34

GitHub - Flight-School/FloatingPointApproximation: A correct way to determine if...

 5 years ago
source link: https://github.com/Flight-School/FloatingPointApproximation
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.

README.md

FloatingPointApproximation

A correct way to determine if two floating-point numbers are approximately equal to one another.

This functionality is discussed in Chapter 3 of Flight School Guide to Swift Numbers.

Requirements

  • Swift 4.0+

Installation

Swift Package Manager

Add the FloatingPointApproximation package to your target dependencies in Package.swift:

import PackageDescription

let package = Package(
  name: "YourProject",
  dependencies: [
    .package(
        url: "https://github.com/Flight-School/FloatingPointApproximation",
        from: "1.0.0"
    ),
  ]
)

Then run the swift build command to build your project.

Carthage

To use FloatingPointApproximation in your Xcode project using Carthage, specify it in Cartfile:

github "Flight-School/FloatingPointApproximation" ~> 1.0.0

Then run the carthage update command to build the framework, and drag the built FloatingPointApproximation.framework into your Xcode project.

Usage

Floating-point arithmetic can produce unexpected results, such as 0.1 + 0.2 != 0.3. The reason for this is that many fractional numbers, including 0.1, 0.2, and 0.3, cannot be precisely expressed in a binary number representation.

A common mistake is to use an arbitrarily small constant (such as .ulpOfOne) to determine whether two floating-point numbers are approximately equal. For example:

let actual = 0.1 + 0.2
let expected = 0.3
abs(expected - actual) < .ulpOfOne // true

However, this doesn't work for large scale numbers:

let actual = 1e25 + 2e25
let expected = 3e25
abs(expected - actual) < .ulpOfOne // false

A better approach for determining approximate equality would be to count how many representable values, or ULPs, exist between two floating-point numbers.

The ==~ operator (and its complement, !=~) defined by this package returns a Boolean value indicating whether two floating-point numbers are approximately equal.

import FloatingPointApproximation

0.1 + 0.2 == 0.3 // false
0.1 + 0.2 ==~ 0.3 // true

Floating-point numbers are defined to be approximately equal if they are within one unit of least precision, or ULP, of one another.

Because of how Swift implements floating-point numbers, the implementation of the ==~ operator is quite simple:

func ==~<T> (lhs: T, rhs: T) -> Bool where T: FloatingPoint  {
    return lhs == rhs || lhs.nextDown == rhs || lhs.nextUp == rhs
}

A more complete approach combines both absolute and relative comparisons. The isApproximatelyEqual(to:within:maximumULPs:) method determines whether a floating-point number is approximately equal to another value by first checking to see if it is within a given absolute margin, if provided, and then checking to see if it falls within a given number of ULPs:

import FloatingPointApproximation

(0.1 + 0.2).isApproximatelyEqual(to: 0.3, within: 1e-12, maximumULPs: 2)

Ultimately, it's your responsibility to determine how to compare two floating-point numbers for approximate equality based on the requirements of your domain.

License

MIT

Contact

Mattt (@mattt)


About Joyk


Aggregate valuable and interesting links.
Joyk means Joy of geeK