2011-12 NBA Top25
許sir
Friday, May 01, 2015
Data Description
2011-12 NBA球季，ESPN選出的前25名球星，包含
- 得分： ppg
- 籃板： rpg
- 助攻： apg
- 火鍋： blk
- 抄截： stl
等數據。我們可以集群分析來將25名球星歸類。
函數說明
#dist() 計算距離矩陣
#hclust()集群分析
#cutree()指定歸類為n群
#pvclust()計算p值的集群分析
#t()行列互換1.設定所需的函式庫(libraries)以及載入資料
#install.packages("pvclust");
library(pvclust) #計算集群分析與P值之套件
setwd("d:/Rdata Practice")
nba<-read.csv("2011NBA.csv", header=T, sep=",")
nba[1:5,]## player division ppg rpg apg blk stl
## 1 LeBron James East 27.1 7.9 6.2 0.81 1.85
## 2 Kevin Durant West 28.0 8.0 3.5 1.17 1.33
## 3 Dwight Howard East 20.6 14.5 1.9 2.15 1.50
## 4 Christ Paul West 19.8 3.6 9.1 0.07 2.53
## 5 Derrick Rose East 21.8 3.4 7.9 0.72 0.902.集群分析
nba.clust<-nba[,-c(1,2)]#排除資料中的player與division文字變數，方便計算歐幾里得距離
means<-apply(nba.clust, 2, mean)#標準化變數，計算每攔變數的平均數，2代表column，1代表row
sds<-apply(nba.clust, 2, sd)#標準化變數，計算每欄變數的標準差，2代表column，1代表row
nba.clust<-scale(nba.clust, center=means, scale=sds)#標準化變數，計算Z分數
nba.dist<-dist(nba.clust,, method="euclidean")#計算歐幾里得距離
nba.fit<-hclust(nba.dist, method="ward.D")#以Ward法進行集群分析
#The "ward" method has been renamed to "ward.D"; note new "ward.D2"
plot(nba.fit, labels=nba$player, main="2011-12 NBA TOP25")#繪製集群分析樹狀圖
rect.hclust(nba.fit, k=5, border="red") #以紅線指定5個集群
cluster5<-cutree(nba.fit, k=5) #指定集群分析為5群
nba$player[cluster5==1] #呼叫屬於第一集群的分析結果## [1] LeBron James Kevin Durant Kobe Bryant Dwyane Wade
## [5] Russell Westbrook
## 25 Levels: Andrew Bynum Blake Griffin Carmelo Anthony ... Tyson Chandlersapply(unique(cluster5), function(a)nba$player[cluster5==a])#呼叫全部集群分析結果## [[1]]
## [1] LeBron James Kevin Durant Kobe Bryant Dwyane Wade
## [5] Russell Westbrook
## 25 Levels: Andrew Bynum Blake Griffin Carmelo Anthony ... Tyson Chandler
##
## [[2]]
## [1] Dwight Howard Andrew Bynum Pau Gasol Tyson Chandler
## [5] Marc Gasol
## 25 Levels: Andrew Bynum Blake Griffin Carmelo Anthony ... Tyson Chandler
##
## [[3]]
## [1] Christ Paul Rajon Rondo
## 25 Levels: Andrew Bynum Blake Griffin Carmelo Anthony ... Tyson Chandler
##
## [[4]]
## [1] Derrick Rose Deron Williams Tony Parker Steve Nash
## [5] Kyrie Irving Manu Ginobili
## 25 Levels: Andrew Bynum Blake Griffin Carmelo Anthony ... Tyson Chandler
##
## [[5]]
## [1] Kevin Love Dirk Nowitzki Blake Griffin Carmelo Anthony
## [5] Chris Bosh LaMarcus Aldridge Kevin Garnett
## 25 Levels: Andrew Bynum Blake Griffin Carmelo Anthony ... Tyson Chandlernba.new<-cbind(nba, cbind(cluster5))#將集群分析產生的新變數cluster5，與原資料nba合併
nba.new$cluster5<-factor(nba.new$cluster5, levels=c(1:5), labels=c("scorer", "defender", "point guard", "combo guard", "power forward"))#將五個集群依序命名
table(nba.new$division, nba.new$cluster5)#比較NBA東西區的集群分布##
## scorer defender point guard combo guard power forward
## East 2 2 1 3 3
## West 3 3 1 3 4nba.new## player division ppg rpg apg blk stl cluster5
## 1 LeBron James East 27.1 7.9 6.2 0.81 1.85 scorer
## 2 Kevin Durant West 28.0 8.0 3.5 1.17 1.33 scorer
## 3 Dwight Howard East 20.6 14.5 1.9 2.15 1.50 defender
## 4 Christ Paul West 19.8 3.6 9.1 0.07 2.53 point guard
## 5 Derrick Rose East 21.8 3.4 7.9 0.72 0.90 combo guard
## 6 Kobe Bryant West 27.9 5.4 4.6 0.31 1.19 scorer
## 7 Kevin Love West 26.0 13.3 2.0 0.51 0.85 power forward
## 8 Dwyane Wade East 22.1 4.6 4.8 1.29 1.67 scorer
## 9 Russell Westbrook West 23.6 4.6 5.5 0.30 1.70 scorer
## 10 Deron Williams East 21.0 3.3 8.7 0.36 1.22 combo guard
## 11 Dirk Nowitzki West 21.6 6.7 2.2 0.48 0.68 power forward
## 12 Rajon Rondo East 11.9 4.9 11.7 0.06 1.79 point guard
## 13 Andrew Bynum West 18.7 11.8 1.4 1.93 0.45 defender
## 14 Blake Griffin West 20.7 10.9 3.2 0.73 0.82 power forward
## 15 Pau Gasol West 17.4 10.4 3.7 1.35 0.57 defender
## 16 Tony Parker West 18.3 2.9 7.7 0.08 0.95 combo guard
## 17 Carmelo Anthony East 22.6 6.3 3.6 0.44 1.13 power forward
## 18 Chris Bosh East 18.0 7.9 1.8 0.79 0.89 power forward
## 19 Steve Nash West 12.5 3.0 10.7 0.13 0.61 combo guard
## 20 LaMarcus Aldridge West 21.7 8.0 2.4 0.82 0.93 power forward
## 21 Kevin Garnett East 15.8 8.2 2.9 1.02 0.93 power forward
## 22 Kyrie Irving East 18.5 3.7 5.4 0.39 1.06 combo guard
## 23 Tyson Chandler East 11.3 9.9 0.9 1.40 0.90 defender
## 24 Marc Gasol West 14.6 8.9 3.1 1.86 0.95 defender
## 25 Manu Ginobili West 12.9 3.4 4.4 0.35 0.71 combo guard我們透過集群分析，將NBA 25位球星分為五種類型，分別是得分主力、防守悍將、控球後衛、雙能衛與強力前鋒。集群分析是希望能使群內同質性高，群間異質性大。可以簡單以變異數分析，看看是否達到集群分析目的。
3.變異數分析
anova(lm(nba.new$ppg~factor(cluster5)))#平均得分與球員類型單因子變異數分析## Analysis of Variance Table
##
## Response: nba.new$ppg
## Df Sum Sq Mean Sq F value Pr(>F)
## factor(cluster5) 4 301.83 75.458 5.8769 0.002698 **
## Residuals 20 256.79 12.840
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 14.集群分析與P值
#pvclust套件，提供pvclust()函數計算au(Approximately Unbiased)P值 及bp((Bootstrap Probability)
nbaclust.pvalue<-pvclust(t(nba.clust), method.hclust="ward.D", method.dist="euclidean") #pvclust集群分析以欄為分類基礎，因此先用t()函數行列互換## Bootstrap (r = 0.4)... Done.
## Bootstrap (r = 0.6)... Done.
## Bootstrap (r = 0.6)... Done.
## Bootstrap (r = 0.8)... Done.
## Bootstrap (r = 0.8)... Done.
## Bootstrap (r = 1.0)... Done.
## Bootstrap (r = 1.0)... Done.
## Bootstrap (r = 1.2)... Done.
## Bootstrap (r = 1.2)... Done.
## Bootstrap (r = 1.4)... Done.plot(nbaclust.pvalue, labels=nba$player, main="2011-12 NBA TOP25 with P-value")