{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# 3.2 线性回归的从零开始实现" ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "1.2.0\n" ] } ], "source": [ "%matplotlib inline\n", "import torch\n", "from IPython import display\n", "from matplotlib import pyplot as plt\n", "import numpy as np\n", "import random\n", "\n", "print(torch.__version__)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.1 生成数据集" ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "collapsed": true }, "outputs": [], "source": [ "num_inputs = 2\n", "num_examples = 1000\n", "true_w = [2, -3.4]\n", "true_b = 4.2\n", "features = torch.randn(num_examples, num_inputs,\n", " dtype=torch.float32)\n", "labels = true_w[0] * features[:, 0] + true_w[1] * features[:, 1] + true_b\n", "labels += torch.tensor(np.random.normal(0, 0.01, size=labels.size()),\n", " dtype=torch.float32)" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "tensor([0.1706, 1.0724]) tensor(0.8927)\n" ] } ], "source": [ "print(features[0], labels[0])" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [ { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "def use_svg_display():\n", " # 用矢量图显示\n", " display.set_matplotlib_formats('svg')\n", "\n", "def set_figsize(figsize=(3.5, 2.5)):\n", " use_svg_display()\n", " # 设置图的尺寸\n", " plt.rcParams['figure.figsize'] = figsize\n", "\n", "# # 在../d2lzh_pytorch里面添加上面两个函数后就可以这样导入\n", "# import sys\n", "# sys.path.append(\"..\")\n", "# from d2lzh_pytorch import * \n", "\n", "set_figsize()\n", "plt.scatter(features[:, 1].numpy(), labels.numpy(), 1);" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.2 读取数据" ] }, { "cell_type": "code", "execution_count": 5, "metadata": { "collapsed": true }, "outputs": [], "source": [ "# 本函数已保存在d2lzh包中方便以后使用\n", "def data_iter(batch_size, features, labels):\n", " num_examples = len(features)\n", " indices = list(range(num_examples))\n", " random.shuffle(indices) # 样本的读取顺序是随机的\n", " for i in range(0, num_examples, batch_size):\n", " j = torch.LongTensor(indices[i: min(i + batch_size, num_examples)]) # 最后一次可能不足一个batch\n", " yield features.index_select(0, j), labels.index_select(0, j)" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "tensor([[-1.5561, 1.5366],\n", " [ 0.7773, -0.4141],\n", " [-0.7296, 0.2837],\n", " [-0.1572, -0.1902],\n", " [-1.1338, 0.0436],\n", " [ 1.5135, -2.5492],\n", " [ 0.5583, 0.2310],\n", " [-1.3505, -1.5909],\n", " [-0.2826, 1.3457],\n", " [ 0.2002, -2.1443]]) \n", " tensor([-4.1393, 7.1701, 1.7999, 4.5387, 1.7729, 15.8816, 4.5391, 6.9128,\n", " -0.9564, 11.9046])\n" ] } ], "source": [ "batch_size = 10\n", "\n", "for X, y in data_iter(batch_size, features, labels):\n", " print(X, '\\n', y)\n", " break" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.3 初始化模型参数" ] }, { "cell_type": "code", "execution_count": 7, "metadata": { "collapsed": true }, "outputs": [], "source": [ "w = torch.tensor(np.random.normal(0, 0.01, (num_inputs, 1)), dtype=torch.float32)\n", "b = torch.zeros(1, dtype=torch.float32)" ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "tensor([0.], requires_grad=True)" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ "w.requires_grad_(requires_grad=True)\n", "b.requires_grad_(requires_grad=True) " ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.4 定义模型" ] }, { "cell_type": "code", "execution_count": 9, "metadata": { "collapsed": true }, "outputs": [], "source": [ "def linreg(X, w, b): # 本函数已保存在d2lzh包中方便以后使用\n", " return torch.mm(X, w) + b" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.5 定义损失函数" ] }, { "cell_type": "code", "execution_count": 10, "metadata": { "collapsed": true }, "outputs": [], "source": [ "def squared_loss(y_hat, y): # 本函数已保存在pytorch_d2lzh包中方便以后使用\n", " return (y_hat - y.view(y_hat.size())) ** 2 / 2" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.6 定义优化算法" ] }, { "cell_type": "code", "execution_count": 11, "metadata": { "collapsed": true }, "outputs": [], "source": [ "def sgd(params, lr, batch_size): # 本函数已保存在d2lzh_pytorch包中方便以后使用\n", " for param in params:\n", " param.data -= lr * param.grad / batch_size # 注意这里更改param时用的param.data" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 3.2.7 训练模型" ] }, { "cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "epoch 1, loss 0.031926\n", "epoch 2, loss 0.000115\n", "epoch 3, loss 0.000054\n" ] } ], "source": [ "lr = 0.03\n", "num_epochs = 3\n", "net = linreg\n", "loss = squared_loss\n", "\n", "for epoch in range(num_epochs): # 训练模型一共需要num_epochs个迭代周期\n", " # 在每一个迭代周期中，会使用训练数据集中所有样本一次（假设样本数能够被批量大小整除）。X\n", " # 和y分别是小批量样本的特征和标签\n", " for X, y in data_iter(batch_size, features, labels):\n", " l = loss(net(X, w, b), y).sum() # l是有关小批量X和y的损失\n", " l.backward() # 小批量的损失对模型参数求梯度\n", " sgd([w, b], lr, batch_size) # 使用小批量随机梯度下降迭代模型参数\n", " \n", " # 不要忘了梯度清零\n", " w.grad.data.zero_()\n", " b.grad.data.zero_()\n", " train_l = loss(net(features, w, b), labels)\n", " print('epoch %d, loss %f' % (epoch + 1, train_l.mean().item()))" ] }, { "cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "[2, -3.4] \n", " tensor([[ 2.0000],\n", " [-3.3996]], requires_grad=True)\n", "4.2 \n", " tensor([4.1990], requires_grad=True)\n" ] } ], "source": [ "print(true_w, '\\n', w)\n", "print(true_b, '\\n', b)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "collapsed": true }, "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": "3.6.2" } }, "nbformat": 4, "nbformat_minor": 2 }