華泰人工智能系列之十四對(duì)抗過(guò)擬合，從時(shí)序交叉驗(yàn)證談起

1、正文目錄正文目錄本文研究導(dǎo)讀................................................................5過(guò)擬合問(wèn)題與交叉驗(yàn)證........................................................6模型的參數(shù)和超參數(shù)......................................................6欠擬合和過(guò)擬合...

2、.......................................................7交叉驗(yàn)證................................................................8簡(jiǎn)單交叉驗(yàn)證.......................................................9K折交叉驗(yàn)證...........................

3、.............................9留一法和留P法.....................................................9時(shí)序交叉驗(yàn)證......................................................10時(shí)序交叉驗(yàn)證應(yīng)用于機(jī)器學(xué)習(xí)公共數(shù)據(jù)集.......................................11數(shù)據(jù)集.....

4、............................................................11北京地區(qū)PM2.5數(shù)據(jù)集..............................................11辦公樓監(jiān)控管理數(shù)據(jù)集..............................................12銀行電話營(yíng)銷數(shù)據(jù)集...............................

5、.................12機(jī)器學(xué)習(xí)方法...........................................................12K折與時(shí)序交叉驗(yàn)證的結(jié)果及比較.........................................13時(shí)間序列數(shù)據(jù)......................................................13非時(shí)間序列數(shù)據(jù).......

6、.............................................14小結(jié)...................................................................15時(shí)序交叉驗(yàn)證應(yīng)用于全A選股數(shù)據(jù)集..........................................16人工智能選股模型測(cè)試流程..............................

7、.................16K折與時(shí)序交叉驗(yàn)證的結(jié)果及比較.........................................18因子的時(shí)序特性....................................................18模型最優(yōu)超參數(shù)....................................................19機(jī)器學(xué)習(xí)模型性能.............

8、.....................................19單因子分層回測(cè)....................................................21構(gòu)建策略組合及回測(cè)分析............................................24小結(jié).............................................................

9、......26總結(jié)和展望.................................................................27風(fēng)險(xiǎn)提示...................................................................28圖表40：時(shí)序交叉驗(yàn)證應(yīng)用于XGBoost模型分層組合績(jī)效分析（20110131～20180928）.................

10、.......................................................................................................................22圖表41：時(shí)序交叉驗(yàn)證應(yīng)用于XGBoost模型分層組合回測(cè)凈值..................23圖表42：時(shí)序驗(yàn)證XGBoost各層組合凈值除以基準(zhǔn)組合凈值示意圖..............2

11、3圖表43：時(shí)序驗(yàn)證XGBoost分層組合1相對(duì)滬深300月超額收益分布............23圖表44：時(shí)序驗(yàn)證XGBoost多空組合月收益率及累積收益率....................23圖表45：兩種交叉驗(yàn)證方法單因子分層回測(cè)結(jié)果對(duì)比...........................24圖表46：基于兩種交叉驗(yàn)證方法構(gòu)建全A選股策略回測(cè)指標(biāo)對(duì)比（邏輯回歸為基學(xué)習(xí)器，回測(cè)期20110131～20180928）....

12、...........................................24圖表47：基于兩種交叉驗(yàn)證方法構(gòu)建全A選股策略回測(cè)指標(biāo)對(duì)比（XGBoost為基學(xué)習(xí)器，回測(cè)期20110131～20180928）...............................................24圖表48：兩種交叉驗(yàn)證方法應(yīng)用于XGBoost全A選股策略表現(xiàn)（個(gè)股權(quán)重偏離上限2%，基準(zhǔn)為滬深300）..........