Subset data frame based on number of rows per group

Question

I have data like this, where some "name" occurs more than three times:

df <- data.frame(name = c("a", "a", "a", "b", "b", "c", "c", "c", "c"), x = 1:9)

  name x
1    a 1
2    a 2
3    a 3
4    b 4
5    b 5
6    c 6
7    c 7
8    c 8
9    c 9

I wish to subset (filter) the data based on number of rows (observations) within each level of the name variable. If a certain level of name occurs more than say 3 times, I want to remove all rows belonging to that level. So in this example, we would drop observations where name == c, since there are > 3 rows in that group:

  name x
1    a 1
2    a 2
3    a 3
4    b 4
5    b 5

I wrote this code, but can't get it to work.

as.data.frame(table(unique(df)$name))
subset(df, name > 3)

Answer 1

First, two base alternatives. One relies on table, and the other on ave and length. Then, two data.table ways.

---

1. table

tt <- table(df$name)

df2 <- subset(df, name %in% names(tt[tt < 3]))
# or
df2 <- df[df$name %in% names(tt[tt < 3]), ]

---

If you want to walk it through step by step:

# count each 'name', assign result to an object 'tt'
tt <- table(df$name)

# which 'name' in 'tt' occur more than three times?
# Result is a logical vector that can be used to subset the table 'tt'
tt < 3

# from the table, select 'name' that occur < 3 times
tt[tt < 3]

# ...their names
names(tt[tt < 3])

# rows of 'name' in the data frame that matches "the < 3 names"
# the result is a logical vector that can be used to subset the data frame 'df'
df$name %in% names(tt[tt < 3])

# subset data frame by a logical vector
# 'TRUE' rows are kept, 'FALSE' rows are removed.
# assign the result to a data frame with a new name
df2 <- subset(df, name %in% names(tt[tt < 3]))
# or
df2 <- df[df$name %in% names(tt[tt < 3]), ]

---

2. ave and length

As suggested by @flodel:

df[ave(df$x, df$name, FUN = length) < 3, ]

---

3. data.table: .N and .SD:

library(data.table)
setDT(df)[, if (.N < 3) .SD, by = name]

---

4. data.table: .N and .I:

setDT(df)
df[df[, .I[.N < 3], name]$V1] 

---

See also the related Q&A Count number of observations/rows per group and add result to data frame.

Answer 2

Using the dplyr package:

df %>%
  group_by(name) %>%
  filter(n() < 4)

# A tibble: 5 x 2
# Groups:   name [2]
  name      x
  <fct> <int>
1 a         1
2 a         2
3 a         3
4 b         4
5 b         5

n() returns the number of observations in the current group, so we can group_by name, and then keep only those rows which are part of a group where the number of rows in that group is less than 4.

Answer 3

Yet another way using the dpylr package is using the count function and then doing a semi join on the original data frame:

library(dplyr)

df %>% 
  count(name) %>%
  filter(n <= 3) %>%
  semi_join(df, ., by = "name")

Answer 4

Package "inops" has some useful infix operators. For this particular case the operator %in#% can select elements based on how many times they occur.

library(inops)

df[df$name %in#% 1:3,]

Which returns:

  name x
1    a 1
2    a 2
3    a 3
4    b 4
5    b 5

Here df$name %in#% 1:3 returns TRUE only for elements that occur 1, 2, or 3 times. If instead we wanted to select elements that occur 4 times we would do:

df[df$name %in#% 4,]

With the following result:

  name x
6    c 6
7    c 7
8    c 8
9    c 9