Transforming Sequences

We've seen that using structural recursion we can create and transform lists. This pattern is simple to use and to understand, but it requires we write the same skeleton time and again. In this section we'll learn that we can replace structural recursion in some cases by using a method on List called map. We'll also see that other useful datatypes provide this method and we can use it as a common interface for manipulating sequences.

Transforming the Elements of a List

In the previous section we saw several examples where we transformed one list to another. For example, we incremented the elements of a list with the following code.

def increment(list: List[Int]): List[Int] =
list match {
case Nil => Nil
case hd :: tl => (hd + 1) :: tl
}

increment(List(1, 2, 3))
// res0: List[Int] = List(2, 2, 3)

In this example the structure of the list doesn't change. If we start with three elements we end with three elements. We can abstract this pattern in a method called map. If we have a list with elements of type A, we pass map a function of type A => B and we get back a list with elements of type B. For example, we can implement increment using map with the function x => x + 1.

def increment(list: List[Int]): List[Int] =
list.map(x => x + 1)

increment(List(1, 2, 3))
// res2: List[Int] = List(2, 3, 4)

Each element is transformed by the function we pass to map, in this case x => x + 1. With map we can transform the elements, but we cannot change the number of elements in the list.

We find a graphical notation helps with understanding map. If we had some type Circle we can draw a List[Circle] as a box containing a circle, as shown in Figure sequences:circle-box.

Now we can draw an equation for map as in Figure sequences:map. If you prefer symbols instead of pictures, the picture is showing that List[Circle].map(Circle => Triangle) = List[Triangle]. One feature of the graphical representation is it nicely illustrates that map cannot create a new "box", which represents the structure of the list---or more concretely the number of elements and their order.

The graphical drawing of map exactly illustrates what holds at the type level for map. If we prefer we can write it down symbolically:

List[A].map(A => B) = List[B]

The left hand side of the equation has the type of the list we map and the function we use to do the mapping. The right hand is the type of the result. We can perform algebra with this representation, substituting in concrete types from our program.

Transforming Sequences of Numbers

We have also written a lot of methods that transform a natural number to a list. We briefly discussed how we can represent a natural number as a list. 3 is equivalent to 1 + 1 + 1 + 0, which in turn we could represent as List(1, 1, 1). So what? Well, it means we could write a lot of the methods that accepts natural numbers as methods that worked on lists.

def fill[A](n: Int, a: A): List[A] =
n match {
case 0 => Nil
case n => a :: fill(n-1, a)
}

fill(3, "Hi")
// res3: List[String] = List("Hi", "Hi", "Hi")

we could write

def fill[A](n: List[Int], a: A): List[A] =
n.map(x => a)

fill(List(1, 1, 1), "Hi")
// res5: List[String] = List("Hi", "Hi", "Hi")

The implementation of this version of fill is more convenient to write, but it is much less convenient for the user to call it with List(1, 1, ,1) than just writing 3.

If we want to work with sequences of numbers we are better off using Ranges. We can create these using the until method of Int.

0 until 10
// res6: Range = Range(0, 1, 2, 3, 4, 5, 6, 7, 8, 9)

Ranges have a by method that allows us to change the step between consecutive elements of the range:

0 until 10 by 2
// res7: Range = Range(0, 2, 4, 6, 8)

Ranges also have a map method just like List

(0 until 3) map (x => x + 1)
// res8: IndexedSeq[Int] = Vector(1, 2, 3)

You'll notice the result has type IndexedSeq and is implemented as a Vector---two types we haven't seen yet. We can treat an IndexedSeq much like a List, but for simplicity we can convert a Range or an IndexedSeq to a List using the toList method.

(0 until 7).toList
(0 until 3).map(x => x + 1).toList
// res9: List[Int] = List(1, 2, 3)

With Ranges in our toolbox we can write fill as

def fill[A](n: Int, a: A): List[A] =
(0 until n).toList.map(x => a)

fill(3, "Hi")
// res11: List[String] = List("Hi", "Hi", "Hi")

Ranges over Doubles

If we try to create a Range over Double we get an error.

0.0 to 10.0 by 1.0
// error:
// value to is not a member of Double
// 0.0 to 10.0 by 1.0
// ^^^^^^
// error:
// Line is indented too far to the left, or a } is missing
// error:
// Line is indented too far to the left, or a } is missing
// error:
// Line is indented too far to the left, or a } is missing
// error:
// Line is indented too far to the left, or a } is missing
// error:
// Line is indented too far to the left, or a } is missing
// error:
// Line is indented too far to the left, or a } is missing

There are two ways around this. We can use an equivalent Range over Int. In this case we could just write

0 to 10 by 1

We can use the .toInt method to convert a Double to an Int if needed.

Alternatively we can write a Range using BigDecimal.

import scala.math.BigDecimal
BigDecimal(0.0) to 10.0 by 1.0

BigDecimal has methods doubleValue and intValue to get Double and Int values respectively.

BigDecimal(10.0).doubleValue()
// res15: Double = 10.0
BigDecimal(10.0).intValue()
// res16: Int = 10

Exercises {-}

Ranges, Lists, and map {-}

Using our new tools, reimplement the following methods.

Write a method called ones that accepts an Int n and returns a List[Int] with length n and every element is 1. For example

ones(3)
// res17: List[Int] = List(1, 1, 1)

<div class="solution"> We can just map over a Range to achieve this.

def ones(n: Int): List[Int] =
(0 until n).toList.map(x => 1)

ones(3)
// res19: List[Int] = List(1, 1, 1)

</div>

Write a method descending that accepts an natural number n and returns a List[Int] containing the natural numbers from n to 1 or the empty list if n is zero. For example

descending(0)
// res20: List[Int] = List()
descending(3)
// res21: List[Int] = List(3, 2, 1)

<div class="solution"> We can use a Range but we have to set the step size or the range will be empty.

def descending(n: Int): List[Int] =
(n until 0 by -1).toList

descending(0)
// res23: List[Int] = List()
descending(3)
// res24: List[Int] = List(3, 2, 1)

</div>

Write a method ascending that accepts a natural number n and returns a List[Int] containing the natural numbers from 1 to n or the empty list if n is zero.

ascending(0)
// res25: List[Int] = List()
ascending(3)
// res26: List[Int] = List(1, 2, 3)

<div class="solution"> Again we can use a Range but we need to take care to start at 0 and increment the elements by 1 so we have the correct number of elements.

def ascending(n: Int): List[Int] =
(0 until n).toList.map(x => x + 1)

ascending(0)
// res28: List[Int] = List()
ascending(3)
// res29: List[Int] = List(1, 2, 3)

</div>

Write a method double that accepts a List[Int] and returns a list with each element doubled.

double(List(1, 2, 3))
// res30: List[Int] = List(2, 4, 6)
double(List(4, 9, 16))
// res31: List[Int] = List(8, 18, 32)

<div class="solution"> This is a straightforward application of map.

def double(list: List[Int]): List[Int] =
list map (x => x * 2)

double(List(1, 2, 3))
// res33: List[Int] = List(2, 4, 6)
double(List(4, 9, 16))
// res34: List[Int] = List(8, 18, 32)

</div>

Polygons, Again! {-}

Using our new tools, rewrite the polygon method from the previous section.

<div class="solution"> Here's one possible implementation. Much easier to read than the previous implementation!

def polygon(sides: Int, size: Int, initialRotation: Angle): Image = {
import Point._
import PathElement._

val step = (Angle.one / sides).toDegrees.toInt
val path =
(0 to 360 by step).toList.map{ deg =>
lineTo(polar(size, initialRotation + deg.degrees))
}

Image.path(ClosedPath(moveTo(polar(size, initialRotation)) :: path))
}

</div>

Challenge Exercise: Beyond Map {-}

Can we use map to replace all uses of structural recursion? If not, can you characterise the problems that we can't implement with map but can implement with general structural recursion over lists?

<div class="solution"> We've seen many examples that we cannot implement with map: the methods product, sum, find, and more in the previous section cannot be implemented with map.

In abstract, methods implemented with map obey the following equation:

List[A] map A => B = List[B]

If the result is not of type List[B] we cannot implement it with map. For example, methods like product and sum transform List[Int] to Int and so cannot be implemented with map.

Map transforms the elements of a list, but cannot change the number of elements in the result. Even if a method fits the equation for map above it cannot be implemented with map if it requires changing the number of elements in the list. </div>

Tools with Ranges

We've seen the until method to construct Ranges, and the by method to change the increment in a Range. There is one more method that will be useful to know about: to. This constructs a Range like until but the Range includes the endpoint. Compare

1 until 5
// res36: Range = Range(1, 2, 3, 4)
1 to 5
// res37: Inclusive = Range(1, 2, 3, 4, 5)

In technical terms, the Range constructed with until is a half-open interval, while the range constructed with to is an open interval.

Exercises {-}

Using Open Intervals {-}

Write a method ascending that accepts a natural number n and returns a List[Int] containing the natural numbers from 1 to n or the empty list if n is zero. Hint: use to

ascending(0)
// res38: List[Int] = List()
ascending(3)
// res39: List[Int] = List(1, 2, 3)

<div class="solution"> Now that we now about to this is trivial to implement.

def ascending(n: Int): List[Int] =
(1 to n).toList

ascending(0)
// res41: List[Int] = List()
ascending(3)
// res42: List[Int] = List(1, 2, 3)

</div>