±à¼ÍƼö: |
±¾ÎÄÖУ¬±¾ÎÄÖ÷Òª½éÉÜÁËRandom ForestË㷨ģÐÍ£¬RF½«baggingÓëdecision
tree½áºÏÆðÀ´£¬Í¨¹ý°ÑÖÚ¶àµÄ¾ö²ßÊ÷×é½øÐÐ×éºÏ£¬¹¹³ÉÉÁÖµÄÐÎʽ£¬ÀûÓÃͶƱ»úÖÆÈÃG±íÏÖ×î¼Ñ£¬·ÖÀàÄ£Ð͸üÎȶ¨¡£
±¾ÎÄÀ´×ÔÓÚ¸öÈËͼÊé¹Ý£¬ÓÉ»ðÁú¹ûÈí¼þAlice±à¼¡¢ÍƼö¡£ |
|
1¡¢Random Forest Algorithm
Ê×ÏÈÎÒÃÇÀ´¸´Ï°Ò»ÏÂ֮ǰ½éÉܹýµÄÁ½¸ö»úÆ÷ѧϰģÐÍ£ºBaggingºÍDecision Tree¡£BaggingÊÇͨ¹ýbootstrapµÄ·½Ê½£¬´ÓÔʼµÄÊý¾Ý¼¯DÖеõ½ÐµÄD^£»È»ºóÔÙʹÓÃһЩbase
algorithm¶Ôÿ¸öD^¶¼µÃµ½ÏàÓ¦µÄgt£»×îºó½«ËùÓеÄgtͨ¹ýͶƱuniformµÄÐÎʽ×éºÏ³ÉÒ»¸öG£¬G¼´ÎªÎÒÃÇ×îÖյõ½µÄÄ£ÐÍ¡£Decision
TreeÊÇͨ¹ýµÝ¹éÐÎʽ£¬ÀûÓ÷ÖÖ§Ìõ¼þ£¬½«ÔʼÊý¾Ý¼¯DÇиî³ÉÒ»¸ö¸ö×ÓÊ÷½á¹¹£¬³¤³ÉÒ»¿ÃÍêÕûµÄÊ÷Ðνṹ¡£Decision
Tree×îÖյõ½µÄG(x)ÊÇÓÉÏàÓ¦µÄ·ÖÖ§Ìõ¼þb(x)ºÍ·ÖÖ§Ê÷Gc(x)µÝ¹é×é³É¡£

BaggingºÍDecison TreeËã·¨¸÷×ÔÓÐÒ»¸öºÜÖØÒªµÄÌØµã¡£Bagging¾ßÓмõÉÙ²»Í¬gtµÄ·½²îvarianceµÄÌØµã¡£ÕâÊÇÒòΪBagging²ÉÓÃͶƱµÄÐÎʽ£¬½«ËùÓÐgt
uniform½áºÏÆðÀ´£¬Æðµ½ÁËÇ󯽾ùµÄ×÷Ó㬴Ӷø½µµÍvariance¡£¶øDecision Tree¾ßÓÐÔö´ó²»Í¬gtµÄ·½²îvarianceµÄÌØµã¡£ÕâÊÇÒòΪDecision
Treeÿ´ÎÇиîµÄ·½Ê½²»Í¬£¬¶øÇÒ·ÖÖ§°üº¬µÄÑù±¾ÊýÔÚÖð½¥¼õÉÙ£¬ËùÒÔËü¶Ô²»Í¬µÄ×ÊÁÏD»á±È½ÏÃô¸ÐһЩ£¬´Ó¶ø²»Í¬µÄD»áµÃµ½±È½Ï´óµÄvariance¡£
ËùÒÔ˵£¬BaggingÄܼõСvariance£¬¶øDecision TreeÄÜÔö´óvariance¡£Èç¹û°ÑÁ½Õß½áºÏÆðÀ´£¬ÄÜ·ñ·¢»Ó¸÷×ÔµÄÓÅÊÆ£¬Æðµ½ÓÅÊÆ»¥²¹µÄ×÷ÓÃÄØ£¿Õâ¾ÍÊÇÎÒÃǽÓÏÂÀ´½«ÒªÌÖÂÛµÄaggregation
of aggregation£¬¼´Ê¹ÓÃBaggingµÄ·½Ê½°ÑÖÚ¶àµÄDecision Tree½øÐÐuniform½áºÏÆðÀ´¡£ÕâÖÖËã·¨¾Í½Ð×öËæ»úÉÁÖ£¨Random
Forest£©£¬Ëü½«ÍêÈ«³¤³ÉµÄC&RT¾ö²ßÊ÷ͨ¹ýbaggingµÄÐÎʽ½áºÏÆðÀ´£¬×îÖյõ½Ò»¸öÅÓ´óµÄ¾ö²ßÄ£ÐÍ¡£

Random ForestËã·¨Á÷³ÌͼÈçÏÂËùʾ£º

Random ForestËã·¨µÄÓŵãÖ÷ÒªÓÐÈý¸ö¡£µÚÒ»£¬²»Í¬¾ö²ßÊ÷¿ÉÒÔÓɲ»Í¬Ö÷»ú²¢ÐÐѵÁ·Éú³É£¬Ð§Âʺܸߣ»µÚ¶þ£¬Ëæ»úÉÁÖËã·¨¼Ì³ÐÁËC&RTµÄÓŵ㣻µÚÈý£¬½«ËùÓеľö²ßÊ÷ͨ¹ýbaggingµÄÐÎʽ½áºÏÆðÀ´£¬±ÜÃâÁ˵¥¸ö¾ö²ßÊ÷Ôì³É¹ýÄâºÏµÄÎÊÌâ¡£

ÒÔÉÏÊÇ»ù±¾µÄRandom ForestËã·¨£¬ÎÒÃÇÔÙÀ´¿´Ò»ÏÂÈçºÎÈÃRandom ForestÖоö²ßÊ÷µÄ½á¹¹¸üÓжàÑùÐÔ¡£BaggingÖУ¬Í¨¹ýbootstrapµÄ·½·¨µÃµ½²»Í¬ÓÚDµÄD¡¯£¬Ê¹ÓÃÕâÐ©Ëæ»ú³éÈ¡µÄ×ÊÁϵõ½²»Í¬µÄgt¡£³ýÁËËæ»ú³éÈ¡×ÊÁÏ»ñµÃ²»Í¬gtµÄ·½Ê½Ö®Í⣬»¹ÓÐÁíÍâÒ»ÖÖ·½·¨£¬¾ÍÊÇËæ»ú³éȡһ²¿·ÖÌØÕ÷¡£ÀýÈ磬ÔÀ´ÓÐ100¸öÌØÕ÷£¬ÏÖÔÚÖ»´ÓÖÐËæ»úѡȡ30¸öÀ´¹¹³É¾ö²ßÊ÷£¬ÄÇôÿһÂֵõ½µÄÊ÷¶¼Óɲ»Í¬µÄ30¸öÌØÕ÷¹¹³É£¬Ã¿¿ÃÊ÷¶¼²»Ò»Ñù¡£¼ÙÉèÔÀ´Ñù±¾Î¬¶ÈÊÇd£¬ÔòֻѡÔñÆäÖеÄd¡¯£¨d¡¯Ð¡ÓÚd£©¸öά¶ÈÀ´½¨Á¢¾ö²ßÊ÷½á¹¹¡£ÕâÀàËÆÊÇÒ»ÖÖ´Ódάµ½d¡¯Î¬µÄÌØÕ÷ת»»£¬Ï൱ÓÚÊÇ´Ó¸ßάµ½µÍάµÄͶӰ£¬Ò²¾ÍÊÇ˵d¡¯Î¬z¿Õ¼äÆäʵ¾ÍÊÇdάx¿Õ¼äµÄÒ»¸öËæ»ú×ӿռ䣨subspace£©¡£Í¨³£Çé¿öÏ£¬d¡¯Ô¶Ð¡ÓÚd£¬´Ó¶ø±£Ö¤Ëã·¨¸üÓÐЧÂÊ¡£Random
ForestËã·¨µÄ×÷Õß½¨ÒéÔÚ¹¹½¨C&RTÿ¸ö·ÖÖ§b(x)µÄʱºò£¬¶¼¿ÉÒÔÖØÐÂÑ¡Ôñ×ÓÌØÕ÷À´ÑµÁ·£¬´Ó¶øµÃµ½¸ü¾ßÓжàÑùÐԵľö²ßÊ÷¡£

ËùÒÔ˵£¬ÕâÖÖÔöÇ¿µÄRandom ForestËã·¨Ôö¼ÓÁËrandom-subspace¡£

ÉÏÃæÎÒÃǽ²µÄÊÇËæ»ú³éÈ¡ÌØÕ÷£¬³ý´ËÖ®Í⣬»¹¿ÉÒÔ½«ÏÖÓеÄÌØÕ÷x£¬Í¨¹ýÊý×ép½øÐÐÏßÐÔ×éºÏ£¬À´±£³Ö¶àÑùÐÔ£º

ÕâÖÖ·½·¨Ê¹Ã¿´Î·ÖÖ§µÃµ½µÄ²»ÔÙÊǵ¥Ò»µÄ×ÓÌØÕ÷¼¯ºÏ£¬¶øÊÇ×ÓÌØÕ÷µÄÏßÐÔ×éºÏ£¨È¨Öز»Îª1£©¡£ºÃ±ÈÔÚ¶þÎ¬Æ½ÃæÉϲ»Ö¹µÃµ½Ë®Æ½Ïߺʹ¹Ö±Ïߣ¬Ò²Äܵõ½¸÷ÖÖбÏß¡£ÕâÖÖ×ö·¨Ê¹×ÓÌØÕ÷Ñ¡Ôñ¸ü¼Ó¶àÑùÐÔ¡£ÖµµÃ×¢ÒâµÄÊÇ£¬²»Í¬·ÖÖ§iϵÄpiÊDz»Í¬µÄ£¬¶øÇÒÏòÁ¿piÖд󲿷ÖÔªËØÎªÁ㣬ÒòΪÎÒÃÇÑ¡ÔñµÄÖ»ÊÇÒ»²¿·ÖÌØÕ÷£¬ÕâÊÇÒ»ÖÖµÍάӳÉä¡£

ËùÒÔ£¬ÕâÀïµÄRandom ForestËã·¨ÓÖÓÐÔöÇ¿£¬ÓÉÔÀ´µÄrandom-subspace±ä³ÉÁËrandom-combination¡£Ë³±ãÌáһϣ¬ÕâÀïµÄrandom-combinationÀàËÆÓÚperceptronÄ£ÐÍ¡£

2¡¢Out-Of-Bag Estimate
ÉÏÒ»²¿·ÖÎÒÃÇÒѾ½éÉÜÁËRandom ForestËã·¨£¬¶øRandom
ForestËã·¨ÖØÒªµÄÒ»µã¾ÍÊÇBagging¡£½ÓÏÂÀ´½«¼ÌÐøÌ½ÌÖbaggingÖеÄbootstrap»úÖÆµ½µ×Ô̺¬ÁËÄÄЩ¿ÉÒÔΪÎÒÃÇËùÓõĶ«Î÷¡£
ͨ¹ýbootstrapµÃµ½ÐµÄÑù±¾¼¯D¡¯£¬ÔÙÓÉD¡¯ÑµÁ·²»Í¬µÄgt¡£ÎÒÃÇÖªµÀD¡¯Öаüº¬ÁËÔÑù±¾¼¯DÖеÄһЩÑù±¾£¬µ«Ò²ÓÐЩÑù±¾Ã»Óк¸Ç½øÈ¥¡£ÈçϱíËùʾ£¬²»Í¬µÄgtÏ£¬ºìÉ«µÄ±íʾûÓÐÕâЩÑù±¾¡£ÀýÈç¶Ôg1À´Ëµ£¬(x2,y2)ºÍ(x3,y4)ûÓаüº¬½øÈ¥£¬¶Ôg2À´Ëµ£¬(x1,y1)ºÍ(x2,y2)ûÓаüº¬½øÈ¥£¬µÈµÈ¡£Ã¿¸ögtÖУ¬ºìÉ«±íʾµÄÑù±¾±»³ÆÎªout-of-bag(OOB)
example¡£

Ê×ÏÈ£¬ÎÒÃÇÀ´¼ÆËãOOBÑù±¾µ½µ×ÓжàÉÙ¡£¼ÙÉèbootstrapµÄÊýÁ¿N¡¯=N£¬ÄÇôij¸öÑù±¾(xn,yn)ÊÇOOBµÄ¸ÅÂÊÊÇ£º

ÆäÖУ¬eÊÇ×ÔÈ»¶ÔÊý£¬NÊÇÔÑù±¾¼¯µÄÊýÁ¿¡£ÓÉÉÏÊöÍÆµ¼¿ÉµÃ£¬Ã¿¸ögtÖУ¬OOBÊýÄ¿´óÔ¼ÊÇN/e£¬¼´´óÔ¼ÓÐÈý·ÖÖ®Ò»µÄÑù±¾Ã»ÓÐÔÚbootstrapÖб»³éµ½¡£
È»ºó£¬ÎÒÃǽ«OOBÓë֮ǰ½éÉܵÄValidation½øÐжԱȣº




3¡¢Feature Selection
Èç¹ûÑù±¾×ÊÁÏÌØÕ÷¹ý¶à£¬¼ÙÈçÓÐ10000¸öÌØÕ÷£¬¶øÎÒÃÇÖ»Ïë´ÓÖÐѡȡ300¸öÌØÕ÷£¬Õâʱºò¾ÍÐèÒªÉáÆú²¿·ÖÌØÕ÷¡£Í¨³£À´Ëµ£¬ÐèÒªÒÆ³ýµÄÌØÕ÷·ÖΪÁ½ÀࣺһÀàÊÇÈßÓàÌØÕ÷£¬¼´ÌØÕ÷³öÏÖÖØ¸´£¬ÀýÈç¡°ÄêÁ䡱ºÍ¡°ÉúÈÕ¡±£»ÁíÒ»ÀàÊDz»Ïà¹ØÌØÕ÷£¬ÀýÈç¼²²¡Ô¤²âµÄʱºòÒýÈëµÄ¡°±£ÏÕ×´¿ö¡±¡£ÕâÖÖ´ÓdÎ¬ÌØÕ÷µ½d¡¯Î¬ÌØÕ÷µÄsubset-transform
¦µ(x)³ÆÎªFeature Selection£¬×îÖÕʹÓÃÕâЩd¡¯Î¬µÄÌØÕ÷½øÐÐÄ£ÐÍѵÁ·¡£

ÌØÕ÷Ñ¡ÔñµÄÓŵãÊÇ£º
1.Ìá¸ßЧÂÊ£¬ÌØÕ÷Ô½ÉÙ£¬Ä£ÐÍÔ½¼òµ¥
2.ÕýÔò»¯£¬·ÀÖ¹ÌØÕ÷¹ý¶à³öÏÖ¹ýÄâºÏ
3.È¥³ýÎÞ¹ØÌØÕ÷£¬±£ÁôÏà¹ØÐÔ´óµÄÌØÕ÷£¬½âÊÍÐÔÇ¿
ͬʱ£¬ÌØÕ÷Ñ¡ÔñµÄȱµãÊÇ£º
1.É¸Ñ¡ÌØÕ÷µÄ¼ÆËãÁ¿½Ï´ó
2.²»Í¬ÌØÕ÷×éºÏ£¬Ò²ÈÝÒ×·¢Éú¹ýÄâºÏ
3.ÈÝÒ×Ñ¡µ½ÎÞ¹ØÌØÕ÷£¬½âÊÍÐÔ²î

ÖµµÃÒ»ÌáµÄÊÇ£¬ÔÚdecision treeÖУ¬ÎÒÃÇʹÓõÄdecision
stumpÇиʽҲÊÇÒ»ÖÖfeature selection¡£
ÄÇô£¬ÈçºÎ¶ÔÐí¶àÎ¬ÌØÕ÷½øÐÐÉ¸Ñ¡ÄØ£¿ÎÒÃÇ¿ÉÒÔͨ¹ý¼ÆËã³öÿ¸öÌØÕ÷µÄÖØÒªÐÔ£¨¼´È¨ÖØ£©£¬È»ºóÔÙ¸ù¾ÝÖØÒªÐÔµÄÅÅÐò½øÐÐÑ¡Ôñ¼´¿É¡£

ÕâÖÖ·½·¨ÔÚÏßÐÔÄ£ÐÍÖбȽÏÈÝÒ×¼ÆËã¡£ÒòΪÏßÐÔÄ£Ð͵ÄscoreÊÇÓÉÿ¸öÌØÕ÷¾¹ý¼ÓȨÇóºÍ¶øµÃµ½µÄ£¬¶ø¼ÓȨϵÊýµÄ¾ø¶ÔÖµ|wi|ÕýºÃ´ú±íÁ˶ÔÓ¦ÌØÕ÷xiµÄÖØÒªÐÔΪ¶àÉÙ¡£|wi|Ô½´ó£¬±íʾ¶ÔÓ¦ÌØÕ÷xiÔ½ÖØÒª£¬Ôò¸ÃÌØÕ÷Ó¦¸Ã±»Ñ¡Ôñ¡£wµÄÖµ¿ÉÒÔͨ¹ý¶ÔÒÑÓеÄÊý¾Ý¼¯(xi,yi)½¨Á¢ÏßÐÔÄ£ÐͶøµÃµ½¡£

È»¶ø£¬¶ÔÓÚ·ÇÏßÐÔÄ£ÐÍÀ´Ëµ£¬ÒòΪ²»Í¬ÌØÕ÷¿ÉÄÜÊÇ·ÇÏßÐÔ½»²æÔÚÒ»ÆðµÄ£¬ËùÒÔ¼ÆËãÿ¸öÌØÕ÷µÄÖØÒªÐԾͱäµÃ±È½Ï¸´ÔÓºÍÀ§ÄÑ¡£ÀýÈ磬Random
Forest¾ÍÊÇÒ»¸ö·ÇÏßÐÔÄ£ÐÍ£¬½ÓÏÂÀ´£¬ÎÒÃǽ«ÌÖÂÛÈçºÎÔÚRFϽøÐÐÌØÕ÷Ñ¡Ôñ¡£
RFÖУ¬ÌØÕ÷Ñ¡ÔñµÄºËÐÄ˼ÏëÊÇrandom test¡£random
testµÄ×ö·¨ÊǶÔÓÚij¸öÌØÕ÷£¬Èç¹ûÓÃÁíÍâÒ»¸öËæ»úÖµÌæ´úËüÖ®ºóµÄ±íÏÖ±È֮ǰ¸ü²î£¬Ôò±íÃ÷¸ÃÌØÕ÷±È½ÏÖØÒª£¬ËùÕ¼µÄÈ¨ÖØÓ¦¸Ã½Ï´ó£¬²»ÄÜÓÃÒ»¸öËæ»úÖµÌæ´ú¡£Ïà·´£¬Èç¹ûËæ»úÖµÌæ´úºóµÄ±íÏÖûÓÐÌ«´ó²î±ð£¬Ôò±íÃ÷¸ÃÌØÕ÷²»ÄÇÃ´ÖØÒª£¬¿ÉÓпÉÎÞ¡£ËùÒÔ£¬Í¨¹ý±È½ÏÄ³ÌØÕ÷±»Ëæ»úÖµÌæ´úǰºóµÄ±íÏÖ£¬¾ÍÄÜÍÆ¶Ï³ö¸ÃÌØÕ÷µÄÈ¨ÖØºÍÖØÒªÐÔ¡£
ÄÇôrandom testÖеÄËæ»úÖµÈçºÎÑ¡ÔñÄØ£¿Í¨³£ÓÐÁ½ÖÖ·½·¨£ºÒ»ÊÇʹÓÃuniform»òÕßgaussian³éÈ¡Ëæ»úÖµÌæ»»ÔÌØÕ÷£»Ò»ÊÇͨ¹ýpermutationµÄ·½Ê½½«ÔÀ´µÄËùÓÐN¸öÑù±¾µÄµÚi¸öÌØÕ÷ÖµÖØÐ´òÂÒ·Ö²¼£¨Ï൱ÓÚÖØÐÂÏ´ÅÆ£©¡£±È½Ï¶øÑÔ£¬µÚ¶þÖÖ·½·¨¸ü¼Ó¿ÆÑ§£¬±£Ö¤ÁËÌØÕ÷Ìæ´úÖµÓëÔÌØÕ÷µÄ·Ö²¼ÊǽüËÆµÄ£¨Ö»ÊÇÖØÐÂÏ´ÅÆ¶øÒÑ£©¡£ÕâÖÖ·½·¨½Ð×öpermutation
test£¨Ëæ»úÅÅÐò²âÊÔ£©£¬¼´ÔÚ¼ÆËãµÚi¸öÌØÕ÷µÄÖØÒªÐÔµÄʱºò£¬½«N¸öÑù±¾µÄµÚi¸öÌØÕ÷ÖØÐÂÏ´ÅÆ£¬È»ºó±È½ÏDºÍD(p)±íÏֵIJîÒìÐÔ¡£Èç¹û²îÒìºÜ´ó£¬Ôò±íÃ÷µÚi¸öÌØÕ÷ÊÇÖØÒªµÄ¡£


4¡¢Random Forest in Action
×îºó£¬ÎÒÃÇͨ¹ýʵ¼ÊµÄÀý×ÓÀ´¿´Ò»ÏÂRFµÄÌØµã¡£Ê×ÏÈ£¬ÈÔÈ»ÊÇÒ»¸ö¶þÔª·ÖÀàµÄÀý×Ó¡£ÈçÏÂͼËùʾ£¬×ó±ßÊÇÒ»¸öC&RTÊ÷ûÓÐʹÓÃbootstrapµÃµ½µÄÄ£ÐÍ·ÖÀàЧ¹û£¬ÆäÖв»Í¬ÌØÕ÷Ö®¼ä½øÐÐÁËËæ»ú×éºÏ£¬ËùÒÔÓÐбÏß×÷Ϊ·ÖÀàÏߣ»ÖмäÊÇÓÉbootstrap£¨N¡¯=N/2£©ºóÉú³ÉµÄÒ»¿Ã¾ö²ßÊ÷×é³ÉµÄËæ»úÉÁÖ£¬Í¼ÖмӴֵĵã±íʾ±»bootstrapÑ¡Öеĵ㣻ÓÒ±ßÊǽ«Ò»¿Ã¾ö²ßÊ÷½øÐÐbaggingºóµÄ·ÖÀàÄ£ÐÍ£¬Ð§¹ûÓëÖмäͼÊÇÒ»ÑùµÄ£¬¶¼ÊÇÒ»¿ÃÊ÷¡£

µ±t=100£¬¼´Ñ¡ÔñÁË100¿ÃÊ÷ʱ£¬ÖмäµÄÄ£ÐÍÊǵÚ100¿Ã¾ö²ßÊ÷¹¹³ÉµÄ£¬»¹ÊÇÖ»ÓÐÒ»¿ÃÊ÷£»ÓұߵÄÄ£ÐÍÊÇÓÉ100¿Ã¾ö²ßÊ÷baggingÆðÀ´µÄ£¬ÈçÏÂͼËùʾ£º

µ±t=500ʱ£º

µ±t=1000ʱ£º

Ëæ×ÅÊ÷ľ¸öÊýµÄÔö¼Ó£¬ÎÒÃÇ·¢ÏÖ£¬·Ö½çÏßÔ½À´Ô½¹â»¬¶øÇҵõ½ÁËlarge-margin-like boundary£¬ÀàËÆÓÚSVMÒ»ÑùµÄЧ¹û¡£Ò²¾ÍÊÇ˵£¬Ê÷ľԽ¶à£¬·ÖÀàÆ÷µÄÖÃÐÅÇø¼äÔ½´ó¡£
È»ºó£¬ÎÒÃÇÔÙÀ´¿´Ò»¸ö±È½Ï¸´ÔÓµÄÀý×Ó£¬¶þÎ¬Æ½ÃæÉÏ·Ö²¼×ÅÐí¶àÀëÉ¢µã£¬·Ö½çÏßÐÎÈçsinº¯Êý¡£µ±Ö»ÓÐÒ»¿ÃÊ÷µÄʱºò£¨t=1£©£¬ÏÂͼ×ó±ß±íʾµ¥Ò»Ê÷×é³ÉµÄRF£¬Óұ߱íʾËùÓÐÊ÷bagging×éºÏÆðÀ´¹¹³ÉµÄRF¡£ÒòΪֻÓÐÒ»¿ÃÊ÷£¬ËùÒÔ×óÓÒÁ½±ßЧ¹ûÒ»Ö¡£

µ±t=6ʱ£º

µ±t=21ʱ£º

¿ÉÒÔ¿´µ½£¬µ±RFÓÉ21¿ÃÊ÷¹¹³ÉµÄʱºò£¬·Ö½çÏß¾Í±È½ÏÆ½»¬ÁË£¬¶øÇÒËüµÄ±ß½ç±Èµ¥Ò»Ê÷¹¹³ÉµÄRFÒªrobustµÃ¶à£¬¸ü¼Óƽ»¬ºÍÎȶ¨¡£
×îºó£¬»ùÓÚÉÏÃæµÄÀý×Ó£¬ÔÙÈÃÎÊÌ⸴ÔÓÒ»µã£ºÔÚÆ½ÃæÉÏÌí¼ÓÒ»Ð©Ëæ»úÔëÉù¡£µ±t=1ʱ£¬ÈçÏÂͼËùʾ£º

µ±t=6ʱ£º

µ±t=21ʱ£º

´ÓÉÏͼÖУ¬ÎÒÃÇ·¢ÏÖ21¿ÃÊ÷µÄʱºò£¬Ëæ»únoiseµÄÓ°Ïì»ù±¾ÉÏÄܹ»ÐÞÕýºÍÏû³ý¡£ÕâÖÖbaggingͶƱµÄ»úÖÆÄܹ»±£Ö¤½ÏºÃµÄ½µÔëÐÔ£¬´Ó¶øµÃµ½±È½ÏÎȶ¨µÄ½á¹û¡£
¾¹ýÒÔÉÏÈý¸öÀý×Ó£¬ÎÒÃÇ·¢ÏÖRFÖУ¬Ê÷µÄ¸öÊýÔ½¶à£¬Ä£ÐÍÔ½Îȶ¨Ô½ÄܱíÏֵúá£ÔÚʵ¼ÊÓ¦ÓÃÖУ¬Ó¦¸Ã¾¡¿ÉÄÜÑ¡Ôñ¸ü¶àµÄÊ÷¡£ÖµµÃÒ»ÌáµÄÊÇ£¬RFµÄ±íÏÖͬʱҲÓërandom
seedÓйأ¬¼´Ëæ»úµÄ³õʼֵҲ»áÓ°ÏìRFµÄ±íÏÖ¡£

5¡¢Summary
RF½«baggingÓëdecision tree½áºÏÆðÀ´£¬Í¨¹ý°ÑÖÚ¶àµÄ¾ö²ßÊ÷×é½øÐÐ×éºÏ£¬¹¹³ÉÉÁÖµÄÐÎʽ£¬ÀûÓÃͶƱ»úÖÆÈÃG±íÏÖ×î¼Ñ£¬·ÖÀàÄ£Ð͸üÎȶ¨¡£ÆäÖÐΪÁËÈÃdecision
treeµÄËæ»úÐÔ¸üǿһЩ£¬¿ÉÒÔ²ÉÓÃrandomly projected subspaces²Ù×÷£¬¼´½«²»Í¬µÄfeaturesÏßÐÔ×éºÏÆðÀ´£¬´Ó¶ø½øÐи÷ʽ¸÷ÑùµÄÇиͬʱ£¬ÎÒÃÇÒ²½éÉÜÁË¿ÉÒÔʹÓÃOOBÑù±¾À´½øÐÐself-validation£¬È»ºó¿ÉÒÔʹÓÃself-validationÀ´¶Ôÿ¸öÌØÕ÷½øÐÐpermutaion
test£¬µÃµ½²»Í¬ÌØÕ÷µÄÖØÒªÐÔ£¬´Ó¶ø½øÐÐfeature selection¡£×ܵÄÀ´Ëµ£¬RFËã·¨Äܹ»µÃµ½±È½Ïƽ»¬µÄ±ß½ç£¬Îȶ¨ÐÔÇ¿£¬Ç°ÌáÊÇÓÐ×ã¹»¶àµÄÊ÷¡£
|