# 設(shè)置工作空間
# 把“數(shù)據(jù)及程序”文件夾拷貝到F盤下，再用setwd設(shè)置工作空間
setwd("E:\\R_workspace\\R語(yǔ)言數(shù)據(jù)分析與挖掘?qū)崙?zhàn)\\chp4")
# 讀取銷售數(shù)據(jù)文件，提取標(biāo)題行
inputfile <- read.csv('./data/catering_sale.csv', header = TRUE) 
View(inputfile)
# 變換變量名
inputfile <- data.frame(sales = inputfile$'銷量', date = inputfile$'日期')
View(inputfile)
# 數(shù)據(jù)截取
inputfile <- inputfile[5:16, ]
View(inputfile)
# 缺失數(shù)據(jù)的識(shí)別
is.na(inputfile) # 判斷是否存在缺失
n <- sum(is.na(inputfile)) # 輸出缺失值個(gè)數(shù)
n
# 異常值識(shí)別
par(mfrow = c(1, 2)) # 將繪圖窗口劃為1行兩列，同時(shí)顯示兩圖
dotchart(inputfile$sales) # 繪制單變量散點(diǎn)圖
boxplot(inputfile$sales, horizontal = TRUE) # 繪制水平箱形圖
# 異常數(shù)據(jù)處理
inputfile$sales[5] = NA # 將異常值處理成缺失值
fix(inputfile) # 表格形式呈現(xiàn)數(shù)據(jù)
# 缺失值的處理
inputfile$date <- as.numeric(inputfile$date) # 將日期轉(zhuǎn)換成數(shù)值型變量
sub <- which(is.na(inputfile$sales)) # 識(shí)別缺失值所在行數(shù)
sub
# 將數(shù)據(jù)集分成完整數(shù)據(jù)和缺失數(shù)據(jù)兩部分
inputfile1 <- inputfile[-sub, ] 
inputfile2 <- inputfile[sub, ]
# 行刪除法處理缺失，結(jié)果轉(zhuǎn)存
result1 <- inputfile1
View(result1)
# 均值替換法處理缺失，結(jié)果轉(zhuǎn)存
avg_sales <- mean(inputfile1$sales) # 求變量未缺失部分的均值
avg_sales
# 用均值替換缺失
inputfile2$sales <- rep(avg_sales,n) 
# 并入完成插補(bǔ)的數(shù)據(jù)
result2 <- rbind(inputfile1, inputfile2) 
View(result2)
# 回歸插補(bǔ)法處理缺失，結(jié)果轉(zhuǎn)存
# 回歸模型擬合
# 注意：因變量~自變量
model <- lm(sales ~ date, data = inputfile1) 
# 模型預(yù)測(cè)
inputfile2$sales <- predict(model, inputfile2) 
result3 <- rbind(inputfile1, inputfile2)
# 多重插補(bǔ)法處理缺失，結(jié)果轉(zhuǎn)存
library(lattice) # 調(diào)入函數(shù)包
library(MASS)
library(nnet)
library(mice) # 前三個(gè)包是mice的基礎(chǔ)
# 4重插補(bǔ)，即生成4個(gè)無(wú)缺失數(shù)據(jù)集
imp <- mice(inputfile, m = 4) 
# 選擇插補(bǔ)模型
# inputfile為原始數(shù)據(jù)，有缺失
fit <- with(imp,lm(sales ~ date, data = inputfile)) 
# m重復(fù)完整數(shù)據(jù)分析結(jié)果池
pooled <- pool(fit)
summary(pooled)
result4 <- complete(imp, action = 3) # 選擇第三個(gè)插補(bǔ)數(shù)據(jù)集作為結(jié)果

airquality <- read.csv(data.csv)
head(airquality)

summary(airairquality)

#install.packages(‘VIM')
library(VIM)
aggr(airquality)

dat1 <- na.omit(airquality)

#平均值填補(bǔ)
airquality$ Ozone[is.na(airquality$Ozone)] <- mean(airquality $ Ozone,na.rm=T)
#中位數(shù)填補(bǔ)
airquality$ Solar.R[is.na(airquality$ Solar.R)] <- median(airquality$ Solar.R,na.rm = T)
#計(jì)算缺失值個(gè)數(shù)，等于0 則不存在缺失值
sum(is.na(airquality))
#相鄰均值填補(bǔ)
airquality <- read.csv(data.csv) #重新讀入數(shù)據(jù)
for (i in 1:length(airquality$ Ozone)) {
airquality$ Ozone[i] <- ifelse(is.na(airquality$ Ozone[i]),
mean(c(airquality$ Ozone[i-1],airquality$ Ozone[i+1]),na.rm=T),
airquality$ Ozone[i])
}

dat1 <- knnImputation(airquality[,c(1:4)],meth = ‘weighAvg',scale = T)

aggr(dat1) #查看缺失值分布

library(missForest)
dat2 <- missForest(airquality,ntree = 100)

head(dat2$ximp)
aggr(dat2$ximp)

dat2$OOBerror

dat3 <- mice(airquality,m=5,method = ‘rf')

summary(dat3)

dat3$ imp$Solar.R