[ArrayDeque]

Scala 2.13 Collections

* https://english.stackexchange.com/questions/226954/how-is-deque-commonly-pronounced

About me

Scala 2.13 collections

https://www.scala-lang.org/blog/2018/06/13/scala-213-collections.html

canbuildfrom

def map[B, That](f: A => B)(implicit bf: CanBuildFrom[Repr, B, That]): That
trait IterableOps[A, CC[_]] {
  def map[B](f: A => B): CC[B]
}

https://www.scala-lang.org/blog/2017/05/30/tribulations-canbuildfrom.html

New APIs

def namesByAge(users: Seq[User]): Map[Int, Seq[String]] =
  users.groupBy(_.age).mapValues(users => users.map(_.name))
def namesByAge(users: Seq[User]): Map[Int, Seq[String]] =
  users.groupMap(_.age)(_.name)

In place mutable api

val users: mutable.ArrayBuffer[User] = ???

users
  .filterInPlace(user => !user.name.startsWith("J"))
  .mapInPlace(user => user.copy(age = user.age + 1))

More

  • Stream vs LazyList
  • Better Views
  • Iterable vs. Traversable/Iterator
  • New collections

ArrayDeque

  • New collection in Scala 2.13
  • Also known as CircularBuffer
  • Replacement for most mutable collections
  • Faster than ArrayBuffer when used as an array
  • Faster than LinkedList when used as a linked list
  • My first contribution to Scala!

Data Structures 101

Data Structures get(idx) update(idx, e) append(e) prepend(e) deleteFirst() deleteLast() insertAt(idx, e) deleteAt(idx)
mutable.ArrayDeque O(1) O(1) O(1) O(1) O(1) O(1) O(min(i, n-i)) O(min(i, n-i))
mutable.ArrayBuffer O(1) O(1) O(1) O(n) O(n) O(1) O(n) O(n)
mutable.Stack O(n) O(n) O(1) (push) O(n) O(n) O(1) (pop) O(n) O(n)
mutable.Queue O(n) O(n) O(1) (enque) O(n) O(1) (deque) O(n) O(n) O(n)
mutable.LinkedList O(n) O(n) O(1) O(1) O(1) O(1) O(n) O(n)
java.util.ArrayList O(1) O(1) O(1) O(n) O(n) O(1) O(n) O(n)
java.util.ArrayDeque N/A O(1) O(1) O(1) O(1) O(1) O(n) O(n)

Array  

Doubly

Ended

Queue 

class ArrayDeque[A] {
  val n = 64
  val array = Array.ofDim[A](n)
  var start, end = 0

  def update(i: Int, a: A): Unit =
    array((start + i)%n) = a

  def apply(i: Int): A =
    array((start + i)%n)
}

class ArrayDeque[A] {
  val n = 64
  val array = Array.ofDim[A](n)
  var start, end = 0

  def update(i: Int, a: A): Unit =
    array((start + i)%n) = a

  def apply(i: Int): A =
    array((start + i)%n)

  def append(a: A): Unit = {
    array(end) = a
    end = (end + 1)%n
  }
}

class ArrayDeque[A] {
  val n = 64
  val array = Array.ofDim[A](n)
  var start, end = 0

  def update(i: Int, a: A): Unit =
    array((start + i)%n) = a

  def apply(i: Int): A =
    array((start + i)%n)

  def append(a: A): Unit = {
    array(end) = a
    end = (end + 1)%n
  }

  def prepend(a: A): Unit = {
    start = (start - 1)%n
    array(start) = a
  }
}

class ArrayDeque[A] {
  val n = 64
  val array = Array.ofDim[A](n)
  var start, end = 0

  def update(i: Int, a: A): Unit =
    array((start + i)%n) = a

  def apply(i: Int): A =
    array((start + i)%n)

  def append(a: A): Unit = {
    array(end) = a
    end = (end + 1)%n
  }

  def prepend(a: A): Unit = {
    start = (start - 1)%n
    array(start) = a
  }

  def deleteLast(): Unit =
    end = (end - 1)%n
}

class ArrayDeque[A] {
  val n = 64
  val array = Array.ofDim[A](n)
  var start, end = 0

  def update(i: Int, a: A): Unit =
    array((start + i)%n) = a

  def apply(i: Int): A =
    array((start + i)%n)

  def append(a: A): Unit = {
    array(end) = a
    end = (end + 1)%n
  }

  def prepend(a: A): Unit = {
    start = (start - 1)%n
    array(start) = a
  }

  def deleteLast(): Unit =
    end = (end - 1)%n

  def deleteFirst(): Unit =
    start = (start + 1)%n
}

class ArrayDeque[A] {
  val n = 64
  val array = Array.ofDim[A](n)
  var start, end = 0

  def update(i: Int, a: A): Unit =
    array((start + i)%n) = a

  def apply(i: Int): A =
    array((start + i)%n)

  def append(a: A): Unit = {
    array(end) = a
    end = (end + 1)%n
  }

  def prepend(a: A): Unit = {
    start = (start - 1)%n
    array(start) = a
  }

  def deleteLast(): Unit =
    end = (end - 1)%n

  def deleteFirst(): Unit =
    start = (start + 1)%n

  def clear(): Unit =
    start = end
}

class ArrayDeque[A] {
  val n = 64
  val array = Array.ofDim[A](n)
  var start, end = 0

  def update(i: Int, a: A): Unit =
    array((start + i)%n) = a

  def apply(i: Int): A =
    array((start + i)%n)

  def append(a: A): Unit = {
    array(end) = a
    end = (end + 1)%n
  }

  def prepend(a: A): Unit = {
    start = (start - 1)%n
    array(start) = a
  }

  def deleteLast(): Unit =
    end = (end - 1)%n

  def deleteFirst(): Unit =
    start = (start + 1)%n

  def clear(): Unit =
    start = end

  def size: Int =
    (end - start)%n
}

DEMO

Scala 2.13

scala.collection.mutable.ArrayDeque

Performace

  • Array.copy (memcpy)
    • insertAt(idx), deleteAt(idx), remove(idx)
    • clone()
    • slice()
    • Pre-emptive allocations:
      • insertAll(), prependAll()
  • Bit hack if n (array.length) = 2^k 
    • i%n == i&(n - 1)

Benchmarks

Operation ArrayBuffer ArrayDeque Speedup
Insert lots of items 2473.36 ms 956.76 ms 2.5x
Drop some items from an head index 7.65 ms 1.25 ms 5x
Drop some items from a tail index 2.54 ms 0.28 ms 10x
Append lots of items one by one 3576.63 ms 2222.13 ms 1.5x
Prepend few items one by one 8699.13 ms 1.33 ms O(n)
Prepend lots of items at once 2124.02 ms 462.76 ms 5x
Random indexing 81.62 ms 84.02 ms -
Insert items near head 2980.46 ms 1429.52 ms 2x
Reversal 491.46 ms 378.69 ms 1.5x
Insert items near tail 8588.98 ms 2504.20 ms 3x
Sliding 1591.47 ms 157.25 ms 10x
toArray 194.55 ms 181.07 ms -
Clear lots of items 48.34 ms 28.62 ms 2x

github.com/stanch/reftree

import reftree.core._
import reftree.render._
import reftree.diagram._
import reftree.util.Reflection._

implicit def renderArrayDeque: ToRefTree[ArrayDeque[Char]] = ToRefTree {ds =>
  val array = ds.privateField[Array[AnyRef]]("array")
  val start = ds.privateField[Int]("start")
  val end = ds.privateField[Int]("end")

  val arrayRef = {
    val arrayFields = array.zipWithIndex map { case (a, i) =>
      val fieldName = {
        var s = i.toString
        if (i == start) s = '↳' + s
        if (i == end) s = s + '↲' 
        s
      }
      val refTree = Option(a) match {
        case Some(c) => RefTree.Val(c.asInstanceOf[Char]).withHighlight(true)
        case None => RefTree.Null().withHighlight(i == end)
      }
      refTree.toField.withName(fieldName)
    }
    RefTree.Ref(array, arrayFields).rename(s"char[${array.length}]")
  }

  RefTree.Ref(ds, Seq(
    start.refTree.withHighlight(true).toField.withName("start"),
    end.refTree.withHighlight(true).toField.withName("end"),
    arrayRef.toField.withName("array")
  ) ++ ds.toArray.zipWithIndex.map({case (a, i) => a.refTree.toField.withName(i.toString)}))
}

Take Aways

  • ArrayDeques are cool
  • Please contribute
    • Contributing to Scala is not scary
    • Data structures > Algorithms
    • Rope, SkipList, Zipper, Heap
  • Visualize your data structures
  • Scala 2.13 awaits!

Thank YOU

github.com/pathikrit/arraydeque-talk

We are hiring

  • Who we are:
  • What we do:
    • Quant Trading
    • Data Science
    • NLP
  • What we love:
    • Functional Programming
    • Algorithms
    • Statistics
    • Data

pbhowmick@coatue.com

  •  Tech stack:
    • Scala
    • Spark
    • PostgreSQL
    • Tableau
    • R
    • Python
    • Docker
    • AWS
  • Where are we:
    • NYC
    • SF/Menlo

ArrayDeque (Scala by the Bay)

By Pathikrit Bhowmick

Made with Slides.com

ArrayDeque (Scala by the Bay)

https://github.com/pathikrit/arraydeque-talk

  • 1,400

More from Pathikrit Bhowmick