predict-house-prices.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 通过神经网络预测房价\n",
    "在这个项目中，我们希望能够构建神经网络来预测房屋的价格"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "首先，我们导入一些必要的库"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import torch\n",
    "from torch import nn\n",
    "\n",
    "%matplotlib inline\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "读取训练集和测试集的数据"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train = pd.read_csv('./all/train.csv')\n",
    "test = pd.read_csv('./all/test.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "可以具体看看前面 5 个训练集长什么样子，可以看到，前面都是这个房屋的属性，最后是房屋的价格"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Id</th>\n",
       "      <th>MSSubClass</th>\n",
       "      <th>MSZoning</th>\n",
       "      <th>LotFrontage</th>\n",
       "      <th>LotArea</th>\n",
       "      <th>Street</th>\n",
       "      <th>Alley</th>\n",
       "      <th>LotShape</th>\n",
       "      <th>LandContour</th>\n",
       "      <th>Utilities</th>\n",
       "      <th>...</th>\n",
       "      <th>PoolArea</th>\n",
       "      <th>PoolQC</th>\n",
       "      <th>Fence</th>\n",
       "      <th>MiscFeature</th>\n",
       "      <th>MiscVal</th>\n",
       "      <th>MoSold</th>\n",
       "      <th>YrSold</th>\n",
       "      <th>SaleType</th>\n",
       "      <th>SaleCondition</th>\n",
       "      <th>SalePrice</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>60</td>\n",
       "      <td>RL</td>\n",
       "      <td>65.0</td>\n",
       "      <td>8450</td>\n",
       "      <td>Pave</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Reg</td>\n",
       "      <td>Lvl</td>\n",
       "      <td>AllPub</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>2008</td>\n",
       "      <td>WD</td>\n",
       "      <td>Normal</td>\n",
       "      <td>208500</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>20</td>\n",
       "      <td>RL</td>\n",
       "      <td>80.0</td>\n",
       "      <td>9600</td>\n",
       "      <td>Pave</td>\n",
       "      <td>NaN</td>\n",
       "      <td>Reg</td>\n",
       "      <td>Lvl</td>\n",
       "      <td>AllPub</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>2007</td>\n",
       "      <td>WD</td>\n",
       "      <td>Normal</td>\n",
       "      <td>181500</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>60</td>\n",
       "      <td>RL</td>\n",
       "      <td>68.0</td>\n",
       "      <td>11250</td>\n",
       "      <td>Pave</td>\n",
       "      <td>NaN</td>\n",
       "      <td>IR1</td>\n",
       "      <td>Lvl</td>\n",
       "      <td>AllPub</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>9</td>\n",
       "      <td>2008</td>\n",
       "      <td>WD</td>\n",
       "      <td>Normal</td>\n",
       "      <td>223500</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>70</td>\n",
       "      <td>RL</td>\n",
       "      <td>60.0</td>\n",
       "      <td>9550</td>\n",
       "      <td>Pave</td>\n",
       "      <td>NaN</td>\n",
       "      <td>IR1</td>\n",
       "      <td>Lvl</td>\n",
       "      <td>AllPub</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>2006</td>\n",
       "      <td>WD</td>\n",
       "      <td>Abnorml</td>\n",
       "      <td>140000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>60</td>\n",
       "      <td>RL</td>\n",
       "      <td>84.0</td>\n",
       "      <td>14260</td>\n",
       "      <td>Pave</td>\n",
       "      <td>NaN</td>\n",
       "      <td>IR1</td>\n",
       "      <td>Lvl</td>\n",
       "      <td>AllPub</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>2008</td>\n",
       "      <td>WD</td>\n",
       "      <td>Normal</td>\n",
       "      <td>250000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 81 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   Id  MSSubClass MSZoning  LotFrontage  LotArea Street Alley LotShape  \\\n",
       "0   1          60       RL         65.0     8450   Pave   NaN      Reg   \n",
       "1   2          20       RL         80.0     9600   Pave   NaN      Reg   \n",
       "2   3          60       RL         68.0    11250   Pave   NaN      IR1   \n",
       "3   4          70       RL         60.0     9550   Pave   NaN      IR1   \n",
       "4   5          60       RL         84.0    14260   Pave   NaN      IR1   \n",
       "\n",
       "  LandContour Utilities    ...     PoolArea PoolQC Fence MiscFeature MiscVal  \\\n",
       "0         Lvl    AllPub    ...            0    NaN   NaN         NaN       0   \n",
       "1         Lvl    AllPub    ...            0    NaN   NaN         NaN       0   \n",
       "2         Lvl    AllPub    ...            0    NaN   NaN         NaN       0   \n",
       "3         Lvl    AllPub    ...            0    NaN   NaN         NaN       0   \n",
       "4         Lvl    AllPub    ...            0    NaN   NaN         NaN       0   \n",
       "\n",
       "  MoSold YrSold  SaleType  SaleCondition  SalePrice  \n",
       "0      2   2008        WD         Normal     208500  \n",
       "1      5   2007        WD         Normal     181500  \n",
       "2      9   2008        WD         Normal     223500  \n",
       "3      2   2006        WD        Abnorml     140000  \n",
       "4     12   2008        WD         Normal     250000  \n",
       "\n",
       "[5 rows x 81 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "接着我们可以看看训练集和测试集分别有多少个样本"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "一共有 1460 个训练集样本\n",
      "一共有 1459 个测试集样本\n"
     ]
    }
   ],
   "source": [
    "print('一共有 {} 个训练集样本'.format(train.shape[0]))\n",
    "print('一共有 {} 个测试集样本'.format(test.shape[0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "接着我们开始对数据进行处理，首先我们取出**第二个特征**到**倒数第二个特征**，这些特征作为我们神经网络的输入特征"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "all_features = pd.concat((train.loc[:, 'MSSubClass':'SaleCondition'],\n",
    "                          test.loc[:, 'MSSubClass':'SaleCondition']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "接着我们需要进行数据标准化，对于所有的数值特征，我们都会减去均值，除以方差"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "numeric_feats = all_features.dtypes[all_features.dtypes != \"object\"].index # 取出所有的数值特征\n",
    "\n",
    "# 减去均值，除以方差\n",
    "all_features[numeric_feats] = all_features[numeric_feats].apply(lambda x: (x - x.mean()) \n",
    "                                                                / (x.std()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 对预测的价格取 log\n",
    "train['SalePrice'] = np.log(train['SalePrice'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "如果你仔细看看上面的特征，你会发现，除了数值特征之外，还有很多非数值特征，这些特征我们没有办法将其转换成数值表示，所以我们通过 pandas 的内置函数将其转换成种类表示\n",
    "\n",
    "比如 **MSZoning** 有两种可能，一种是 RL，一种是 RM，那么我们就将这个特征变成两个新的特征，RL 和 RM，如果这个数据在 **MSZoning** 上是 RL，那么 RL 取 1，RM 取 0；反之如果这个特征是 RM，那么 RL 取 0，RM 取 1.\n",
    "\n",
    "| RL | RM |\n",
    "|-|-|\n",
    "| 0 | 1 |\n",
    "| 1 | 0 |"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "all_features = pd.get_dummies(all_features, dummy_na=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "除此之外，我们会发现整个数据中有一些丢失数据，这些丢失数据都是 'NA'，我们没有办法将这些数据输入到网络中，所以需要对这些丢失数据进行赋值，这里我们将数据的均值填入到丢失数据中"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "all_features = all_features.fillna(all_features.mean())\n",
    "feat_dim = all_features.shape[1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "前面我们已经做好了数据的预处理，下面我们将所有的训练集和验证集都取出成为一个 numpy 的数组"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "num_train = int(0.9 * train.shape[0]) # 划分训练样本和验证集样本\n",
    "indices = np.arange(train.shape[0])\n",
    "np.random.shuffle(indices)  # shuffle 顺序\n",
    "train_indices = indices[:num_train]\n",
    "valid_indices = indices[num_train:]\n",
    "\n",
    "# 提取训练集和验证集的特征\n",
    "train_features = all_features.iloc[train_indices].values.astype(np.float32)\n",
    "train_features = torch.from_numpy(train_features)\n",
    "valid_features = all_features.iloc[valid_indices].values.astype(np.float32)\n",
    "valid_features = torch.from_numpy(valid_features)\n",
    "train_valid_features = all_features[:train.shape[0]].values.astype(np.float32)\n",
    "train_valid_features = torch.from_numpy(train_valid_features)\n",
    "\n",
    "# 提取训练集和验证集的label\n",
    "train_labels = train['SalePrice'].values[train_indices, None].astype(np.float32)\n",
    "train_labels = torch.from_numpy(train_labels)\n",
    "valid_labels = train['SalePrice'].values[valid_indices, None].astype(np.float32)\n",
    "valid_labels = torch.from_numpy(valid_labels)\n",
    "train_valid_labels = train['SalePrice'].values[:, None].astype(np.float32)\n",
    "train_valid_labels = torch.from_numpy(train_valid_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "test_features = all_features[train.shape[0]:].values.astype(np.float32)\n",
    "test_features = torch.from_numpy(test_features)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "下面是构建神经网络的地方，可以构建任意想要的神经网络"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sequential(\n",
      "  (0): Linear(in_features=331, out_features=1, bias=True)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "def get_model():\n",
    "    net = nn.Sequential(\n",
    "        nn.Linear(feat_dim, 1)\n",
    "    )\n",
    "    return net\n",
    "\n",
    "net = get_model()\n",
    "print(net)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "在评估模型的时候，为了保证大的价格和小的价格对模型都有着近似相同的影响，我们不会直接使用前面定义的均方误差作为最后的评价函数，我们会对预测的价格和真实的价格取 log，然后计算他们之间均方误差的平方根来作为评价指标，这里的指标我们已经在 `utils.py` 中实现了，感兴趣的同学可以去看看。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from utils import train_model, pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 10, train rmse: 0.181, valid rmse: 0.186\n",
      "\n",
      "epoch: 20, train rmse: 0.141, valid rmse: 0.160\n",
      "\n",
      "epoch: 30, train rmse: 0.123, valid rmse: 0.154\n",
      "\n",
      "epoch: 40, train rmse: 0.112, valid rmse: 0.153\n",
      "\n",
      "epoch: 50, train rmse: 0.106, valid rmse: 0.152\n",
      "\n",
      "epoch: 60, train rmse: 0.102, valid rmse: 0.152\n",
      "\n",
      "epoch: 70, train rmse: 0.100, valid rmse: 0.158\n",
      "\n",
      "epoch: 80, train rmse: 0.097, valid rmse: 0.158\n",
      "\n",
      "epoch: 90, train rmse: 0.097, valid rmse: 0.164\n",
      "\n",
      "epoch: 100, train rmse: 0.095, valid rmse: 0.167\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlsAAAFACAYAAACLPLm0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x21a9d5a7320>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 可以调整的超参\n",
    "batch_size = 128\n",
    "epochs = 100\n",
    "lr = 0.01\n",
    "wd = 0\n",
    "use_gpu = False\n",
    "\n",
    "net = get_model()\n",
    "train_model(net, train_features, train_labels, valid_features, valid_labels, epochs, \n",
    "            batch_size, lr, wd, use_gpu)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "当我们构建好了训练的过程，下面就开始了不断地调参尝试，最后得到一个效果最好的模型"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch: 10, train rmse: 0.181\n",
      "\n",
      "epoch: 20, train rmse: 0.140\n",
      "\n",
      "epoch: 30, train rmse: 0.123\n",
      "\n",
      "epoch: 40, train rmse: 0.114\n",
      "\n",
      "epoch: 50, train rmse: 0.108\n",
      "\n",
      "epoch: 60, train rmse: 0.105\n",
      "\n",
      "epoch: 70, train rmse: 0.102\n",
      "\n",
      "epoch: 80, train rmse: 0.100\n",
      "\n",
      "epoch: 90, train rmse: 0.099\n",
      "\n",
      "epoch: 100, train rmse: 0.098\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlsAAAFACAYAAACLPLm0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYQsAACAhwhYAAEBChC0AAICE/j/jxzUeU6JzfgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x21aa462db70>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "net = get_model()\n",
    "train_model(net, train_valid_features, train_valid_labels, None, None, epochs, \n",
    "            batch_size, lr, wd, use_gpu)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "运行下面的代码，可以通过训练好的模型预测 testset 的结果，会在当前目录生成 `submission.csv` 用于提交"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred(net, test, test_features)"
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}