自動(dòng)機(jī)器學(xué)習(xí)：方法、系統(tǒng)與挑戰(zhàn)（AutoML）

目 錄

自動(dòng)機(jī)器學(xué)習(xí)方法

第1章 超參優(yōu)化 ··································2

1.1 引言 ··············································2

1.2 問(wèn)題定義 ·······································4

1.2.1 優(yōu)化替代方案：集成與邊緣化 ·············5

1.2.2 多目標(biāo)優(yōu)化 ···········································5

1.3 黑盒超參優(yōu)化 ·······························6

1.3.1 免模型的黑盒優(yōu)化方法 ························6

1.3.2 貝葉斯優(yōu)化 ···········································8

1.4 多保真度優(yōu)化 ······························13

1.4.1 基于學(xué)習(xí)曲線預(yù)測(cè)的早停法 ··············14

1.4.2 基于Bandit的選擇方法 ·····················15

1.4.3 保真度的適應(yīng)性選擇 ··························17

1.5 AutoML的相關(guān)應(yīng)用 ····················18

1.6 探討與展望 ··································20

1.6.1 基準(zhǔn)測(cè)試和基線模型 ··························21

1.6.2 基于梯度的優(yōu)化 ··································22

1.6.3 可擴(kuò)展性 ·············································22

1.6.4 過(guò)擬合和泛化性 ··································23

1.6.5 任意尺度的管道構(gòu)建 ··························24

參考文獻(xiàn)···············································25

第2章 元學(xué)習(xí) ···································36

2.1 引言 ·············································36

2.2 模型評(píng)估中學(xué)習(xí) ··························37

2.2.1 獨(dú)立于任務(wù)的推薦 ······························38

2.2.2 配置空間的設(shè)計(jì) ··································39

2.2.3 配置遷移 ·············································39

2.2.4 學(xué)習(xí)曲線 ·············································42

2.3 任務(wù)特性中學(xué)習(xí) ··························43

2.3.1 元特征 ·················································43

2.3.2 元特征的學(xué)習(xí) ·····································44

2.3.3 基于相似任務(wù)熱啟動(dòng)優(yōu)化過(guò)程 ···········46

2.3.4 元模型 ·················································48

2.3.5 管道合成 ·············································49

2.3.6 調(diào)優(yōu)與否 ·············································50

2.4 先前模型中學(xué)習(xí) ··························50

第一篇



XVI

2.4.1 遷移學(xué)習(xí) ·············································51

2.4.2 針對(duì)神經(jīng)網(wǎng)絡(luò)的元學(xué)習(xí) ······················51

2.4.3 小樣本學(xué)習(xí) ·········································52

2.4.4 不止于監(jiān)督學(xué)習(xí) ··································54

2.5 總結(jié) ·············································55

參考文獻(xiàn)···············································56

第3章 神經(jīng)網(wǎng)絡(luò)架構(gòu)搜索 ··················68

3.1 引言 ·············································68

3.2 搜索空間 ······································69

3.3 搜索策略 ······································73

3.4 性能評(píng)估策略 ······························76

3.5 未來(lái)方向 ······································78

參考文獻(xiàn)···············································80

自動(dòng)機(jī)器學(xué)習(xí)系統(tǒng)

第4章 Auto-WEKA ···························86

4.1 引言 ·············································86

4.2 準(zhǔn)備工作 ······································88

4.2.1 模型選擇 ·············································88

4.2.2 超參優(yōu)化 ·············································88

4.3 算法選擇與超參優(yōu)化結(jié)合
（CASH） ···································89

4.4 Auto-WEKA ·································91

4.5 實(shí)驗(yàn)評(píng)估 ······································93

4.5.1 對(duì)比方法 ·············································94

4.5.2 交叉驗(yàn)證性能 ·····································96

4.5.3 測(cè)試性能 ·············································96

4.6 總結(jié) ·············································98

參考文獻(xiàn)···············································98

第5章 Hyperopt-sklearn ·················101

5.1 引言 ···········································101

5.2 Hyperopt背景 ····························102

5.3 Scikit-Learn模型選擇 ···············103

5.4 使用示例 ····································105

5.5 實(shí)驗(yàn) ···········································109

5.6 討論與展望 ································111

5.7 總結(jié) ···········································114

參考文獻(xiàn)·············································114

第6章 Auto-sklearn ························116

6.1 引言 ···········································116

6.2 CASH問(wèn)題 ································118

6.3 改進(jìn) ···········································119

6.3.1 元學(xué)習(xí)步驟 ········································119

6.3.2 集成的自動(dòng)構(gòu)建 ································121

6.4 Auto-sklearn系統(tǒng) ······················121

6.5 Auto-sklearn的對(duì)比試驗(yàn) ···········125

6.6 Auto-sklearn改進(jìn)項(xiàng)的評(píng)估 ·······127

6.7 Auto-sklearn組件的詳細(xì)分析 ···129

6.8 討論與總結(jié) ································134

6.8.1 討論 ···················································134

第二篇



XVII

6.8.2 使用示例 ···········································134

6.8.3 Auto-sklearn的擴(kuò)展 ··························135

6.8.4 總結(jié)與展望 ·······································136

參考文獻(xiàn)·············································136

第7章 Auto-Net ······························140

7.1 引言 ···········································140

7.2 Auto-Net 1.0 ·······························142

7.3 Auto-Net 2.0 ·······························144

7.4 實(shí)驗(yàn) ···········································151

7.4.1 基線評(píng)估 ···········································151

7.4.2 AutoML競(jìng)賽上的表現(xiàn) ·····················152

7.4.3 Auto-Net 1.0與Auto-Net 2.0的對(duì)比····154

7.5 總結(jié) ···········································155

參考文獻(xiàn)·············································156

第8章 TPOT ··································160

8.1 引言 ···········································160

8.2 方法 ···········································161

8.2.1 機(jī)器學(xué)習(xí)管道算子 ····························161

8.2.2 構(gòu)建基于樹的管道 ····························162

8.2.3 優(yōu)化基于樹的管道 ····························163

8.2.4 基準(zhǔn)測(cè)試數(shù)據(jù) ···································163

8.3 實(shí)驗(yàn)結(jié)果 ····································164

8.4 總結(jié)與展望 ································167

參考文獻(xiàn)·············································168

第9章 自動(dòng)統(tǒng)計(jì) ······························170

9.1 引言 ···········································170

9.2 自動(dòng)統(tǒng)計(jì)項(xiàng)目的基本結(jié)構(gòu) ·········172

9.3 應(yīng)用于時(shí)序數(shù)據(jù)的自動(dòng)統(tǒng)計(jì) ·····173

9.3.1 核函數(shù)上的語(yǔ)法 ································173

9.3.2 搜索和評(píng)估過(guò)程 ································175

9.3.3 生成自然語(yǔ)言性的描述 ····················175

9.3.4 與人類比較 ·······································177

9.4 其他自動(dòng)統(tǒng)計(jì)系統(tǒng) ····················178

9.4.1 核心組件 ···········································178

9.4.2 設(shè)計(jì)挑戰(zhàn) ···········································179

9.5 總結(jié) ···········································180

參考文獻(xiàn)·············································180

自動(dòng)機(jī)器學(xué)習(xí)挑戰(zhàn)賽

第10章 自動(dòng)機(jī)器學(xué)習(xí)挑戰(zhàn)賽分析 ···186

10.1 引言··········································187

10.2 問(wèn)題形式化和概述 ···················190

10.2.1 問(wèn)題的范圍 ·····································190

10.2.2 全模型選擇 ·····································191

10.2.3 超參優(yōu)化 ·········································192

10.2.4 模型搜索策略 ·································193

10.3 數(shù)據(jù)··········································197

10.4 挑戰(zhàn)賽協(xié)議 ······························201

10.4.1 時(shí)間預(yù)算和計(jì)算資源 ······················201

10.4.2 評(píng)分標(biāo)準(zhǔn) ·········································202

10.4.3 挑戰(zhàn)賽2015/2016中的輪次和階段 ····205

第三篇



10.4.4 挑戰(zhàn)賽2018中的階段 ····················206

10.5 結(jié)果··········································207

10.5.1 挑戰(zhàn)賽2015/2016上的得分 ···········207

10.5.2 挑戰(zhàn)賽2018上的得分 ····················209

10.5.3 數(shù)據(jù)集/任務(wù)的難度 ·······················210

10.5.4 超參優(yōu)化 ·········································217

10.5.5 元學(xué)習(xí) ·············································217

10.5.6 挑戰(zhàn)賽中使用的方法 ······················219

10.6 討論··········································224

10.7 總結(jié)··········································226

參考文獻(xiàn)·············································229