{ "cells": [ { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "'1.0.0'" ] }, "execution_count": 1, "metadata": {}, "output_type": "execute_result" } ], "source": [ "import torch\n", "import torch.nn as nn\n", "import torch.nn.functional as F\n", "import torch.optim as optim\n", "from torchvision import datasets, transforms\n", "torch.__version__" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# 3.2 MNIST数据集手写数字识别\n", "\n", "## 3.2.1 数据集介绍\n", "MNIST 包括6万张28x28的训练样本，1万张测试样本，很多教程都会对它”下手”几乎成为一个 “典范”，可以说它就是计算机视觉里面的Hello World。所以我们这里也会使用MNIST来进行实战。\n", "\n", "前面在介绍卷积神经网络的时候说到过LeNet-5，LeNet-5之所以强大就是因为在当时的环境下将MNIST数据的识别率提高到了99%，这里我们也自己从头搭建一个卷积神经网络，也达到99%的准确率" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.2 手写数字识别\n", "首先，我们定义一些超参数" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [], "source": [ "BATCH_SIZE=512 #大概需要2G的显存\n", "EPOCHS=20 # 总共训练批次\n", "DEVICE = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\") # 让torch判断是否使用GPU，建议使用GPU环境，因为会快很多" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "因为Pytorch里面包含了MNIST的数据集，所以我们这里直接使用即可。\n", "如果第一次执行会生成data文件夹，并且需要一些时间下载，如果以前下载过就不会再次下载了\n", "\n", "由于官方已经实现了dataset，所以这里可以直接使用DataLoader来对数据进行读取" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz\n", "Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz\n", "Downloading http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz\n", "Downloading http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz\n", "Processing...\n", "Done!\n" ] } ], "source": [ "train_loader = torch.utils.data.DataLoader(\n", " datasets.MNIST('data', train=True, download=True, \n", " transform=transforms.Compose([\n", " transforms.ToTensor(),\n", " transforms.Normalize((0.1307,), (0.3081,))\n", " ])),\n", " batch_size=BATCH_SIZE, shuffle=True)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "测试集" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "test_loader = torch.utils.data.DataLoader(\n", " datasets.MNIST('data', train=False, transform=transforms.Compose([\n", " transforms.ToTensor(),\n", " transforms.Normalize((0.1307,), (0.3081,))\n", " ])),\n", " batch_size=BATCH_SIZE, shuffle=True)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "下面我们定义一个网络，网络包含两个卷积层，conv1和conv2，然后紧接着两个线性层作为输出，最后输出10个维度，这10个维度我们作为0-9的标识来确定识别出的是那个数字\n", "\n", "在这里建议大家将每一层的输入和输出维度都作为注释标注出来，这样后面阅读代码的会方便很多" ] }, { "cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [], "source": [ "class ConvNet(nn.Module):\n", " def __init__(self):\n", " super().__init__()\n", " # batch*1*28*28（每次会送入batch个样本，输入通道数1（黑白图像），图像分辨率是28x28）\n", " # 下面的卷积层Conv2d的第一个参数指输入通道数，第二个参数指输出通道数，第三个参数指卷积核的大小\n", " self.conv1 = nn.Conv2d(1, 10, 5) # 输入通道数1，输出通道数10，核的大小5\n", " self.conv2 = nn.Conv2d(10, 20, 3) # 输入通道数10，输出通道数20，核的大小3\n", " # 下面的全连接层Linear的第一个参数指输入通道数，第二个参数指输出通道数\n", " self.fc1 = nn.Linear(20*10*10, 500) # 输入通道数是2000，输出通道数是500\n", " self.fc2 = nn.Linear(500, 10) # 输入通道数是500，输出通道数是10，即10分类\n", " def forward(self,x):\n", " in_size = x.size(0) # 在本例中in_size=512，也就是BATCH_SIZE的值。输入的x可以看成是512*1*28*28的张量。\n", " out = self.conv1(x) # batch*1*28*28 -> batch*10*24*24（28x28的图像经过一次核为5x5的卷积，输出变为24x24）\n", " out = F.relu(out) # batch*10*24*24（激活函数ReLU不改变形状））\n", " out = F.max_pool2d(out, 2, 2) # batch*10*24*24 -> batch*10*12*12（2*2的池化层会减半）\n", " out = self.conv2(out) # batch*10*12*12 -> batch*20*10*10（再卷积一次，核的大小是3）\n", " out = F.relu(out) # batch*20*10*10\n", " out = out.view(in_size, -1) # batch*20*10*10 -> batch*2000（out的第二维是-1，说明是自动推算，本例中第二维是20*10*10）\n", " out = self.fc1(out) # batch*2000 -> batch*500\n", " out = F.relu(out) # batch*500\n", " out = self.fc2(out) # batch*500 -> batch*10\n", " out = F.log_softmax(out, dim=1) # 计算log(softmax(x))\n", " return out" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "我们实例化一个网络，实例化后使用.to方法将网络移动到GPU\n", "\n", "优化器我们也直接选择简单暴力的Adam" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [], "source": [ "model = ConvNet().to(DEVICE)\n", "optimizer = optim.Adam(model.parameters())" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "下面定义一下训练的函数，我们将训练的所有操作都封装到这个函数中" ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [], "source": [ "def train(model, device, train_loader, optimizer, epoch):\n", " model.train()\n", " for batch_idx, (data, target) in enumerate(train_loader):\n", " data, target = data.to(device), target.to(device)\n", " optimizer.zero_grad()\n", " output = model(data)\n", " loss = F.nll_loss(output, target)\n", " loss.backward()\n", " optimizer.step()\n", " if(batch_idx+1)%30 == 0: \n", " print('Train Epoch: {} [{}/{} ({:.0f}%)]\\tLoss: {:.6f}'.format(\n", " epoch, batch_idx * len(data), len(train_loader.dataset),\n", " 100. * batch_idx / len(train_loader), loss.item()))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "测试的操作也一样封装成一个函数" ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [], "source": [ "def test(model, device, test_loader):\n", " model.eval()\n", " test_loss = 0\n", " correct = 0\n", " with torch.no_grad():\n", " for data, target in test_loader:\n", " data, target = data.to(device), target.to(device)\n", " output = model(data)\n", " test_loss += F.nll_loss(output, target, reduction='sum').item() # 将一批的损失相加\n", " pred = output.max(1, keepdim=True)[1] # 找到概率最大的下标\n", " correct += pred.eq(target.view_as(pred)).sum().item()\n", "\n", " test_loss /= len(test_loader.dataset)\n", " print('\\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\\n'.format(\n", " test_loss, correct, len(test_loader.dataset),\n", " 100. * correct / len(test_loader.dataset)))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "下面开始训练，这里就体现出封装起来的好处了，只要写两行就可以了" ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Train Epoch: 1 [14848/60000 (25%)]\tLoss: 0.272529\n", "Train Epoch: 1 [30208/60000 (50%)]\tLoss: 0.235455\n", "Train Epoch: 1 [45568/60000 (75%)]\tLoss: 0.101858\n", "\n", "Test set: Average loss: 0.1018, Accuracy: 9695/10000 (97%)\n", "\n", "Train Epoch: 2 [14848/60000 (25%)]\tLoss: 0.057989\n", "Train Epoch: 2 [30208/60000 (50%)]\tLoss: 0.083935\n", "Train Epoch: 2 [45568/60000 (75%)]\tLoss: 0.051921\n", "\n", "Test set: Average loss: 0.0523, Accuracy: 9825/10000 (98%)\n", "\n", "Train Epoch: 3 [14848/60000 (25%)]\tLoss: 0.045383\n", "Train Epoch: 3 [30208/60000 (50%)]\tLoss: 0.049402\n", "Train Epoch: 3 [45568/60000 (75%)]\tLoss: 0.061366\n", "\n", "Test set: Average loss: 0.0408, Accuracy: 9866/10000 (99%)\n", "\n", "Train Epoch: 4 [14848/60000 (25%)]\tLoss: 0.035253\n", "Train Epoch: 4 [30208/60000 (50%)]\tLoss: 0.038444\n", "Train Epoch: 4 [45568/60000 (75%)]\tLoss: 0.036877\n", "\n", "Test set: Average loss: 0.0433, Accuracy: 9859/10000 (99%)\n", "\n", "Train Epoch: 5 [14848/60000 (25%)]\tLoss: 0.038996\n", "Train Epoch: 5 [30208/60000 (50%)]\tLoss: 0.020670\n", "Train Epoch: 5 [45568/60000 (75%)]\tLoss: 0.034658\n", "\n", "Test set: Average loss: 0.0339, Accuracy: 9885/10000 (99%)\n", "\n", "Train Epoch: 6 [14848/60000 (25%)]\tLoss: 0.067320\n", "Train Epoch: 6 [30208/60000 (50%)]\tLoss: 0.016328\n", "Train Epoch: 6 [45568/60000 (75%)]\tLoss: 0.017037\n", "\n", "Test set: Average loss: 0.0348, Accuracy: 9881/10000 (99%)\n", "\n", "Train Epoch: 7 [14848/60000 (25%)]\tLoss: 0.022150\n", "Train Epoch: 7 [30208/60000 (50%)]\tLoss: 0.009608\n", "Train Epoch: 7 [45568/60000 (75%)]\tLoss: 0.012742\n", "\n", "Test set: Average loss: 0.0346, Accuracy: 9895/10000 (99%)\n", "\n", "Train Epoch: 8 [14848/60000 (25%)]\tLoss: 0.010173\n", "Train Epoch: 8 [30208/60000 (50%)]\tLoss: 0.019482\n", "Train Epoch: 8 [45568/60000 (75%)]\tLoss: 0.012159\n", "\n", "Test set: Average loss: 0.0323, Accuracy: 9886/10000 (99%)\n", "\n", "Train Epoch: 9 [14848/60000 (25%)]\tLoss: 0.007792\n", "Train Epoch: 9 [30208/60000 (50%)]\tLoss: 0.006970\n", "Train Epoch: 9 [45568/60000 (75%)]\tLoss: 0.004989\n", "\n", "Test set: Average loss: 0.0294, Accuracy: 9909/10000 (99%)\n", "\n", "Train Epoch: 10 [14848/60000 (25%)]\tLoss: 0.003764\n", "Train Epoch: 10 [30208/60000 (50%)]\tLoss: 0.005944\n", "Train Epoch: 10 [45568/60000 (75%)]\tLoss: 0.001866\n", "\n", "Test set: Average loss: 0.0361, Accuracy: 9902/10000 (99%)\n", "\n", "Train Epoch: 11 [14848/60000 (25%)]\tLoss: 0.002737\n", "Train Epoch: 11 [30208/60000 (50%)]\tLoss: 0.014134\n", "Train Epoch: 11 [45568/60000 (75%)]\tLoss: 0.001365\n", "\n", "Test set: Average loss: 0.0309, Accuracy: 9905/10000 (99%)\n", "\n", "Train Epoch: 12 [14848/60000 (25%)]\tLoss: 0.003344\n", "Train Epoch: 12 [30208/60000 (50%)]\tLoss: 0.003090\n", "Train Epoch: 12 [45568/60000 (75%)]\tLoss: 0.004847\n", "\n", "Test set: Average loss: 0.0318, Accuracy: 9902/10000 (99%)\n", "\n", "Train Epoch: 13 [14848/60000 (25%)]\tLoss: 0.001278\n", "Train Epoch: 13 [30208/60000 (50%)]\tLoss: 0.003016\n", "Train Epoch: 13 [45568/60000 (75%)]\tLoss: 0.001328\n", "\n", "Test set: Average loss: 0.0358, Accuracy: 9906/10000 (99%)\n", "\n", "Train Epoch: 14 [14848/60000 (25%)]\tLoss: 0.002219\n", "Train Epoch: 14 [30208/60000 (50%)]\tLoss: 0.003487\n", "Train Epoch: 14 [45568/60000 (75%)]\tLoss: 0.014429\n", "\n", "Test set: Average loss: 0.0376, Accuracy: 9896/10000 (99%)\n", "\n", "Train Epoch: 15 [14848/60000 (25%)]\tLoss: 0.003042\n", "Train Epoch: 15 [30208/60000 (50%)]\tLoss: 0.002974\n", "Train Epoch: 15 [45568/60000 (75%)]\tLoss: 0.000871\n", "\n", "Test set: Average loss: 0.0346, Accuracy: 9909/10000 (99%)\n", "\n", "Train Epoch: 16 [14848/60000 (25%)]\tLoss: 0.000618\n", "Train Epoch: 16 [30208/60000 (50%)]\tLoss: 0.003164\n", "Train Epoch: 16 [45568/60000 (75%)]\tLoss: 0.007245\n", "\n", "Test set: Average loss: 0.0357, Accuracy: 9905/10000 (99%)\n", "\n", "Train Epoch: 17 [14848/60000 (25%)]\tLoss: 0.001874\n", "Train Epoch: 17 [30208/60000 (50%)]\tLoss: 0.013951\n", "Train Epoch: 17 [45568/60000 (75%)]\tLoss: 0.000729\n", "\n", "Test set: Average loss: 0.0322, Accuracy: 9922/10000 (99%)\n", "\n", "Train Epoch: 18 [14848/60000 (25%)]\tLoss: 0.002581\n", "Train Epoch: 18 [30208/60000 (50%)]\tLoss: 0.001396\n", "Train Epoch: 18 [45568/60000 (75%)]\tLoss: 0.015521\n", "\n", "Test set: Average loss: 0.0389, Accuracy: 9914/10000 (99%)\n", "\n", "Train Epoch: 19 [14848/60000 (25%)]\tLoss: 0.000283\n", "Train Epoch: 19 [30208/60000 (50%)]\tLoss: 0.001385\n", "Train Epoch: 19 [45568/60000 (75%)]\tLoss: 0.011184\n", "\n", "Test set: Average loss: 0.0383, Accuracy: 9901/10000 (99%)\n", "\n", "Train Epoch: 20 [14848/60000 (25%)]\tLoss: 0.000472\n", "Train Epoch: 20 [30208/60000 (50%)]\tLoss: 0.003306\n", "Train Epoch: 20 [45568/60000 (75%)]\tLoss: 0.018017\n", "\n", "Test set: Average loss: 0.0393, Accuracy: 9899/10000 (99%)\n", "\n" ] } ], "source": [ "for epoch in range(1, EPOCHS + 1):\n", " train(model, DEVICE, train_loader, optimizer, epoch)\n", " test(model, DEVICE, test_loader)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "我们看一下结果，准确率99%，没问题" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "如果你的模型连MNIST都搞不定，那么你的模型没有任何的价值\n", "\n", "即使你的模型搞定了MNIST，你的模型也可能没有任何的价值\n", "\n", "MNIST是一个很简单的数据集，由于它的局限性只能作为研究用途，对实际应用带来的价值非常有限。但是通过这个例子，我们可以完全了解一个实际项目的工作流程\n", "\n", "我们找到数据集，对数据做预处理，定义我们的模型，调整超参数，测试训练，再通过训练结果对超参数进行调整或者对模型进行调整。\n", "\n", "并且通过这个实战我们已经有了一个很好的模板，以后的项目都可以以这个模板为样例" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "pytorch 1.0", "language": "python", "name": "pytorch1" }, "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" } }, "nbformat": 4, "nbformat_minor": 2 }