æ¬æ¥ã¯ãæ©æ¢°åŠç¿ïŒä»¥äžãMLïŒãçšããŠæ°Žå§ç Žç ïŒHFïŒã®åè£äºæžã®æ€çŽ¢ã·ã¹ãã ãã©ã®ããã«éçºãããããããŠãã®çµæããäŒãããŸãããªãæ²¹å§ãã©ã¯ãã£ãªã³ã°ãå¿ èŠãªã®ããMLããããšäœã®é¢ä¿ãããã®ãââããããŠãªãç§ãã¡ã®çµéšãç³æ²¹æ¥è ã ãã§ãªã圹ç«ã€ã®ããç解ããŸãããã
ã«ããã®äžã§ãåé¡ã®è©³çŽ°ãªèª¬æãITãœãªã¥ãŒã·ã§ã³ã®èª¬æãã¡ããªãã¯ã®éžæãMLãã€ãã©ã€ã³ã®äœæã補åã§ã¢ãã«ããªãªãŒã¹ããããã®ã¢ãŒããã¯ãã£ã®éçº
ãã©ã¯ãã£ãªã³ã°ãè¡ãããçç±ã«ã€ããŠã¯ã以åã®èšäºã®ãããšããã«æžããŠããŸãã
ãªãããã§æ©æ¢°åŠç¿ãããŠããã®ã§ããïŒæ²¹å§ãã©ã¯ãã£ãªã³ã°ã¯ãããªã«ãããå®äŸ¡ã§ãããããã§ãã³ã¹ããããããŸãããŸãããã¹ãŠã®ãŠã§ã«ã§ãããè¡ãããšã¯äžå¯èœã§ãããå¹æã¯ãããŸãããå°éã®å°è³ªåŠè ãé©åãªå Žæãæ¢ããŠããŸããéå¶äŒç€Ÿã®æ°ãå€ãïŒæ°äžïŒãããéžæè¢ãèŠèœãšããã¡ã§ãããå©çãåŸãããšãã§ããŸãããæ©æ¢°åŠç¿ã䜿çšãããšãæ å ±ã®åæãå€§å¹ ã«ã¹ããŒãã¢ããã§ããŸãããã ããMLã¢ãã«ã®äœæã¯æŠãã®ååã«ãããŸããããããã³ã³ã¹ã¿ã³ãã¢ãŒãã§åäœãããããŒã¿ãµãŒãã¹ã«æ¥ç¶ããçŸããã€ã³ã¿ãŒãã§ã€ã¹ãæç»ããŠããŠãŒã¶ãŒãã¢ããªã±ãŒã·ã§ã³ã«å ¥ãã2åã®ã¯ãªãã¯ã§åé¡ã解決ããã®ã«äŸ¿å©ã«ãªãããã«ããå¿ èŠããããŸãã
ç³æ²¹æ¥çããæœè±¡åãããšããã¹ãŠã®äŒæ¥ã§åæ§ã®ã¿ã¹ã¯ã解決ãããŠããããšãããããŸãã ïŒèª°ããããã
A.èªååã倧èŠæš¡ãªããŒã¿ã»ã¹ããªãŒã ã®åŠçããã³åæãã
B.ã³ã¹ããåæžããã¡ãªãããéããªãããã«ããŸãã
C.ãã®ãããªã·ã¹ãã ãé«éãã€å¹ççã«ããŸãã
ãã®èšäºããããã®ãããªã·ã¹ãã ãã©ã®ããã«å®è£ ããããã©ã®ããŒã«ã䜿çšããããMLãæ¬çªç°å¢ã«å°å ¥ãããšããåä»ãªéã®ãã§ã©ã®ãããªåé¡ãçºçããããåŠã³ãŸããç§ãã¡ã®çµéšã¯ã掻åã®åéã«é¢ä¿ãªããã«ãŒãã³ãèªååããããã¹ãŠã®äººã«ãšã£ãŠèå³æ·±ããã®ã«ãªããšç¢ºä¿¡ããŠããŸãã
ãåŸæ¥ã®ãæ¹æ³ã§æ°Žå§ç Žç çšã«ãŠã§ã«ãéžæããæ¹æ³
æ²¹å§ãã©ã¯ãã£ãªã³ã°ã®åè£ãŠã§ã«ãéžæãããšãããªã€ã«ãã³ã¯åœŒã®è±å¯ãªçµéšã«äŸåããããŸããŸãªã°ã©ããšè¡šãèŠãŠããã®åŸãæ²¹å§ãã©ã¯ãã£ãªã³ã°ãå®è¡ããå Žæãäºæž¬ããŸãããããã確ãã«ãå°äžãèŠãã®ã¯ããã»ã©ç°¡åã§ã¯ãªãã®ã§ãæ°åã¡ãŒãã«ã®æ·±ãã§äœãèµ·ãã£ãŠããã®ã誰ãç¥ããŸããïŒåã®èšäºã§ãã£ãšèªãããšãã§ããŸãïŒã ãåŸæ¥ã®ãæ¹æ³ã«ããããŒã¿åæã«ã¯ããªãã®äººä»¶è²»ãå¿ èŠã§ãããæ®å¿µãªãããæ°Žå§ç Žç ã®çµæãæ£ç¢ºã«äºæž¬ããããšã¯ä¿èšŒãããŸããïŒã¹ãã€ã©ãŒ-MLã䜿çšïŒã
æ²¹å§ãã©ã¯ãã£ã®åè£ãŠã§ã«ãç¹å®ããçŸåšã®ããã»ã¹ã説æãããšãäŒæ¥æ å ±ã·ã¹ãã ããã®ãŠã§ã«ããŒã¿ã®ã¢ã³ããŒããååŸããããŒã¿ã®åŠçãå°é家ã«ããåæã®å®æœã解決çã®åæãæ²¹å§ãã©ã¯ãã£ã®å®æœãçµæã®åæã®æ®µéã§æ§æãããŸããã·ã³ãã«ã«èŠããŸãããå®å šã§ã¯ãããŸããã
åè£ãŠã§ã«
ã®éžæã®çŸåšã®ããã»ã¹ãã®ãæåãã¢ãããŒãã®äž»ãªæ¬ ç¹ã¯ãå€ãã®ã«ãŒãã³ãããªã¥ãŒã ã®å¢å ã人ã ãä»äºã«æººãå§ããããšãããã»ã¹ãšæ¹æ³ã«éææ§ããªãããšã§ãã
åé¡ã®å®åŒå
2019幎ãåœç€Ÿã®ããŒã¿åæããŒã ã¯ãæ°Žå§ç Žç ã®åè£äºæžãéžæããããã®èªåã·ã¹ãã ãäœæãããšãã課é¡ã«çŽé¢ããŸãããç§ãã¡ã«ãšã£ãŠã¯ããã¹ãŠã®äºæžã®ç¶æ ãã·ãã¥ã¬ãŒãããããã«èãããŸãã-çŸæç¹ã§ãããã«æ²¹å§ãã©ã¯ãã£ãªã³ã°ãå®è¡ããå¿ èŠããããšä»®å®ããŠã次ã«ãç³æ²¹çç£ã®æ倧ã®å¢å ã«ãã£ãŠäºæžãã©ã³ã¯ä»ãããèŠéã移åãããããNäºæžãéžæããç³æ²¹ååãå¢ããããã®å¯Ÿçãè¬ããŸãã
MLã¢ãã«ã䜿çšããŠãç¹å®ã®äºæžã§ã®æ°Žåç Žç ã®å®çŸå¯èœæ§ã瀺ãææšã圢æãããŸããèšç»ãããæ°Žåç Žç åŸã®ç³æ²¹çç£ãšãã®ã€ãã³ãã®æåã§ãã
ç§ãã¡ã®å Žåãç³æ²¹çç£çã¯ã1ãæãããã®ç«æ¹ã¡ãŒãã«ã§çç£ãããç³æ²¹ã®éã§ãããã®ææšã¯ã液äœæµéãšãŠã©ãŒã¿ãŒã«ããã®2ã€ã®å€ã«åºã¥ããŠèšç®ãããŸããç³æ²¹äŒç€Ÿã¯æ¶²äœãæ²¹ãšæ°Žã®æ··åç©ãšåŒãã§ããŸã-äºæžã®ç£ç©ã§ããã®ã¯ãã®æ··åç©ã§ãããããŠããŠã©ãŒã¿ãŒã«ããã¯ãç¹å®ã®æ··åç©ã®æ°Žåå«æéã®å²åã§ããç Žç åŸã®äºæ³ãããç³æ²¹çç£çãèšç®ããããã«ã2ã€ã®ååž°ã¢ãã«ã䜿çšãããŸãã1ã€ã¯ç Žç åŸã®æµäœã®æµéãäºæž¬ãããã1ã€ã¯ãŠã©ãŒã¿ãŒã«ãããäºæž¬ããŸããã¢ãã«ããŒã¿ã«ãã£ãŠè¿ãããå€ã䜿çšããŠãç³æ²¹çç£äºæž¬ã¯æ¬¡ã®åŒã䜿çšããŠèšç®ãããŸãïŒ
ãã©ã¯ãã£ãªã³ã°ã®æåã¯ããã€ããªã¿ãŒã²ããå€æ°ã§ããããã¯ãæ°Žå§ç Žç åŸã«åŸãããç³æ²¹çç£çã®å¢å ã®å®éã®å€ã䜿çšããŠæ±ºå®ãããŸããã²ã€ã³ããã¡ã€ã³é åã®å°é家ã«ãã£ãŠæ±ºå®ãããç¹å®ã®ãããå€ããã倧ããå Žåãæåå±æ§ã®å€ã¯1ã«çããããã以å€ã®å Žåã¯ãŒãã«çãããªããŸãããããã£ãŠãåé¡åé¡ã解決ããããã®ããŒã¯ã¢ãããäœæããŸãã
ææšã«ã€ããŠã¯...ææšã¯ããžãã¹ããã®ãã®ã§ããã顧客ã®é¢å¿ãåæ ããŠããå¿ èŠããããŸããæ©æ¢°åŠç¿ã³ãŒã¹ã¯ãã¹ãŠæããŠãããŸããç§ãã¡ã®æèŠã§ã¯ããããæ©æ¢°åŠç¿ãããžã§ã¯ãã®äž»ãªæåãŸãã¯å€±æã®åå ã§ããããŒã¿ãµã€ãšã³ãã£ã¹ãã®ã°ã«ãŒãã¯ãã¢ãã«ã®å質ã奜ããªã ãæ¹åã§ããŸãããããã顧客ã®ããžãã¹äŸ¡å€ããŸã£ããåäžãããªãå Žåããã®ãããªã¢ãã«ã¯éåœã¥ããããŸããçµå±ã®ãšãããæ°Žå§ç Žç åŸã®åäºæ§èœãã©ã¡ãŒã¿ã®ãç©ççãäºæž¬ãåããæ£ç¢ºãªåè£ã顧客ãååŸããããšãéèŠã§ããã
ååž°åé¡ã«ã€ããŠã¯ã次ã®ã¡ããªãã¯ãéžæãããŸããã
ãªã1ã€ã®ã¡ããªãã¯ããªãã®ããããªãã¯å°ããŸã-ãããããç¬èªã®çå®ãåæ ããŠããŸããå¹³åçç£çãé«ãåéã§ã¯ãMAEã倧ãããMAPEãå°ãããªããŸããå¹³åçç£çã®äœãåéããšããšãéã«ãªããŸãã
åé¡åé¡ã«ã¯ã次ã®ã¡ããªãã¯ãéžæãããŸããïŒ
ïŒwikiïŒã
ROCã®äžã®é åâ AUCæ²ç·ïŒwikiïŒã
ãšã©ãŒ
ïŒ1-ãã¹ãŠã®ãã£ãŒã«ãã«å¯ŸããŠ1ã€ã®ãŠãããŒãµã«ã¢ãã«ãæ§ç¯ããããã«çºçãããšã©ãŒã
ããŒã¿ã»ãããåæããåŸãããŒã¿ããããã£ãŒã«ãããå¥ã®ãã£ãŒã«ãã«å€åããããšãæããã«ãªããŸãããå ç©ç©ã¯ååãšããŠç°ãªãå°è³ªæ§é ãæã£ãŠããã®ã§ãããã¯é©ãã¹ãããšã§ã¯ãããŸããã
ãã¬ãŒãã³ã°ã«å©çšã§ãããã¹ãŠã®ããŒã¿ãã¢ãã«ã«åã蟌ãã§é§åãããšãããèªäœãå°è³ªæ§é ã®èŠåæ§ãæããã«ãããšããç§ãã¡ã®ä»®å®ã¯å€±æããŸãããç¹å®ã®ãã£ãŒã«ãã®ããŒã¿ã§ãã¬ãŒãã³ã°ãããã¢ãã«ã¯ã䜿çšå¯èœãªãã¹ãŠã®ãã£ãŒã«ãã«é¢ããæ å ±ã䜿çšããŠäœæãããã¢ãã«ãããé«ãå質ã®äºæž¬ã瀺ããŸããã
åãã£ãŒã«ãã«ã€ããŠãããŸããŸãªæ©æ¢°åŠç¿ã¢ã«ãŽãªãºã ããã¹ããããçžäºæ€èšŒã®çµæã«åºã¥ããŠãMAPEãæãäœããã®ãéžæãããŸããã
ééãïŒ2-ããŒã¿ã®æ·±ãç解ã®æ¬ åŠã
å®éã®ç©çããã»ã¹ã«é©ããæ©æ¢°åŠç¿ã¢ãã«ãäœæããå Žåã¯ããã®ããã»ã¹ãã©ã®ããã«è¡ãããããç解ããŠãã ããã
åœåãç§ãã¡ã®ããŒã ã«ã¯ãã¡ã€ã³ãšãã¹ããŒããããªãã£ããããç¡ç§©åºã«ç§»åããŸãããæ®å¿µãªãããäºæž¬ãåæãããšãã«ã¢ãã«ã®ãšã©ãŒã«æ°ä»ããªãã£ããããçµæã«åºã¥ããŠèª€ã£ãçµè«ãå°ãåºãããŸããã
ééãïŒ3-ã€ã³ãã©ã¹ãã©ã¯ãã£ã®æ¬ åŠã
æåã«ãããŸããŸãªãã£ãŒã«ããšããŸããŸãªãã©ã¡ãŒã¿çšã«ããŸããŸãªcsvãã¡ã€ã«ãããŠã³ããŒãããŸãããããæç¹ã§ãèããããªãã»ã©å€æ°ã®ãã¡ã€ã«ãšã¢ãã«ãèç©ãããŸããããã§ã«å®æœããå®éšãåçŸããããšãã§ããªããªãããã¡ã€ã«ã倱ãããæ··ä¹±ãçããŸããã
1.æè¡ããŒã
ä»æ¥ãåè£è ã®èªåéžæã·ã¹ãã ã¯æ¬¡ã®ããã«ãªã£ãŠããŸãã
åã³ã³ããŒãã³ãã¯ãç¹å®ã®æ©èœãå®è¡ããåé¢ãããã³ã³ãããŒã§ãã
2.1 ETL =ããŒã¿ããŒã
ããã¯ãã¹ãŠããŒã¿ããå§ãŸããŸããç¹ã«ãæ©æ¢°åŠç¿ã¢ãã«ãæ§ç¯ãããå Žåãçµ±åã·ã¹ãã ãšããŠPentahoDataIntegrationãéžæããŸããã
å€æã®1ã€ã®ã¹ã¯ãªãŒã³ã·ã§ãã
äž»ãªå©ç¹ïŒ
- ç¡æã·ã¹ãã ;
- ããŸããŸãªããŒã¿ãœãŒã¹ã«æ¥ç¶ããããŒã¿ã¹ããªãŒã ãå€æããããã®ã³ã³ããŒãã³ãã®å¹ åºãéžæè¢ã
- Webã€ã³ã¿ãŒãã§ãŒã¹ã®å¯çšæ§ã
- RESTAPIãä»ããŠç®¡çããæ©èœã
- ãã®ã³ã°ã
äžèšã®ãã¹ãŠã«å ããŠããã®è£œåã®çµ±åã®éçºã«ãããŠè±å¯ãªçµéšããããŸãã MLãããžã§ã¯ãã§ããŒã¿çµ±åãå¿ èŠãªã®ã¯ãªãã§ããïŒããŒã¿ã»ãããæºåããããã»ã¹ã§ã¯ãè€éãªèšç®ãå®è£ ããããŒã¿ãåäžã®åœ¢åŒã«ãããéäžã§ãæ°ããææšãèšç®ããæéã®çµéã«äŒŽããã©ã¡ãŒã¿ã®å€åãªã©ãåžžã«å¿ èŠãšããŸãã
æ²¹å§ãã©ã¯ãã£ã®åäºå®ã«ã€ããŠããã®æç¹ã§ã®äºæžã®æäœã説æãã400ãè¶ ãããã©ã¡ãŒã¿ãã¢ã³ããŒããããŸãã掻åãé£æ¥ããäºæžã®æäœãããã³ä»¥åã«å®è¡ãããæ°Žåç Žç ã«é¢ããæ å ±ãããã«ãããŒã¿å€æãšååŠçãè¡ãããŸãã
åŠçãããããŒã¿ã®ãªããžããªãšããŠPostgreSQLãéžæããŸããã jsonãæäœããããã®ã¡ãœããã®ã»ããããããããããŸããæçµçãªããŒã¿ã»ããããã®åœ¢åŒã§ä¿åãããããããã決å®çãªèŠå ã«ãªããŸããã
æ©æ¢°åŠç¿ãããžã§ã¯ãã¯ãæ°æ©èœã®è¿œå ã«ããå ¥åããŒã¿ã®çµ¶ãéãªãå€åã«é¢é£ä»ããããŠãããããData VaultãããŒã¿ããŒã¹ã¹ããŒããšããŠäœ¿çšãããŸãïŒwikiãžã®ãªã³ã¯ïŒããã®ã¹ãã¬ãŒãžèšèšã¹ããŒã ã«ããããªããžã§ã¯ãã«é¢ããæ°ããããŒã¿ããã°ããè¿œå ã§ããããŒãã«ãšã¯ãšãªã®æŽåæ§ã«éåããããšã¯ãããŸããã
2.2ããŒã¿ããã³ã¢ãã«ãµãŒãã¹
å¿ èŠãªææšãçµã¿åãããŠèšç®ããåŸãããŒã¿ã¯ããŒã¿ããŒã¹ã«ã¢ããããŒããããŸãããããã¯ããã«ä¿åãããããŒã¿ã·ã³ã¿ãŒãMLã¢ãã«ãäœæããããã«ããããååŸããã®ãåŸ ã£ãŠããŸãããã®ããã«ãDataServiceããããŸããããã¯Pythonã§èšè¿°ãããgRPCãããã³ã«ã䜿çšãããµãŒãã¹ã§ããããã«ãããããŒã¿ã»ãããšãã®ã¡ã¿ããŒã¿ïŒæ©èœã®ã¿ã€ãã説æãããŒã¿ã»ãããµã€ãºãªã©ïŒã®ååŸãäºæž¬ã®ããŒããšã¢ã³ããŒãããã¬ã€ã³/ãã¹ãã«ãããã£ã«ã¿ãªã³ã°ãšåå²ãã©ã¡ãŒã¿ãŒã®ç®¡çãå¯èœã«ãªããŸããããŒã¿ããŒã¹å ã®äºæž¬ã¯json圢åŒã§ä¿åãããŸããããã«ãããããŒã¿ããã°ããåä¿¡ããäºæž¬å€ã ãã§ãªãããã®ç¹å®ã®äºæž¬ã«å¯Ÿããåæ©èœã®åœ±é¿ãä¿åã§ããŸãã
ããŒã¿ãµãŒãã¹ã®ãµã³ãã«ããããã¡ã€ã«ã
ã¢ãã«ãäœæããããä¿åããå¿ èŠããããŸãããã®ç®çã®ããã«ãModelServiceã䜿çšãããŸãããããgRPCã䜿çšããŠPythonã§èšè¿°ãããŠããŸãããã®ãµãŒãã¹ã®æ©èœã¯ãã¢ãã«ã®ä¿åãšããŒãã«éå®ãããŸãããããã«ãã¡ããªãã¯ãæ©èœã®éèŠæ§ãç£èŠã§ããæ°ããããŒã¿ã衚瀺ããããšãã«äºæž¬ãèªåçã«äœæããããã®ã¢ãã«ãšããŒã¿ã»ããã®æ¥ç¶ãå®è£ ã§ããŸãã
ã¢ãã«ãµãŒãã¹ã®æ§é ã¯æ¬¡ã®ããã«ãªããŸãã
2.3 MLã¢ãã«
ããæç¹ã§ãç§ãã¡ã®ããŒã ã¯ãèªååãMLã¢ãã«ã®äœæã«ã圱é¿ãäžããã¯ãã§ããããšã«æ°ã¥ããŸããããã®å¿ èŠæ§ã¯ãäºæž¬ãè¡ãã仮説ããã¹ãããããã»ã¹ãã¹ããŒãã¢ããããå¿ èŠæ§ã«ãããã®ã§ããããããŠãç¬èªã®AutoMLã©ã€ãã©ãªãéçºããŠãã€ãã©ã€ã³ã«å®è£ ããããšã決å®ããŸããã
åœåãæ¢è£œã®AutoMLã©ã€ãã©ãªã䜿çšããå¯èœæ§ãæ€èšãããŸããããæ¢åã®ãœãªã¥ãŒã·ã§ã³ã¯ã¿ã¹ã¯ã«å¯ŸããŠååãªæè»æ§ããªããå¿ èŠãªãã¹ãŠã®æ©èœãäžåºŠã«åããŠããŸããã§ããïŒã¯ãŒã«ãŒã®èŠæ±ã«å¿ããŠãAutoMLã«é¢ããå¥ã®èšäºãæžãããšãã§ããŸãïŒãç§ãã¡ãéçºãããã¬ãŒã ã¯ãŒã¯ã«ã¯ãããŒã¿ã»ããã®ååŠçãæ©èœã®çæãšéžæã«äœ¿çšãããã¯ã©ã¹ãå«ãŸããŠããããšã«æ³šæããŠãã ãããæ©æ¢°åŠç¿ã¢ãã«ãšããŠããããŸã§ã§æãæåããã¢ã«ãŽãªãºã ã®ã»ããã䜿çšããŸããxgboostããã®åŸé ããŒã¹ãã®å®è£ ãcatboostã©ã€ãã©ãªãSklearnããã®ã©ã³ãã ãã©ã¬ã¹ããPytorchäžã®å®å šã«æ¥ç¶ããããã¥ãŒã©ã«ãããã¯ãŒã¯ãªã©ããã¬ãŒãã³ã°åŸãAutoMLã¯ä»¥äžãå«ãsklearnãã€ãã©ã€ã³ãè¿ããŸããäžèšã®ã¯ã©ã¹ãããã³MLã¢ãã«ãããã¯ãéžæããã¡ããªãã¯ã®çžäºæ€èšŒã§æè¯ã®çµæã瀺ããŸããã
ã¢ãã«ã«å ããŠãç¹å®ã®äºæž¬ã«å¯Ÿããå åã®åœ±é¿ã«é¢ããã¬ããŒããäœæãããŸãããã®ãããªã¬ããŒãã«ãããå°è³ªåŠè ã¯ç¥ç§çãªãã©ãã¯ããã¯ã¹ã®å éšãèŠãããšãã§ããŸãããããã£ãŠãAutoMLã¯DataServiceã䜿çšããŠã¿ã°ä»ãããŒã¿ã»ãããåãåãããã¬ãŒãã³ã°åŸãæçµã¢ãã«ã圢æããŸãã次ã«ããã¹ãããŒã¿ã»ãããããŒãããäºæž¬ãçæããå質ã¡ããªãã¯ãèšç®ããããšã«ãããã¢ãã«ã®å質ã®æçµçãªèŠç©ãããååŸã§ããŸããæçµæ®µéã§ã¯ãçæãããã¢ãã«ã®ãã€ããªãã¡ã€ã«ããã®èª¬æãã¡ããªãã¯ãModelServiceã«ã¢ããããŒãããæ©èœã®åœ±é¿ã«é¢ããäºæž¬ãšæ å ±ãDataServiceã«è¿ããŸãã
ãããã£ãŠãã¢ãã«ã¯ãã¹ããã¥ãŒãã«é 眮ããã補åã«æå ¥ããæºåãæŽããŸãããã€ã§ããããã䜿çšããŠãæ°ããé¢é£ããŒã¿ã«åºã¥ããŠäºæž¬ãçæã§ããŸãã
2.4ã€ã³ã¿ãŒãã§ãŒã¹
ç§ãã¡ã®è£œåã®ãšã³ããŠãŒã¶ãŒã¯å°è³ªåŠè ã§ããã圌ã¯ã©ããããããMLã¢ãã«ãšå¯Ÿè©±ããå¿ èŠããããŸãã圌ã«ãšã£ãŠæã䟿å©ãªæ¹æ³ã¯ãå°çšãœãããŠã§ã¢ã®ã¢ãžã¥ãŒã«ã§ããå®è£ ããŸããã
ãŠãŒã¶ãŒãå©çšã§ããããã³ããšã³ãã¯ãªã³ã©ã€ã³ã¹ãã¢ã®ããã«èŠããŸããç®çã®ãã£ãŒã«ããéžæããŠãæåããå¯èœæ§ãæãé«ããŠã§ã«ã®ãªã¹ããååŸã§ããŸãããŠã§ã«ã«ãŒãã§ã¯ããŠãŒã¶ãŒã¯æ°Žå§ç Žç åŸã«äºæž¬ãããæé·ã確èªãããããããã¹ã±ãããã«è¿œå ããŠãã詳现ã«æ€èšãããã©ãããèªåã§æ±ºå®ããŸãã
ã¢ããªã±ãŒã·ã§ã³ã®ã¢ãžã¥ãŒã«ã€ã³ã¿ãŒãã§ã€ã¹ã
ããã¯ãä»é²ã®ãŠã§ã«ã«ãŒãã®å€èŠ³ã§ãã
äºæž¬ãããæ²¹ãšæ¶²äœã®å¢å ã«å ããŠããŠãŒã¶ãŒã¯ã©ã®æ©èœãææ¡ãããçµæã«åœ±é¿ãäžããããç¥ãããšãã§ããŸããæ©èœã®éèŠæ§ã¯ãshapã¡ãœããã䜿çšããŠã¢ãã«ãäœæãã段éã§èšç®ãããDataServiceã䜿çšããŠãœãããŠã§ã¢ã€ã³ã¿ãŒãã§ã€ã¹ã«ããŒããããŸãã
ãã®ã¢ããªã±ãŒã·ã§ã³ã¯ãã¢ãã«ã®äºæž¬ã«æãéèŠãªæ©èœãæ確ã«ç€ºããŠããŸãã
ãŠãŒã¶ãŒã¯ãé¢å¿ã®ããäºæžã®é¡äŒŒç©ãèŠãããšãã§ããŸããã¢ããã°ã®æ€çŽ¢ã¯ãKdããªãŒã¢ã«ãŽãªãºã ã䜿çšããŠã¯ã©ã€ã¢ã³ãåŽã§å®è£ ãããŸãã
ã¢ãžã¥ãŒã«ã¯ãåæ§ã®å°è³ªåŠçãã©ã¡ãŒã¿ãæã€äºæžã衚瀺ããŸãã
2. MLã¢ãã«
ãã©ã®ããã«æ¹åãããå©çšå¯èœãªããŒã¿ã«å¯ŸããŠAutoMLãå®è¡ãã䟡å€ãããããã«æãããç§ãã¡ã¯æºè¶³ããŠããŸããããããèªåçã«ååŸãããäºæž¬ã®å質ãããŒã¿ã€ã³ã¿ãŒã³ã®çµæãšæ¯èŒã§ããªãå ŽåããããŸããäºå®ãã¢ããªã¹ãã¯ã¢ãã«ãæ¹åããããã«ããŸããŸãªä»®èª¬ãæ瀺ããŠãã¹ãããããšããããããŸããã¢ã€ãã¢ãå®éã®ããŒã¿ã®äºæž¬ã®ç²ŸåºŠãåäžãããå Žåãããã¯AutoMLã«å®è£ ãããŸãããããã£ãŠãæ°ããæ©èœãè¿œå ããããšã«ãããã¢ããªã¹ãã®é¢äžãæå°éã«æããŠã¢ãã«ãšäºæž¬ã®äœæã«é²ãã®ã«ååãªèªåäºæž¬ãæ¹åãããŸããã AutoMLã§ãã¹ãããã³å®è£ ãããããã€ãã®ä»®èª¬ã次ã«ç€ºããŸã
ã1ãå å¡«æ¹æ³ã®å€æŽ
æåã®ã¢ãã«ã§ã¯ãã«ããŽãªãé€ããŠãç¹æ§ã®ã»ãšãã©ãã¹ãŠã®ã®ã£ãããå¹³åã§åããŸããããããã«ã€ããŠã¯ãæãäžè¬çãªæå³ã䜿çšãããŸããããã®åŸãã¢ããªã¹ããšå°é家ã®å ±åäœæ¥ã«ããããã¡ã€ã³é åã§ãæ©èœã®80ïŒ ã®ã®ã£ãããåããããã«æé©ãªå€ãéžæããããšãå¯èœã«ãªããŸããããŸããsklearnã©ã€ãã©ãªãšmissingpyã©ã€ãã©ãªã䜿çšããŠãããã«ããã€ãã®å ¥åæ¹æ³ãè©ŠããŸãããæè¯ã®çµæã¯ãäžå®ã®å å¡«ãšKNNImputerïŒæ倧5ïŒ ã®MAPEïŒã§åŸãããŸããã
ããŸããŸãªæ¹æ³ã§ã®ã£ãããåããå®éšã®çµæã
2.æ©èœã®çæ
æ°ããæ©èœã®è¿œå ã¯ãç§ãã¡ã«ãšã£ãŠå埩çãªããã»ã¹ã§ããã¢ãã«ãæ¹åããããã«ãç§åŠèšäºã®çµéšãšããŒã¿ããã®ç¬èªã®çµè«ã«åºã¥ããŠããã¡ã€ã³ã®å°é家ã®æšå¥šã«åºã¥ããŠæ°ããæ©èœãè¿œå ããããšããŠããŸãã
ããŒã ã«ãã£ãŠææ¡ããã仮説ããã¹ãããããšã¯ãæ°ããæ©èœãå°å ¥ããã®ã«åœ¹ç«ã¡ãŸãã
æåã®1ã€ã¯ãã¯ã©ã¹ã¿ãªã³ã°ã«åºã¥ããŠèå¥ãããæ©èœã§ãããå®éãå°ççãã©ã¡ãŒã¿ã«åºã¥ããŠããŒã¿ã»ããå ã®ã¯ã©ã¹ã¿ãŒãéžæããã¯ã©ã¹ã¿ãŒã«åºã¥ããŠä»ã®ç¹æ§ã®åºæ¬çãªçµ±èšãçæããã ãã§ããããã«ãããå質ããããã«åäžããŸããã
ã¯ã©ã¹ã¿ãŒã®éžæã«åºã¥ããŠæ©èœãäœæããããã»ã¹ã
ãŸããç颚çãšãªã¢ã«æµžãããšãã«çºæããå åãè¿œå ããŸããã环ç©ç³æ²¹çç£éã¯ãŠã§ã«ã®å¹Žéœ¢ïŒæïŒã«æ£èŠåããã环ç©æ³šå ¥ã¯ãŠã§ã«ã®å¹Žéœ¢ïŒæïŒã«æ£èŠåããããã©ã¡ãŒã¿ãŒã¯DupuisåŒã«å«ãŸããŠããŸããããããsklearnã®PolynomialFeaturesããæšæºã»ãããçæããŠããå質ã¯åäžããŸããã§ããã
3.
æ©èœã®éžææ©èœã®éžæãè€æ°åå®è¡ããŸããããã¡ã€ã³ãšãã¹ããŒããšäžç·ã«æåã§è¡ãå Žåãšãæšæºã®æ©èœéžææ¹æ³ã䜿çšããå Žåã®äž¡æ¹ã§ããæ°åç¹°ãè¿ããåŸãã¿ãŒã²ããã«åœ±é¿ãäžããªãããã€ãã®æ©èœãããŒã¿ããåé€ããããšã«ããŸããããã®ãããåãå質ãç¶æããªããããŒã¿ã»ããã®ãµã€ãºãçž®å°ããããšãã§ããã¢ãã«ã®äœæãå€§å¹ ã«é«éåããããšãã§ããŸããã
ãããŠä»ãåãåã£ãã¡ããªãã¯ã«ã€ããŠ...
ããåéã§ã¯ã以äžã®ã¢ãã«å質ææšãååŸããŸããã
æ²¹å§ç Žå£ã®çµæã¯ãäºæž¬ãããŠããªãå€ãã®å€éšèŠå ã«ãäŸåããããšã«æ³šæããŠãã ããããããã£ãŠãMAPEã0ã«æžããããšã«ã€ããŠè©±ãããšã¯ã§ããŸããã
çµè«
MLã䜿çšããæ°Žå§ç Žç ã®åè£äºæžã®éžæã¯ãããŒã¿ãšã³ãžãã¢ãããŒã¿ãµã³ã¿ãŒããã¡ã€ã³ãšãã¹ããŒãããããŒãžã£ãŒã®7人ãéãŸã£ãéå¿çãªãããžã§ã¯ãã§ããä»æ¥ããããžã§ã¯ãã¯å®éã«ç«ã¡äžãã®æºåãã§ããŠããããã§ã«äŒç€Ÿã®ããã€ãã®åäŒç€Ÿã§ãã¹ããããŠããŸãã
å瀟ã¯å®éšã«ãªãŒãã³ã§ããããããªã¹ãããçŽ20ã®ãŠã§ã«ãéžæãããç Žå£ãããŸãããéå§ç³æ²¹çç£çïŒMAPEïŒã®å®éã®å€ãšã®äºæž¬ã®åå·®ã¯çŽ10ïŒ ã§ããããããŠãããã¯éåžžã«è¯ãçµæã§ãïŒ
ç¡çŸã§ã¯ãããŸãããç¹ã«åæ段éã§ã¯ãææ¡ãããäºæžã®ããã€ãã¯äžé©åãªéžæè¢ã§ããããšãå€æããŸããã
質åãã³ã¡ã³ããæžããŠãã ãã-ç§ãã¡ã¯ãããã«çããããšããŸãã
ç§ãã¡ã®ããã°ã賌èªããŠãã ãããç§ãã¡ã¯ééããªãç§ãã¡ãæžãã§ããããã£ãšå€ãã®èå³æ·±ãã¢ã€ãã¢ããããžã§ã¯ããæã£ãŠããŸãïŒ