In Scala how do I remove duplicates from a list?

Question

Suppose I have

val dirty = List(\"a\", \"b\", \"a\", \"c\")

Is there a list operation that returns \"a\", \"b\", \"c\"

Answer 1

scala.collection.immutable.List now has a .distinct method.

So calling dirty.distinct is now possible without converting to a Set or Seq.

Answer 2

For Already-Sorted Lists

If you happen to want the distinct items of a list that you know is already sorted, as I have often needed, the following performs about twice the speed as .distinct:

  def distinctOnSorted[V](seq: List[V]): List[V] =
    seq.foldLeft(List[V]())((result, v) =>
      if (result.isEmpty || v != result.head) v :: result else result)
    .reverse

Performance results on a list of 100,000,000 random Ints from 0-99:

distinct        : 0.6655373s
distinctOnSorted: 0.2848134s

Performance with MutableList or ListBuffer

While it would seem that a more mutable / non-functional programming approach might be faster than prepending to an immutable list, practice shows otherwise. The immutable implementation consistently performs better. My guess for the reason is that scala focuses its compiler optimizations on immutable collections, and does a good job at it. (I welcome others to submit better implementations.)

List size 1e7, random 0 to 1e6
------------------------------
distinct            : 4562.2277ms
distinctOnSorted    : 201.9462ms
distinctOnSortedMut1: 4399.7055ms
distinctOnSortedMut2: 246.099ms
distinctOnSortedMut3: 344.0758ms
distinctOnSortedMut4: 247.0685ms

List size 1e7, random 0 to 100
------------------------------
distinct            : 88.9158ms
distinctOnSorted    : 41.0373ms
distinctOnSortedMut1: 3283.8945ms
distinctOnSortedMut2: 54.4496ms
distinctOnSortedMut3: 58.6073ms
distinctOnSortedMut4: 51.4153ms

Implementations:

object ListUtil {
  def distinctOnSorted[V](seq: List[V]): List[V] =
    seq.foldLeft(List[V]())((result, v) =>
      if (result.isEmpty || v != result.head) v :: result else result)
    .reverse

  def distinctOnSortedMut1[V](seq: List[V]): Seq[V] = {
    if (seq.isEmpty) Nil
    else {
      val result = mutable.MutableList[V](seq.head)
      seq.zip(seq.tail).foreach { case (prev, next) =>
        if (prev != next) result += next
      }
      result //.toList
    }
  }

  def distinctOnSortedMut2[V](seq: List[V]): Seq[V] = {
    val result = mutable.MutableList[V]()
    if (seq.isEmpty) return Nil
    result += seq.head
    var prev = seq.head
    for (v <- seq.tail) {
      if (v != prev) result += v
      prev = v
    }
    result //.toList
  }

  def distinctOnSortedMut3[V](seq: List[V]): List[V] = {
    val result = mutable.MutableList[V]()
    if (seq.isEmpty) return Nil
    result += seq.head
    var prev = seq.head
    for (v <- seq.tail) {
      if (v != prev) v +=: result
      prev = v
    }
    result.reverse.toList
  }

  def distinctOnSortedMut4[V](seq: List[V]): Seq[V] = {
    val result = ListBuffer[V]()
    if (seq.isEmpty) return Nil
    result += seq.head
    var prev = seq.head
    for (v <- seq.tail) {
      if (v != prev) result += v
      prev = v
    }
    result //.toList
  }
}

Test:

import scala.util.Random

class ListUtilTest extends UnitSpec {
  "distinctOnSorted" should "return only the distinct elements in a sorted list" in {
    val bigList = List.fill(1e7.toInt)(Random.nextInt(100)).sorted

    val t1 = System.nanoTime()
    val expected = bigList.distinct
    val t2 = System.nanoTime()
    val actual = ListUtil.distinctOnSorted[Int](bigList)
    val t3 = System.nanoTime()
    val actual2 = ListUtil.distinctOnSortedMut1(bigList)
    val t4 = System.nanoTime()
    val actual3 = ListUtil.distinctOnSortedMut2(bigList)
    val t5 = System.nanoTime()
    val actual4 = ListUtil.distinctOnSortedMut3(bigList)
    val t6 = System.nanoTime()
    val actual5 = ListUtil.distinctOnSortedMut4(bigList)
    val t7 = System.nanoTime()

    actual should be (expected)
    actual2 should be (expected)
    actual3 should be (expected)
    actual4 should be (expected)
    actual5 should be (expected)

    val distinctDur = t2 - t1
    val ourDur = t3 - t2

    ourDur should be < (distinctDur)

    print(s"distinct            : ${distinctDur / 1e6}ms\n")
    print(s"distinctOnSorted    : ${ourDur / 1e6}ms\n")
    print(s"distinctOnSortedMut1: ${(t4 - t3) / 1e6}ms\n")
    print(s"distinctOnSortedMut2: ${(t5 - t4) / 1e6}ms\n")
    print(s"distinctOnSortedMut3: ${(t6 - t5) / 1e6}ms\n")
    print(s"distinctOnSortedMut4: ${(t7 - t6) / 1e6}ms\n")
  }
}

Answer 3

You can also use recursion and pattern matching:

def removeDuplicates[T](xs: List[T]): List[T] = xs match {
  case Nil => xs
  case head :: tail => head :: removeDuplicates(for (x <- tail if x != head) yield x)
}

Answer 4

Using Set in the first place is the right way to do it, of course, but:

scala> List("a", "b", "a", "c").toSet.toList
res1: List[java.lang.String] = List(a, b, c)

Works. Or just toSet as it supports the Seq Traversable interface.

Answer 5

The algorithmic way...

def dedupe(str: String): String = {
  val words = { str split " " }.toList

  val unique = words.foldLeft[List[String]] (Nil) {
    (l, s) => {
      val test = l find { _.toLowerCase == s.toLowerCase } 
      if (test == None) s :: l else l
    }
  }.reverse

  unique mkString " "
}

Answer 6

Before using Kitpon's solution, think about using a Set rather than a List, it ensures each element is unique.

As most list operations (foreach, map, filter, ...) are the same for sets and lists, changing collection could be very easy in the code.