Swift 中十进制到分数的转换

Question

I am building a calculator and want it to automatically convert every decimal into a fraction. So if the user calculates an expression for which the answer is "0.333333...", it would return "1/3". For "0.25" it would return "1/4". Using GCD, as found here ( Decimal to fraction conversion ), I have figured out how to convert any rational, terminating decimal into a decimal, but this does not work on any decimal that repeats (like .333333).

Every other function for this on stack overflow is in Objective-C. But I need a function in my swift app! So a translated version of this ( https://stackoverflow.com/a/13430237/5700898 ) would be nice!

Any ideas or solutions on how to convert a rational or repeating/irrational decimal to a fraction (ie convert "0.1764705882..." to 3/17) would be great!

Answer 1

If you want to display the results of calculations as rational numbers then the only 100% correct solution is to use rational arithmetic throughout all calculations, ie all intermediate values are stored as a pair of integers (numerator, denominator) , and all additions, multiplications, divisions, etc are done using the rules for rational numbers.

As soon as a result is assigned to a binary floating point number such as Double , information is lost. For example,

let x : Double = 7/10

stores in x an approximation of 0.7 , because that number cannot be represented exactly as a Double . From

print(String(format:"%a", x)) // 0x1.6666666666666p-1

one can see that x holds the value

0x16666666666666 * 2^(-53) = 6305039478318694 / 9007199254740992
                           ≈ 0.69999999999999995559107901499373838305

So a correct representation of x as a rational number would be 6305039478318694 / 9007199254740992 , but that is of course not what you expect. What you expect is 7/10 , but there is another problem:

let x : Double = 69999999999999996/100000000000000000

assigns exactly the same value to x , it is indistinguishable from 0.7 within the precision of a Double .

So should x be displayed as 7/10 or as 69999999999999996/100000000000000000 ?

As said above, using rational arithmetic would be the perfect solution. If that is not viable, then you can convert the Double back to a rational number with a given precision . (The following is taken from Algorithm for LCM of doubles in Swift .)

Continued Fractions are an efficient method to create a (finite or infinite) sequence of fractions h _n /k _n that are arbitrary good approximations to a given real number x , and here is a possible implementation in Swift:

typealias Rational = (num : Int, den : Int)

func rationalApproximationOf(x0 : Double, withPrecision eps : Double = 1.0E-6) -> Rational {
    var x = x0
    var a = floor(x)
    var (h1, k1, h, k) = (1, 0, Int(a), 1)

    while x - a > eps * Double(k) * Double(k) {
        x = 1.0/(x - a)
        a = floor(x)
        (h1, k1, h, k) = (h, k, h1 + Int(a) * h, k1 + Int(a) * k)
    }
    return (h, k)
}

Examples:

rationalApproximationOf(0.333333) // (1, 3)
rationalApproximationOf(0.25)     // (1, 4)
rationalApproximationOf(0.1764705882) // (3, 17)

The default precision is 1.0E-6, but you can adjust that to your needs:

rationalApproximationOf(0.142857) // (1, 7)
rationalApproximationOf(0.142857, withPrecision: 1.0E-10) // (142857, 1000000)

rationalApproximationOf(M_PI) // (355, 113)
rationalApproximationOf(M_PI, withPrecision: 1.0E-7) // (103993, 33102)
rationalApproximationOf(M_PI, withPrecision: 1.0E-10) // (312689, 99532)

---

Swift 3 version:

typealias Rational = (num : Int, den : Int)

func rationalApproximation(of x0 : Double, withPrecision eps : Double = 1.0E-6) -> Rational {
    var x = x0
    var a = x.rounded(.down)
    var (h1, k1, h, k) = (1, 0, Int(a), 1)

    while x - a > eps * Double(k) * Double(k) {
        x = 1.0/(x - a)
        a = x.rounded(.down)
        (h1, k1, h, k) = (h, k, h1 + Int(a) * h, k1 + Int(a) * k)
    }
    return (h, k)
}

Examples:

rationalApproximation(of: 0.333333) // (1, 3)
rationalApproximation(of: 0.142857, withPrecision: 1.0E-10) // (142857, 1000000)

Or – as suggested by @brandonscript – with a struct Rational and an initializer:

struct Rational {
    let numerator : Int
    let denominator: Int

    init(numerator: Int, denominator: Int) {
        self.numerator = numerator
        self.denominator = denominator
    }

    init(approximating x0: Double, withPrecision eps: Double = 1.0E-6) {
        var x = x0
        var a = x.rounded(.down)
        var (h1, k1, h, k) = (1, 0, Int(a), 1)

        while x - a > eps * Double(k) * Double(k) {
            x = 1.0/(x - a)
            a = x.rounded(.down)
            (h1, k1, h, k) = (h, k, h1 + Int(a) * h, k1 + Int(a) * k)
        }
        self.init(numerator: h, denominator: k)
    }
}

Example usage:

print(Rational(approximating: 0.333333))
// Rational(numerator: 1, denominator: 3)

print(Rational(approximating: .pi, withPrecision: 1.0E-7))
// Rational(numerator: 103993, denominator: 33102)

Answer 2

So a little late here, but I had a similar problem and ended up building Swift FractionFormatter . This works because most of the irrational numbers you care about are part of the set of vulgar, or common fractions and are easy to validate proper transformation. The rest may or may not round, but you get very close on any reasonable fraction your user might generate. It is designed to be a drop in replacement for NumberFormatter.

Answer 3

As Martin R said, the Only way to have (99.99%)exact calculations, is to calculate everything with rational numbers, from beginning to the end.

the reason behind the creation of this class was also the fact that i needed to have very accurate calculations, and that was not possible with the swift-provided types. so i created my own type.

here is the code, i'll explain it below.

class Rational {

   var alpha = 0
   var beta = 0

   init(_ a: Int, _ b: Int) {
       if (a > 0 && b > 0) || (a < 0 && b < 0) {
           simplifier(a,b,"+")
       }
       else {
           simplifier(a,b,"-")
       }
   }

   init(_ double: Double, accuracy: Int = -1) {
       exponent(double, accuracy)
   }

   func exponent(_ double: Double, _ accuracy: Int) {
       //Converts a double to a rational number, in which the denominator is of power of 10.

       var exp = 1
       var double = double

       if accuracy != -1 {
           double = Double(NSString(format: "%.\(accuracy)f" as NSString, double) as String)!
       }

       while (double*Double(exp)).remainder(dividingBy: 1) != 0 {
           exp *= 10
       }

       if double > 0 {
           simplifier(Int(double*Double(exp)), exp, "+")
       }
       else {
           simplifier(Int(double*Double(exp)), exp, "-")
       }

   }

   func gcd(_ alpha: Int, _ beta: Int) -> Int {
       // Calculates 'Greatest Common Divisor'

       var inti: [Int] = []
       var multi = 1
       var a = Swift.min(alpha,beta)
       var b = Swift.max(alpha,beta)

           for idx in 2...a {
               if idx != 1 {
                   while (a%idx == 0 && b%idx == 0) {
                       a = a/idx
                       b = b/idx
                       inti.append(idx)
                   }
               }
           }
       inti.map{ multi *= $0 }
       return multi
   }


   func simplifier(_ alpha: Int, _ beta: Int, _ posOrNeg: String) {
       //Simplifies nominator and denominator (alpha and beta) so they are 'prime' to one another.

       let alpha = alpha > 0 ? alpha : -alpha
       let beta = beta > 0 ? beta : -beta

       let greatestCommonDivisor = gcd(alpha,beta)

       self.alpha = posOrNeg == "+" ? alpha/greatestCommonDivisor : -alpha/greatestCommonDivisor
       self.beta = beta/greatestCommonDivisor
   }

}

typealias Rnl = Rational

func *(a: Rational, b: Rational) -> Rational {

   let aa = a.alpha*b.alpha
   let bb = a.beta*b.beta

   return Rational(aa, bb)

}

func /(a: Rational, b: Rational) -> Rational {

   let aa = a.alpha*b.beta
   let bb = a.beta*b.alpha

   return Rational(aa, bb)

}

func +(a: Rational, b: Rational) -> Rational {

   let aa = a.alpha*b.beta + a.beta*b.alpha
   let bb = a.beta*b.beta

   return Rational(aa, bb)

}

func -(a: Rational, b: Rational) -> Rational {

   let aa = a.alpha*b.beta - a.beta*b.alpha
   let bb = a.beta*b.beta

   return Rational(aa, bb)

}

extension Rational {

   func value() -> Double {
       return Double(self.alpha) / Double(self.beta)
   }

}

extension Rational {

   func rnlValue() -> String {

       if self.beta == 1 {
           return "\(self.alpha)"
       }
       else if self.alpha == 0  {
           return "0"
       }
       else {
           return "\(self.alpha) / \(self.beta)"
       }
   }

}

// examples:

let first = Rnl(120,45)
let second = Rnl(36,88)
let third = Rnl(2.33435, accuracy: 2)
let forth = Rnl(2.33435)

print(first.alpha, first.beta, first.value(), first.rnlValue()) // prints  8   3   2.6666666666666665   8 / 3
print((first*second).rnlValue()) // prints  12 / 11
print((first+second).rnlValue()) // prints  203 / 66
print(third.value(), forth.value()) // prints  2.33   2.33435

First of all, we have the class itself. the class can be initialised in two ways:

in the Rational class, alpha ~= nominator & beta ~= denominator

The First way is initialising the class using two integers, first of with is the nominator, and the second one is the denominator. the class gets those two integers, and then reduces them to the least numbers possible. eg reduces (10,5) to (2,1) or as another example, reduces (144, 60) to (12,5). this way, always the simplest numbers are stored. this is possible using the gcd (greatest common divisor) function and simplifier function, which are not hard to understand from the code. the only thing is that the class faces some issues with negative numbers, so it always saves whether the final rational number is negative or positive, and if its negative it makes the nominator negative.

The Second way to initialise the class, is with a double, and with an optional parameter called 'accuracy'. the class gets the double, and also the accuracy of how much numbers after decimal point you need, and converts the double to a nominator/denominator form, in which the denominator is of power of 10. eg 2.334 will be 2334/1000 or 342.57 will be 34257/100. then tries to simplify the rational numbers using the same method which was explained in the #1 way.

After the class definition, there is type-alias 'Rnl', which you can obviously change it as you wish.

Then there are 4 functions, for the 4 main actions of math: * / + -, which i defined so eg you can easily multiply two numbers of type Rational.

After that, there are 2 extensions to Rational type, first of which ('value') gives you the double value of a Rational number, the second one ('rnlValue') gives you the the Rational number in form of a human-readable string: "nominator / denominator"

At last, you can see some examples of how all these work.