问题描述
我有一个看起来像此可重复代码的数据框,并且我想删除每一列的离群值(在我们的情况下,数据点低于或高于均值的2.5个标准差),而不会删除整个主题/行。
Subj <- c("A","B","C","D","E","F","G","H","I","J")
Var1 <- c("1","5","100","0.1","3","2","2.5","4")
Var2 <- sample(1:10,10,replace=TRUE)
Var3 <- runif(10,min=0,max=700)
Var4 <- c("0.5","23","0.2","0.4","0.6","0.12","0.3","0.25","-75")
df <- as.data.frame(cbind(Subj,Var1,Var2,Var3,Var4))
df$Var1_scale <- scale(as.numeric(df$Var1),scale = TRUE)
df$Var2_scale <- scale(as.numeric(df$Var2))
df$Var3_scale <- scale(as.numeric(df$Var3))
df$Var4_scale <- scale(as.numeric(df$Var4))
我想基于缩放后的变量消除两个数据点-Var1为100,Var4为-75。最好的方法是什么?我总是将其视为消除行,但这不是这里的目的。
输出看起来像这样(即用空格代替异常值)
Subj Var1 Var2 Var3 Var4 Var1_scale Var2_scale Var3_scale Var4_scale
1 A 1 9 82.5652090134099 0.5 -0.3757658 0.8660254 -1.2116275 0.2128018
2 B 5 2 606.970524415374 0.1 -0.2457431 -1.1547005 1.2318109 0.1971919
3 C 9 422.833283618093 23 2.8422981 0.8660254 0.3738333 1.0908581
4 D 0.1 10 100.154890632257 0.2 -0.4050210 1.1547005 -1.1296693 0.2010944
5 E 3 4 144.251625519246 0.4 -0.3107545 -0.5773503 -0.9242029 0.2088993
6 F 5 2 310.489796195179 0.6 -0.2457431 -1.1547005 -0.1496251 0.2167043
7 G 2 8 624.485966027714 0.12 -0.3432602 0.5773503 1.3134231 0.1979724
8 H 3 3 617.240970185958 0.3 -0.3107545 -0.8660254 1.2796654 0.2049969
9 I 2.5 10 293.290452379733 0.25 -0.3270073 1.1547005 -0.2297645 0.2030456
10 J 4 3 223.737383470871 -0.2782488 -0.8660254 -0.5538433 -2.7335648
解决方法
不要将它们替换为空白,而应将它们替换为NA
,以便维护类。
cols <- paste0('Var',1:4)
mat <- sapply(df[cols],function(x) {
mn <- mean(x,na.rm = TRUE)
sd <- sd(x,na.rm = TRUE)
(x > mn + sd * 2.5) | (x < mn - sd * 2.5)
})
df[cols][mat] <- NA
df
# Subj Var1 Var2 Var3 Var4
#1 A 1.0 2 383.35261 0.50
#2 B 5.0 8 498.22071 0.10
#3 C NA 6 272.23357 23.00
#4 D 0.1 1 70.61119 0.20
#5 E 3.0 2 649.11146 0.40
#6 F 5.0 7 198.26275 0.60
#7 G 2.0 7 413.40121 0.12
#8 H 3.0 4 77.25242 0.30
#9 I 2.5 9 588.35492 0.25
#10 J 4.0 8 222.57458 NA
数据
您以一种将数字更改为字符的方式创建了数据集,这使得很难对它们执行任何数学计算。我使用type.convert
将其更改为原始类。
df <- type.convert(df)