第2åããããããã¹ããã¤ãã³ã°åå¼·ä¼ã§çºè¡¨ããk-meansã«é¢ããè³æã§ããWeniger lesen
ã¨ããããã§è¡ã£ã¦ãã¾ããã第ï¼å ãã¼ã¿ãã¤ãã³ã°+WEB åå¼·ä¼ï¼ æ±äº¬ã( TokyoWebmining 9)?1st Week? 大è¦æ¨¡è§£æã»æ©æ¢°å¦ç¿ã»ã¯ãªã³ã ç¥ã? : ATNDFirst Weekã£ã¦ãâ 大è¦æ¨¡è§£æï¼1. Mahout Canopy Clustering (è¬å¸«ï¼@hamadakoichi)(çºè¡¨30åï¼è°è«60å)ãCanopy Clusteringã¯é常ã®å¤ãã®ææ³ã¨ç°ãªããã¯ã©ã¹ã¿æ°æå®ãå¿ è¦ã¨ãããæå®è·é¢ é¢ããã¯ã©ã¹ã¿ç®åºãå®ç¾ããããHadoopä¸ã§åä½ãã大è¦æ¨¡ãã¼ã¿ãã¤ãã³ã°ã»æ©æ¢°å¦ç¿ã©ã¤ãã©ãª Mahoutã§ã®å®è¡æ³ãå«ãã話ããã¾ã2. æ©æ¢°å¦ç¿ï¼æ©æ¢°ã®ä»£ããã«äººéãå¦ç¿ (è¬å¸«ï¼@shuyo))(çºè¡¨20åï¼è°è«40å)ãGihyo.jp ã§ãæ©æ¢°å¦ç¿ã®é£è¼ã裾éãåºããæ´»åãããã¦ãã @shuyo ããããä»åãæ©æ¢°å¦ç¿ã®æ´å²ãå°éå¤
ã¯ã©ã¹ã¿ãªã³ã°ãã¼ã« bayon ãã¨ã¦ã¤ããªãç´ æ´ãããã§ãï¼ 2009-06-10-5 [Algorithm][Software] mixi ã® fujisawa ããã«ãããC++ ã§æ¸ãããã¯ã©ã¹ã¿ãªã³ã°ãã¼ã« bayon ãã·ã³ãã«ã¤ããã§ççã«ç´ æ´ãããã¦ã¯ã¼ã«ã§ãã - 軽éãã¼ã¿ã¯ã©ã¹ã¿ãªã³ã°ãã¼ã«bayon (mixi Engineers' Blog) http://alpha.mixi.co.jp/blog/?p=1049 - ãã¥ã¼ããªã¢ã«ï¼Tutorial_ja - bayonï¼ http://code.google.com/p/bayon/wiki/Tutorial_ja 詳細ã¯ä¸è¨URLãè¦ã¦ãããã¨ãã¦ã ãã¾ãã¾æå ã«250ä¸ä»¶ã®ãã¼ã¿ï¼ã©ãã«ï¼ç¹å¾´èªãªã¹ãï¼ããã£ãã®ã§ãã£ãã試ãã¦ã¿ã¾ããã ããã¥ã¡ã³ãæ°250ä¸ä»¶ã åããã¥ã¡ã³ãã®ç¹å¾´ãç¾ããã¼ã®å¹³
New in version 2.2. Redisã§ã¯ãå®éã«ã¯ã©ã¤ã¢ã³ãã¨éä¿¡ããããã¼ã以å¤ã«ããµã¼ãã¹ç¨ãã£ã³ãã«ãéãã¦ãã¹ã¦ã®ãã¼ãã¯ãã®ãã£ã³ãã«ã使ã£ã¦ç´æ¥æ¥ç¶ãããã¾ããTCPã®ãã¼ã¹ã®ãã¼ã+4000çªã使ããããããã¯ã©ã¤ã¢ã³ãåãã«6379çªã®ãã¼ãã使ç¨ãã¦ããå ´åã«ã¯ã10379çªã使ç¨ãã¾ãããã¼ãéã®éä¿¡ã«ã¯ããã³ãå¹ ã¨é度ã«æé©åããããã¤ããªãããã³ã«ã使ç¨ãã¾ããã¯ã©ã¤ã¢ã³ãã¨ã®éä¿¡ã§ã¯é常ã¯ã¢ã¹ãã¼ãããã³ã«ã使ç¨ãã¾ãããããã¤ãå¤æ´ãããã¾ãããã¼ãå士ã¯ãã¯ã¨ãªã¼ããããã·ã¼(転é)ãããã¨ã¯ããã¾ããã ãã¼ãéã®éä¿¡å 容¶ åãã¼ãã¯ãPING/PONGã¨åæã«ãèªåãéä¿¡ããä¸ã§å¤æããå¨è¾ºãã¼ãæ å ±(Gossip/ããã話)ãä¸ç·ã«éä¿¡ãã¾ãã
K-means and KD-trees resources Read the K-means paper (PS), or K-means paper (PDF) . Note: recently a similar, though independent, result, was brought to our attention. It predates our work. For completeness, you can read that too. Read the X-means paper (PS) or X-means paper (PDF). The X-means and K-means implementation in binary form is now available for download! Currently, there are versions f
ãã£ããã¯Pythonã§pLSA(pLSI)ãå®è£ ãã¦ãã人ãããã®ã§ãææ¸ã¯ã©ã¹ã¿ãªã³ã°ã«ä½¿ããªãããªãã¨èãããã¨ã§ãããsatomacoto: Pythonã§PLSAãå®è£ ãã¦ã¿ããã pLSIã¯ããããã½ããã¯ã©ã¹ã¿ãªã³ã°ã§ã以ä¸ã®ãããªåé¡ãããã¾ããã çµæã®è§£éãç´æ¥ãã¥ãã å¦çæéãããããã®ãã¡1ã¤ç®ã®åé¡ã¯ææ¸ãä¸ããããã¨ãã®ã¯ã©ã¹ã¿ã¸ã®æå±ç¢ºç P(z|d) ã®æ大å¤ãåã£ã¦ãã¼ããªã¯ã©ã¹ã¿ãªã³ã°ã®çµæãå¾ãã°è§£æ±ºãã¾ããããããå¦çæéã®ã»ãã¯EMã¢ã«ã´ãªãºã ã®éä¸ã§0ã§ãªããã©ã¡ã¼ã¿ãå¤ããã¨ãåå ãªã®ã§ãåã«P(z|d)ããã¼ãåãã¦ãä»ã®ãã©ã¡ã¼ã¿ ï¼P(z|w,d)ãªã©ï¼ ããªããªããªãã®ã§ã¤ãã¤ãã§ããããã㧠P(z|w,d), P(w|z), P(d|z) ã®åãã©ã¡ã¼ã¿ããã¼ãåãããã¨ã§é«éåã§ããªãããªã¨èããã®ã§ãããçµæã¯ããªãå£åãã¦ãã¾
bayonãCLUTOãçéãªçç± - download_takeshiâs diaryãèªãã§ãããã«ã¯æãç«ã¤ãã©ããåãããªãã£ãã®ã§è¨¼æãã¦ã¿ãã ä¸ã®è¨äºã§è¿°ã¹ããã¦ãããã¨ã¯ã¯ã©ã¹ã¿ä¸ã®ãã¯ãã«ã¨ãã®ä¸å¿ãã¯ãã«ã®ã³ãµã¤ã³é¡ä¼¼åº¦ã®åã¨ãã¯ã©ã¹ã¿ä¸ã®ãã¯ãã«ãå ¨ã¦è¶³ãããã¯ãã«ã®ãã«ã ãä¸è´ããã¨ããã§ããã ãã ãããã§ã¯ã©ã¹ã¿ä¸ã®è¦ç´ ãã¯ãã«ã¯ãã¹ã¦å¤§ãã1ã®è¦æ ¼åããããã¯ãã«ã§ããã¨ããã 証æ ä»ã¯ã©ã¹ã¿å ã«å«ã¾ãããã¯ãã«ã ã¨ããã ãã®ã¨ãå ¨ãã¯ãã«ã足ãããã è¤åãã¯ãã«ã ã¨ãããã¾ããã®ã¯ã©ã¹ã¿ã®ã»ã³ããã¤ã㯠ã¨ãªãããã®ã¨ãã»ã³ããã¤ãã¨åãã¯ãã«ã¨ã®ã³ãµã¤ã³é¡ä¼¼åº¦ã¯ [tex: s_i = \frac{}{||C|| ||x_i||} = \frac{}{||{C}||}] ã¨ãªããããã§ã¨æ£è¦åããã¦ãããã¨ãç¨ããããã®é¡ä¼¼åº¦ã®åè¨ã¯ [tex:
ã¯ã©ã¹ã¿ãªã³ã°ãã¼ã«bayonã使ã£ã¦ãã¦ã常ã ãã©ããã¦ãããªã«é«éã«å¦çã§ããã®ããªããã¨çåã«æãã¦ãã¾ãããrepeated bisectionã¨ããææ³èªä½ãk-meansæ³ãªã©ã¨æ¯ã¹ãã¨å¹çãããã®ã§ãããããã«ãã¦ããããã ãã§ã¯èª¬æãã¤ããªãã»ã©çéãªããã§ãã ãã¾ãä¾ãããã¾ããããèªåã§k-meansã®ã¹ã¯ãªãããæ¸ãã¦æ¯ã¹ã¦ã¿ãã¨ãèªè»¢è»ã¨æ°å¹¹ç·ãããã¡ããã¨ããå°è±¡ã§ããã¯ããã¦CLUTOã触ã£ãæãæ°ä¸ä»¶ç¨è¦æ¨¡ã®ã¯ã©ã¹ã¿ãªã³ã°å¦çãæ¬å½ã«ããã£ãã¨ããéã«çµãã£ã¦ãã¾ããã³ã£ããããè¨æ¶ãããã¾ãã ãã£ã¨å®è£ é¢ã§ãªã«ãç¹æ®ãªãã¨ããããã ãããªã¨æããmixiã¨ã³ã¸ãã¢ããã°ã§bayonã®è¨äºãæ¹ãã¦èªã¿æ¼ã£ã¦ãããã以ä¸ã®é¨åãç®ã«æ¢ã¾ãã¾ããã ãã®ã¯ã©ã¹ã¿ã®è©ä¾¡ã¯ãã¯ã©ã¹ã¿ã®åè¦ç´ ã¨ã¯ã©ã¹ã¿ã®ä¸å¿ã¨ã®cosineé¡ä¼¼åº¦ã®åã¨ãã¦ãã¾ãããã®åã大ããã»ã©
æ©æ¢°å¦ç¿ç³»ã®ã¨ã³ããªãç¶ãã¦æ¸ãã¦ã¿ããã¯ã©ã¹ã¿ãªã³ã°ã«ã¤ãã¦ç¥ããªã人ã¯ä»¥ä¸ã®ã¨ã³ããªèªãã§ãã¡ãã·ãããã·ãã ã¨æãã®ã§ãã¯ã©ã¹ã¿ãªã³ã°ã¨ããæ¦å¿µã«ã¤ãã¦ã¯ç¥ã£ã¦ãããã®ã¨ããã ããã§ãä»æ¥ã¯ã¹ãã¯ãã©ã«ã¯ã©ã¹ã¿ãªã³ã°ã®è©±ãèªç¶è¨èªå¦ç以å¤ã§ãå©ç¨ããã¦ããããããã¯ã°ã©ãã®ã¹ãã¯ãã«ã«åºã¥ãã¯ã©ã¹ã¿ãªã³ã°ã®ææ³ã§ãåæ師ããå¦ç¿ã¸ã®æ¡å¼µããããããã®ãå©ç¹ããªã«ããããã¨ããã¨ã¯ã©ã¹ã¿ãªã³ã°ãã°ã©ãã®åå²åé¡(çã§ããã¨ãã¸ãã«ãã)ã«å¸°çãã¦è§£ãææ³ã§ãã©ããããµãã«åå²ãããã«ãã£ã¦ Normalized cut (Ncut) ã¨ã Min-max cut (Mcut) ã¨ãããããããã å®å ¨ã«ã°ã©ããåå²ã§ããå ´åã¯ããã§ãã§ãããã§ãããªã®ã ããå®ä¸çã®ã°ã©ãã¯ãããªç°¡åã«åããªããã¨ãå¾ã ã«ãã¦ãããããã§è¿ä¼¼ãã¦ãã®ã°ã©ãåå²åé¡ã解ãã®ã ããNormalized c
æ¨æ¥ã®Seasar Conference 2009 Autumnã§çºè¡¨ããã¦ããã ãããBlogopolisã®è£å´ãã®è³æãå ¬éãã¾ãã Blogopolisã®è£å´View more documents from kaiseh. è³æã®28æç®ã«ãéã¿ä»ããããã¤å³ã®éå¿ãã¼ã¹ã¬ã¤ã¢ã¦ãã®èª¬æç¨åç»ãããã¾ããããã®åç»ã¯ä»¥ä¸ã«ã¢ãããã¾ããã è¬æ¼è ã®çãããéå¶ã®çæ§ãæ¬å½ã«ãç²ãæ§ã§ããï¼ è¿½è¨ id:mi-changãã p14ã§ã§ã¦ããé ç¹æ°ãããå¤è§å½¢æ°ãã£ã¦ä½ãæå³ãã¦ããã ãã?é ç¹æ°ãå¤ãã¨ãããã¨ã¯ããå¤ãã®ã¿ã°ã¨çµã³ã¤ãã¦ããã£ã¦ãã¨? ããã¯ã1ã¤1ã¤ã®ã¨ã³ããªã¼ãããã°ãå°åºï¼ã«ãã´ãªï¼ã«å¯¾å¿ããåå°ã®å¹¾ä½ãã¼ã¿ã®ãã¨ã§ããä¾ãã°ã5è§å½¢ã®åå°ã®å ´åã¯5åã®é ç¹åº§æ¨ãå¿ è¦ã«ãªãã¾ããåå°ã®é ç¹æ°ã¯ã¬ã¤ã¢ã¦ãä¸ã®çç±ã§æ±ºã¾ããã®ã§ãã¿ã°ã¨ã¯ç´æ¥é¢ä¿ã¯ããã¾ããã
ãç®æ¬¡ã ï¼ï¼ï¼£è¨èªåºç¤ ï¼ï¼ï¼ï¼æ¬å½ã®åºç¤ ï¼ï¼ï¼ï¼é åã¨ãã¤ã³ã¿ ï¼ï¼ï¼ï¼æååæä½ã»ãã¡ã¤ã«æä½ ï¼ï¼ç»ååºç¤ ï¼ï¼ï¼ï¼ç»åãã©ã¼ããã ï¼ï¼ï¼ï¼ããã¹ãã¨ãã¤ã㪠ï¼ï¼ï¼ï¼é åã¨ãã¤ã³ã¿ ï¼ï¼ç»åå¦çåºç¤ ï¼ï¼ï¼ï¼ã¨ãã¸å¦ç ï¼ï¼ï¼ï¼èæ¯å·®åå¦ç ï¼ï¼ã°ã©ãæç»åºç¤ ï¼ï¼ï¼ï¼ï½ï½ï½ï½ï½ï½ï½ ï¼ï¼ï¼ï¼æãç·ã°ã©ã ï¼ï¼ï¼ï¼ãã¹ãã°ã©ã 表示 ï¼ï¼ã¢ã«ã´ãªãºã åºç¤ ï¼ï¼ï¼ï¼ï½ï¼å¹³åã¢ã«ã´ãªãºã ï¼ï¼ï¼ï¼ï¼¥ï¼ã¢ã«ã´ãªãºã ï¼ï¼ç»å表示åºç¤ ï¼ï¼ï¼ï¼ï¼¯ï½ï½ ï½ï¼§ï¼¬ ï¼ï¼ï¼ï¼ï¼¯ï½ï½ ï½ï¼§ï¼¬ã«ããäºæ¬¡å 表示 ï¼ï¼ï¼ï¼ï¼¯ï½ï½ ï½ï¼§ï¼¬ã«ããä¸æ¬¡å 表示 ã¯ããã« ããããç»åå¦çã»èªèã®ç 究ãå§ãããã¨ãã人ã対象ã¨ããå ¥éæ¸ãä½ã£ã¦ãã¾ãï¼å¯¾è±¡ã¯ç 究室ã«é å±ãããã°ããã®æ å ±ç³»å¤§å¦ã®ï¼å¹´çãæ³å®ãã¦ãã¾ããï¼èª°ãèªãã§ãåããããã«å¿ããã¦ããã¤ããã§ãï¼èªã¿é²ããªãã課é¡ã解ãã¦ãããã¡ã«ç»åã®åºç¤ç¥è
å¤ä¸ã®3æåéãã§ãããä¹ ãã¶ãã«ãªããæ¸ããã¨æãã¾ãã ã¡ãã£ã¨åã«mixiã®fujisawaããã¨ããæ¹ãããããã¤ã¹ãªã½ããã¦ã§ã¢ããªãªã¼ã¹ãã¦ããã¾ããã ã軽éãã¼ã¿ã¯ã©ã¹ã¿ãªã³ã°ãã¼ã«bayonã http://alpha.mixi.co.jp/blog/?p=1049 ä»ã¾ã§ã«ãCLUTOã¨ãããããé«ç²¾åº¦ãªã¯ã©ã¹ã¿ãªã³ã°ãã¼ã«ãããã¾ããããããã¤ã¯ã©ã¤ã»ã³ã¹çã«ã¡ãã£ã¨ã¤ã±ãºãªæãã§ãããããã«bayonãã¹ã¼ãã¼ãã³ã®ããã«ç»å ´ãã¦ããã¾ããï¼ãåç¨å©ç¨OKã ããã¨ãããã¨ã§ãä»äºã®ä¸ã§ã®æ©ã¿ã解決ãã¾ãããæ¬å½ã«ããããããã¨ã§ãã ãã¦ãã¦ãæ©é使ã£ã¦ã¿ãããã§ãããããã°ã«æ¸ãã®ã«ã¡ããã©ããé¡æããªãã£ãã®ã§ã以åã«èªåãæ¸ããã¨ã³ããªãããã¼ã¿ãæã£ã¦ãããã¨ã«ãã¾ããã ãè¸è½äººã®ç¸é¢é¢ä¿ãæ¢ã£ã¦ã¿ãã¹ã¯ãªããã http://d.hatena.ne.jp
éåç¥ããã°ã©ãã³ã° ãèªãã§ããããK-means æ³ï¼Kå¹³åæ³ï¼ã®èª¬æãåºã¦ãã¾ãããK-means æ³ã¯ã¯ã©ã¹ã¿ãªã³ã°ãè¡ãããã®å®çªã®ã¢ã«ã´ãªãºã ãããã§ããåå¨ã¯ç¥ã£ã¦ãããã ãã©ãã¾ãã¡ãã³ã¨ãã¦ããªãã£ãã®ã§ãåä½ãç解ããããã«ãµã³ãã«ãä½ã£ã¦ã¿ã¾ãããã¯ãªãã¯ããã¨ï¼ã¹ããããã¤åãããã¨ãã§ãã¾ããã¯ã©ã¹ã¿ã®æ°ãç¹ã®æ°ãå¤æ´ãã¦ãRESET ãæ¼ãã¨å¥½ããªãã©ã¡ã¼ã¿ã§è©¦ããã¨ãã§ãã¾ãããããã£ã¦ï¼ã¹ããããã¤ç¢ºèªããªããåããã¦ã¿ãã¨ãæå¤ã«åç´ãªä»çµã¿ãªã®ãå®æã§ãã¾ãããK-means æ³ã¨ã¯Kå¹³åæ³ - Wikipedia ã«è©³ããæ¸ãã¦ãããã©ããããããã¶ãã¯ãªã¨æ¸ãã¨ãããªã¤ã¡ã¼ã¸ã«ãªãã¾ããåç¹ã«ã©ã³ãã ã«ã¯ã©ã¹ã¿ãå²ãå½ã¦ãã¯ã©ã¹ã¿ã®éå¿ãè¨ç®ãããç¹ã®ã¯ã©ã¹ã¿ããä¸çªè¿ãéå¿ã®ã¯ã©ã¹ã¿ã«å¤æ´ããå¤åããªããã°çµäºãå¤åãããéã㯠2. ã«æ»ãããã
ãã³ãã³åç»ãã¼ã¿åæç 究çºè¡¨ä¼ã¨ããã®ãéå¬ããã¦ããããã ã ã¿ã¤ãã«ã説ææã¯ãã¤ã¸ã¼ãªã®ã§ãåç»ã«ã¤ããããã¿ã°ã使ãã¨å²ã¨ããããªãã¼ã¿ã¨ãã¦å¯è¦åãããã§ãããã¨ãã話ã¯ãã¯ã¦ãªããã¯ãã¼ã¯ã®é¢é£ã¨ã³ããªã¼æ©è½ã®ã¨ããèãããããªè©±ã§ãåºæ¬çã«ã¯ã¤ã³ã¿ã¼ãããã¦ã¼ã¶ã«ç¡æã§ãã¼ã¿ã®ã¿ã°ä»ãããã¦ããã£ã¦ãããã¨ãã話ãªãã ãããªãã¨æãã以åç´¹ä»ããRion Snow ã®è«æ (å½¼ã¯2005å¹´ã« Microsoft Research ã§ã¤ã³ã¿ã¼ã³ãã2006å¹´ã« Powerset (ç¾å¨ã¯ Microsoft ã«è²·åæ¸ã¿)ã2007å¹´ã«ã¯ Google ã§ã¤ã³ã¿ã¼ã³ãã人ç©ãACL ã¨ããèªç¶è¨èªå¦çã®ãããã«ã³ãã¡ã¬ã³ã¹ã§2006å¹´ã«ãã¹ããã¼ãã¼åè³)ã§ã ä»å¹´ã® Rion Snow ã®ãã¼ã¯ã¯ãAmazon Mechanical Turkã¨ããã·ã¹ãã ã使ã£ã¦ãé常ã«å®ä¾¡
http://d.hatena.ne.jp/kaiseh/20090113/1231864089 ä¸ã®è¨äºãè¦ã¦ãk-means++ãé¢ç½ããã ã£ãã®ã§ãã¡ãã£ã¨ã ã試ãã¦ã¿ãã k-meansã¯åæå¤ã«å¤§ããä¾åããã¨ãããå«ããåæå¤ã¸ã®ä¾å度ã軽æ¸ããããã«ãåæå¤ãå¤ãã¦ä½åã試è¡ãã¦ãã®ä¸ã§ä¸çªè¯ãçµæã®ãã®ã使ç¨ããããªãã¦ãã¨ãããªãã¨ãããªãããã®ããå¦çæéã馬鹿ã«ãªããªããªã£ã¦ãã¾ãã®ã§ãã¡ãã£ã¨ããããããªãâ¦ã¨ãããã¨ã§ä½¿ã£ã¦ãªãã£ãã ã§ãä»åã®k-means++ã¯åæå¤ããã¾ãæ±ãããã¨ã§ã精度ã¨é度ã®åä¸ãå¾ãããããããããã¯ããããï¼ è«æèè ã®ãã¼ã¸ã«ãµã³ãã«ã³ã¼ãããã£ãã®ã§è©¦ãã¦ã¿ããã¨æã£ããã ãã©ãMFCã使ã£ã¦ããã¿ããã§åã®ç°å¢ã§ã¯ã³ã³ãã¤ã«ã§ããâ¦ã http://www.stanford.edu/~darthur/kMeansppTest
ã¯ã©ã¹ã¿ãªã³ã° (clustering) ã¨ã¯ï¼åé¡å¯¾è±¡ã®éåãï¼å ççµå (internal cohesion) ã¨å¤çåé¢ (external isolation) ãéæããããããªé¨åéåã«åå²ããã㨠[Everitt 93, å¤§æ© 85] ã§ãï¼çµ±è¨è§£æãå¤å¤é解æã®åéã§ã¯ã¯ã©ã¹ã¿ã¼åæ (cluster analysis) ã¨ãå¼ã°ãï¼åºæ¬çãªãã¼ã¿è§£æææ³ã¨ãã¦ãã¼ã¿ãã¤ãã³ã°ã§ãé »ç¹ã«å©ç¨ããã¦ãã¾ãï¼ åå²å¾ã®åé¨åéåã¯ã¯ã©ã¹ã¿ã¨å¼ã°ãã¾ãï¼åå²ã®æ¹æ³ã«ãå¹¾ã¤ãã®ç¨®é¡ãããï¼å ¨ã¦ã®åé¡å¯¾è±¡ãã¡ããã©ä¸ã¤ã ãã®ã¯ã©ã¹ã¿ã®è¦ç´ ã¨ãªãå ´å(ãã¼ããªãããã¯ï¼ã¯ãªã¹ããªã¯ã©ã¹ã¿ã¨ããã¾ã)ãï¼éã«ä¸ã¤ã®ã¯ã©ã¹ã¿ãè¤æ°ã®ã¯ã©ã¹ã¿ã«åæã«é¨åçã«æå±ããå ´å(ã½ããï¼ã¾ãã¯ï¼ãã¡ã¸ã£ãªã¯ã©ã¹ã¿ã¨ããã¾ã)ãããã¾ãï¼ããã§ã¯åè ã®ãã¼ããªå ´åã®ã¯ã©ã¹ã¿ãªã³ã°ã«ã¤ãã¦è¿°ã¹ã¾ãï¼
K-meansæ³ã¯ãå ¥åãã¼ã¿ããKåã®ã©ã³ãã ãªåä½ãåæã¯ã©ã¹ã¿ã®ä¸å¿ã¨ãã¦é¸æãã以éãã¯ã©ã¹ã¿ã®éå¿ã移åãããã¹ããããç¹°ãè¿ããã¨ã§ã¯ã©ã¹ã¿ãªã³ã°ãè¡ãéé層çææ³ã§ããK-meansæ³ã¯ã·ã³ãã«ã§é«éã§ãããåæå¤ä¾åã大ããã®ãå¼±ç¹ã§ãä¸é©åãªåæå¤é¸æãããã¨ééã£ã解ã«åæãã¦ãã¾ãã¾ãã 以ä¸ã¯ãIntroduction to Information Retrievalã®16ç« ã«åºã¦ããä¾ã§ãã {d1, d2, ..., d6}ãK=2ã§ã¯ã©ã¹ã¿ãªã³ã°ããå ´åã{{d1, d2, d4, d5}, {d3, d6}}ã大åæé©è§£ã§ãããåæã¯ã©ã¹ã¿ã®ä¸å¿ãd2, d5ã§ä¸ããã¨ã{{d1, d2, d3}, {d4, d5, d6}}ã¨ãã誤ã£ã解ã«åæãã¦ãã¾ãã¾ãã ãã®åé¡ãæ¹åããK-means++ã¨ããææ³ãè¦ã¤ããã®ã§ã試ãã¦ã¿ã¾ããã K-means+
ãã°ããéã空ãã¦ãã¾ãã¾ãããIntroduction to Information Retrieval 輪èªä¼ 16ç« ã®å¾©ç¿è³æã以ä¸ã«ã¢ãããã¼ããã¾ããã http://bloghackers.net/~naoya/iir/ppt/iir_16.ppt 16ç« ã®ãã¼ãã¯ã"Flat Clustering" ã§è©±é¡ã¯ã¯ã©ã¹åé¡ããã¯ã©ã¹ã¿ãªã³ã°ã¸ã¨ç§»ãã¾ãã16ç« ã§ã¯ã¯ã©ã¹ã¿ã¨ã¯ã©ã¹ã¿ã®éã«é¢ä¿æ§ããªããã©ããã¯ã©ã¹ã¿ãªã³ã°ãæ±ããç¶ã 17ç« ã§ã¯ã¯ã©ã¹ã¿éã«é層çæ§é ãè¦åºãé層åã¯ã©ã¹ã¿ãªã³ã° (Hierachical clustering) ãæ±ãã¾ãã ã¯ã©ã¹ã¿ãªã³ã° 13ç« ãã15ç« ã¾ã§ã¯ Naive Bayes ã SVM ãªã©ã«ãã "Classification" ã話ã®ä¸»é¡ã§ãããã¯ã©ã¹ã¿ãªã³ã°ãåæ§ã«æ å ±ã®ã°ã«ã¼ãã³ã°ãè¡ããã®ã§ãããClassification
ãªãªã¼ã¹ãé害æ å ±ãªã©ã®ãµã¼ãã¹ã®ãç¥ãã
ææ°ã®äººæ°ã¨ã³ããªã¼ã®é ä¿¡
å¦çãå®è¡ä¸ã§ã
j次ã®ããã¯ãã¼ã¯
kåã®ããã¯ãã¼ã¯
lãã¨ã§èªã
eã³ã¡ã³ãä¸è¦§ãéã
oãã¼ã¸ãéã
{{#tags}}- {{label}}
{{/tags}}