倧人ã«ãªã£ãŠãããã£ãšãç§ã¯ãšããžãŒããªã³ã¯ã飲ãã§ããŸããïŒããããä»ã¯çãããæ§è³ªãæã€é£²ã¿ç©ã«ã€ããŠè©±ããŠããã®ã§ã¯ãããŸããïŒã
ç§ã¯æ å ±æè¡ã®äžçã«ç¹ã«èå³ãæã£ãããšã¯ãªããçŽã®äžã§è¡åãæããããšããã»ãšãã©ã§ããŸããã ç§ã®ä»äºã®è©³çŽ°ãå°ãã§ãç解ããŠããã ãããã«ãç§ã¯ãããå¿ èŠãšããŸããã§ãããçŽ æŽããã話ãå ±æã§ããŸãã 以åãç§ã¯ååã« Excel ã¹ãã¬ããã·ãŒãã§ä»äºãããããã«é Œãã ããšããããŸãããã®ãšããå€åæ¥ã®ååãéããã®ã§ç§ã圌ãã®ãšããã«è¡ããšã圌ãã¯åº§ã£ãŠé»åãããããã¿ã³ã®ä»ããæ®éã®é»ãé»åã§ããŒã¿ãéèšããŠããŸããã ããŠããã®åŸã¯ã©ã®ãããªãã¥ãŒã©ã«ãããã¯ãŒã¯ã®è©±ãã§ããã®ã§ãããã? ãšããããšã§ãITã®äžçã«æµžãã«ããã£ãŠç¹å¥ãªåææ¡ä»¶ã¯ãããŸããã§ããã ãããã圌ããèšãããã«ããç§ãã¡ãããªããšããã¯è¯ãããšã§ãããšç§ã®å人ãã¡ã¯ãæ¡åŒµçŸå®ããã¥ãŒã©ã«ãããã¯ãŒã¯ãããã°ã©ãã³ã°èšèªïŒäž»ã«Pythonã«ã€ããŠïŒã«ã€ããŠç§ã®è³ãããããããŸããã
èšèã§èšããšãšãŠãç°¡åããã«èŠããŸããããç§ã¯ãã®éæ³ã®èžè¡ããã¹ã¿ãŒããŠãèªåã®æŽ»ååéã«å¿çšããŠã¿ãããšæããŸããã
ãã®èšäºã§ã¯ãPython ã®åºç€ããã¹ã¿ãŒããããšããç§ã®è©Šã¿ã¯çç¥ããUdacity ã®ç¡æ TensorFlow ã³ãŒã¹ã®ææ³ãå ±æããŸãã
å°å ¥
ãŸãã泚ç®ã«å€ããã®ã¯ããšãã«ã®ãŒæ¥ç㧠11 幎éå€åããããããããšãç¥ã£ãŠããŠãïŒè²¬ä»»ã«å¿ããŠïŒããã«å°ãã§ãã§ããããã«ãªããšãæ ¹æ¬çã«æ°ããããšãåŠã¶ããšã¯ãäžæ¹ã§ã¯å€§ããªç±æãåŒãèµ·ãããšããããšã§ãããããäžæ¹ã§ã¯ããé ã®äžã®æ¯è»ããèäœçãªçã¿ã«å€ãããŸãã
ç§ã¯ãŸã ããã°ã©ãã³ã°ãšæ©æ¢°åŠç¿ã®åºæ¬æŠå¿µããã¹ãŠå®å šã«ç解ããŠããããã§ã¯ãªãã®ã§ãããŸãå³ããå€æããªãã§ãã ããã ç§ã®èšäºããç§ã®ãããªãœãããŠã§ã¢éçºãšã¯çžé ã人ã ã«ãšã£ãŠèå³æ·±ã圹ç«ã€ããšãé¡ã£ãŠããŸãã
ã³ãŒã¹ã®æŠèŠã«é²ãåã«ãã³ãŒã¹ãåŠç¿ããã«ã¯å°ãªããšãæå°éã® Python ã®ç¥èãå¿ èŠã§ããããšãèšã£ãŠãããŸãã ãããŒçšã®æ¬ãäœåãèªãã§ã¿ãŠãã ãã (ç§ã Stepic ã®ã³ãŒã¹ãåè¬ãå§ããŸãããããŸã å®å šã«ã¯ãã¹ã¿ãŒããŠããŸãã)ã
TensorFlow ã³ãŒã¹èªäœã«ã¯è€éãªæ§æèŠçŽ ã¯å«ãŸããŸããããã©ã€ãã©ãªãã€ã³ããŒããããçç±ãé¢æ°ãã©ã®ããã«å®çŸ©ãããããããã³ã©ã€ãã©ãªã«äœãã眮æãããçç±ãç解ããå¿ èŠããããŸãã
TensorFlow ãš Udacity ã䜿çšããçç±
ç§ã®ãã¬ãŒãã³ã°ã®äž»ãªç®æšã¯ããã¥ãŒã©ã« ãããã¯ãŒã¯ã䜿çšããŠé»æ°èšåèŠçŽ ã®åçãèªèããããšããé¡æã§ããã
TensorFlow ãéžãã ã®ã¯ãå人ããèããããã§ãã ãããŠãç§ãç解ããŠããããã«ããã®ã³ãŒã¹ã¯éåžžã«äººæ°ããããŸãã
å
¬åŒããåŠã³å§ããŠã¿ã
ãããŠãXNUMX ã€ã®åé¡ã«ééããŸããã
- ææã¯ããããããããã®çš®é¡ãè±å¯ã§ãã ç»åèªèåé¡ã解決ããããã®å°ãªããšãå€ããå°ãªããå®å šãªå šäœåãäœæããããšã¯ãç§ã«ãšã£ãŠéåžžã«å°é£ã§ããã
- ç§ãå¿ èŠãšããèšäºã®ã»ãšãã©ã¯ãã·ã¢èªã«ç¿»èš³ãããŠããŸããã ããŸããŸãç§ã¯åäŸã®é ã«ãã€ãèªãåŠãã ã®ã§ãããä»ã§ã¯å€ãã®ãœé£ã®åäŸãã¡ãšåãããã«ããã€ãèªãè±èªãç¥ããŸããã ãã¡ããã倧人ã«ãªã£ãŠãããè±èªããã¹ã¿ãŒããããšåªåããŸããããçµµã«æãããããªçµæã«ãªããŸããã
å
¬åŒãŠã§ããµã€ãã調ã¹ãŠã¿ããšãããæšå¥šäºé
ãèŠã€ããŸãã
ç§ã®ç解ã§ã¯ãCoursera ã®ã³ãŒã¹ã¯ææã§ãããã³ãŒã¹ã¯
è¬åº§ã®å 容
ãã®ã³ãŒã¹ã¯9ã€ã®ã¬ãã¹ã³ã§æ§æãããŠããŸãã
äžçªæåã®ã»ã¯ã·ã§ã³ã¯å ¥éã§ãããåççã«ãªããããå¿ èŠãªã®ãã説æããŸãã
ã¬ãã¹ã³ #2 ãç§ã®ãæ°ã«å ¥ãã«ãªããŸããã ãšãŠãåããããããç§åŠã®çŽ æŽããããåãããŸããã ã€ãŸãããã®ã¬ãã¹ã³ã§ã¯ããã¥ãŒã©ã« ãããã¯ãŒã¯ã«é¢ããåºæ¬æ å ±ã«å ããŠãäœæè ã¯ã枩床ãè¯æ°ããææ°ã«å€æããåé¡ã解決ããããã«åå±€ãã¥ãŒã©ã« ãããã¯ãŒã¯ã䜿çšããæ¹æ³ã瀺ããŸãã
ããã¯ç¢ºãã«éåžžã«ããããããäŸã§ãã ç§ã¯ä»ãããã«åº§ã£ãŠãåæ§ã®åé¡ãã©ããã£ãŠèãåºããŠè§£æ±ºããããèããŠããŸãããããã¯é»æ°æåž«ã«éã£ã話ã§ãã
æ®å¿µãªãããç§ã¯ããã«å€±éããŠããŸããŸããããªãã¿ã®ãªãèšèªã§ç解ã§ããªãããšãåŠã¶ã®ã¯éåžžã«é£ããããã§ãã ç§ãæã£ãŠãããã®ã¯ãããã¬ã§èŠã€ãããã®ã§ãã
翻蚳ã¯é«å質ã§ãColab ããŒãããã¯ã翻蚳ãããŠããã®ã§ãåæãšç¿»èš³ã®äž¡æ¹ãèŠãŸããã
å®éãã¬ãã¹ã³ 3 ã¯ãå ¬åŒ TensorFlow ãã¥ãŒããªã¢ã«ã®è³æãæ¹å€ãããã®ã§ãã ãã®ãã¥ãŒããªã¢ã«ã§ã¯ãå€å±€ãã¥ãŒã©ã« ãããã¯ãŒã¯ã䜿çšããŠãè¡£æã®åç (ãã¡ãã·ã§ã³ MNIST ããŒã¿ã»ãã) ãåé¡ããæ¹æ³ãåŠç¿ããŸãã
ã¬ãã¹ã³ No. 4 ãã No. 7 ããã¥ãŒããªã¢ã«ãå¿çšãããã®ã§ãã ãã ãããããã¯æ£ããé 眮ãããŠãããããåŠç¿ã®é åºãèªåã§ç解ããå¿ èŠã¯ãããŸããã ãããã®ã¬ãã¹ã³ã§ã¯ãè¶ é«ç²ŸåºŠãã¥ãŒã©ã« ãããã¯ãŒã¯ããã¬ãŒãã³ã°ã®ç²ŸåºŠãé«ããæ¹æ³ãã¢ãã«ãä¿åããæ¹æ³ã«ã€ããŠç°¡åã«èª¬æããŸãã åæã«ãç»åå ã®ç«ãšç¬ãåé¡ããåé¡ãåæã«è§£æ±ºããŸãã
ã¬ãã¹ã³ No. 8 ã¯å®å šã«å¥ã®ã³ãŒã¹ã§ãå çãç°ãªããã³ãŒã¹èªäœãããªãå å®ããŠããŸãã ã¬ãã¹ã³ã¯æç³»åã«ã€ããŠã§ãã ãŸã èå³ããªãã®ã§æãã«ã¹ãã£ã³ããŠã¿ãŸããã
ããã¯ãTensorFlow lite ã®ç¡æã³ãŒã¹ãžã®æåŸ ã§ããã¬ãã¹ã³ #9 ã§çµäºããŸãã
奜ãã ã£ãããšãå«ãã ã£ãããš
é·æããå§ããŸãã
- ã³ãŒã¹ã¯ç¡æã§ã
- ãã®ã³ãŒã¹ã¯ TensorFlow 2 ã䜿çšããŠããŸããç§ãèŠãæç§æžãã€ã³ã¿ãŒãããäžã®äžéšã®ã³ãŒã¹ã¯ TensorFlow 1 ã䜿çšããŠããŸããã倧ããªéãããããã©ããã¯ããããŸããããçŸåšã®ããŒãžã§ã³ãåŠã¶ã®ã¯è¯ãããšã§ãã
- ãããªã®äžã®æåž«ã¯ãã£ãšããã人ã§ã¯ãããŸããïŒãã ãããã·ã¢èªçã§ã¯åäœã»ã©éœæ°ã«æ¬ãèªãã§ããŸããïŒã
- ã³ãŒã¹ã¯ããã»ã©æéã¯ããããŸãã
- ãã®ã³ãŒã¹ã§ã¯ãæ²ããæ°æã¡ã«ãªã£ããã絶æçãªæ°æã¡ã«ãªã£ããããããšã¯ãããŸããã ã³ãŒã¹å ã®ã¿ã¹ã¯ã¯ã·ã³ãã«ã§ãäœãäžæãªç¹ãããå Žåã¯ãæ£ãã解決çã瀺ããã³ããåžžã« Colab 圢åŒã§è¡šç€ºãããŸã (ãããŠãã¿ã¹ã¯ã®ããªãã®éšåãç§ã«ã¯ããããŸããã§ãã)ã
- äœãã€ã³ã¹ããŒã«ããå¿ èŠã¯ãªããã³ãŒã¹ã®ãã¹ãŠã®å®éšäœæ¥ã¯ãã©ãŠã¶ãŒã§å®è¡ã§ããŸãã
ä»çæïŒ
- å®è³ªçã«å¶åŸ¡ææã¯ãããŸããã ãã¹ããã¿ã¹ã¯ã¯ãªããã³ãŒã¹ã®ç¿ç床ã確èªãããã®ã¯äœããããŸãã
- ç§ã®ã¡ã¢åž³ã®ãã¹ãŠãæ£åžžã«æ©èœããããã§ã¯ãããŸããã è±èªã®å ã®ã³ãŒã¹ã® XNUMX çªç®ã®ã¬ãã¹ã³ã§ Colab ããšã©ãŒãã¹ããŒããŠãããšæããŸãããã©ã察åŠããã°ããã®ãããããŸããã§ãã
- ããœã³ã³ã ãã§èŠèŽã§ããã®ã§äŸ¿å©ã§ãã ããç解ã§ããŠããªãã£ãã®ãããããŸããããã¹ããŒããã©ã³ã§ã¯ Udacity ã¢ããªãèŠã€ãããŸããã§ããã ãŸãããµã€ãã®ã¢ãã€ã«çã¯å¿çæ§ãæªããç»é¢é åã®ã»ãŒå šäœãããã²ãŒã·ã§ã³ ã¡ãã¥ãŒã§å ããããŠããŸãããã¡ã€ã³ ã³ã³ãã³ãã衚瀺ããã«ã¯ã衚瀺é åãè¶ ããŠå³ã«ã¹ã¯ããŒã«ããå¿ èŠããããŸãã ãŸããæºåž¯é»è©±ã§ã¯åç»ãèŠèŽããããšã¯ã§ããŸããã 6 ã€ã³ã匷ã®ç»é¢ã«ã¯å®éã«ã¯äœãèŠããŸããã
- ã³ãŒã¹å ã®ããã€ãã®å 容ã¯äœåºŠãåã¿ç ãããŸãããåæã«ãç³ã¿èŸŒã¿ãããã¯ãŒã¯èªäœã§æ¬åœã«å¿ èŠãªãã®ã¯ã³ãŒã¹å ã§åã¿ç ãããŸããã ããã€ãã®æŒç¿ã®å šäœçãªç®ç (ããšãã°ãMax Pooling ã®ç®çãªã©) ããŸã ç解ã§ããŸããã§ããã
ãµããªãŒ
確ãã«ããªãã¯ãã§ã«å¥è·¡ãèµ·ãããªãã£ãããšããã§ã«æšæž¬ããŠããŸããã ãããŠããã®çãã³ãŒã¹ãå®äºãããšããŠãããã¥ãŒã©ã« ãããã¯ãŒã¯ãã©ã®ããã«æ©èœããããçã«ç解ããããšã¯äžå¯èœã§ãã
ãã¡ããããã®åŸãééè£ çœ®ã®ã¹ã€ããããã¿ã³ã®åçã®åé¡ã«é¢ããåé¡ãèªåã§è§£æ±ºããããšã¯ã§ããŸããã§ããã
ããããå šäœçã«ã¯ãã®ã³ãŒã¹ã¯åœ¹ã«ç«ã¡ãŸãã TensorFlow ã§äœãã§ãããããããŠæ¬¡ã«ã©ã®ãããªæ¹åã«é²ãã¹ããã瀺ããŸãã
ãŸã㯠Python ã®åºç€ãåŠã³ããã¥ãŒã©ã« ãããã¯ãŒã¯ã®ä»çµã¿ã«ã€ããŠãã·ã¢èªã®æ¬ãèªãã§ãããTensorFlow ã«ææŠããå¿ èŠããããšæããŸãã
æåŸã«ãHabr ã«é¢ããæåã®èšäºãæžãããç§ãåŸæŒãããèšäºã®äœè£ãæŽããŠãããå人ãã¡ã«æè¬ã®æãè¡šããããšæããŸãã
PS ããªãã®ã³ã¡ã³ãã建èšçãªæ¹å€ãåãã§æèŠãããŠããã ããŸãã
åºæïŒ habr.com