动手学人工智能-多层感知机10-实战Kaggle比赛：预测房价

本文我们将详细介绍数据预处理、模型设计和超参数选择。通过亲身实践，你将获得一手经验，这些经验将有益开发者的职业成长。

1. 下载和缓存数据集

python 复制代码

# -*- coding: utf-8 -*-
"""
@File    : 4.10.实战Kaggle比赛：预测房价.py
@Time    : 2024/12/5 16:40
@Desc    : 
"""
import hashlib
import os
import tarfile
import zipfile

import requests

# @save
DATA_HUB = dict()
DATA_URL = 'http://d2l-data.s3-accelerate.amazonaws.com/'


def download(name, cache_dir=os.path.join('..', 'data')):  # @save
    """下载一个DATA_HUB中的文件，返回本地文件名"""
    assert name in DATA_HUB, f"{name} 不存在于 {DATA_HUB}"
    url, sha1_hash = DATA_HUB[name]
    os.makedirs(cache_dir, exist_ok=True)
    fname = os.path.join(cache_dir, url.split('/')[-1])
    if os.path.exists(fname):
        sha1 = hashlib.sha1()
        with open(fname, 'rb') as f:
            while True:
                data = f.read(1048576)
                if not data:
                    break
                sha1.update(data)
        if sha1_hash == sha1.hexdigest():
            return fname
    print(f"正在从{url}下载{fname}")
    r = requests.get(url, stream=True, verify=True)
    with open(fname, 'wb') as f:
        f.write(r.content)
    return fname


def download_extract(name, folder=None):  # @save
    """下载并解压zip/tar文件"""
    fname = download(name)
    base_dir = os.path.dirname(fname)
    data_dir, ext = os.path.splitext(fname)
    if ext == '.zip':
        fp = zipfile.ZipFile(fname, 'r')
    elif ext in ('.tar', '.gz'):
        fp = tarfile.open(fname, 'r')
    else:
        assert False, '只有zip/tar文件可以被解压缩'
    fp.extract(base_dir)
    return os.path.join(base_dir, folder) if folder else data_dir


def download_all():  # @save
    """下载DATA_HUB中的所有文件"""
    for name in DATA_HUB:
        download(name)


DATA_HUB['kaggle_house_train'] = (
    DATA_URL + 'kaggle_house_pred_train.csv',
    '585e9cc93e70b39160e7921475f9bcd7d31219ce'
)

DATA_HUB['kaggle_house_test'] = (
    DATA_URL + 'kaggle_house_pred_test.csv',
    'fa19780a7b011d9b009e8bff8e99922a8ee2eb90'
)

if __name__ == '__main__':
    download_all()

2. Kaggle

Kaggle 是一个全球知名的数据科学和机器学习平台，提供丰富的数据集、代码示例和竞赛资源。用户可以通过参与实际问题的比赛提升技能，与全球数据科学家交流合作。无论是新手还是专家，Kaggle 都是学习、实践和展示数据科学能力的理想选择。

3. 访问和读取数据集

python 复制代码

import numpy as np

import pandas as pd
import torch
from torch import nn

import d2l

train_data = pd.read_csv(d2l.download('kaggle_house_train'))
test_data = pd.read_csv(d2l.download('kaggle_house_test'))

print(train_data.shape)  # (1460, 81)
print(test_data.shape)  # (1459, 80)

# 让我们看看前五行前四个和最后两个特征，以及相应标签（房价）。
print(train_data.iloc[0:5, [0, 1, 2, 3, -3, -2, -1]])
"""
   Id  MSSubClass MSZoning  LotFrontage SaleType SaleCondition  SalePrice
0   1          60       RL         65.0       WD        Normal     208500
1   2          20       RL         80.0       WD        Normal     181500
2   3          60       RL         68.0       WD        Normal     223500
3   4          70       RL         60.0       WD       Abnorml     140000
4   5          60       RL         84.0       WD        Normal     250000
"""

# 第一个特征是ID，不携带任何用于预测的信息，在将数据提供给模型之前，我们将其从数据集中删除
all_features = pd.concat((train_data.iloc[:, 1:-1], test_data.iloc[:, 1:]))

print(all_features.shape)  # (2919, 79)

4. 数据预处理

在开始建模之前，我们需要对数据进行预处理。首先，我们将所有缺失的值替换为相应特征的平均值。然后，为了将所有特征放在一个共同的尺度上，我们通过将特征重新缩放到零均值和单位方差来标准化数据：
<math xmlns="http://www.w3.org/1998/Math/MathML" display="block"> x ← x − μ σ x \leftarrow \frac{x - \mu}{\sigma} </math>x←σx−μ

其中 <math xmlns="http://www.w3.org/1998/Math/MathML"> μ \mu </math>μ 和 <math xmlns="http://www.w3.org/1998/Math/MathML"> σ \sigma </math>σ 分别表示均值和标准差。

python 复制代码

# 若无法获得测试数据，则可根据训练数据计算均值和标准差
numeric_features = all_features.dtypes[all_features.dtypes != 'object'].index
all_features[numeric_features] = all_features[numeric_features].apply(
    lambda x: (x - x.mean()) / x.std()
)
# 在标准化数据之后，所有均值为0，因此我们可以将缺失值设置为0
all_features[numeric_features] = all_features[numeric_features].fillna(0)

print(all_features.iloc[0:5, [0, 1, 2, 3, -3, -2, -1]])
"""
   MSSubClass MSZoning  LotFrontage   LotArea    YrSold SaleType SaleCondition
0    0.067320       RL    -0.184443 -0.217841  0.157619       WD        Normal
1   -0.873466       RL     0.458096 -0.072032 -0.602858       WD        Normal
2    0.067320       RL    -0.055935  0.137173  0.157619       WD        Normal
3    0.302516       RL    -0.398622 -0.078371 -1.363335       WD       Abnorml
4    0.067320       RL     0.629439  0.518814  0.157619       WD        Normal
"""

我们用独热编码替换它们，方法与前面将多类别标签转换为向量的方式相同。

python 复制代码

# "Dummy_na=True"将"na"（缺失值）视为有效的特征值，并为其创建指示符特征
all_features = pd.get_dummies(all_features, dummy_na=True)

print(all_features.shape)  # (2919, 330)

# 我们可以 从pandas格式中提取NumPy格式，并将其转换为张量表示用于训练
n_train = train_data.shape[0]
train_features = torch.tensor(all_features[:n_train].values.astype(np.float32), dtype=torch.float32)
test_features = torch.tensor(all_features[n_train:].values.astype(np.float32), dtype=torch.float32)
train_labels = torch.tensor(train_data.SalePrice.values.reshape(-1, 1), dtype=torch.float32)

print(train_features.shape, test_features.shape, train_labels.shape)
# torch.Size([1460, 330]) torch.Size([1459, 330]) torch.Size([1460, 1])

5. 训练

python 复制代码

loss = nn.MSELoss()  # 均方误差损失函数，reduction参数默认是'mean'
in_features = train_features.shape[1]


def get_net():
    return nn.Sequential(nn.Linear(in_features, 1))

房价就像股票价格一样，我们关心的是相对数量，而不是绝对数量。因此，我们更关心相对误差 <math xmlns="http://www.w3.org/1998/Math/MathML"> y − y ^ y \frac{y-\hat{y}}{y} </math>yy−y^，而不是绝对误差 <math xmlns="http://www.w3.org/1998/Math/MathML"> y − y ^ y-\hat{y} </math>y−y^。解决这个问题的一种方法是用价格预测的对数来衡量差异。
<math xmlns="http://www.w3.org/1998/Math/MathML" display="block"> 1 n ∑ i = 1 n ( log ⁡ y i − log ⁡ y ^ i ) 2 \sqrt{\frac{1}{n}\sum_{i=1}^{n}(\log{y_i}-\log{\hat{y}_i})^2} </math>n1i=1∑n(logyi−logy^i)2

python 复制代码

def log_rmse(net, features, labels):
    # 为了在取对数时进一步稳定该值，将输出值裁剪到1到正无穷之间
    clipped_preds = torch.clamp(net(features), 1, float('inf'))
    rmse = torch.sqrt(loss(torch.log(clipped_preds), torch.log(labels)))
    return rmse.item()

我们的训练函数将借助Adam优化器（我们将在后面更详细地描述它）。

python 复制代码

def train(net, train_features, train_labels, test_features, test_labels,
          num_epochs, learning_rate, weight_decay, batch_size):
    train_ls, test_ls = [], []
    train_iter = d2l.load_array((train_features, train_labels), batch_size)
    # 这里使用的是Adam优化算法
    optimizer = torch.optim.Adam(net.parameters(), lr=learning_rate, weight_decay=weight_decay)
    for epoch in range(num_epochs):
        for X, y in train_iter:
            optimizer.zero_grad()
            l = loss(net(X), y)
            l.backward()
            optimizer.step()
        train_ls.append(log_rmse(net, train_features, train_labels))
        if test_labels is not None:
            test_ls.append(log_rmse(net, test_features, test_labels))
    return train_ls, test_ls

6. K折交叉验证

我们首先需要定义一个函数，在 <math xmlns="http://www.w3.org/1998/Math/MathML"> K K </math>K 折交叉验证过程中返回第 <math xmlns="http://www.w3.org/1998/Math/MathML"> i i </math>i 折的数据。具体地说，它选择第 <math xmlns="http://www.w3.org/1998/Math/MathML"> i i </math>i 个切片作为验证数据，其余部分作为训练数据。

python 复制代码

def get_k_fold_data(k, i, X, y):
    assert k > 1
    fold_size = X.shape[0] // k
    X_train, y_train, X_valid, y_valid = None, None, None, None
    for j in range(k):
        idx = slice(j * fold_size, j * fold_size + fold_size)
        X_part, y_part = X[idx, :], y[idx]
        if j == i:
            X_valid, y_valid = X_part, y_part
        elif X_train is None:
            X_train, y_train = X_part, y_part
        else:
            X_train = torch.cat([X_train, X_part], 0)
            y_train = torch.cat([y_train, y_part], 0)
    return X_train, y_train, X_valid, y_valid

当我们在 <math xmlns="http://www.w3.org/1998/Math/MathML"> K K </math>K 折交叉验证中训练 <math xmlns="http://www.w3.org/1998/Math/MathML"> K K </math>K 次后，返回训练和验证误差的平均值。

python 复制代码

def k_fold(k, X_train, y_train, num_epochs, learning_rate, weight_decay, batch_size):
    train_l_sum, valid_l_sum = 0, 0
    for i in range(k):
        data = get_k_fold_data(k, i, X_train, y_train)
        net = get_net()
        train_ls, valid_ls = train(net, *data, num_epochs, learning_rate, weight_decay, batch_size)
        train_l_sum += train_ls[-1]
        valid_l_sum += valid_ls[-1]
        if i == 0:
            d2l.plot(list(range(1, num_epochs + 1)), [train_ls, valid_ls], xlabel="训练轮数",
                     ylabel="对数均方根误差", xlim=[1, num_epochs], legend=["训练集", "验证集"], yscale='log')
        print(f'{i + 1}折，训练 log rmse = {float(train_ls[-1]):f}, '
              f'验证 log rmse = {float(valid_ls[-1]):f}')
    return train_l_sum / k, valid_l_sum / k

7. 模型选择

在本例中，我们选择了一组未调优的超参数，并将其留给读者来改进模型。

python 复制代码

if __name__ == '__main__':
    k, num_epochs, lr, weight_decay, batch_size = 5, 100, 5, 0, 64
    train_l, valid_l = k_fold(k, train_features, train_labels, num_epochs,
                              lr, weight_decay, batch_size)
    print(f'{k}-折验证: 平均训练 log rmse: {float(train_l):f}, '
          f'平均验证 log rmse: {float(valid_l):f}')

plaintext 复制代码

1折，训练 log rmse = 0.169811, 验证 log rmse = 0.156870
2折，训练 log rmse = 0.162617, 验证 log rmse = 0.194116
3折，训练 log rmse = 0.163898, 验证 log rmse = 0.168392
4折，训练 log rmse = 0.167716, 验证 log rmse = 0.154894
5折，训练 log rmse = 0.162885, 验证 log rmse = 0.182460
5-折验证: 平均训练 log rmse: 0.165385, 平均验证 log rmse: 0.171346

8. 提交Kaggle预测

我们不妨使用所有数据对其进行训练（而不是仅使用交叉验证中使用的 <math xmlns="http://www.w3.org/1998/Math/MathML"> 1 − 1 K 1-\frac{1}{K} </math>1−K1 的数据）。然后，我们通过这种方式获得的模型可以应用于测试集。将预测保存在CSV文件中可以简化将结果上传到Kaggle的过程。

python 复制代码

def train_and_pred(train_features, test_features, train_labels, test_data,
                   num_epochs, lr, weight_decay, batch_size):
    net = get_net()
    train_ls, _ = train(net, train_features, train_labels, None, None,
                        num_epochs, lr, weight_decay, batch_size)
    d2l.plot(list(range(1, num_epochs + 1)), [train_ls], xlabel="训练轮数",
             ylabel="log rmse", xlim=[1, num_epochs], yscale='log')
    print(f'训练log rmse：{float(train_ls[-1]):f}')
    # 将网络应用于测试集
    preds = net(test_features).detach().numpy()
    # 将其重新格式化以导出到Kaggle
    test_data['SalePrice'] = pd.Series(preds.reshape(1, -1)[0])
    submission = pd.concat([test_data['Id'], test_data['SalePrice']], axis=1)
    submission.to_csv('submission.csv', index=False)
    
if __name__ == '__main__':
    k, num_epochs, lr, weight_decay, batch_size = 5, 100, 5, 0, 64
    train_and_pred(train_features, test_features, train_labels, test_data,
                   num_epochs, lr, weight_decay, batch_size)

plaintext 复制代码

训练log rmse：0.162367

完整代码

python 复制代码

import numpy as np
import pandas as pd
import torch
from torch import nn

import d2l

train_data = pd.read_csv(d2l.download('kaggle_house_train'))
test_data = pd.read_csv(d2l.download('kaggle_house_test'))

# 第一个特征是ID，不携带任何用于预测的信息，在将数据提供给模型之前，我们将其从数据集中删除
all_features = pd.concat((train_data.iloc[:, 1:-1], test_data.iloc[:, 1:]))

# 若无法获得测试数据，则可根据训练数据计算均值和标准差
numeric_features = all_features.dtypes[all_features.dtypes != 'object'].index
all_features[numeric_features] = all_features[numeric_features].apply(
    lambda x: (x - x.mean()) / x.std()
)
# 在标准化数据之后，所有均值为0，因此我们可以将缺失值设置为0
all_features[numeric_features] = all_features[numeric_features].fillna(0)

# "Dummy_na=True"将"na"（缺失值）视为有效的特征值，并为其创建指示符特征
all_features = pd.get_dummies(all_features, dummy_na=True)

# 我们可以 从pandas格式中提取NumPy格式，并将其转换为张量表示用于训练
n_train = train_data.shape[0]
train_features = torch.tensor(all_features[:n_train].values.astype(np.float32), dtype=torch.float32)
test_features = torch.tensor(all_features[n_train:].values.astype(np.float32), dtype=torch.float32)
train_labels = torch.tensor(train_data.SalePrice.values.reshape(-1, 1), dtype=torch.float32)

loss = nn.MSELoss()  # 均方误差损失函数，reduction参数默认是'mean'
in_features = train_features.shape[1]


def get_net():
    return nn.Sequential(nn.Linear(in_features, 1))


def log_rmse(net, features, labels):
    # 为了在取对数时进一步稳定该值，将输出值裁剪到1到正无穷之间
    clipped_preds = torch.clamp(net(features), 1, float('inf'))
    rmse = torch.sqrt(loss(torch.log(clipped_preds), torch.log(labels)))
    return rmse.item()


def train(net, train_features, train_labels, test_features, test_labels,
          num_epochs, learning_rate, weight_decay, batch_size):
    train_ls, test_ls = [], []
    train_iter = d2l.load_array((train_features, train_labels), batch_size)
    # 这里使用的是Adam优化算法
    optimizer = torch.optim.Adam(net.parameters(), lr=learning_rate, weight_decay=weight_decay)
    for epoch in range(num_epochs):
        for X, y in train_iter:
            optimizer.zero_grad()
            l = loss(net(X), y)
            l.backward()
            optimizer.step()
        train_ls.append(log_rmse(net, train_features, train_labels))
        if test_labels is not None:
            test_ls.append(log_rmse(net, test_features, test_labels))
    return train_ls, test_ls


def get_k_fold_data(k, i, X, y):
    assert k > 1
    fold_size = X.shape[0] // k
    X_train, y_train, X_valid, y_valid = None, None, None, None
    for j in range(k):
        idx = slice(j * fold_size, j * fold_size + fold_size)
        X_part, y_part = X[idx, :], y[idx]
        if j == i:
            X_valid, y_valid = X_part, y_part
        elif X_train is None:
            X_train, y_train = X_part, y_part
        else:
            X_train = torch.cat([X_train, X_part], 0)
            y_train = torch.cat([y_train, y_part], 0)
    return X_train, y_train, X_valid, y_valid


def k_fold(k, X_train, y_train, num_epochs, learning_rate, weight_decay, batch_size):
    train_l_sum, valid_l_sum = 0, 0
    for i in range(k):
        data = get_k_fold_data(k, i, X_train, y_train)
        net = get_net()
        train_ls, valid_ls = train(net, *data, num_epochs, learning_rate, weight_decay, batch_size)
        train_l_sum += train_ls[-1]
        valid_l_sum += valid_ls[-1]
        if i == 0:
            d2l.plot(list(range(1, num_epochs + 1)), [train_ls, valid_ls], xlabel="训练轮数",
                     ylabel="对数均方根误差", xlim=[1, num_epochs], legend=["训练集", "验证集"], yscale='log')
        print(f'{i + 1}折，训练 log rmse = {float(train_ls[-1]):f}, '
              f'验证 log rmse = {float(valid_ls[-1]):f}')
    return train_l_sum / k, valid_l_sum / k


def train_and_pred(train_features, test_features, train_labels, test_data,
                   num_epochs, lr, weight_decay, batch_size):
    net = get_net()
    train_ls, _ = train(net, train_features, train_labels, None, None,
                        num_epochs, lr, weight_decay, batch_size)
    d2l.plot(list(range(1, num_epochs + 1)), [train_ls], xlabel="训练轮数",
             ylabel="log rmse", xlim=[1, num_epochs], yscale='log')
    print(f'训练log rmse：{float(train_ls[-1]):f}')
    # 将网络应用于测试集
    preds = net(test_features).detach().numpy()
    # 将其重新格式化以导出到Kaggle
    test_data['SalePrice'] = pd.Series(preds.reshape(1, -1)[0])
    submission = pd.concat([test_data['Id'], test_data['SalePrice']], axis=1)
    submission.to_csv('submission.csv', index=False)


if __name__ == '__main__':
    k, num_epochs, lr, weight_decay, batch_size = 5, 100, 5, 0, 64
    # train_l, valid_l = k_fold(k, train_features, train_labels, num_epochs,
    #                           lr, weight_decay, batch_size)
    # print(f'{k}-折验证: 平均训练 log rmse: {float(train_l):f}, '
    #       f'平均验证 log rmse: {float(valid_l):f}')
    train_and_pred(train_features, test_features, train_labels, test_data,
                   num_epochs, lr, weight_decay, batch_size)