第1 章 什么是机器学习？这是一种常识，唯一特别之处在于由计算机完成···········1
1.1 我是否需要掌握大量的数学和编程背景知识才能理解机器学习············ 2
1.2 机器学习究竟是什么······· 3
1.3 如何让机器根据数据做出决策？记忆-制定-预测框架············· 6
1.4 本章小结····················· 12
第2 章 机器学习类型·····················15
2.1 标签数据和无标签数据的区别··························· 17
2.2 监督学习：处理标签数据的机器学习分支··············· 18
2.3 无监督学习：处理无标签
数据的机器学习分支······ 21
2.4 什么是强化学习············ 28
2.5 本章小结····················· 30
2.6 练习··························· 31
第3 章 在点附近画一条线：线性
回归····································33
3.1 问题：预测房屋的价格··· 35
3.2 解决方案：建立房价回归模型··························· 35
3.3 如何让计算机绘制出这
条线：线性回归算法······ 41
3.4 如何衡量结果？误差函数··························· 54
3.5 实际应用：使用Turi Create预测房价····················· 61
3.6 如果数据不在一行怎么办？
多项式回归·················· 63
3.7 参数和超参数··············· 64
3.8 回归应用····················· 65
3.9 本章小结····················· 66
3.10 练习························· 66
第4 章 优化训练过程：欠拟合、过拟合、测试和正则化······ 69
4.1 使用多项式回归的欠拟合和过拟合示例··············· 71
4.2 如何让计算机选择正确的模型？测试·················· 73
4.3 我们在哪里打破了黄金法则，如何解决呢？验证集························ 75
4.4 一种决定模型复杂度的数值方法：模型复杂度图··························· 76
4.5 避免过拟合的另一种选择：正则化························ 77
4.6 使用Turi Create 进行多项式回归、测试和正则化······ 85
4.7 本章小结····················· 89
4.8 练习··························· 90
第5 章 使用线来划分点: 感知器算法····································93
5.1 问题：我们在一个外星球上，听不懂外星人的语言······ 95
5.2 如何确定分类器的好坏？误差函数··················· 108
5.3 如何找到一个好的分类器？感知器算法················ 115
5.4 感知器算法编程实现···· 123
5.5 感知器算法的应用······· 128
5.6 本章小结··················· 129
5.7 练习························· 130
第6 章 划分点的连续方法：逻辑分类器··································133
6.1 逻辑分类器：连续版感知器分类器··················· 134
6.2 如何找到一个好的逻辑分类器？逻辑回归算法······· 144
6.3 对逻辑回归算法进行编程························· 150
6.4 实际应用：使用Turi Create对IMDB 评论进行分类························· 154
6.5 多分类：softmax 函数·· 156
6.6 本章小结··················· 157
6.7 练习························· 158
第7 章 如何衡量分类模型？准确率和其他相关概念·················· 159
7.1 准确率：模型的正确频率是多少······················ 160
7.2 如何解决准确率问题？定义不同类型的误差以及如何进行衡量············· 161
7.3 一个有用的模型评价工具
ROC 曲线·················· 170
7.4 本章小结··················· 179
7.5 练习························· 181
第8 章 使用概率最大化：
朴素贝叶斯模型··············· 183
8.1 生病还是健康？以贝叶斯定理为主角的故事······· 184
8.2 用例：垃圾邮件检测模型························· 188
8.3 使用真实数据构建垃圾邮件检测模型············· 201
8.4 本章小结··················· 204
8.5 练习························· 205
第9 章 通过提问划分数据：决策树····························· 207
9.1 问题：需要根据用户可能下载的内容向用户推荐
应用························· 213
9.2 解决方案：构建应用推荐系统························· 214
9.3 超出“是”或“否”之类的问题················ 228
9.4 决策树的图形边界······· 231
9.5 实际应用：使用Scikit-Learn 构建招生模型····· 234
9.6 用于回归的决策树······· 238
9.7 应用························· 241
9.8 本章小结··················· 242
9.9 练习························· 242
第10 章 组合积木以获得更多力量：
神经网络························245
10.1 以更复杂的外星球为例，开
启神经网络学习········ 247
10.2 训练神经网络··········· 258
10.3 Keras 中的神经网络编程······················· 264
10.4 用于回归的神经网络·· 272
10.5 用于更复杂数据集的其他架构················· 273
10.6 本章小结················· 275
10.7 练习······················· 276
第11 章 用风格寻找界限：支持向量机和内核方法··········279
11.1 使用新的误差函数构建更好的分类器··········· 281
11.2 Scikit-Learn 中的SVM编程······················· 287
11.3 训练非线性边界的SVM：
内核方法················· 289
11.4 本章小结················· 308
11.5 练习······················· 309
第12 章 组合模型以最大化结果：
集成学习························311
12.1 获取朋友的帮助········ 312
12.2 bagging：随机组合弱学习器以构建强学习器····· 314
12.3 AdaBoost：以智能方式组合弱学习器以构建强学习器···················· 319
12.4 梯度提升：使用决策树构建强学习器··········· 327
12.5 XGBoost：一种梯度提升
的极端方法·············· 332
12.6 集成方法的应用········ 340
12.7 本章小结················· 341
12.8 练习······················· 341
第13 章 理论付诸实践：数据工程和
机器学习真实示例········· 343
13.1 泰坦尼克号数据集····· 344
13.2 清洗数据集：缺失值及其处理方法·············· 348
13.3 特征工程：在训练模型之前转换数据集中的特征······················· 350
13.4 训练模型················· 355
13.5 调整超参数以找到最佳模型：网格搜索········ 359
13.6 使用k 折交叉验证来重用训练和验证数据········ 362
13.7 本章小结················· 363
13.8 练习······················· 364
以下内容可扫封底二维码下载
附录A 习题解答·························· 365
附录B 梯度下降背后的数学原理：
使用导数和斜率下山········ 398
附录C 参考资料·························· 416
