评分卡风险视图和异常值处理（R语言）

1、风险视图

# 风险视图 library(ggplot2) library(plyr) homedir <- "C:/Users/Administrator/Desktop/rrr1/rrr" #设置路径 setwd(homedir) a <- read.csv('C:/Users/Administrator/Desktop/bianliang.csv') #读取大宽表 b <- sub("Min. :|1st Qu.:|Median :|Mean :|3rd Qu.:|Max. :","",summary(a[, -length(a)])) # 将summary得出的各分位数转化为数据框格式 table1 <- as.data.frame(matrix(b[1:length(b)],nrow = length(row.names(b)), ncol = length(colnames(b)))) row.names(table1) = c('min','P25','P50','mean','P75','max') colnames(table1) = colnames(b) table1 <- table1[-4,] #区间为左开右闭，故在最大值后加0.0001，防止大宽表中变量的最大值溢出 table1 <- apply(table1, c(1,2), as.character) table1 <- as.data.frame(apply(table1, c(1,2), as.numeric)) table1[5,] <- table1[5,]+0.0001 lis1 <- list() label <- list() #循环作图，并将图像保存在文件夹中 for(i in 1:ncol(table1)){ lis1[[i]] <- cbind(a[,i], as.character(cut(a[,i], unique(table1[,i]),right = FALSE)), cut(a[,i], unique(table1[,i]),right = FALSE), a$y) lis1[[i]] <- as.data.frame(lis1[[i]]) rbi <- ddply(lis1[[i]], .(V3, V4), nrow) ce <- ddply(rbi, .(V3), transform, percent = V1/sum(V1)*100) p <- ggplot(ce, aes(x = V3, y = percent, fill = V4)) + geom_bar(stat = 'identity') png(filename=paste0("graphs/", i, "_category.png"), width=640, height=480) print(p) dev.off() }

2、异常值处理

最小值一般均为0，故对过大值进行处理。选取了3个分位数：99%，99.5%，99.9%，分别用后一个除以前一个，得到两个比值，选取比值较大的作为‘拐点’，通过观察，大部分变量的‘拐点’为后者，故将大于99.5%分位数的值定义为异常值，用99.5%分位数代替。

# 替换99.5%分位数 homedir <- 'C:/Users/Administrator/Desktop' setwd(homedir) dic <- read.csv('dic.csv') #各变量的99.5%分位数 try <- read.csv('try.csv') #大宽表 for(i in colnames(try)){ try[,which(colnames(try) == i)][which(try[,which(colnames(try) == i)]>dic[which(dic$var == i),]$extreme)] <- dic[which(dic$var == i),]$extreme } list_new <- list() for(i in 1:ncol(try)){ list_new[[i]] <- try[,i][which(try[,i]<=dic[i,]$extreme)] }

3、groupby功能

#1、aggregate #2、grpub_by %>% summarise require(dplyr) data%>% group_by(a)%>% summarise(AveMPG=mean(b),std=sd(b)) #3、ddply library(plyr) #风险视图 rbi <- ddply(data, .(V3, V4), nrow) ce <- ddply(rbi, .(V3), transform, percent = V1/sum(V1)*100)