如何根据最小样本大小对R中的数据帧进行子集化

Question

Let's say you have a data frame with two levels of factors that looks like this:

Factor1    Factor2    Value
A          1          0.75
A          1          0.34
A          2          1.21   
A          2          0.75 
A          2          0.53
B          1          0.42
B          2          0.21  
B          2          0.18
B          2          1.42

etc.

How do I subset this data frame ("df", if you will) based on the condition that the combination of Factor1 and Factor2 (Fact1*Fact2) has more than, say, 2 observations? Can you use the length argument in subset to do this?

Answer 1

library(data.table)

dt = data.table(your_df)

dt[, if(.N > 2) .SD, list(Factor1, Factor2)]
#   Factor1 Factor2 Value
#1:       A       2  1.21
#2:       A       2  0.75
#3:       A       2  0.53
#4:       B       2  0.21
#5:       B       2  0.18
#6:       B       2  1.42

Answer 2

You can use interaction and table to see the number of observation for each interaction (mydata is your data) and then use %in% to subset the data.

 mydata$inter<-with(mydata,interaction(Factor1,Factor2))
 table(mydata$inter)
A.1 B.1 A.2 B.2 
  2   1   3   3 

mydata[!mydata$inter %in% c("A.1","B.1"), ]
  Factor1 Factor2 Value inter
3       A       2  1.21   A.2
4       A       2  0.75   A.2
5       A       2  0.53   A.2
7       B       2  0.21   B.2
8       B       2  0.18   B.2
9       B       2  1.42   B.2

Updated as per @Ananda's comment :You can use following one line code after creating the interaction variable.

mydata[mydata$inter %in% names(which(table(mydata$inter) > 2)), ]

Answer 3

Assuming your data.frame is called mydf , you can use ave to create a logical vector to help subset:

mydf[with(mydf, as.logical(ave(Factor1, Factor1, Factor2, 
                           FUN = function(x) length(x) > 2))), ]
#   Factor1 Factor2 Value
# 3       A       2  1.21
# 4       A       2  0.75
# 5       A       2  0.53
# 7       B       2  0.21
# 8       B       2  0.18
# 9       B       2  1.42

---

Here's ave counting up your combinations. Notice that ave returns an object the same length as the number of rows in your data.frame (this makes it convenient for subsetting).

> with(mydf, ave(Factor1, Factor1, Factor2, FUN = length))
[1] "2" "2" "3" "3" "3" "1" "3" "3" "3"

The next step is to compare that length to your threshold. For that we need an anonymous function for our FUN argument.

> with(mydf, ave(Factor1, Factor1, Factor2, FUN = function(x) length(x) > 2))
[1] "FALSE" "FALSE" "TRUE"  "TRUE"  "TRUE"  "FALSE" "TRUE"  "TRUE"  "TRUE" 

Almost there... but since the first item was a character vector, our output is also a character vector. We want it as.logical so we can directly use it for subsetting.

---

ave doesn't work on objects of class factor , in which case you'll need to do something like:

mydf[with(mydf, as.logical(ave(as.character(Factor1), Factor1, Factor2, 
                               FUN = function(x) length(x) > 2))),]