機器學習圖解

第1 章 什麽是機器學習？這是一種常識，唯一特別之處在於由電腦完成···········1

1.1 我是否需要掌握大量的數學和編程背景知識才能理解機器學習············ 2

1.2 機器學習究竟是什麽······· 3

1.3 如何讓機器根據數據做出決策？記憶-制定-預測框架············· 6

1.4 本章小結····················· 12

第2 章 機器學習類型·····················15

2.1 標簽數據和無標簽數據的區別··························· 17

2.2 監督學習：處理標簽數據的機器學習分支··············· 18

2.3 無監督學習：處理無標簽

數據的機器學習分支······ 21

2.4 什麽是強化學習············ 28

2.5 本章小結····················· 30

2.6 練習··························· 31

第3 章 在點附近畫一條線：線性

回歸····································33

3.1 問題：預測房屋的價格··· 35

3.2 解決方案：建立房價回歸模型··························· 35

3.3 如何讓電腦繪制出這

條線：線性回歸算法······ 41

3.4 如何衡量結果？誤差函數··························· 54

3.5 實際應用：使用Turi Create預測房價····················· 61

3.6 如果數據不在一行怎麽辦？

多項式回歸·················· 63

3.7 參數和超參數··············· 64

3.8 回歸應用····················· 65

3.9 本章小結····················· 66

3.10 練習························· 66

第4 章 優化訓練過程：欠擬合、過擬合、測試和正則化······ 69

4.1 使用多項式回歸的欠擬合和過擬合示例··············· 71

4.2 如何讓電腦選擇正確的模型？測試·················· 73

4.3 我們在哪裡打破了黃金法則，如何解決呢？驗證集························ 75

4.4 一種決定模型復雜度的數值方法：模型復雜度圖··························· 76

4.5 避免過擬合的另一種選擇：正則化························ 77

4.6 使用Turi Create 進行多項式回歸、測試和正則化······ 85

4.7 本章小結····················· 89

4.8 練習··························· 90

第5 章 使用線來劃分點: 感知器算法····································93

5.1 問題：我們在一個外星球上，聽不懂外星人的語言······ 95

5.2 如何確定分類器的好壞？誤差函數··················· 108

5.3 如何找到一個好的分類器？感知器算法················ 115

5.4 感知器算法編程實現···· 123

5.5 感知器算法的應用······· 128

5.6 本章小結··················· 129

5.7 練習························· 130

第6 章 劃分點的連續方法：邏輯分類器··································133

6.1 邏輯分類器：連續版感知器分類器··················· 134

6.2 如何找到一個好的邏輯分類器？邏輯回歸算法······· 144

6.3 對邏輯回歸算法進行編程························· 150

6.4 實際應用：使用Turi Create對IMDB 評論進行分類························· 154

6.5 多分類：softmax 函數·· 156

6.6 本章小結··················· 157

6.7 練習························· 158

第7 章 如何衡量分類模型？準確率和其他相關概念·················· 159

7.1 準確率：模型的正確頻率是多少······················ 160

7.2 如何解決準確率問題？定義不同類型的誤差以及如何進行衡量············· 161

7.3 一個有用的模型評價工具

ROC 曲線·················· 170

7.4 本章小結··················· 179

7.5 練習························· 181

第8 章 使用概率最大化：

樸素貝葉斯模型··············· 183

8.1 生病還是健康？以貝葉斯定理為主角的故事······· 184

8.2 用例：垃圾郵件檢測模型························· 188

8.3 使用真實數據構建垃圾郵件檢測模型············· 201

8.4 本章小結··················· 204

8.5 練習························· 205

第9 章 通過提問劃分數據：決策樹····························· 207

9.1 問題：需要根據用戶可能下載的內容向用戶推薦

應用························· 213

9.2 解決方案：構建應用推薦系統························· 214

9.3 超出“是”或“否”之類的問題················ 228

9.4 決策樹的圖形邊界······· 231

9.5 實際應用：使用Scikit-Learn 構建招生模型····· 234

9.6 用於回歸的決策樹······· 238

9.7 應用························· 241

9.8 本章小結··················· 242

9.9 練習························· 242

第10 章 組合積木以獲得更多力量：

神經網絡························245

10.1 以更復雜的外星球為例，開

啟神經網絡學習········ 247

10.2 訓練神經網絡··········· 258

10.3 Keras 中的神經網絡編程······················· 264

10.4 用於回歸的神經網絡·· 272

10.5 用於更復雜數據集的其他架構················· 273

10.6 本章小結················· 275

10.7 練習······················· 276

第11 章 用風格尋找界限：支持向量機和內核方法··········279

11.1 使用新的誤差函數構建更好的分類器··········· 281

11.2 Scikit-Learn 中的SVM編程······················· 287

11.3 訓練非線性邊界的SVM：

內核方法················· 289

11.4 本章小結················· 308

11.5 練習······················· 309

第12 章 組合模型以最大化結果：

集成學習························311

12.1 獲取朋友的幫助········ 312

12.2 bagging：隨機組合弱學習器以構建強學習器····· 314

12.3 AdaBoost：以智能方式組合弱學習器以構建強學習器···················· 319

12.4 梯度提升：使用決策樹構建強學習器··········· 327

12.5 XGBoost：一種梯度提升

的極端方法·············· 332

12.6 集成方法的應用········ 340

12.7 本章小結················· 341

12.8 練習······················· 341

第13 章 理論付諸實踐：數據工程和

機器學習真實示例········· 343

13.1 泰坦尼克號數據集····· 344

13.2 清洗數據集：缺失值及其處理方法·············· 348

13.3 特徵工程：在訓練模型之前轉換數據集中的特徵······················· 350

13.4 訓練模型················· 355

13.5 調整超參數以找到最佳模型：網格搜索········ 359

13.6 使用k 折交叉驗證來重用訓練和驗證數據········ 362

13.7 本章小結················· 363

13.8 練習······················· 364

以下內容可掃封底二維碼下載

附錄A 習題解答·························· 365

附錄B 梯度下降背後的數學原理：

使用導數和斜率下山········ 398

附錄C 參考資料·························· 416