更多大数据分析、建模等内容请关注gongzhonghao《bigdatamodeling》

将代码封装在函数PlotKS_N里，Pred_Var是预测结果，可以是评分或概率形式；labels_Var是好坏标签，取值为1或0，1代表坏客户，0代表好客户；descending用于控制数据按违约概率降序排列，如果Pred_Var是评分，则descending=0，如果Pred_Var是概率形式，则descending=1；N表示在将数据按风险降序排列后，等分N份后计算KS值。

PlotKS_N函数返回的结果为一列表，列表中的元素依次为KS最大值、KS取最大值的人数百分位置、KS曲线对象、KS数据框。

代码如下：

 1 ####################   PlotKS_N ################################2 PlotKS_N<-function(Pred_Var, labels_Var, descending, N){3   # Pred_Var is prop: descending=14   # Pred_Var is score: descending=05   library(dplyr)6  7   df<- data.frame(Pred=Pred_Var, labels=labels_Var)8  9   if (descending==1){
10     df1<-arrange(df, desc(Pred), labels)
11   }else if (descending==0){
12     df1<-arrange(df, Pred, labels)
13   }
14
15   df1$good1<-ifelse(df1$labels==0,1,0)
16   df1$bad1<-ifelse(df1$labels==1,1,0)
17   df1$cum_good1<-cumsum(df1$good1)
18   df1$cum_bad1<-cumsum(df1$bad1)
19   df1$rate_good1<-df1$cum_good1/sum(df1$good1)
20   df1$rate_bad1<-df1$cum_bad1/sum(df1$bad1)
21
22   if (descending==1){
23     df2<-arrange(df, desc(Pred), desc(labels))
24   }else if (descending==0){
25     df2<-arrange(df, Pred, desc(labels))
26   }
27
28   df2$good2<-ifelse(df2$labels==0,1,0)
29   df2$bad2<-ifelse(df2$labels==1,1,0)
30   df2$cum_good2<-cumsum(df2$good2)
31   df2$cum_bad2<-cumsum(df2$bad2)
32   df2$rate_good2<-df2$cum_good2/sum(df2$good2)
33   df2$rate_bad2<-df2$cum_bad2/sum(df2$bad2)
34
35   rate_good<-(df1$rate_good1+df2$rate_good2)/2
36   rate_bad<-(df1$rate_bad1+df2$rate_bad2)/2
37   df_ks<-data.frame(rate_good,rate_bad)
38
39   df_ks$KS<-df_ks$rate_bad-df_ks$rate_good
40
41   L<- nrow(df_ks)
42   if (N>L) N<- L
43   df_ks$tile<- 1:L
44   qus<- quantile(1:L, probs = seq(0,1, 1/N))[-1]
45   qus<- ceiling(qus)
46   df_ks<- df_ks[df_ks$tile%in%qus,]
47   df_ks$tile<- df_ks$tile/L
48   df_0<-data.frame(rate_good=0,rate_bad=0,KS=0,tile=0)
49   df_ks<-rbind(df_0, df_ks)
50
51   M_KS<-max(df_ks$KS)
52   Pop<-df_ks$tile[which(df_ks$KS==M_KS)]
53   M_good<-df_ks$rate_good[which(df_ks$KS==M_KS)]
54   M_bad<-df_ks$rate_bad[which(df_ks$KS==M_KS)]
55
56   library(ggplot2)
57   PlotKS<-ggplot(df_ks)+
58     geom_line(aes(tile,rate_bad),colour="red2",size=1.2)+
59     geom_line(aes(tile,rate_good),colour="blue3",size=1.2)+
60     geom_line(aes(tile,KS),colour="forestgreen",size=1.2)+
61
62     geom_vline(xintercept=Pop,linetype=2,colour="gray",size=0.6)+
63     geom_hline(yintercept=M_KS,linetype=2,colour="forestgreen",size=0.6)+
64     geom_hline(yintercept=M_good,linetype=2,colour="blue3",size=0.6)+
65     geom_hline(yintercept=M_bad,linetype=2,colour="red2",size=0.6)+
66
67     annotate("text", x = 0.5, y = 1.05, label=paste("KS=", round(M_KS, 4), "at Pop=", round(Pop, 4)), size=4, alpha=0.8)+
68
69     scale_x_continuous(breaks=seq(0,1,.2))+
70     scale_y_continuous(breaks=seq(0,1,.2))+
71
72     xlab("of Total Population")+
73     ylab("of Total Bad/Good")+
74
75     ggtitle(label="KS - Chart")+
76
77     theme_bw()+
78
79     theme(
80       plot.title=element_text(colour="gray24",size=12,face="bold"),
81       plot.background = element_rect(fill = "gray90"),
82       axis.title=element_text(size=10),
83       axis.text=element_text(colour="gray35")
84     )
85
86   result<-list(M_KS=M_KS,Pop=Pop,PlotKS=PlotKS,df_ks=df_ks)
87   return(result)
88 }

作图如下：