Repository: TommyZihao/zihaopytorch Branch: master Commit: ca6c2e1434fc Files: 3 Total size: 97.9 KB Directory structure: gitextract_c9uf5xzq/ ├── LICENSE ├── README.md └── 对Fashion-MNIST数据集中的时尚物品进行分类.ipynb ================================================ FILE CONTENTS ================================================ ================================================ FILE: LICENSE ================================================ MIT License Copyright (c) 2019 Tommy in Tongji Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. ================================================ FILE: README.md ================================================ # zihaopytorch simple tutorial of pytorch ================================================ FILE: 对Fashion-MNIST数据集中的时尚物品进行分类.ipynb ================================================ { "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# 用pytorch搭建全连接神经网络，对Fashion-MNIST数据集中的时尚物品进行分类" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 概要\n", "本课，你将通过业界知名的深度学习框架pytorch搭建四层全连接神经网络，分析Fashion-MNIST数据集中的六万张训练集图片和一万张测试集图片，观察训练误差和验证误差随训练代数提高的变化。\n", "首先搭建普通的四层全连接神经网络，训练后发现过拟合明显。\n", "通过Dropout方法，减少过拟合。\n", "你将掌握图像多分类、准确率与误差分析、搭建全连接神经网络并通过Adam算法进行梯度下降训练，最重要的是，熟悉pytorch这一简洁明快的深度学习工具。\n", "\n", "Github：[zihaopytorch](https://github.com/TommyZihao/zihaopytorch)
\n", "\n", "![Fashion-MNIST时尚物品数据集](https://upload-images.jianshu.io/upload_images/13714448-e0d2b1be2bb1702e.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)\n", "\n", "## 作者\n", "\n", "张子豪 同济大学研究生 2019-06-06\n", "\n", "Bilibili视频教程：[同济子豪兄](https://space.bilibili.com/1900783/#/)
\n", "\n", "知乎专栏：[人工智能小技巧](https://zhuanlan.zhihu.com/c_1032626015746502656)
\n", "\n", "简书专栏：[人工智能小技巧](https://www.jianshu.com/u/38cccf09b515)
\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 粉丝答疑交流QQ群：953712961\n", "\n", "\n", "# 微信赞赏码\n", "\n", "![子豪兄的赞赏码](https://upload-images.jianshu.io/upload_images/13714448-bec288cb077c7f08.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 0.安装pytorch" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 0.1下载pytorch\n", "打开pytorch[官网](https://pytorch.org/get-started/locally/)，选择对应操作系统、python版本、python包管理器和CUDA版本的pytorch，界面上会自动生成你需要运行的命令行。\n", "例如，我是Windows64位、python3.7，使用pip作为python包管理器，不需要CUDA，我的界面长这个样子\n", "![pytorch下载界面](https://upload-images.jianshu.io/upload_images/13714448-579e4ad805844c95.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "按界面提示，我需要在操作系统命令行中运行以下两条命令：
\n", "\n", "`pip3 install https://download.pytorch.org/whl/cpu/torch-1.1.0-cp37-cp37m-win_amd64.whl`\n", "\n", "`pip3 install https://download.pytorch.org/whl/cpu/torchvision-0.3.0-cp37-cp37m-win_amd64.whl`" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 0.2验证安装成功\n", "\n", "运行下面的代码单元，如果正常就说明pytorch安装成功啦！" ] }, { "cell_type": "code", "execution_count": 35, "metadata": {}, "outputs": [], "source": [ "import torch" ] }, { "cell_type": "code", "execution_count": 36, "metadata": { "scrolled": true }, "outputs": [ { "data": { "text/plain": [ "'1.1.0'" ] }, "execution_count": 36, "metadata": {}, "output_type": "execute_result" } ], "source": [ "torch.__version__" ] }, { "cell_type": "code", "execution_count": 37, "metadata": {}, "outputs": [], "source": [ "# 上面单元格的运行结果：pytorch的版本号" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 1 下载Fasion-MNIST数据集" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "`Fashion-MNIST`是一个替代原始的[MNIST手写数字数据集](http://yann.lecun.com/exdb/mnist/)的`另一个`图像数据集。 它是由Zalando（一家德国的时尚科技公司）旗下的[研究部门](https://research.zalando.com/)提供。其涵盖了来自10种类别的共7万个不同商品的正面图片。Fashion-MNIST的大小、格式和训练集/测试集划分与原始的MNIST完全一致。60000/10000的训练测试数据划分，28x28的灰度图片。你可以直接用它来测试你的机器学习和深度学习算法性能，且**不需要**改动任何的代码。\n", "\n", "Fashion-MNIST 数据集的[中文文档说明](https://github.com/zalandoresearch/fashion-mnist/blob/master/README.zh-CN.md)\n" ] }, { "cell_type": "code", "execution_count": 38, "metadata": {}, "outputs": [], "source": [ "import torch # 导入pytorch\n", "from torch import nn, optim # 导入神经网络与优化器对应的类\n", "import torch.nn.functional as F \n", "from torchvision import datasets, transforms ## 导入数据集与数据预处理的方法" ] }, { "cell_type": "code", "execution_count": 39, "metadata": {}, "outputs": [], "source": [ "# 数据预处理：标准化图像数据，使得灰度数据在-1到+1之间\n", "transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.5,), (0.5,))])\n", "\n", "# 下载Fashion-MNIST训练集数据，并构建训练集数据载入器trainloader,每次从训练集中载入64张图片，每次载入都打乱顺序\n", "trainset = datasets.FashionMNIST('dataset/', download=True, train=True, transform=transform)\n", "trainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)\n", "\n", "# 下载Fashion-MNIST测试集数据，并构建测试集数据载入器trainloader,每次从测试集中载入64张图片，每次载入都打乱顺序\n", "testset = datasets.FashionMNIST('dataset/', download=True, train=False, transform=transform)\n", "testloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=True)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 显示图片" ] }, { "cell_type": "code", "execution_count": 40, "metadata": { "scrolled": false }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "这张图片对应的标签是 外套\n" ] }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "image, label = next(iter(trainloader))\n", "\n", "# image图片中有64张图片，我们查看索引为2的图片\n", "imagedemo = image[3]\n", "imagedemolabel = label[3]\n", "\n", "imagedemo = imagedemo.reshape((28,28))\n", "\n", "import matplotlib.pyplot as plt\n", "%matplotlib inline\n", "plt.imshow(imagedemo)\n", "\n", "labellist = ['T恤','裤子','套衫','裙子','外套','凉鞋','汗衫','运动鞋','包包','靴子']\n", "print(f'这张图片对应的标签是 {labellist[imagedemolabel]}')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 2 打开数据集中的图片" ] }, { "cell_type": "code", "execution_count": 41, "metadata": {}, "outputs": [], "source": [ "image, label = next(iter(trainloader))" ] }, { "cell_type": "code", "execution_count": 42, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "torch.Size([64, 1, 28, 28])" ] }, "execution_count": 42, "metadata": {}, "output_type": "execute_result" } ], "source": [ "image.shape" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "image包含了64张28 * 28的灰度图片，1代表单通道，也就是灰度" ] }, { "cell_type": "code", "execution_count": 43, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "tensor([7, 0, 2, 6, 4, 5, 4, 2, 5, 1, 2, 2, 9, 8, 3, 2, 4, 3, 3, 7, 0, 7, 3, 7,\n", " 9, 9, 3, 0, 5, 9, 2, 6, 0, 0, 2, 7, 2, 4, 5, 2, 0, 2, 6, 2, 0, 6, 0, 9,\n", " 3, 5, 2, 2, 5, 5, 8, 2, 4, 0, 1, 0, 8, 4, 7, 8])" ] }, "execution_count": 43, "metadata": {}, "output_type": "execute_result" } ], "source": [ "label" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "label包含了image里面64张图片对应的标签\n", "\n", "| 标注编号 | 描述 |\n", "| --- | --- |\n", "| 0 | T-shirt/top（T恤）|\n", "| 1 | Trouser（裤子）|\n", "| 2 | Pullover（套衫）|\n", "| 3 | Dress（裙子）|\n", "| 4 | Coat（外套）|\n", "| 5 | Sandal（凉鞋）|\n", "| 6 | Shirt（汗衫）|\n", "| 7 | Sneaker（运动鞋）|\n", "| 8 | Bag（包）|\n", "| 9 | Ankle boot（踝靴）|" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 打开训练集中的一张图片看看" ] }, { "cell_type": "code", "execution_count": 44, "metadata": {}, "outputs": [], "source": [ "imagedemo = image[2]\n", "imagedemolabel = label[2]" ] }, { "cell_type": "code", "execution_count": 45, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "torch.Size([1, 28, 28])" ] }, "execution_count": 45, "metadata": {}, "output_type": "execute_result" } ], "source": [ "imagedemo.shape" ] }, { "cell_type": "code", "execution_count": 46, "metadata": {}, "outputs": [], "source": [ "imagedemo = imagedemo.reshape((28,28))" ] }, { "cell_type": "code", "execution_count": 47, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "torch.Size([28, 28])" ] }, "execution_count": 47, "metadata": {}, "output_type": "execute_result" } ], "source": [ "imagedemo.shape" ] }, { "cell_type": "code", "execution_count": 48, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "tensor(2)" ] }, "execution_count": 48, "metadata": {}, "output_type": "execute_result" } ], "source": [ "imagedemolabel" ] }, { "cell_type": "code", "execution_count": 49, "metadata": {}, "outputs": [], "source": [ "import matplotlib.pyplot as plt\n", "%matplotlib inline" ] }, { "cell_type": "code", "execution_count": 50, "metadata": { "scrolled": true }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "这张图片对应的标签是 套衫\n" ] }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "plt.imshow(imagedemo)\n", "\n", "labellist = ['T恤','裤子','套衫','裙子','外套','凉鞋','汗衫','运动鞋','包包','靴子']\n", "print(f'这张图片对应的标签是 {labellist[imagedemolabel]}')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 查看一张图片和标签完整的代码" ] }, { "cell_type": "code", "execution_count": 51, "metadata": { "scrolled": false }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "这张图片对应的标签是 外套\n" ] }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "image, label = next(iter(trainloader))\n", "\n", "# image图片中有64张图片，我们查看索引为2的图片\n", "imagedemo = image[3]\n", "imagedemolabel = label[3]\n", "\n", "imagedemo = imagedemo.reshape((28,28))\n", "\n", "import matplotlib.pyplot as plt\n", "%matplotlib inline\n", "plt.imshow(imagedemo)\n", "\n", "labellist = ['T恤','裤子','套衫','裙子','外套','凉鞋','汗衫','运动鞋','包包','靴子']\n", "print(f'这张图片对应的标签是 {labellist[imagedemolabel]}')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 2 搭建并训练四层全连接神经网络\n", "\n", "神经网络的输入为28 * 28 = 784 个像素
\n", "第一个隐含层包含256个神经元
\n", "第二个隐含层包含128个神经元
\n", "第三个隐含层包含64个神经元
\n", "输出层输出10个结果，对应图片的10种分类
" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "![全连接神经网络](https://upload-images.jianshu.io/upload_images/13714448-c8b5fa11504798bf.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)" ] }, { "cell_type": "code", "execution_count": 52, "metadata": {}, "outputs": [], "source": [ "from torch import nn, optim\n", "import torch.nn.functional as F\n", "\n", "class Classifier(nn.Module):\n", " def __init__(self):\n", " super().__init__()\n", " self.fc1 = nn.Linear(784, 256)\n", " self.fc2 = nn.Linear(256, 128)\n", " self.fc3 = nn.Linear(128, 64)\n", " self.fc4 = nn.Linear(64, 10)\n", " \n", " def forward(self, x):\n", " # make sure input tensor is flattened\n", " x = x.view(x.shape[0], -1)\n", " \n", " x = F.relu(self.fc1(x))\n", " x = F.relu(self.fc2(x))\n", " x = F.relu(self.fc3(x))\n", " x = F.log_softmax(self.fc4(x), dim=1)\n", " \n", " return x" ] }, { "cell_type": "code", "execution_count": 53, "metadata": { "scrolled": false }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "开始训练\n", "训练集学习次数: 1/15.. 训练误差: 0.515.. 测试误差: 0.436.. 模型分类准确率: 0.844\n", "训练集学习次数: 2/15.. 训练误差: 0.389.. 测试误差: 0.395.. 模型分类准确率: 0.856\n", "训练集学习次数: 3/15.. 训练误差: 0.353.. 测试误差: 0.395.. 模型分类准确率: 0.858\n", "训练集学习次数: 4/15.. 训练误差: 0.331.. 测试误差: 0.395.. 模型分类准确率: 0.858\n", "训练集学习次数: 5/15.. 训练误差: 0.316.. 测试误差: 0.368.. 模型分类准确率: 0.874\n", "训练集学习次数: 6/15.. 训练误差: 0.301.. 测试误差: 0.364.. 模型分类准确率: 0.870\n", "训练集学习次数: 7/15.. 训练误差: 0.288.. 测试误差: 0.375.. 模型分类准确率: 0.867\n", "训练集学习次数: 8/15.. 训练误差: 0.282.. 测试误差: 0.380.. 模型分类准确率: 0.872\n", "训练集学习次数: 9/15.. 训练误差: 0.270.. 测试误差: 0.373.. 模型分类准确率: 0.876\n", "训练集学习次数: 10/15.. 训练误差: 0.264.. 测试误差: 0.380.. 模型分类准确率: 0.869\n", "训练集学习次数: 11/15.. 训练误差: 0.257.. 测试误差: 0.361.. 模型分类准确率: 0.884\n", "训练集学习次数: 12/15.. 训练误差: 0.250.. 测试误差: 0.379.. 模型分类准确率: 0.868\n", "训练集学习次数: 13/15.. 训练误差: 0.249.. 测试误差: 0.372.. 模型分类准确率: 0.873\n", "训练集学习次数: 14/15.. 训练误差: 0.237.. 测试误差: 0.347.. 模型分类准确率: 0.885\n", "训练集学习次数: 15/15.. 训练误差: 0.232.. 测试误差: 0.377.. 模型分类准确率: 0.884\n" ] } ], "source": [ "# 对上面定义的Classifier类进行实例化\n", "model = Classifier()\n", "\n", "# 定义损失函数为负对数损失函数\n", "criterion = nn.NLLLoss()\n", "\n", "# 优化方法为Adam梯度下降方法，学习率为0.003\n", "optimizer = optim.Adam(model.parameters(), lr=0.003)\n", "\n", "# 对训练集的全部数据学习15遍，这个数字越大，训练时间越长\n", "epochs = 15\n", "\n", "# 将每次训练的训练误差和测试误差存储在这两个列表里，后面绘制误差变化折线图用\n", "train_losses, test_losses = [], []\n", "\n", "print('开始训练')\n", "for e in range(epochs):\n", " running_loss = 0\n", " \n", " # 对训练集中的所有图片都过一遍\n", " for images, labels in trainloader:\n", " # 将优化器中的求导结果都设为0，否则会在每次反向传播之后叠加之前的\n", " optimizer.zero_grad()\n", " \n", " # 对64张图片进行推断，计算损失函数，反向传播优化权重，将损失求和\n", " log_ps = model(images)\n", " loss = criterion(log_ps, labels)\n", " loss.backward()\n", " optimizer.step()\n", " running_loss += loss.item()\n", " \n", " # 每次学完一遍数据集，都进行以下测试操作\n", " else:\n", " test_loss = 0\n", " accuracy = 0\n", " # 测试的时候不需要开自动求导和反向传播\n", " with torch.no_grad():\n", " # 关闭Dropout\n", " model.eval()\n", " \n", " # 对测试集中的所有图片都过一遍\n", " for images, labels in testloader:\n", " # 对传入的测试集图片进行正向推断、计算损失，accuracy为测试集一万张图片中模型预测正确率\n", " log_ps = model(images)\n", " test_loss += criterion(log_ps, labels)\n", " ps = torch.exp(log_ps)\n", " top_p, top_class = ps.topk(1, dim=1)\n", " equals = top_class == labels.view(*top_class.shape)\n", " \n", " # 等号右边为每一批64张测试图片中预测正确的占比\n", " accuracy += torch.mean(equals.type(torch.FloatTensor))\n", " # 恢复Dropout\n", " model.train()\n", " # 将训练误差和测试误差存在两个列表里，后面绘制误差变化折线图用\n", " train_losses.append(running_loss/len(trainloader))\n", " test_losses.append(test_loss/len(testloader))\n", "\n", " print(\"训练集学习次数: {}/{}.. \".format(e+1, epochs),\n", " \"训练误差: {:.3f}.. \".format(running_loss/len(trainloader)),\n", " \"测试误差: {:.3f}.. \".format(test_loss/len(testloader)),\n", " \"模型分类准确率: {:.3f}\".format(accuracy/len(testloader)))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 3 验证模型效果" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 绘制训练误差和测试误差随学习次数增加的变化" ] }, { "cell_type": "code", "execution_count": 54, "metadata": {}, "outputs": [], "source": [ "import matplotlib.pyplot as plt\n", "%matplotlib inline" ] }, { "cell_type": "code", "execution_count": 55, "metadata": { "scrolled": false }, "outputs": [ { "data": { "text/plain": [ "" ] }, "execution_count": 55, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "plt.plot(train_losses, label='Training loss')\n", "plt.plot(test_losses, label='Validation loss')\n", "plt.legend()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "可以看到，虽然训练误差一直在下降，但测试误差居高不下，我们的神经网络仿佛一个高分低能的同学，平时把所有课后题答案都死记硬背下来，一到考试见到新题的时候就不会做了。\n", "\n", "虽然高分低能，但大部分时候依旧能做出正确判断，但是有时候预测概率只有百分之三四十的把握，不能做到十有八九的确定。" ] }, { "cell_type": "code", "execution_count": 95, "metadata": { "scrolled": true }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "神经网络猜测图片里是 套衫，概率为49.677035212516785%\n" ] }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "dataiter = iter(testloader)\n", "images, labels = dataiter.next()\n", "img = images[0]\n", "img = img.reshape((28,28)).numpy()\n", "plt.imshow(img)\n", "\n", "# 将测试图片转为一维的列向量\n", "img = torch.from_numpy(img)\n", "img = img.view(1, 784)\n", "\n", "# 进行正向推断，预测图片所在的类别\n", "with torch.no_grad():\n", " output = model.forward(img)\n", "ps = torch.exp(output)\n", "\n", "top_p, top_class = ps.topk(1, dim=1)\n", "labellist = ['T恤','裤子','套衫','裙子','外套','凉鞋','汗衫','运动鞋','包包','靴子']\n", "prediction = labellist[top_class]\n", "probability = float(top_p)\n", "print(f'神经网络猜测图片里是 {prediction}，概率为{probability*100}%')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 4 采用Dropout方法防止过拟合" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "我们可以采用Dropout的方法，也就是在每次正向推断训练神经元的时候随机“掐死”一部分神经元，阻断其输入输出，这样可以起到正则化的作用。\n", "\n", "可以理解为，皇上雨露均沾，今天受宠，明天可能就被打入冷宫，这样就防止了杨贵妃那样的“三千宠爱在一身”，从而防止了某些神经元一家独大，成为话题领袖，只手遮天。\n", "\n", "所有神经元处于平等地位，防止过拟合。" ] }, { "cell_type": "code", "execution_count": 57, "metadata": {}, "outputs": [], "source": [ "class Classifier(nn.Module):\n", " def __init__(self):\n", " super().__init__()\n", " self.fc1 = nn.Linear(784, 256)\n", " self.fc2 = nn.Linear(256, 128)\n", " self.fc3 = nn.Linear(128, 64)\n", " self.fc4 = nn.Linear(64, 10)\n", " \n", " # 构造Dropout方法，在每次训练过程中都随机“掐死”百分之二十的神经元，防止过拟合。\n", " self.dropout = nn.Dropout(p=0.2)\n", " \n", " def forward(self, x):\n", " # 确保输入的tensor是展开的单列数据，把每张图片的通道、长度、宽度三个维度都压缩为一列\n", " x = x.view(x.shape[0], -1)\n", " \n", " # 在训练过程中对隐含层神经元的正向推断使用Dropout方法\n", " x = self.dropout(F.relu(self.fc1(x)))\n", " x = self.dropout(F.relu(self.fc2(x)))\n", " x = self.dropout(F.relu(self.fc3(x)))\n", " \n", " # 在输出单元不需要使用Dropout方法\n", " x = F.log_softmax(self.fc4(x), dim=1)\n", " \n", " return x" ] }, { "cell_type": "code", "execution_count": 58, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "开始训练\n", "训练集学习次数: 1/15.. 训练误差: 0.604.. 测试误差: 0.488.. 模型分类准确率: 0.821\n", "训练集学习次数: 2/15.. 训练误差: 0.480.. 测试误差: 0.439.. 模型分类准确率: 0.844\n", "训练集学习次数: 3/15.. 训练误差: 0.449.. 测试误差: 0.422.. 模型分类准确率: 0.848\n", "训练集学习次数: 4/15.. 训练误差: 0.428.. 测试误差: 0.424.. 模型分类准确率: 0.852\n", "训练集学习次数: 5/15.. 训练误差: 0.425.. 测试误差: 0.406.. 模型分类准确率: 0.850\n", "训练集学习次数: 6/15.. 训练误差: 0.409.. 测试误差: 0.392.. 模型分类准确率: 0.862\n", "训练集学习次数: 7/15.. 训练误差: 0.403.. 测试误差: 0.395.. 模型分类准确率: 0.857\n", "训练集学习次数: 8/15.. 训练误差: 0.399.. 测试误差: 0.388.. 模型分类准确率: 0.863\n", "训练集学习次数: 9/15.. 训练误差: 0.394.. 测试误差: 0.381.. 模型分类准确率: 0.866\n", "训练集学习次数: 10/15.. 训练误差: 0.386.. 测试误差: 0.387.. 模型分类准确率: 0.865\n", "训练集学习次数: 11/15.. 训练误差: 0.384.. 测试误差: 0.407.. 模型分类准确率: 0.857\n", "训练集学习次数: 12/15.. 训练误差: 0.377.. 测试误差: 0.392.. 模型分类准确率: 0.862\n", "训练集学习次数: 13/15.. 训练误差: 0.368.. 测试误差: 0.386.. 模型分类准确率: 0.871\n", "训练集学习次数: 14/15.. 训练误差: 0.375.. 测试误差: 0.391.. 模型分类准确率: 0.869\n", "训练集学习次数: 15/15.. 训练误差: 0.373.. 测试误差: 0.390.. 模型分类准确率: 0.867\n" ] } ], "source": [ "# 对上面定义的Classifier类进行实例化\n", "model = Classifier()\n", "\n", "# 定义损失函数为负对数损失函数\n", "criterion = nn.NLLLoss()\n", "\n", "# 优化方法为Adam梯度下降方法，学习率为0.003\n", "optimizer = optim.Adam(model.parameters(), lr=0.003)\n", "\n", "# 对训练集的全部数据学习15遍，这个数字越大，训练时间越长\n", "epochs = 15\n", "\n", "# 将每次训练的训练误差和测试误差存储在这两个列表里，后面绘制误差变化折线图用\n", "train_losses, test_losses = [], []\n", "\n", "print('开始训练')\n", "for e in range(epochs):\n", " running_loss = 0\n", " \n", " # 对训练集中的所有图片都过一遍\n", " for images, labels in trainloader:\n", " # 将优化器中的求导结果都设为0，否则会在每次反向传播之后叠加之前的\n", " optimizer.zero_grad()\n", " \n", " # 对64张图片进行推断，计算损失函数，反向传播优化权重，将损失求和\n", " log_ps = model(images)\n", " loss = criterion(log_ps, labels)\n", " loss.backward()\n", " optimizer.step()\n", " running_loss += loss.item()\n", " \n", " # 每次学完一遍数据集，都进行以下测试操作\n", " else:\n", " test_loss = 0\n", " accuracy = 0\n", " # 测试的时候不需要开自动求导和反向传播\n", " with torch.no_grad():\n", " # 关闭Dropout\n", " model.eval()\n", " \n", " # 对测试集中的所有图片都过一遍\n", " for images, labels in testloader:\n", " # 对传入的测试集图片进行正向推断、计算损失，accuracy为测试集一万张图片中模型预测正确率\n", " log_ps = model(images)\n", " test_loss += criterion(log_ps, labels)\n", " ps = torch.exp(log_ps)\n", " top_p, top_class = ps.topk(1, dim=1)\n", " equals = top_class == labels.view(*top_class.shape)\n", " \n", " # 等号右边为每一批64张测试图片中预测正确的占比\n", " accuracy += torch.mean(equals.type(torch.FloatTensor))\n", " # 恢复Dropout\n", " model.train()\n", " # 将训练误差和测试误差存在两个列表里，后面绘制误差变化折线图用\n", " train_losses.append(running_loss/len(trainloader))\n", " test_losses.append(test_loss/len(testloader))\n", "\n", " print(\"训练集学习次数: {}/{}.. \".format(e+1, epochs),\n", " \"训练误差: {:.3f}.. \".format(running_loss/len(trainloader)),\n", " \"测试误差: {:.3f}.. \".format(test_loss/len(testloader)),\n", " \"模型分类准确率: {:.3f}\".format(accuracy/len(testloader)))" ] }, { "cell_type": "code", "execution_count": 59, "metadata": {}, "outputs": [], "source": [ "import matplotlib.pyplot as plt\n", "%matplotlib inline" ] }, { "cell_type": "code", "execution_count": 60, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "" ] }, "execution_count": 60, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "plt.plot(train_losses, label='Training loss')\n", "plt.plot(test_losses, label='Validation loss')\n", "plt.legend()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "可以看到，训练误差和测试误差都随学习次数增加逐渐降低，没有出现“高分低能”和“死记硬背”的过拟合现象，这其实是Dropout正则化的功劳。" ] }, { "cell_type": "code", "execution_count": 90, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "神经网络猜测图片里是 套衫，概率为99.89591240882874%\n" ] }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "model.eval()\n", "dataiter = iter(testloader)\n", "images, labels = dataiter.next()\n", "img = images[0]\n", "img = img.reshape((28,28)).numpy()\n", "plt.imshow(img)\n", "\n", "# 将测试图片转为一维的列向量\n", "img = torch.from_numpy(img)\n", "img = img.view(1, 784)\n", "\n", "# 进行正向推断，预测图片所在的类别\n", "with torch.no_grad():\n", " output = model.forward(img)\n", "ps = torch.exp(output)\n", "\n", "top_p, top_class = ps.topk(1, dim=1)\n", "labellist = ['T恤','裤子','套衫','裙子','外套','凉鞋','汗衫','运动鞋','包包','靴子']\n", "prediction = labellist[top_class]\n", "probability = float(top_p)\n", "print(f'神经网络猜测图片里是 {prediction}，概率为{probability*100}%')" ] }, { "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": "3.7.2" } }, "nbformat": 4, "nbformat_minor": 2 }