Last active
June 4, 2021 21:42
-
-
Save toddreed/a734589cadd92814c394ecca5004cd53 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
// Copyright 2017 Todd Reed | |
// | |
// Permission is hereby granted, free of charge, to any person obtaining a copy of this software | |
// and associated documentation files (the “Software”), to deal in the Software without | |
// restriction, including without limitation the rights to use, copy, modify, merge, publish, | |
// distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the | |
// Software is furnished to do so, subject to the following conditions: | |
// | |
// The above copyright notice and this permission notice shall be included in all copies or | |
// substantial portions of the Software. | |
// | |
// THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING | |
// BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, | |
// DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | |
// This code is part of the article at <https://www.toddreed.name/articles/line-fitting/>. | |
import Foundation | |
/// A line expressed in Hessian normal form: ρ = x cos θ + y sin θ. | |
/// | |
/// `rho` is the length of the vector n from the origin to the line that is perpendicular to the | |
/// line. `theta` is the angle of between the x-axis and the vector n. | |
struct Line { | |
let rho: Double | |
let theta: Double | |
init(rho: Double, theta: Double) { | |
// Normalize rho and theta so that rho > 0 and theta in [0, 2π] | |
var r = rho | |
var t = theta | |
if (r < 0) { | |
r = -r | |
t += Double.pi | |
} | |
if (t < 0 || t > 2*Double.pi) { | |
t = normalizeAngle(t) | |
} | |
if (r == 0 && t > Double.pi) { | |
t -= Double.pi | |
} | |
self.rho = r | |
self.theta = t | |
} | |
} | |
struct Point { | |
let x: Double | |
let y: Double | |
} | |
extension Point { | |
static prefix func -(p: Point) -> Point { | |
return Point(x: -p.x, y: -p.y) | |
} | |
static func +(a: Point, b: Point) -> Point { | |
return Point(x: a.x+b.x, y: a.y+b.y) | |
} | |
static func -(a: Point, b: Point) -> Point { | |
return Point(x: a.x-b.x, y: a.y-b.y) | |
} | |
static func *(s: Double, p: Point) -> Point { | |
return Point(x: s*p.x, y: s*p.y) | |
} | |
} | |
/// Returns the arithmetic mean of a sequence. | |
/// | |
/// This implements the numerically stable algorithm detailed here: | |
/// https://diego.assencio.com/?index=c34d06f4f4de2375658ed41f70177d59. Thanks to Alan Nunns for | |
/// directing me to this algorithm, which avoids floating-point errors caused by the accumulation | |
/// of a large sum. | |
func mean<T>(_ values: T) -> T.Element where T: Sequence, T.Element: FloatingPoint { | |
var mean: T.Element = 0 | |
for (i, v) in values.enumerated() { | |
mean += (v-mean)/T.Element(i+1) | |
} | |
return mean | |
} | |
/// Normalizes an angle `theta` so that is lies in the 2π interval [center-π, center+π]. | |
func normalizeAngle(_ theta: Double, center: Double = Double.pi) -> Double | |
{ | |
return theta-2*Double.pi*floor((theta+Double.pi-center)/(2*Double.pi)) | |
} | |
/// Returns the best least-squares line fit to the provided data points using perpendicular | |
/// distances. | |
/// | |
/// Returns nil if not enough points are provided or no unique solution exists. | |
func fitLine(points: [Point]) -> Line? { | |
if points.count < 2 { | |
return nil | |
} | |
let centroid = Point(x: mean(points.map { $0.x }), y: mean(points.map { $0.y })) | |
let translatedPoints = points.map { p in p-centroid } | |
// Compute a = ∑ (yᵢ²-xᵢ²). Instead of a direct summation, we use the factored form of | |
// yᵢ²-xᵢ²: (yᵢ-xᵢ)×(yᵢ-xᵢ). This avoids “catastrophic cancellation”. Thanks to Alan | |
// Nunns for identifying this improvement. Also see: | |
// | |
// Goldberg, David. 1991. “What Every Computer Scientist Should Know About | |
// Floating-Point Arithmetic.” Computing Surveys (CSUR 23 (1): 5–48. | |
// doi:10.1145/103162.103163. | |
let a = translatedPoints | |
.map({ p in (p.y-p.x) * (p.y+p.x) }) | |
.reduce(0.0, +) | |
let b = translatedPoints | |
.map({ p in p.x * p.y }) | |
.reduce(0.0, +) | |
if (a == 0 && b == 0) { | |
return nil | |
} else { | |
let beta = 2*b | |
let gamma = hypot(a, beta) | |
// Avoid subtracting two nearby floating-points to avoid “catastrophic | |
// cancellation”. This could happen when a > 0 and a² ≫ b²; then a ≈ gamma. | |
let theta = a > 0 ? atan2(-beta, a+gamma) : atan2(a-gamma, beta); | |
let rho = centroid.x*cos(theta) + centroid.y*sin(theta) | |
return Line(rho: rho, theta: theta) | |
} | |
} | |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment