Repository Public
1import Foundation23extension Descriptive {4 /// k-th central moment (divisor n).5 public static func centralMoment(_ x: [Double], _ k: Int) -> Double {6 guard !x.isEmpty else { return .nan }7 let m = mean(x)8 return x.reduce(0) { $0 + Foundation.pow($1 - m, Double(k)) } / Double(x.count)9 }1011 /// Skewness.12 /// - `sample: true` (default): the adjusted Fisher–Pearson coefficient G₁ = √(n(n−1))/(n−2) · g₁ — Excel SKEW, SciPy `skew(bias=False)`. Needs n ≥ 3.13 /// - `sample: false`: the population (moment) coefficient g₁ = m₃ / m₂^{3/2} — Excel SKEW.P, SciPy `skew()`.14 public static func skewness(_ x: [Double], sample: Bool = true) -> Double {15 let n = Double(x.count)16 guard x.count >= (sample ? 3 : 1) else { return .nan }17 let m2 = centralMoment(x, 2), m3 = centralMoment(x, 3)18 guard m2 > 0 else { return .nan }19 let g1 = m3 / Foundation.pow(m2, 1.5)20 return sample ? (n * (n - 1)).squareRoot() / (n - 2) * g1 : g121 }2223 /// Excess kurtosis (0 for a normal distribution).24 /// - `sample: true` (default): G₂ = (n−1)/((n−2)(n−3)) · ((n+1)·g₂ + 6) — Excel KURT, SciPy `kurtosis(bias=False)`. Needs n ≥ 4.25 /// - `sample: false`: the population coefficient g₂ = m₄ / m₂² − 3 — SciPy `kurtosis()`.26 public static func kurtosis(_ x: [Double], sample: Bool = true) -> Double {27 let n = Double(x.count)28 guard x.count >= (sample ? 4 : 1) else { return .nan }29 let m2 = centralMoment(x, 2), m4 = centralMoment(x, 4)30 guard m2 > 0 else { return .nan }31 let g2 = m4 / (m2 * m2) - 332 return sample ? (n - 1) / ((n - 2) * (n - 3)) * ((n + 1) * g2 + 6) : g233 }3435 /// Standard error of the mean: s / √n with the sample standard deviation.36 public static func standardError(_ x: [Double]) -> Double {37 x.count > 1 ? standardDeviation(x) / Double(x.count).squareRoot() : .nan38 }3940 /// Coefficient of variation s / mean (sample standard deviation by default). NaN when the mean is 0.41 public static func coefficientOfVariation(_ x: [Double], sample: Bool = true) -> Double {42 let m = mean(x)43 return m == 0 ? .nan : standardDeviation(x, sample: sample) / m44 }4546 /// Geometric mean of strictly positive data (NaN otherwise).47 public static func geometricMean(_ x: [Double]) -> Double {48 guard !x.isEmpty, x.allSatisfy({ $0 > 0 }) else { return .nan }49 return exp(sum(x.map { log($0) }) / Double(x.count))50 }5152 /// Harmonic mean of strictly positive data (NaN otherwise).53 public static func harmonicMean(_ x: [Double]) -> Double {54 guard !x.isEmpty, x.allSatisfy({ $0 > 0 }) else { return .nan }55 return Double(x.count) / sum(x.map { 1 / $0 })56 }5758 /// Mean after removing the fraction `proportion` of observations from EACH tail59 /// (SciPy `trim_mean`: ⌊proportion·n⌋ values cut per side).60 public static func trimmedMean(_ x: [Double], proportion: Double) -> Double {61 guard !x.isEmpty, proportion >= 0, proportion < 0.5 else { return .nan }62 let s = x.sorted()63 let cut = Int((proportion * Double(s.count)).rounded(.down))64 return mean(Array(s[cut..<(s.count - cut)]))65 }6667 /// z-scores (x − mean) / sd with the sample standard deviation by default.68 public static func standardized(_ x: [Double], sample: Bool = true) -> [Double] {69 let m = mean(x), s = standardDeviation(x, sample: sample)70 return x.map { ($0 - m) / s }71 }72}73