r-statistics.co by Selva Prabhakaran

About

dplyr arrange(), slice() & top_n(): Sort, Sample & Filter by Rank

arrange() sorts rows by column values. slice() picks rows by position. slice_max() and slice_min() get the top or bottom N rows by a value — with optional grouping for "top N per group."

arrange(): Sort Rows

library(dplyr) # Ascending (default) mtcars |> arrange(mpg) |> select(mpg, hp, wt) |> head(5)


  








  


    
library(dplyr)

# Descending
mtcars |> arrange(desc(mpg)) |> select(mpg, hp, wt) |> head(5)


    


      

      

    


    


  








  


    
library(dplyr)

# Multiple columns: cyl ascending, then mpg descending within each cyl
mtcars |>
  arrange(cyl, desc(mpg)) |>
  select(cyl, mpg, hp) |>
  head(10)


    


      

      

    


    


  






slice(): Pick Rows by Position





  


    
library(dplyr)

mtcars |> slice(1:3) |> select(mpg, hp)          # First 3
mtcars |> slice(c(1, 10, 20)) |> select(mpg, hp) # Specific rows


    


      

      

    


    


  






slice_head() and slice_tail()





  


    
library(dplyr)

cat("First 3:\n")
mtcars |> slice_head(n = 3) |> select(mpg, hp)

cat("\nLast 3:\n")
mtcars |> slice_tail(n = 3) |> select(mpg, hp)


    


      

      

    


    


  






slice_max() and slice_min(): Top/Bottom by Value





  


    
library(dplyr)

# Top 5 most fuel-efficient
mtcars |>
  mutate(car = rownames(mtcars)) |>
  slice_max(mpg, n = 5) |>
  select(car, mpg, hp)


    


      

      

    


    


  








  


    
library(dplyr)

# Bottom 3 by horsepower (least powerful)
mtcars |>
  mutate(car = rownames(mtcars)) |>
  slice_min(hp, n = 3) |>
  select(car, hp, mpg, cyl)


    


      

      

    


    


  






slice_sample(): Random Rows





  


    
library(dplyr)

set.seed(42)
# Random 5 rows
mtcars |>
  mutate(car = rownames(mtcars)) |>
  slice_sample(n = 5) |>
  select(car, mpg, hp)


    


      

      

    


    


  








  


    
library(dplyr)

set.seed(42)
# 20% random sample
mtcars |> slice_sample(prop = 0.2) |> select(mpg, hp) |> nrow()


    


      

      

    


    


  






Grouped Slicing: Top N Per Group



The real power: combine group_by() with slice functions to get the top/bottom N within each group.





  


    
library(dplyr)

# Top 2 most efficient cars PER cylinder group
mtcars |>
  mutate(car = rownames(mtcars)) |>
  group_by(cyl) |>
  slice_max(mpg, n = 2) |>
  select(car, cyl, mpg) |>
  ungroup()


    


      

      

    


    


  








  


    
library(dplyr)

# Heaviest car per cylinder group
mtcars |>
  mutate(car = rownames(mtcars)) |>
  group_by(cyl) |>
  slice_max(wt, n = 1) |>
  select(car, cyl, wt, mpg) |>
  ungroup()


    


      

      

    


    


  






Stratified Random Sample





  


    
library(dplyr)

set.seed(123)
# 2 random cars per cylinder group
mtcars |>
  mutate(car = rownames(mtcars)) |>
  group_by(cyl) |>
  slice_sample(n = 2) |>
  select(car, cyl, mpg) |>
  ungroup()


    


      

      

    


    


  






Handling Ties



By default, slice_max/slice_min include ties — you might get more than N rows. Use with_ties = FALSE for exactly N.





  


    
library(dplyr)

# With ties (default) — may return more than 3
mtcars |> slice_min(cyl, n = 3) |> nrow()

# Without ties — exactly 3
mtcars |> slice_min(cyl, n = 3, with_ties = FALSE) |> nrow()


    


      

      

    


    


  






Practice Exercises



Exercise 1: Top Performers by Group



Find the 2 most powerful cars (highest hp) in each cylinder group.





  


    
library(dplyr)



    


      

      

    


    


  







Click to reveal solution

```r


  

    
library(dplyr)

mtcars |>
  mutate(car = rownames(mtcars)) |>
  group_by(cyl) |>
  slice_max(hp, n = 2) |>
  select(car, cyl, hp, mpg) |>
  ungroup()

    

      
      
    

    

  








Exercise 2: Stratified Sample for Training Data



Take 60% of each iris Species as a training set.





  


    
library(dplyr)

set.seed(42)



    


      

      

    


    


  







Click to reveal solution

```r


  

    
library(dplyr)

set.seed(42)
train <- iris |>
  group_by(Species) |>
  slice_sample(prop = 0.6) |>
  ungroup()

cat("Training set:", nrow(train), "rows\n")
train |> count(Species)

    

      
      
    

    

  




**Explanation:** `slice_sample(prop = 0.6)` inside `group_by(Species)` takes 60% from each group — a stratified sample that preserves class proportions.





Summary







Function
Purpose
Example






arrange(col)
Sort ascending
arrange(mpg)




arrange(desc(col))
Sort descending
arrange(desc(mpg))




slice(rows)
Pick by position
slice(1:5)




slice_head(n=)
First N
slice_head(n = 3)




slice_tail(n=)
Last N
slice_tail(n = 3)




slice_max(col, n=)
Top N by value
slice_max(mpg, n = 5)




slice_min(col, n=)
Bottom N by value
slice_min(hp, n = 3)




slice_sample(n=)
Random N
slice_sample(n = 10)




slice_sample(prop=)
Random %
slice_sample(prop = 0.2)







FAQ



Is arrange() stable (preserves original order for ties)?



Yes. dplyr's arrange is a stable sort — rows with identical sort values keep their original relative order.



What replaced top_n()?



slice_max() and slice_min() replaced top_n() in dplyr 1.0. They have a clearer API: slice_max(col, n = 5) vs the old top_n(5, col) where argument order was confusing.



Can I sort by a computed expression?



Yes: arrange(desc(hp / wt)) sorts by power-to-weight ratio without creating a new column first.



What's Next?




        


        
      


      

        

        
© 2016- Selva Prabhakaran.
          This work is licensed under the Creative Commons License.