5-7 ����������

Old Chinese version

¨©¦¡¤ÀÃþªk¡]Bayes classifier¡^¤D¬O®Ú¾Ú¨©¤ó©w²z¡]Bayes' theorem¡^¬°°ò¦¡A¥Î¥H§PÂ_¥¼ª¾Ãþ§Oªº¸ê®ÆÀ³¸Ó³Ì±µªñ­þ¤@­ÓÃþ§O¡C¾ã­Ó¨©¦¡¤ÀÃþªkªº¥Ø¼Ð¬O§Æ±æ¯à³z¹L¾÷²v²Î­pªº¤ÀªR¡A¹F¨ì³Ì¤p»~®tªº¤@ºØ¤ÀÃþ¤è¦¡¡C

°²³]²{¦b¦s¦b¬Y­Ó¯S¼x­Èx¤Î¬Y­ÓÃþ§O C¡AP(x) ªí¥Ü¸Ó¯S¼x­È¥X²{ªº¦ô´ú¾÷²v¡AP(C) ªí¥Ü¥ô·NÂǥѶüƨú¥Xªº¯S¼x­È«ê¥©¸¨©óÃþ§O C ªº¾÷²v¡A§Ú­Ì±N¤§ºÙ¬°¨Æ«e¾÷²v¡]prior probability¡^¡A«h®Ú¾Ú±ø¥ó¾÷²v¡]conditional probability¡^¡A¨©¦¡©w²z¥i¥Hªí¥Ü¬°¡G

P(C|x) = P(C¡äx)/P(x) = P(C)P(x|C)/P(x)
¨ä¤¤¡AP(C|x) ªí¥Ü x ¸Ó¯S¼x­È¥X²{®É¡A¤S«ê¥©¸¨©óÃþ§OCªº¾÷²v¡A§Ú­Ì±N¥LºÙ¬°¨Æ«á¾÷²v¡]posterior probability¡^¡F¦Ü©ó P(x|C) «hªí¥Ü¸¨©óÃþ§O C ¤¤ªº¸ê®ÆÂI¤¤¡A¤S«ê¥©µo¥Í¯S¼x­È¬° x ªº¾÷²v¡C

°²³]¸ÓªÅ¶¡¤¤¥i¯à¥X²{ªºÃþ§OÁ`¦@¦³ k ­Ó {C1, C2, ¡K, Ck}¡A¥B¨C­ÓÃþ§O©¼¦¹§¡¤¬¥¸¡A«h§Ú­Ì¥i¥H±o¨ì¤U¦C¤èµ{¦¡¡G
P(x)=P(x¡äC1) + P(x¡äC2) + ... + P(x¡äCk)
=P(C1)P(x|C1) + P(C2)P(x|C2) + ... + P(Ck)P(x|Ck)
½Ð°Ñ¦Ò¤U¦C¿W¥ß¨Æ¥ó¾÷²v¤À§G¥Ü·N¹Ï¡G

¥Ñ«e­z¤èµ{¦¡¡A§Ú­Ì¥i¥H±oª¾¡G
P(Ci|x) = P(Ci)P(x|Ci)/P(x)
­Y¨Ï¥Î¤W­z¤èµ{¦¡¡A§Ú­Ì¥i¥H±o¨ìÀ³¥Î©ók­ÓÃþ§Oªº¨©¦¡©w²z¡G
P(Ci|x) = P(Ci)P(x|Ci)
¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w
P(C1)P(x|C1) + P(C2)P(x|C2) + ... + P(Ck)P(x|Ck)
·í§Ú­Ì­n§PÂ_¬Y¯S¼x­Èx¨s³ºÄÝ©ó­þ¤@­ÓÃþ§O®É¡A«h§Ú­Ì¶È»Ý¦ôºâÃþ§OCi»PÃþ§OCj¤§¶¡ªº¬Û¦ü²v¡]likelihood ratio¡^R¡G
R = P(Ci|x) = P(Ci)P(x|Ci)
¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w
P(Cj|x)P(Cj)P(x|Cj)
°²¦p R > 1¡Aªí¥Ü x ¤ñ¸û°¾¦VÃþ§O Ci¡F¤Ï¤§¡A°²¦p R < 1¡Aªí¥Ü x ¤ñ¸û°¾¦VÃþ§O Cj¡C

¦b¹ê»Ú¹Bºâ®É¡AP(Ci) ¬O²Ä i Ãþ¸ê®Æ¦ûÁ`¼Ë¥»¸ê®Æªº¾÷²v¡A¦Ó P(x|Ci) «h¬O¥Ñ²ÄiÃþ¸ê®ÆÂI©Ò¦ô´ú¥X¨Óªº¤@­Ó¾÷²v±K«×¨ç¼Æ¡]¨Ò¦p°ª´µ¤À§G¡^¡C

§Ú­Ì¥i¥H±N¨©¦¡©w²z¦A©¹¤U±Àºt¡A°²¦p²{¦b§PÂ_ªº±ø¥ó¤£¤î¤@­Ó¯S¼x­È¡A¦Ó¬O¤@²Õ©¼¦¹¤¬¬Û¿W¥ßªº¯S¼x­È (x1, x2, ¡K, xd)¡A«h·íµ¹©w¬Y­ÓÃþ§O Ci ®É¡A¨ä±ø¥ó¾÷²v¥i¥Hªí¥Ü¬°¡G

P(x1, x2, ¡K, xd|Ci) = P(x1|Ci)P(x2|Ci) ... P(xd|Ci)
¦pªG±N¤èµ{¦¡¡]3-2.6¡^ªºµ²ªG¥N¤J¤èµ{¦¡¡]3-2.4¡^¤¤¡A«h§Ú­Ì¥i¥H±o¨ìk­ÓÃþ§O¤¤¡A¥]§td­Ó¯S¼x­Èªº¨©¦¡©w²z¡G
P(Ci|x1, x2, ¡K, xd) = P(Ci) P(x1|Ci)P(x2|Ci) ... P(xd|Ci)
¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w¢w
Si=1k P(Ci) P(x1|Ci)P(x2|Ci) ... P(xd|Ci)

Data Clustering and Pattern Recognition (¸ê®Æ¤À¸s»P¼Ë¦¡¿ë»{)