross.item()) # loss是个scalar"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "反向传播之前 conv1.bias的梯度\n",
      "None\n",
      "反向传播之后 conv1.bias的梯度\n",
      "tensor([ 0.0051, -0.0045,  0.0044,  0.0038,  0.0018,  0.0060])\n"
     ]
    }
   ],
   "source": [
    "# 运行.backward，观察调用之前和调用之后的grad\n",
    "netforcross.zero_grad() # 把net中所有可学习参数的梯度清零\n",
    "print('反向传播之前 conv1.bias的梯度')\n",
    "print(netforcross.conv1.bias.grad)\n",
    "lossforcross.backward()\n",
    "print('反向传播之后 conv1.bias的梯度')\n",
    "print(netforcross.conv1.bias.grad)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 优化器"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "在反向传播计算完所有参数的梯度后，还需要使用优化方法来更新网络的权重和参数。\n",
    "\n",
    "例如随机梯度下降法(SGD)的更新策略如下：\n",
    "```\n",
    "weight = weight - learning_rate * gradient\n",
    "```\n",
    "手动实现如下："
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial net begin\n",
      "input size: torch.Size([5, 1, 32, 32])\n",
      "output size: torch.Size([5, 10])\n",
      "target: tensor([4, 8, 6, 2, 1]), size: torch.Size([5])\n",
      "Initial net end\n",
      "通过手动实现SGD: weight = weight - learning_rate * gradient\n",
      "Train epoch: 1, loss value: 2.309054374694824\n",
      "Train epoch: 2, loss value: 2.3034679889678955\n",
      "Train epoch: 3, loss value: 2.2980825901031494\n",
      "Train epoch: 4, loss value: 2.2927603721618652\n",
      "Train epoch: 5, loss value: 2.287485361099243\n",
      "Train epoch: 6, loss value: 2.282428026199341\n",
      "Train epoch: 7, loss value: 2.277522087097168\n",
      "Train epoch: 8, loss value: 2.272451400756836\n",
      "Train epoch: 9, loss value: 2.267273187637329\n",
      "Train epoch: 10, loss value: 2.262345790863037\n",
      "Train epoch: 11, loss value: 2.2573435306549072\n",
      "Train epoch: 12, loss value: 2.2522966861724854\n",
      "Train epoch: 13, loss value: 2.2474045753479004\n",
      "Train epoch: 14, loss value: 2.242506504058838\n",
      "Train epoch: 15, loss value: 2.2373647689819336\n",
      "Train epoch: 16, loss value: 2.2321648597717285\n",
      "Train epoch: 17, loss value: 2.226863384246826\n",
      "Train epoch: 18, loss value: 2.221585273742676\n",
      "Train epoch: 19, loss value: 2.216405153274536\n",
      "Train epoch: 20, loss value: 2.2110602855682373\n",
      "Train epoch: 21, loss value: 2.2054991722106934\n",
      "Train epoch: 22, loss value: 2.1996090412139893\n",
      "Train epoch: 23, loss value: 2.1935226917266846\n",
      "Train epoch: 24, loss value: 2.1870219707489014\n",
      "Train epoch: 25, loss value: 2.1801388263702393\n",
      "Train epoch: 26, loss value: 2.1730358600616455\n",
      "Train epoch: 27, loss value: 2.165714740753174\n",
      "Train epoch: 28, loss value: 2.158118724822998\n",
      "Train epoch: 29, loss value: 2.150303363800049\n",
      "Train epoch: 30, loss value: 2.14216947555542\n",
      "Train epoch: 31, loss value: 2.1337456703186035\n",
      "Train epoch: 32, loss value: 2.124967575073242\n",
      "Train epoch: 33, loss value: 2.1159701347351074\n",
      "Train epoch: 34, loss value: 2.1064019203186035\n",
      "Train epoch: 35, loss value: 2.0965590476989746\n",
      "Train epoch: 36, loss value: 2.086120367050171\n",
      "Train epoch: 37, loss value: 2.075251817703247\n",
      "Train epoch: 38, loss value: 2.063805103302002\n",
      "Train epoch: 39, loss value: 2.052138090133667\n",
      "Train epoch: 40, loss value: 2.039886951446533\n",
      "Train epoch: 41, loss value: 2.027174711227417\n",
      "Train epoch: 42, loss value: 2.013892889022827\n",
      "Train epoch: 43, loss value: 2.0003323554992676\n",
      "Train epoch: 44, loss value: 1.9864013195037842\n",
      "Train epoch: 45, loss value: 1.9719855785369873\n",
      "Train epoch: 46, loss value: 1.9571058750152588\n",
      "Train epoch: 47, loss value: 1.9419790506362915\n",
      "Train epoch: 48, loss value: 1.926627516746521\n",
      "Train epoch: 49, loss value: 1.9108959436416626\n",
      "Train epoch: 50, loss value: 1.8949114084243774\n",
      "Train epoch: 51, loss value: 1.8788995742797852\n",
      "Train epoch: 52, loss value: 1.8625074625015259\n",
      "Train epoch: 53, loss value: 1.8465092182159424\n",
      "Train epoch: 54, loss value: 1.8303569555282593\n",
      "Train epoch: 55, loss value: 1.8146240711212158\n",
      "Train epoch: 56, loss value: 1.798802137374878\n",
      "Train epoch: 57, loss value: 1.7827284336090088\n",
      "Train epoch: 58, loss value: 1.767000436782837\n",
      "Train epoch: 59, loss value: 1.751916527748108\n",
      "Train epoch: 60, loss value: 1.7378911972045898\n",
      "Train epoch: 61, loss value: 1.7243473529815674\n",
      "Train epoch: 62, loss value: 1.711503267288208\n",
      "Train epoch: 63, loss value: 1.6993589401245117\n",
      "Train epoch: 64, loss value: 1.6878712177276611\n",
      "Train epoch: 65, loss value: 1.6767085790634155\n",
      "Train epoch: 66, loss value: 1.6662778854370117\n",
      "Train epoch: 67, loss value: 1.6565395593643188\n",
      "Train epoch: 68, loss value: 1.6468467712402344\n",
      "Train epoch: 69, loss value: 1.6384830474853516\n",
      "Train epoch: 70, loss value: 1.629875898361206\n",
      "Train epoch: 71, loss value: 1.6218887567520142\n",
      "Train epoch: 72, loss value: 1.6146787405014038\n",
      "Train epoch: 73, loss value: 1.607593297958374\n",
      "Train epoch: 74, loss value: 1.6009471416473389\n",
      "Train epoch: 75, loss value: 1.594679594039917\n",
      "Train epoch: 76, loss value: 1.5888642072677612\n",
      "Train epoch: 77, loss value: 1.5831350088119507\n",
      "Train epoch: 78, loss value: 1.577640175819397\n",
      "Train epoch: 79, loss value: 1.5726252794265747\n",
      "Train epoch: 80, loss value: 1.5676841735839844\n",
      "Train epoch: 81, loss value: 1.563631296157837\n",
      "Train epoch: 82, loss value: 1.558897614479065\n",
      "Train epoch: 83, loss value: 1.5543992519378662\n",
      "Train epoch: 84, loss value: 1.550126314163208\n",
      "Train epoch: 85, loss value: 1.5462672710418701\n",
      "Train epoch: 86, loss value: 1.5421158075332642\n",
      "Train epoch: 87, loss value: 1.5382537841796875\n",
      "Train epoch: 88, loss value: 1.534472942352295\n",
      "Train epoch: 89, loss value: 1.5309312343597412\n",
      "Train epoch: 90, loss value: 1.5271894931793213\n",
      "Train epoch: 91, loss value: 1.523879051208496\n",
      "Train epoch: 92, loss value: 1.5202029943466187\n",
      "Train epoch: 93, loss value: 1.5165616273880005\n",
      "Train epoch: 94, loss value: 1.5132811069488525\n",
      "Train epoch: 95, loss value: 1.510145664215088\n",
      "Train epoch: 96, loss value: 1.5066378116607666\n",
      "Train epoch: 97, loss value: 1.503173828125\n",
      "Train epoch: 98, loss value: 1.4999114274978638\n",
      "Train epoch: 99, loss value: 1.496513843536377\n",
      "Train epoch: 100, loss value: 1.4933445453643799\n"
     ]
    }
   ],
   "source": [
    "# batch-size x Channels x Height x Width\n",
    "print('Initial net begin')\n",
    "t.manual_seed(88)\n",
    "inputbatchsize5 = t.randn(5, 1, 32, 32)\n",
    "print('input size: {0}'.format(inputbatchsize5.size()))\n",
    "netforcross = Net()\n",
    "outputforcross = netforcross(inputbatchsize5)\n",
    "print('output size: {0}'.format(outputforcross.size()))\n",
    "targetforcross = t.empty(5, dtype=t.long).random_(10)\n",
    "print('target: {0}, size: {1}'.format(targetforcross, targetforcross.size()))\n",
    "criterionforcross = nn.CrossEntropyLoss()\n",
    "print('Initial net end')\n",
    "\n",
    "learning_rate = 0.01\n",
    "epochs = 100\n",
    "print('通过手动实现SGD: weight = weight - learning_rate * gradient')\n",
    "for epoch in range(epochs):\n",
    "    netforcross.zero_grad() # 把net中所有可学习参数的梯度清零\n",
    "    outputforcross = netforcross(inputbatchsize5)\n",
    "    lossforcross = criterionforcross(outputforcross, targetforcross)\n",
    "    print('Train epoch: {0}, loss value: {1}'.format(epoch + 1, lossforcross.item())) # loss是个scalar\n",
    "    lossforcross.backward()\n",
    "    # 以下代码相当于optimizer.step()\n",
    "    for index, f in enumerate(netforcross.parameters()):\n",
    "        f.data.sub_(f.grad.data * learning_rate)# inplace 减法"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "torch.optim中实现了深度学习中绝大多数的优化方法，例如RMSProp、Adam、SGD等，更便于使用。\n",
    "\n",
    "以SGD为例，上述手动优化器的实现方式，等价于："
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial net begin\n",
      "input size: torch.Size([5, 1, 32, 32])\n",
      "output size: torch.Size([5, 10])\n",
      "target: tensor([4, 8, 6, 2, 1]), size: torch.Size([5])\n",
      "Initial net end\n",
      "Create optimizer as SGD, learning rate is 0.01\n",
      "Train epoch: 1, loss value: 2.309054374694824\n",
      "Train epoch: 2, loss value: 2.3034679889678955\n",
      "Train epoch: 3, loss value: 2.2980825901031494\n",
      "Train epoch: 4, loss value: 2.2927603721618652\n",
      "Train epoch: 5, loss value: 2.287485361099243\n",
      "Train epoch: 6, loss value: 2.282428026199341\n",
      "Train epoch: 7, loss value: 2.277522087097168\n",
      "Train epoch: 8, loss value: 2.272451400756836\n",
      "Train epoch: 9, loss value: 2.267273187637329\n",
      "Train epoch: 10, loss value: 2.262345790863037\n",
      "Train epoch: 11, loss value: 2.2573435306549072\n",
      "Train epoch: 12, loss value: 2.2522966861724854\n",
      "Train epoch: 13, loss value: 2.2474045753479004\n",
      "Train epoch: 14, loss value: 2.242506504058838\n",
      "Train epoch: 15, loss value: 2.2373647689819336\n",
      "Train epoch: 16, loss value: 2.2321648597717285\n",
      "Train epoch: 17, loss value: 2.226863384246826\n",
      "Train epoch: 18, loss value: 2.221585273742676\n",
      "Train epoch: 19, loss value: 2.216405153274536\n",
      "Train epoch: 20, loss value: 2.2110602855682373\n",
      "Train epoch: 21, loss value: 2.2054991722106934\n",
      "Train epoch: 22, loss value: 2.1996090412139893\n",
      "Train epoch: 23, loss value: 2.1935226917266846\n",
      "Train epoch: 24, loss value: 2.1870219707489014\n",
      "Train epoch: 25, loss value: 2.1801388263702393\n",
      "Train epoch: 26, loss value: 2.1730358600616455\n",
      "Train epoch: 27, loss value: 2.165714740753174\n",
      "Train epoch: 28, loss value: 2.158118724822998\n",
      "Train epoch: 29, loss value: 2.150303363800049\n",
      "Train epoch: 30, loss value: 2.14216947555542\n",
      "Train epoch: 31, loss value: 2.1337456703186035\n",
      "Train epoch: 32, loss value: 2.124967575073242\n",
      "Train epoch: 33, loss value: 2.1159701347351074\n",
      "Train epoch: 34, loss value: 2.1064019203186035\n",
      "Train epoch: 35, loss value: 2.0965590476989746\n",
      "Train epoch: 36, loss value: 2.086120367050171\n",
      "Train epoch: 37, loss value: 2.075251817703247\n",
      "Train epoch: 38, loss value: 2.063805103302002\n",
      "Train epoch: 39, loss value: 2.052138090133667\n",
      "Train epoch: 40, loss value: 2.039886951446533\n",
      "Train epoch: 41, loss value: 2.027174711227417\n",
      "Train epoch: 42, loss value: 2.013892889022827\n",
      "Train epoch: 43, loss value: 2.0003323554992676\n",
      "Train epoch: 44, loss value: 1.9864013195037842\n",
      "Train epoch: 45, loss value: 1.9719855785369873\n",
      "Train epoch: 46, loss value: 1.9571058750152588\n",
      "Train epoch: 47, loss value: 1.9419790506362915\n",
      "Train epoch: 48, loss value: 1.926627516746521\n",
      "Train epoch: 49, loss value: 1.9108959436416626\n",
      "Train epoch: 50, loss value: 1.8949114084243774\n",
      "Train epoch: 51, loss value: 1.8788995742797852\n",
      "Train epoch: 52, loss value: 1.8625074625015259\n",
      "Train epoch: 53, loss value: 1.8465092182159424\n",
      "Train epoch: 54, loss value: 1.8303569555282593\n",
      "Train epoch: 55, loss value: 1.8146240711212158\n",
      "Train epoch: 56, loss value: 1.798802137374878\n",
      "Train epoch: 57, loss value: 1.7827284336090088\n",
      "Train epoch: 58, loss value: 1.767000436782837\n",
      "Train epoch: 59, loss value: 1.751916527748108\n",
      "Train epoch: 60, loss value: 1.7378911972045898\n",
      "Train epoch: 61, loss value: 1.7243473529815674\n",
      "Train epoch: 62, loss value: 1.711503267288208\n",
      "Train epoch: 63, loss value: 1.6993589401245117\n",
      "Train epoch: 64, loss value: 1.6878712177276611\n",
      "Train epoch: 65, loss value: 1.6767085790634155\n",
      "Train epoch: 66, loss value: 1.6662778854370117\n",
      "Train epoch: 67, loss value: 1.6565395593643188\n",
      "Train epoch: 68, loss value: 1.6468467712402344\n",
      "Train epoch: 69, loss value: 1.6384830474853516\n",
      "Train epoch: 70, loss value: 1.629875898361206\n",
      "Train epoch: 71, loss value: 1.6218887567520142\n",
      "Train epoch: 72, loss value: 1.6146787405014038\n",
      "Train epoch: 73, loss value: 1.607593297958374\n",
      "Train epoch: 74, loss value: 1.6009471416473389\n",
      "Train epoch: 75, loss value: 1.594679594039917\n",
      "Train epoch: 76, loss value: 1.5888642072677612\n",
      "Train epoch: 77, loss value: 1.5831350088119507\n",
      "Train epoch: 78, loss value: 1.577640175819397\n",
      "Train epoch: 79, loss value: 1.5726252794265747\n",
      "Train epoch: 80, loss value: 1.5676841735839844\n",
      "Train epoch: 81, loss value: 1.563631296157837\n",
      "Train epoch: 82, loss value: 1.558897614479065\n",
      "Train epoch: 83, loss value: 1.5543992519378662\n",
      "Train epoch: 84, loss value: 1.550126314163208\n",
      "Train epoch: 85, loss value: 1.5462672710418701\n",
      "Train epoch: 86, loss value: 1.5421158075332642\n",
      "Train epoch: 87, loss value: 1.5382537841796875\n",
      "Train epoch: 88, loss value: 1.534472942352295\n",
      "Train epoch: 89, loss value: 1.5309312343597412\n",
      "Train epoch: 90, loss value: 1.5271894931793213\n",
      "Train epoch: 91, loss value: 1.523879051208496\n",
      "Train epoch: 92, loss value: 1.5202029943466187\n",
      "Train epoch: 93, loss value: 1.5165616273880005\n",
      "Train epoch: 94, loss value: 1.5132811069488525\n",
      "Train epoch: 95, loss value: 1.510145664215088\n",
      "Train epoch: 96, loss value: 1.5066378116607666\n",
      "Train epoch: 97, loss value: 1.503173828125\n",
      "Train epoch: 98, loss value: 1.4999114274978638\n",
      "Train epoch: 99, loss value: 1.496513843536377\n",
      "Train epoch: 100, loss value: 1.4933445453643799\n"
     ]
    }
   ],
   "source": [
    "import torch.optim as optim\n",
    "# batch-size x Channels x Height x Width\n",
    "print('Initial net begin')\n",
    "t.manual_seed(88)\n",
    "inputbatchsize5 = t.randn(5, 1, 32, 32)\n",
    "print('input size: {0}'.format(inputbatchsize5.size()))\n",
    "netforcross = Net()\n",
    "outputforcross = netforcross(inputbatchsize5)\n",
    "print('output size: {0}'.format(outputforcross.size()))\n",
    "targetforcross = t.empty(5, dtype=t.long).random_(10)\n",
    "print('target: {0}, size: {1}'.format(targetforcross, targetforcross.size()))\n",
    "criterionforcross = nn.CrossEntropyLoss()\n",
    "print('Initial net end')\n",
    "\n",
    "#新建一个优化器，指定要调整的参数和学习率\n",
    "print('Create optimizer as SGD, learning rate is 0.01')\n",
    "optimizer = optim.SGD(netforcross.parameters(), lr = 0.01)\n",
    "\n",
    "learning_rate = 0.01\n",
    "epochs = 100\n",
    "for epoch in range(epochs):\n",
    "    optimizer.zero_grad() # 把net中所有可学习参数的梯度清零\n",
    "    outputforcross = netforcross(inputbatchsize5)\n",
    "    lossforcross = criterionforcross(outputforcross, targetforcross)\n",
    "    print('Train epoch: {0}, loss value: {1}'.format(epoch + 1, lossforcross.item())) # loss是个scalar\n",
    "    lossforcross.backward()\n",
    "    optimizer.step()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 数据加载与预处理"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "在深度学习中数据加载及预处理是非常复杂繁琐的，但PyTorch提供了一些可极大简化和加快数据处理流程的工具。\n",
    "\n",
    "同时，对于常用的数据集，PyTorch也提供了封装好的接口供用户快速调用，这些数据集主要保存在torchvison中。\n",
    "\n",
    "torchvision实现了常用的图像数据加载功能，例如Imagenet、CIFAR10、MNIST等，以及常用的数据转换操作，这极大地方便了数据加载，并且代码具有可重用性。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 小试牛刀：CIFAR-10分类"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下面我们来尝试实现对CIFAR-10数据集的分类，步骤如下:\n",
    "\n",
    "- 使用torchvision加载并预处理CIFAR-10数据集\n",
    "- 定义网络\n",
    "- 定义损失函数和优化器\n",
    "- 训练网络并更新网络参数\n",
    "- 测试网络"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### CIFAR-10数据加载及预处理"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "[CIFAR-10](http://www.cs.toronto.edu/~kriz/cifar.html)是一个常用的彩色图片数据集，它有10个类别: 'airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck'。每张图片都是$3\\times32\\times32$，也即3-通道彩色图片，分辨率为$32\\times32$。\n",
    "<img src='./image/cifar10.png' />\n",
    "\n",
    "文件可以从上述地址下载，文件大小约161M，也可以通过torchvision自动下载，但需要花费一定时间。\n",
    "\n",
    "我们把数据存放在./cifar10/data下面"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src='./image/cifar10datapath.png' />"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "datapath = './cifar10/data'\n",
    "if not os.path.exists(datapath):\n",
    "    os.makedirs(datapath)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torchvision as tv\n",
    "import torch as t\n",
    "import torchvision.transforms as transforms\n",
    "from torchvision.transforms import ToPILImage\n",
    "import matplotlib.pyplot as plt\n",
    "import os\n",
    "show = ToPILImage() # 可以把Tensor转成Image，方便可视化"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "我们把数据下载之后，即可尝试进行数据的预处理与加载"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Files already downloaded and verified\n",
      "Files already downloaded and verified\n"
     ]
    }
   ],
   "source": [
    "# 第一次运行程序torchvision会自动下载CIFAR-10数据集，\n",
    "# 大约100M，需花费一定的时间，\n",
    "# 如果已经下载有CIFAR-10，可通过root参数指定\n",
    "\n",
    "# 定义对数据的预处理\n",
    "transform = transforms.Compose([\n",
    "        transforms.ToTensor(), # 转为Tensor\n",
    "        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)), # 归一化\n",
    "                             ])\n",
    "\n",
    "# 训练集\n",
    "trainset = tv.datasets.CIFAR10(root=datapath,\n",
    "                               train=True,\n",
    "                               download=True,\n",
    "                               transform=transform)\n",
    "\n",
    "trainloader = t.utils.data.DataLoader(trainset,\n",
    "                                      batch_size=4,\n",
    "                                      shuffle=True)\n",
    "\n",
    "# 测试集\n",
    "testset = tv.datasets.CIFAR10(datapath,\n",
    "                              train=False,\n",
    "                              download=True,\n",
    "                              transform=transform)\n",
    "\n",
    "testloader = t.utils.data.DataLoader(testset,\n",
    "                                     batch_size=4,\n",
    "                                     shuffle=False)\n",
    "\n",
    "classes = ('plane', 'car', 'bird', 'cat',\n",
    "           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trainset size: 50000\n",
      "Testset size: 10000\n"
     ]
    }
   ],
   "source": [
    "print('Trainset size: {0}'.format(len(trainset)))\n",
    "print('Testset size: {0}'.format(len(testset)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Dataset对象是一个数据集，可以按下标访问，返回形如(data, label)的数据。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "horse\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.Image.Image image mode=RGB size=50x50 at 0x1526156A668>"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "(data, label) = trainset[2000]\n",
    "print(classes[label])\n",
    "\n",
    "# (data + 1) / 2是为了还原被归一化的数据\n",
    "show((data + 1) / 2).resize((50, 50))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Dataloader是一个可迭代的对象，它将dataset返回的每一条数据拼接成一个batch，并提供多线程加速优化和数据打乱等操作。\n",
    "\n",
    "当程序对dataset的所有数据遍历完一遍之后，相应的对Dataloader也完成了一次迭代。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "      horse        frog        deer       horse\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<PIL.Image.Image image mode=RGB size=400x100 at 0x15207818A90>"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataiter = iter(trainloader)\n",
    "images, labels = dataiter.next() # 返回4张图片及标签\n",
    "print(' '.join('%11s'%classes[labels[j]] for j in range(4)))\n",
    "show(tv.utils.make_grid((images+1)/2)).resize((400,100))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 定义网络"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "拷贝上面的LeNet网络，修改self.conv1第一个参数为3通道，因CIFAR-10是3通道彩图。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Net(\n",
      "  (conv1): Conv2d(3, 6, kernel_size=(5, 5), stride=(1, 1))\n",
      "  (conv2): Conv2d(6, 16, kernel_size=(5, 5), stride=(1, 1))\n",
      "  (fc1): Linear(in_features=400, out_features=120, bias=True)\n",
      "  (fc2): Linear(in_features=120, out_features=84, bias=True)\n",
      "  (fc3): Linear(in_features=84, out_features=10, bias=True)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "class Net(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(Net, self).__init__()\n",
    "        self.conv1 = nn.Conv2d(3, 6, 5) \n",
    "        self.conv2 = nn.Conv2d(6, 16, 5)  \n",
    "        self.fc1   = nn.Linear(16*5*5, 120)  \n",
    "        self.fc2   = nn.Linear(120, 84)\n",
    "        self.fc3   = nn.Linear(84, 10)\n",
    "\n",
    "    def forward(self, x): \n",
    "        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2)) \n",
    "        x = F.max_pool2d(F.relu(self.conv2(x)), 2) \n",
    "        x = x.view(x.size()[0], -1) \n",
    "        x = F.relu(self.fc1(x))\n",
    "        x = F.relu(self.fc2(x))\n",
    "        x = self.fc3(x)        \n",
    "        return x\n",
    "\n",
    "\n",
    "net = Net()\n",
    "print(net)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 定义损失函数和优化器(loss和optimizer)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "以交叉熵为损失函数\n",
    "\n",
    "以Adam为优化函数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch import optim\n",
    "criterion = nn.CrossEntropyLoss() # 交叉熵损失函数\n",
    "optimizer = t.optim.Adam(net.parameters())\n",
    "# optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 训练网络"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "所有网络的训练流程都是类似的，不断地执行如下流程：\n",
    "\n",
    "- 输入数据\n",
    "- 前向传播+反向传播\n",
    "- 更新参数"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12500\n",
      "50000\n"
     ]
    }
   ],
   "source": [
    "print(len(trainloader))\n",
    "print(len(trainset))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 233,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 1 begin\n",
      "[1,  2000] loss: 0.293\n",
      "[1,  4000] loss: 0.296\n",
      "[1,  6000] loss: 0.294\n",
      "[1,  8000] loss: 0.285\n",
      "[1, 10000] loss: 0.286\n",
      "[1, 12000] loss: 0.283\n",
      "Train Loss: 0.289488, Acc: 0.589340\n",
      "save model for best accuate: 0.58934 begin\n",
      "save model for best accuate: 0.58934 end\n",
      "Epoch: 1 end\n",
      "Epoch: 2 begin\n",
      "[2,  2000] loss: 0.265\n",
      "[2,  4000] loss: 0.272\n",
      "[2,  6000] loss: 0.273\n",
      "[2,  8000] loss: 0.270\n",
      "[2, 10000] loss: 0.272\n",
      "[2, 12000] loss: 0.270\n",
      "Train Loss: 0.270589, Acc: 0.616080\n",
      "save model for best accuate: 0.61608 begin\n",
      "save model for best accuate: 0.61608 end\n",
      "Epoch: 2 end\n",
      "Epoch: 3 begin\n",
      "[3,  2000] loss: 0.256\n",
      "[3,  4000] loss: 0.254\n",
      "[3,  6000] loss: 0.257\n",
      "[3,  8000] loss: 0.258\n",
      "[3, 10000] loss: 0.257\n",
      "[3, 12000] loss: 0.259\n",
      "Train Loss: 0.256798, Acc: 0.636320\n",
      "save model for best accuate: 0.63632 begin\n",
      "save model for best accuate: 0.63632 end\n",
      "Epoch: 3 end\n",
      "Epoch: 4 begin\n",
      "[4,  2000] loss: 0.242\n",
      "[4,  4000] loss: 0.244\n",
      "[4,  6000] loss: 0.243\n",
      "[4,  8000] loss: 0.248\n",
      "[4, 10000] loss: 0.251\n",
      "[4, 12000] loss: 0.251\n",
      "Train Loss: 0.246549, Acc: 0.653200\n",
      "save model for best accuate: 0.6532 begin\n",
      "save model for best accuate: 0.6532 end\n",
      "Epoch: 4 end\n",
      "Epoch: 5 begin\n",
      "[5,  2000] loss: 0.233\n",
      "[5,  4000] loss: 0.228\n",
      "[5,  6000] loss: 0.234\n",
      "[5,  8000] loss: 0.241\n",
      "[5, 10000] loss: 0.242\n",
      "[5, 12000] loss: 0.246\n",
      "Train Loss: 0.237677, Acc: 0.663560\n",
      "save model for best accuate: 0.66356 begin\n",
      "save model for best accuate: 0.66356 end\n",
      "Epoch: 5 end\n",
      "Epoch: 6 begin\n",
      "[6,  2000] loss: 0.232\n",
      "[6,  4000] loss: 0.226\n",
      "[6,  6000] loss: 0.233\n",
      "[6,  8000] loss: 0.232\n",
      "[6, 10000] loss: 0.234\n",
      "[6, 12000] loss: 0.239\n",
      "Train Loss: 0.232507, Acc: 0.672100\n",
      "save model for best accuate: 0.6721 begin\n",
      "save model for best accuate: 0.6721 end\n",
      "Epoch: 6 end\n",
      "Epoch: 7 begin\n",
      "[7,  2000] loss: 0.220\n",
      "[7,  4000] loss: 0.217\n",
      "[7,  6000] loss: 0.226\n",
      "[7,  8000] loss: 0.233\n",
      "[7, 10000] loss: 0.239\n",
      "[7, 12000] loss: 0.228\n",
      "Train Loss: 0.227266, Acc: 0.679040\n",
      "save model for best accuate: 0.67904 begin\n",
      "save model for best accuate: 0.67904 end\n",
      "Epoch: 7 end\n",
      "Epoch: 8 begin\n",
      "[8,  2000] loss: 0.214\n",
      "[8,  4000] loss: 0.217\n",
      "[8,  6000] loss: 0.222\n",
      "[8,  8000] loss: 0.219\n",
      "[8, 10000] loss: 0.225\n",
      "[8, 12000] loss: 0.226\n",
      "Train Loss: 0.220878, Acc: 0.688600\n",
      "save model for best accuate: 0.6886 begin\n",
      "save model for best accuate: 0.6886 end\n",
      "Epoch: 8 end\n",
      "Epoch: 9 begin\n",
      "[9,  2000] loss: 0.206\n",
      "[9,  4000] loss: 0.216\n",
      "[9,  6000] loss: 0.217\n",
      "[9,  8000] loss: 0.224\n",
      "[9, 10000] loss: 0.215\n",
      "[9, 12000] loss: 0.226\n",
      "Train Loss: 0.217180, Acc: 0.696180\n",
      "save model for best accuate: 0.69618 begin\n",
      "save model for best accuate: 0.69618 end\n",
      "Epoch: 9 end\n",
      "Epoch: 10 begin\n",
      "[10,  2000] loss: 0.207\n",
      "[10,  4000] loss: 0.207\n",
      "[10,  6000] loss: 0.215\n",
      "[10,  8000] loss: 0.214\n",
      "[10, 10000] loss: 0.220\n",
      "[10, 12000] loss: 0.218\n",
      "Train Loss: 0.213508, Acc: 0.698240\n",
      "save model for best accuate: 0.69824 begin\n",
      "save model for best accuate: 0.69824 end\n",
      "Epoch: 10 end\n",
      "Epoch: 11 begin\n",
      "[11,  2000] loss: 0.198\n",
      "[11,  4000] loss: 0.204\n",
      "[11,  6000] loss: 0.210\n",
      "[11,  8000] loss: 0.210\n",
      "[11, 10000] loss: 0.213\n",
      "[11, 12000] loss: 0.219\n",
      "Train Loss: 0.209104, Acc: 0.705500\n",
      "save model for best accuate: 0.7055 begin\n",
      "save model for best accuate: 0.7055 end\n",
      "Epoch: 11 end\n",
      "Epoch: 12 begin\n",
      "[12,  2000] loss: 0.194\n",
      "[12,  4000] loss: 0.198\n",
      "[12,  6000] loss: 0.207\n",
      "[12,  8000] loss: 0.214\n",
      "[12, 10000] loss: 0.208\n",
      "[12, 12000] loss: 0.214\n",
      "Train Loss: 0.206412, Acc: 0.711660\n",
      "save model for best accuate: 0.71166 begin\n",
      "save model for best accuate: 0.71166 end\n",
      "Epoch: 12 end\n",
      "Epoch: 13 begin\n",
      "[13,  2000] loss: 0.195\n",
      "[13,  4000] loss: 0.203\n",
      "[13,  6000] loss: 0.207\n",
      "[13,  8000] loss: 0.205\n",
      "[13, 10000] loss: 0.210\n",
      "[13, 12000] loss: 0.211\n",
      "Train Loss: 0.205310, Acc: 0.713300\n",
      "save model for best accuate: 0.7133 begin\n",
      "save model for best accuate: 0.7133 end\n",
      "Epoch: 13 end\n",
      "Epoch: 14 begin\n",
      "[14,  2000] loss: 0.193\n",
      "[14,  4000] loss: 0.199\n",
      "[14,  6000] loss: 0.207\n",
      "[14,  8000] loss: 0.200\n",
      "[14, 10000] loss: 0.207\n",
      "[14, 12000] loss: 0.209\n",
      "Train Loss: 0.202963, Acc: 0.715280\n",
      "save model for best accuate: 0.71528 begin\n",
      "save model for best accuate: 0.71528 end\n",
      "Epoch: 14 end\n",
      "Epoch: 15 begin\n",
      "[15,  2000] loss: 0.193\n",
      "[15,  4000] loss: 0.197\n",
      "[15,  6000] loss: 0.199\n",
      "[15,  8000] loss: 0.204\n",
      "[15, 10000] loss: 0.205\n",
      "[15, 12000] loss: 0.204\n",
      "Train Loss: 0.200400, Acc: 0.720300\n",
      "save model for best accuate: 0.7203 begin\n",
      "save model for best accuate: 0.7203 end\n",
      "Epoch: 15 end\n",
      "Epoch: 16 begin\n",
      "[16,  2000] loss: 0.190\n",
      "[16,  4000] loss: 0.196\n",
      "[16,  6000] loss: 0.196\n",
      "[16,  8000] loss: 0.202\n",
      "[16, 10000] loss: 0.198\n",
      "[16, 12000] loss: 0.205\n",
      "Train Loss: 0.198217, Acc: 0.723040\n",
      "save model for best accuate: 0.72304 begin\n",
      "save model for best accuate: 0.72304 end\n",
      "Epoch: 16 end\n",
      "Epoch: 17 begin\n",
      "[17,  2000] loss: 0.188\n",
      "[17,  4000] loss: 0.194\n",
      "[17,  6000] loss: 0.196\n",
      "[17,  8000] loss: 0.199\n",
      "[17, 10000] loss: 0.203\n",
      "[17, 12000] loss: 0.203\n",
      "Train Loss: 0.196911, Acc: 0.724940\n",
      "save model for best accuate: 0.72494 begin\n",
      "save model for best accuate: 0.72494 end\n",
      "Epoch: 17 end\n",
      "Epoch: 18 begin\n",
      "[18,  2000] loss: 0.184\n",
      "[18,  4000] loss: 0.192\n",
      "[18,  6000] loss: 0.196\n",
      "[18,  8000] loss: 0.196\n",
      "[18, 10000] loss: 0.198\n",
      "[18, 12000] loss: 0.206\n",
      "Train Loss: 0.195447, Acc: 0.728160\n",
      "save model for best accuate: 0.72816 begin\n",
      "save model for best accuate: 0.72816 end\n",
      "Epoch: 18 end\n",
      "Epoch: 19 begin\n",
      "[19,  2000] loss: 0.179\n",
      "[19,  4000] loss: 0.189\n",
      "[19,  6000] loss: 0.196\n",
      "[19,  8000] loss: 0.196\n",
      "[19, 10000] loss: 0.201\n",
      "[19, 12000] loss: 0.198\n",
      "Train Loss: 0.193414, Acc: 0.729520\n",
      "save model for best accuate: 0.72952 begin\n",
      "save model for best accuate: 0.72952 end\n",
      "Epoch: 19 end\n",
      "Epoch: 20 begin\n",
      "[20,  2000] loss: 0.181\n",
      "[20,  4000] loss: 0.185\n",
      "[20,  6000] loss: 0.197\n",
      "[20,  8000] loss: 0.188\n",
      "[20, 10000] loss: 0.200\n",
      "[20, 12000] loss: 0.200\n",
      "Train Loss: 0.192055, Acc: 0.733800\n",
      "save model for best accuate: 0.7338 begin\n",
      "save model for best accuate: 0.7338 end\n",
      "Epoch: 20 end\n",
      "Epoch: 21 begin\n",
      "[21,  2000] loss: 0.177\n",
      "[21,  4000] loss: 0.189\n",
      "[21,  6000] loss: 0.198\n",
      "[21,  8000] loss: 0.189\n",
      "[21, 10000] loss: 0.191\n",
      "[21, 12000] loss: 0.200\n",
      "Train Loss: 0.191171, Acc: 0.735000\n",
      "save model for best accuate: 0.735 begin\n",
      "save model for best accuate: 0.735 end\n",
      "Epoch: 21 end\n",
      "Epoch: 22 begin\n",
      "[22,  2000] loss: 0.182\n",
      "[22,  4000] loss: 0.182\n",
      "[22,  6000] loss: 0.190\n",
      "[22,  8000] loss: 0.183\n",
      "[22, 10000] loss: 0.198\n",
      "[22, 12000] loss: 0.190\n",
      "Train Loss: 0.187858, Acc: 0.737340\n",
      "save model for best accuate: 0.73734 begin\n",
      "save model for best accuate: 0.73734 end\n",
      "Epoch: 22 end\n",
      "Epoch: 23 begin\n",
      "[23,  2000] loss: 0.177\n",
      "[23,  4000] loss: 0.178\n",
      "[23,  6000] loss: 0.189\n",
      "[23,  8000] loss: 0.185\n",
      "[23, 10000] loss: 0.196\n",
      "[23, 12000] loss: 0.198\n",
      "Train Loss: 0.187845, Acc: 0.739040\n",
      "save model for best accuate: 0.73904 begin\n",
      "save model for best accuate: 0.73904 end\n",
      "Epoch: 23 end\n",
      "Epoch: 24 begin\n",
      "[24,  2000] loss: 0.178\n",
      "[24,  4000] loss: 0.184\n",
      "[24,  6000] loss: 0.179\n",
      "[24,  8000] loss: 0.183\n",
      "[24, 10000] loss: 0.194\n",
      "[24, 12000] loss: 0.192\n",
      "Train Loss: 0.185613, Acc: 0.742700\n",
      "save model for best accuate: 0.7427 begin\n",
      "save model for best accuate: 0.7427 end\n",
      "Epoch: 24 end\n",
      "Epoch: 25 begin\n",
      "[25,  2000] loss: 0.172\n",
      "[25,  4000] loss: 0.170\n",
      "[25,  6000] loss: 0.188\n",
      "[25,  8000] loss: 0.191\n",
      "[25, 10000] loss: 0.191\n",
      "[25, 12000] loss: 0.190\n",
      "Train Loss: 0.183947, Acc: 0.745300\n",
      "save model for best accuate: 0.7453 begin\n",
      "save model for best accuate: 0.7453 end\n",
      "Epoch: 25 end\n",
      "Epoch: 26 begin\n",
      "[26,  2000] loss: 0.174\n",
      "[26,  4000] loss: 0.184\n",
      "[26,  6000] loss: 0.179\n",
      "[26,  8000] loss: 0.186\n",
      "[26, 10000] loss: 0.181\n",
      "[26, 12000] loss: 0.193\n",
      "Train Loss: 0.183487, Acc: 0.747620\n",
      "save model for best accuate: 0.74762 begin\n",
      "save model for best accuate: 0.74762 end\n",
      "Epoch: 26 end\n",
      "Epoch: 27 begin\n",
      "[27,  2000] loss: 0.170\n",
      "[27,  4000] loss: 0.179\n",
      "[27,  6000] loss: 0.182\n",
      "[27,  8000] loss: 0.190\n",
      "[27, 10000] loss: 0.185\n",
      "[27, 12000] loss: 0.184\n",
      "Train Loss: 0.182112, Acc: 0.750480\n",
      "save model for best accuate: 0.75048 begin\n",
      "save model for best accuate: 0.75048 end\n",
      "Epoch: 27 end\n",
      "Epoch: 28 begin\n",
      "[28,  2000] loss: 0.175\n",
      "[28,  4000] loss: 0.175\n",
      "[28,  6000] loss: 0.180\n",
      "[28,  8000] loss: 0.176\n",
      "[28, 10000] loss: 0.181\n",
      "[28, 12000] loss: 0.189\n",
      "Train Loss: 0.182558, Acc: 0.747280\n",
      "Epoch: 28 end\n",
      "Epoch: 29 begin\n",
      "[29,  2000] loss: 0.178\n",
      "[29,  4000] loss: 0.176\n",
      "[29,  6000] loss: 0.184\n",
      "[29,  8000] loss: 0.179\n",
      "[29, 10000] loss: 0.181\n",
      "[29, 12000] loss: 0.178\n",
      "Train Loss: 0.179470, Acc: 0.754380\n",
      "save model for best accuate: 0.75438 begin\n",
      "save model for best accuate: 0.75438 end\n",
      "Epoch: 29 end\n",
      "Epoch: 30 begin\n",
      "[30,  2000] loss: 0.164\n",
      "[30,  4000] loss: 0.173\n",
      "[30,  6000] loss: 0.174\n",
      "[30,  8000] loss: 0.193\n",
      "[30, 10000] loss: 0.181\n",
      "[30, 12000] loss: 0.197\n",
      "Train Loss: 0.180765, Acc: 0.752320\n",
      "Epoch: 30 end\n",
      "Epoch: 31 begin\n",
      "[31,  2000] loss: 0.169\n",
      "[31,  4000] loss: 0.172\n",
      "[31,  6000] loss: 0.183\n",
      "[31,  8000] loss: 0.186\n",
      "[31, 10000] loss: 0.182\n",
      "[31, 12000] loss: 0.190\n",
      "Train Loss: 0.180588, Acc: 0.751460\n",
      "Epoch: 31 end\n",
      "Epoch: 32 begin\n",
      "[32,  2000] loss: 0.167\n",
      "[32,  4000] loss: 0.178\n",
      "[32,  6000] loss: 0.181\n",
      "[32,  8000] loss: 0.185\n",
      "[32, 10000] loss: 0.183\n",
      "[32, 12000] loss: 0.177\n",
      "Train Loss: 0.178702, Acc: 0.755560\n",
      "save model for best accuate: 0.75556 begin\n",
      "save model for best accuate: 0.75556 end\n",
      "Epoch: 32 end\n",
      "Epoch: 33 begin\n",
      "[33,  2000] loss: 0.175\n",
      "[33,  4000] loss: 0.171\n",
      "[33,  6000] loss: 0.175\n",
      "[33,  8000] loss: 0.181\n",
      "[33, 10000] loss: 0.175\n",
      "[33, 12000] loss: 0.193\n",
      "Train Loss: 0.178711, Acc: 0.754780\n",
      "Epoch: 33 end\n",
      "Epoch: 34 begin\n",
      "[34,  2000] loss: 0.163\n",
      "[34,  4000] loss: 0.170\n",
      "[34,  6000] loss: 0.176\n",
      "[34,  8000] loss: 0.182\n",
      "[34, 10000] loss: 0.180\n",
      "[34, 12000] loss: 0.181\n",
      "Train Loss: 0.175098, Acc: 0.756000\n",
      "save model for best accuate: 0.756 begin\n",
      "save model for best accuate: 0.756 end\n",
      "Epoch: 34 end\n",
      "Epoch: 35 begin\n",
      "[35,  2000] loss: 0.163\n",
      "[35,  4000] loss: 0.170\n",
      "[35,  6000] loss: 0.175\n",
      "[35,  8000] loss: 0.180\n",
      "[35, 10000] loss: 0.172\n",
      "[35, 12000] loss: 0.180\n",
      "Train Loss: 0.173263, Acc: 0.761600\n",
      "save model for best accuate: 0.7616 begin\n",
      "save model for best accuate: 0.7616 end\n",
      "Epoch: 35 end\n",
      "Epoch: 36 begin\n",
      "[36,  2000] loss: 0.161\n",
      "[36,  4000] loss: 0.168\n",
      "[36,  6000] loss: 0.166\n",
      "[36,  8000] loss: 0.178\n",
      "[36, 10000] loss: 0.178\n",
      "[36, 12000] loss: 0.180\n",
      "Train Loss: 0.172249, Acc: 0.764020\n",
      "save model for best accuate: 0.76402 begin\n",
      "save model for best accuate: 0.76402 end\n",
      "Epoch: 36 end\n",
      "Epoch: 37 begin\n",
      "[37,  2000] loss: 0.161\n",
      "[37,  4000] loss: 0.169\n",
      "[37,  6000] loss: 0.166\n",
      "[37,  8000] loss: 0.181\n",
      "[37, 10000] loss: 0.184\n",
      "[37, 12000] loss: 0.172\n",
      "Train Loss: 0.172209, Acc: 0.764340\n",
      "save model for best accuate: 0.76434 begin\n",
      "save model for best accuate: 0.76434 end\n",
      "Epoch: 37 end\n",
      "Epoch: 38 begin\n",
      "[38,  2000] loss: 0.155\n",
      "[38,  4000] loss: 0.181\n",
      "[38,  6000] loss: 0.170\n",
      "[38,  8000] loss: 0.169\n",
      "[38, 10000] loss: 0.178\n",
      "[38, 12000] loss: 0.175\n",
      "Train Loss: 0.171767, Acc: 0.765180\n",
      "save model for best accuate: 0.76518 begin\n",
      "save model for best accuate: 0.76518 end\n",
      "Epoch: 38 end\n",
      "Epoch: 39 begin\n",
      "[39,  2000] loss: 0.157\n",
      "[39,  4000] loss: 0.166\n",
      "[39,  6000] loss: 0.167\n",
      "[39,  8000] loss: 0.179\n",
      "[39, 10000] loss: 0.170\n",
      "[39, 12000] loss: 0.171\n",
      "Train Loss: 0.169038, Acc: 0.768780\n",
      "save model for best accuate: 0.76878 begin\n",
      "save model for best accuate: 0.76878 end\n",
      "Epoch: 39 end\n",
      "Epoch: 40 begin\n",
      "[40,  2000] loss: 0.156\n",
      "[40,  4000] loss: 0.164\n",
      "[40,  6000] loss: 0.177\n",
      "[40,  8000] loss: 0.171\n",
      "[40, 10000] loss: 0.180\n",
      "[40, 12000] loss: 0.175\n",
      "Train Loss: 0.171263, Acc: 0.767980\n",
      "Epoch: 40 end\n",
      "Epoch: 41 begin\n",
      "[41,  2000] loss: 0.162\n",
      "[41,  4000] loss: 0.161\n",
      "[41,  6000] loss: 0.167\n",
      "[41,  8000] loss: 0.175\n",
      "[41, 10000] loss: 0.178\n",
      "[41, 12000] loss: 0.190\n",
      "Train Loss: 0.172052, Acc: 0.767560\n",
      "Epoch: 41 end\n",
      "Epoch: 42 begin\n",
      "[42,  2000] loss: 0.155\n",
      "[42,  4000] loss: 0.175\n",
      "[42,  6000] loss: 0.166\n",
      "[42,  8000] loss: 0.171\n",
      "[42, 10000] loss: 0.167\n",
      "[42, 12000] loss: 0.170\n",
      "Train Loss: 0.167549, Acc: 0.774160\n",
      "save model for best accuate: 0.77416 begin\n",
      "save model for best accuate: 0.77416 end\n",
      "Epoch: 42 end\n",
      "Epoch: 43 begin\n",
      "[43,  2000] loss: 0.172\n",
      "[43,  4000] loss: 0.164\n",
      "[43,  6000] loss: 0.169\n",
      "[43,  8000] loss: 0.164\n",
      "[43, 10000] loss: 0.172\n",
      "[43, 12000] loss: 0.174\n",
      "Train Loss: 0.170145, Acc: 0.769200\n",
      "Epoch: 43 end\n",
      "Epoch: 44 begin\n",
      "[44,  2000] loss: 0.154\n",
      "[44,  4000] loss: 0.156\n",
      "[44,  6000] loss: 0.171\n",
      "[44,  8000] loss: 0.173\n",
      "[44, 10000] loss: 0.175\n",
      "[44, 12000] loss: 0.168\n",
      "Train Loss: 0.167611, Acc: 0.772380\n",
      "Epoch: 44 end\n",
      "Epoch: 45 begin\n",
      "[45,  2000] loss: 0.153\n",
      "[45,  4000] loss: 0.164\n",
      "[45,  6000] loss: 0.207\n",
      "[45,  8000] loss: 0.174\n",
      "[45, 10000] loss: 0.180\n",
      "[45, 12000] loss: 0.180\n",
      "Train Loss: 0.175624, Acc: 0.765120\n",
      "Epoch: 45 end\n",
      "Epoch: 46 begin\n",
      "[46,  2000] loss: 0.156\n",
      "[46,  4000] loss: 0.161\n",
      "[46,  6000] loss: 0.176\n",
      "[46,  8000] loss: 0.170\n",
      "[46, 10000] loss: 0.169\n",
      "[46, 12000] loss: 0.166\n",
      "Train Loss: 0.166098, Acc: 0.776780\n",
      "save model for best accuate: 0.77678 begin\n",
      "save model for best accuate: 0.77678 end\n",
      "Epoch: 46 end\n",
      "Epoch: 47 begin\n",
      "[47,  2000] loss: 0.160\n",
      "[47,  4000] loss: 0.169\n",
      "[47,  6000] loss: 0.159\n",
      "[47,  8000] loss: 0.166\n",
      "[47, 10000] loss: 0.168\n",
      "[47, 12000] loss: 0.187\n",
      "Train Loss: 0.168385, Acc: 0.772060\n",
      "Epoch: 47 end\n",
      "Epoch: 48 begin\n",
      "[48,  2000] loss: 0.162\n",
      "[48,  4000] loss: 0.165\n",
      "[48,  6000] loss: 0.155\n",
      "[48,  8000] loss: 0.163\n",
      "[48, 10000] loss: 0.168\n",
      "[48, 12000] loss: 0.177\n",
      "Train Loss: 0.165222, Acc: 0.775500\n",
      "Epoch: 48 end\n",
      "Epoch: 49 begin\n",
      "[49,  2000] loss: 0.159\n",
      "[49,  4000] loss: 0.160\n",
      "[49,  6000] loss: 0.169\n",
      "[49,  8000] loss: 0.188\n",
      "[49, 10000] loss: 0.161\n",
      "[49, 12000] loss: 0.179\n",
      "Train Loss: 0.169898, Acc: 0.773220\n",
      "Epoch: 49 end\n",
      "Epoch: 50 begin\n",
      "[50,  2000] loss: 0.152\n",
      "[50,  4000] loss: 0.158\n",
      "[50,  6000] loss: 0.160\n",
      "[50,  8000] loss: 0.182\n",
      "[50, 10000] loss: 0.169\n",
      "[50, 12000] loss: 0.183\n",
      "Train Loss: 0.167347, Acc: 0.774140\n",
      "Epoch: 50 end\n",
      "Finished Training\n"
     ]
    }
   ],
   "source": [
    "t.set_num_threads(8)\n",
    "# 设置checkpoint的模型存放路径\n",
    "modelfolder = './cifar10/model'\n",
    "if not os.path.exists(modelfolder):\n",
    "    os.makedirs(modelfolder)\n",
    "modelfile = os.path.join(modelfolder, 'cifar10_lenet.pt')\n",
    "best_acc = 0.\n",
    "for epoch in range(50):\n",
    "    print('Epoch: {0} begin'.format(epoch + 1))\n",
    "    net.train()\n",
    "    running_loss = 0.0\n",
    "    train_loss = 0.\n",
    "    train_acc = 0.\n",
    "    for i, (inputs, labels) in enumerate(trainloader):\n",
    "        # 梯度清零\n",
    "        optimizer.zero_grad()\n",
    "        # forward + backward \n",
    "        outputs = net(inputs)\n",
    "        loss = criterion(outputs, labels)\n",
    "        train_loss += loss.data.item()\n",
    "        pred = t.max(outputs, 1)[1]\n",
    "        # 通过torch.sum统计预测正确的数量\n",
    "        train_correct = t.sum(pred == labels)\n",
    "        # train_correct是Tensor类型，需要通过item()，得到数值\n",
    "        train_acc += train_correct.item()\n",
    "#         print('train_correct is {0}/4, total acc: {1}'.format(train_correct, train_acc))\n",
    "        loss.backward()   \n",
    "        # 更新参数 \n",
    "        optimizer.step()\n",
    "        # 打印log信息\n",
    "        # loss 是一个scalar,需要使用loss.item()来获取数值，不能使用loss[0]\n",
    "        running_loss += loss.item()\n",
    "        if i % 2000 == 1999: # 每2000个batch打印一下训练状态\n",
    "            print('[%d, %5d] loss: %.3f' \\\n",
    "                  % (epoch+1, i+1, running_loss / (2000 * 4)))\n",
    "            running_loss = 0.0\n",
    "    accurate = train_acc / (len(trainset))\n",
    "    print('Train Loss: {:.6f}, Acc: {:.6f}'.format(train_loss / (len(trainset)), accurate))\n",
    "    # 如果当前epoch的准确率是最高的，则保存模型。\n",
    "    # 模型的权重信息是通过state_dict()获得的，这也是模型的最关键信息\n",
    "    if accurate > best_acc:\n",
    "        print('save model for best accuate: {0} begin'.format(accurate))\n",
    "        best_acc = accurate\n",
    "        t.save({\n",
    "            'epoch': epoch + 1,\n",
    "            'state_dict': net.state_dict(),\n",
    "            'best_acc': best_acc,\n",
    "        }, modelfile)\n",
    "        print('save model for best accuate: {0} end'.format(accurate))\n",
    "    print('Epoch: {0} end'.format(epoch + 1))\n",
    "print('Finished Training')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "将测试图片输入到网络中，计算它的label，然后与实际的label进行比较。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 308,
   "metadata": {},
   "outputs": [],
   "source": [
    "def loadmodel():\n",
    "    modelfolder = './cifar10/model'\n",
    "    modelfile = os.path.join(modelfolder, 'cifar10_lenet.pt')\n",
    "    if not os.path.exists(modelfile):\n",
    "        print('model not exist!')\n",
    "        return None, None\n",
    "    print('Loading best model: {0} begin'.format(modelfile))\n",
    "    mymodel = Net()\n",
    "    checkpoint = t.load(modelfile)\n",
    "    epoch = checkpoint['epoch']\n",
    "    best_acc = checkpoint['best_acc']\n",
    "    mymodel.load_state_dict(checkpoint['state_dict'])\n",
    "    print('Loading best model end, epoch is {0}, training accuracy is {1}'.format(epoch, best_acc))\n",
    "    return mymodel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 304,
   "metadata": {},
   "outputs": [],
   "source": [
    "def getpilimage(image):\n",
    "    return show((image + 1) / 2).resize((100, 100))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下面是如何使用matplotlib输出图片"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 305,
   "metadata": {},
   "outputs": [],
   "source": [
    "def outputimage(images):\n",
    "    plt.figure(\"cifar10\")\n",
    "    initplost = 141\n",
    "    for i in range(4):\n",
    "        plt.subplot(141 + i)\n",
    "        plt.imshow(getpilimage(images[i]))\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 309,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading best model: ./cifar10/model\\cifar10_lenet.pt begin\n",
      "Loading best model end, epoch is 46, training accuracy is 0.77678\n",
      "Prediction: 1 / 10\n",
      "实际的label:    cat  ship  ship plane\n",
      "预测结果:    cat plane plane plane\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 2 / 10\n",
      "实际的label:   frog  frog   car  frog\n",
      "预测结果:   frog   cat   dog  frog\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 3 / 10\n",
      "实际的label:    cat   car plane truck\n",
      "预测结果:    dog   car plane truck\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 4 / 10\n",
      "实际的label:    dog horse truck  ship\n",
      "预测结果:   frog horse truck  ship\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 5 / 10\n",
      "实际的label:    dog horse  ship  frog\n",
      "预测结果:    dog   cat  ship  frog\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 6 / 10\n",
      "实际的label:  horse plane  deer truck\n",
      "预测结果:  horse plane  bird   car\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 7 / 10\n",
      "实际的label:    dog  bird  deer plane\n",
      "预测结果:    dog  bird  deer plane\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 8 / 10\n",
      "实际的label:  truck  frog  frog   dog\n",
      "预测结果:  truck  frog  frog   dog\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 9 / 10\n",
      "实际的label:   deer   dog truck  bird\n",
      "预测结果:   bird   cat truck   cat\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Prediction: 10 / 10\n",
      "实际的label:   deer   car truck   dog\n",
      "预测结果:   deer   car truck   dog\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "mymodel = loadmodel()\n",
    "if mymodel:\n",
    "    mymodel.eval()\n",
    "    dataiter = iter(testloader)\n",
    "    predictcount = 10\n",
    "    with t.no_grad():\n",
    "        for i in range(predictcount):\n",
    "            print('Prediction: {0} / {1}'.format(i+1, predictcount))\n",
    "            images, labels = dataiter.next() # 一个batch返回4张图片\n",
    "            print('实际的label: ', ' '.join(\\\n",
    "                        '%05s'%classes[labels[j]] for j in range(4)))\n",
    "            # 计算图片在每个类别上的分数\n",
    "            outputs = mymodel(images)\n",
    "    #         outputs = mymodel(images)\n",
    "            # 得分最高的那个类\n",
    "            _, predicted = t.max(outputs.data, 1)\n",
    "            print('预测结果: ', ' '.join('%5s'\\\n",
    "                        % classes[predicted[j]] for j in range(4)))\n",
    "            outputimage(images)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "再来看看在整个测试集上的效果。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 310,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading best model: ./cifar10/model\\cifar10_lenet.pt begin\n",
      "Loading best model end, epoch is 46, training accuracy is 0.77678\n",
      "10000张测试集的准确率为: 60%\n"
     ]
    }
   ],
   "source": [
    "correct = 0 # 预测正确的图片数\n",
    "total = 0 # 总共的图片数\n",
    "mymodel = loadmodel()\n",
    "if mymodel:\n",
    "    mymodel.eval()\n",
    "    # 由于测试的时候不需要求导，可以暂时关闭autograd，提高速度，节约内存\n",
    "    with t.no_grad():\n",
    "        for data in testloader:\n",
    "            images, labels = data\n",
    "            outputs = mymodel(images)\n",
    "            _, predicted = t.max(outputs, 1)\n",
    "            total += labels.size(0)\n",
    "            correct += (predicted == labels).sum()\n",
    "\n",
    "    print('{0}张测试集的准确率为: {1}%'.format(total, 100 * correct / total))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "如果想获得更好的效果，理论上就是进行更多epoch的训练，即训练更多次。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "以上就是PyTorch的基础用法，以及相当简单的通过CNN的LeNet模型，训练及使用Cifar10数据集的例子。"
   ]
  },
  {
   "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": "3.6.7"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}