func (pdf *PdfBergstrom) ScaledValue(x, alpha, beta float64) (float64, error) {
var err error
zeta := beta * math.Tan(0.5*math.Pi*alpha)
eps := pdf.eps / x * math.Pi
done := false
n := 1
sum := 0.0
for !done {
a := 1.0
if n%2 == 0 {
a = -1.0
}
a *= math.Gamma(float64(n)*alpha+1) / math.Gamma(float64(n)+1)
a *= math.Pow(1+zeta*zeta, 0.5*float64(n))
a *= math.Sin(float64(n) * (0.5*math.Pi*alpha + math.Atan(zeta)))
delta := a * math.Pow(x, -alpha*float64(n))
sum += delta
if math.Abs(delta) < eps {
done = true
}
if n >= pdf.limit {
done = true
err = fmt.Errorf("Iteration limit in tail approximation exceeded (%d)", pdf.limit)
}
n++
}
sum /= x * math.Pi
return sum, err
}
func (b *Beta) CalcPDF(x float64) (float64, error) {
if x <= 0.0 || x >= 1.0 {
return 0, &RangeError{
Offender: x,
Min: 0.0,
Max: 1.0,
}
}
if b.A <= 0.0 {
return 0.0, &RangeError{
ValueName: "Alpha",
Offender: b.A,
Min: 0,
Max: math.MaxFloat64,
}
}
if b.B <= 0.0 {
return 0.0, &RangeError{
ValueName: "Beta",
Offender: b.B,
Min: 0.0,
Max: math.MaxFloat64,
}
}
p1 := math.Gamma(b.A+b.B) / (math.Gamma(b.A) + math.Gamma(b.B))
p2 := math.Pow(x, b.A-1.0) * math.Pow(1.0-x, b.B-1.0)
return p1 * p2, nil
}
func Test_2dinteg02(tst *testing.T) {
//verbose()
chk.PrintTitle("2dinteg02. bidimensional integral")
// Γ(1/4, 1)
gamma_1div4_1 := 0.2462555291934987088744974330686081384629028737277219
x := utl.LinSpace(0, 1, 11)
y := utl.LinSpace(0, 1, 11)
m, n := len(x), len(y)
f := la.MatAlloc(m, n)
for i := 0; i < m; i++ {
for j := 0; j < n; j++ {
f[i][j] = 8.0 * math.Exp(-math.Pow(x[i], 2)-math.Pow(y[j], 4))
}
}
dx, dy := x[1]-x[0], y[1]-y[0]
Vt := Trapz2D(dx, dy, f)
Vs := Simps2D(dx, dy, f)
Vc := math.Sqrt(math.Pi) * math.Erf(1) * (math.Gamma(1.0/4.0) - gamma_1div4_1)
io.Pforan("Vt = %v\n", Vt)
io.Pforan("Vs = %v\n", Vs)
io.Pfgreen("Vc = %v\n", Vc)
chk.Scalar(tst, "Vt", 0.0114830435645548, Vt, Vc)
chk.Scalar(tst, "Vs", 1e-4, Vs, Vc)
}
// Upper incomplete gamma.
func ugamma(x, s float64, regularized bool) float64 {
if x <= 1.1 || x <= s {
if regularized {
return 1 - lgamma(x, s, regularized)
}
return math.Gamma(s) - lgamma(x, s, regularized)
}
f := 1.0 + x - s
C := f
D := 0.0
var a, b, chg float64
for i := 1; i < 10000; i++ {
a = float64(i) * (s - float64(i))
b = float64(i<<1) + 1.0 + x - s
D = b + a*D
C = b + a/C
D = 1.0 / D
chg = C * D
f *= chg
if math.Abs(chg-1) < eps {
break
}
}
if regularized {
logg, _ := math.Lgamma(s)
return math.Exp(s*math.Log(x) - x - logg - math.Log(f))
}
return math.Exp(s*math.Log(x) - x - math.Log(f))
}
// Chisquare returns the p-value of Pr(X^2 > cv).
// Compare this value to the significance level assumed. If chisquare < sigval, then we cannot
// accept the null hypothesis and thus the two variables are dependent.
//
// Thanks to Jacob F. W. for a tutorial on chi-square distributions.
// Source: http://www.codeproject.com/Articles/432194/How-to-Calculate-the-Chi-Squared-P-Value
func Chisquare(df int, cv float64) float64 {
//fmt.Println("Running chi-square...")
if cv < 0 || df < 1 {
return 0.0
}
k := float64(df) / 2.0
x := cv / 2.0
//if df == 1 {
//return math.Exp(-x/2.0) / (math.Sqrt2 * math.SqrtPi * math.Sqrt(x))
//return (math.Pow(x, (k/2.0)-1.0) * math.Exp(-x/2.0)) / (math.Pow(2, k/2.0) * math.Gamma(k/2.0))
//return lgamma(k/2.0, x/2.0, false) / math.Gamma(k/2.0)
//} else if df == 2 {
if df == 2 {
return math.Exp(-x)
}
//fmt.Println("Computing incomplete lower gamma function...")
pval := lgamma(x, k, false)
if math.IsNaN(pval) || math.IsInf(pval, 0) || pval <= 1e-8 {
return 1e-14
}
//fmt.Println("Computing gamma function...")
pval /= math.Gamma(k)
//fmt.Println("Returning chi-square value...")
return 1.0 - pval
}
// Lower incomplete gamma.
func lgamma(x, s float64, regularized bool) float64 {
if x == 0 {
return 0
}
if x < 0 || s <= 0 {
return math.NaN()
}
if x > 1.1 && x > s {
if regularized {
return 1.0 - ugamma(x, s, regularized)
}
return math.Gamma(s) - ugamma(x, s, regularized)
}
var ft float64
r := s
c := 1.0
pws := 1.0
if regularized {
logg, _ := math.Lgamma(s)
ft = s*math.Log(x) - x - logg
} else {
ft = s*math.Log(x) - x
}
ft = math.Exp(ft)
for c/pws > eps {
r++
c *= x / r
pws += c
}
return pws * ft / s
}
// check that the integration function works
func TestIntegrateMid(t *testing.T) {
tests := []struct {
fn smoothFn
x1, x2 float64
Tot float64
}{
// linear
{func(x float64) float64 { return 0.5 * x }, 0.0, 1.0, 0.25},
// normal distribution
{func(x float64) float64 { return 1 / math.Sqrt(2*math.Pi) * math.Exp(-(x*x)/2) }, -100, 100, 1.0},
// normal distribution half
{func(x float64) float64 { return 1 / math.Sqrt(2*math.Pi) * math.Exp(-(x*x)/2) }, -100, 0, 0.5},
// normal distribution segment
{func(x float64) float64 { return 1 / math.Sqrt(2*math.Pi) * math.Exp(-(x*x)/2) }, -2, -1, .1359051219835},
// scaled gamma distribution (similar to my dissertation experiment 3)
{func(x float64) float64 {
k, theta, a := 1.5, 2.0, 1.0/600
return a / (math.Gamma(k) * math.Pow(theta, k)) * math.Sqrt(x*a) * math.Exp(-x*a/2)
}, 0, 2400, 0.73853606463},
}
for i, test := range tests {
got := integrateMid(test.fn, test.x1, test.x2, 10000)
if diff := math.Abs(got - test.Tot); diff > 1e-10 {
t.Errorf("case %v (integral from %v to %v): got %v, want %v", i+1, test.x1, test.x2, got, test.Tot)
}
}
}
func main() {
for true {
r := bufio.NewReader(os.Stdin)
s, err := r.ReadString('\n')
if err == os.EOF {
break
}
s = strings.TrimRight(s, "\n")
a := strings.Split(s, " ")
f := a[0]
x, err := strconv.Atof64(a[1])
switch f {
case "erf":
fmt.Println(math.Erf(x))
case "expm1":
fmt.Println(math.Expm1(x))
case "phi":
fmt.Println(phi.Phi(x))
case "NormalCDFInverse":
fmt.Println(normal_cdf_inverse.NormalCDFInverse(x))
case "Gamma":
fmt.Println(math.Gamma(x))
case "LogGamma":
r, _ := math.Lgamma(x)
fmt.Println(r)
case "LogFactorial":
fmt.Println(log_factorial.LogFactorial(int(x)))
default:
fmt.Println("Unknown function: " + f)
return
}
}
}
func lgammafn(x float64) float64 {
/* For IEEE double precision DBL_EPSILON = 2^-52 = 2.220446049250313e-16 :
xmax = DBL_MAX / log(DBL_MAX) = 2^1024 / (1024 * log(2)) = 2^1014 / log(2)
dxrel = sqrt(DBL_EPSILON) = 2^-26 = 5^26 * 1e-26 (is *exact* below !)
*/
const (
xmax = 2.5327372760800758e+305
dxrel = 1.490116119384765696e-8
)
if isNaN(x) {
return x
}
if x <= 0 && x == trunc(x) { /* Negative integer argument */
return posInf /* +Inf, since lgamma(x) = log|gamma(x)| */
}
y := abs(x)
if y < 1e-306 { // denormalized range
return -log(x)
}
if y <= 10 {
return log(abs(math.Gamma(x)))
}
// ELSE y = |x| > 10
if y > xmax {
return posInf
}
if x > 0 { /* i.e. y = x > 10 */
if x > 1e17 {
return (x * (log(x) - 1))
} else if x > 4934720. {
return (lnSqrt2π + (x-0.5)*log(x) - x)
} else {
return lnSqrt2π + (x-0.5)*log(x) - x + lgammacor(x)
}
}
/* else: x < -10; y = -x */
sinpiy := abs(sin(π * y))
if sinpiy == 0 { // Negative integer argument
// Now UNNECESSARY: caught above, should NEVER happen!
return nan
}
ans := lnSqrtπd2 + (x-0.5)*log(y) - x - log(sinpiy) - lgammacor(y)
if abs((x-trunc(x-0.5))*ans/x) < dxrel {
panic("precision")
}
return ans
}
func main() {
fmt.Println("Running...")
start := time.Now()
A2 = make([]float64, 2)
A3 = make([]float64, 3)
A1 = 1
A2[0] = 1.0 / (3.0 * LaguerreD(2, z2[0]) * Laguerre(3, z2[0]))
A2[1] = 1.0 / (3.0 * LaguerreD(2, z2[1]) * Laguerre(3, z2[1]))
A3[0] = 1.0 / (4.0 * LaguerreD(3, z3[0]) * Laguerre(4, z3[0]))
A3[1] = 1.0 / (4.0 * LaguerreD(3, z3[1]) * Laguerre(4, z3[1]))
A3[2] = 1.0 / (4.0 * LaguerreD(3, z3[2]) * Laguerre(4, z3[2]))
pt = make(plotter.XYs, nPlot)
x = make([]float64, nPlot)
dx := (xmax - xmin) / float64(nPlot-1)
for i := range x {
x[i] = dx*float64(i) + xmin
pt[i].X = x[i]
pt[i].Y = (math.Gamma(x[i]))
}
var p1, p2, p3 plotter.XYs
p1 = make(plotter.XYs, nPlot)
p2 = make(plotter.XYs, nPlot)
p3 = make(plotter.XYs, nPlot)
for i := range x {
p1[i].X = x[i]
p2[i].X = x[i]
p3[i].X = x[i]
p1[i].Y = (A1 * math.Pow(z1, x[i]-1))
p2[i].Y = (A2[0]*math.Pow(z2[0], x[i]-1) + A2[1]*math.Pow(z2[1], x[i]-1))
p3[i].Y = (A3[0]*math.Pow(z3[0], x[i]-1) + A3[1]*math.Pow(z3[1], x[i]-1) + A3[2]*math.Pow(z3[2], x[i]-1))
}
p, err := plot.New()
if err != nil {
panic(err)
}
p.Title.Text = fmt.Sprintf("Gamma Function Approximations")
p.Y.Label.Text = "Log(y)"
p.X.Label.Text = "x"
plotutil.AddLinePoints(p, "Log(Gamma)", pt, "m=0", p1, "m=1", p2, "m=2", p3)
// Save the plot to a PNG file.
if err := p.Save(6, 4, "gammaLow.png"); err != nil {
panic(err)
}
fmt.Println(time.Since(start))
fmt.Println("...program terminated successfully!")
}
func chiSquaredPdf(k float64, x float64) float64 {
if x < 0 {
return 0
}
top := math.Pow(x, (k/2)-1) * math.Exp(-x/2)
bottom := math.Pow(2, k/2) * math.Gamma(k/2)
return top / bottom
}
// Probability density function
func Gamma_PDF(k float64, θ float64) func(x float64) float64 {
return func(x float64) float64 {
if x < 0 {
return 0
}
return math.Pow(x, k-1) * math.Exp(-x/θ) / (math.Gamma(k) * math.Pow(θ, k))
}
}
// LnBeta returns the value of the log beta function. Translation of the Fortran code by W. Fullerton of Los Alamos Scientific Laboratory.
func LnBeta(a, b float64) float64 {
var corr float64
if isNaN(a) || isNaN(b) {
return a + b
}
q := a
p := q
if b < p {
p = b
}
if b > q {
q = b
}
/* both arguments must be >= 0 */
if p < 0 {
return nan
} else if p == 0 {
return posInf
} else if isInf(q, 0) { /* q == +Inf */
return negInf
}
if p >= 10 {
/* p and q are big. */
corr = lgammacor(p) + lgammacor(q) - lgammacor(p+q)
return log(q)*-0.5 + lnSqrt2π + corr + (p-0.5)*log(p/(p+q)) + q*log1p(-p/(p+q))
} else if q >= 10 {
/* p is small, but q is big. */
corr = lgammacor(q) - lgammacor(p+q)
return lgammafn(p) + corr + p - p*log(p+q) + (q-0.5)*log1p(-p/(p+q))
}
/* p and q are small: p <= q < 10. */
if p < 1e-306 {
return LnΓ(p) + (LnΓ(q) - LnΓ(p+q))
}
return log(math.Gamma(p) * (math.Gamma(q) / math.Gamma(p+q)))
}
func gammaIncQ(a, x float64) float64 {
aa1 := a - 1
var f ifctn = func(t float64) float64 {
return math.Pow(t, aa1) * math.Exp(-t)
}
y := aa1
h := 1.5e-2
for f(y)*(x-y) > 2e-8 && y < x {
y += .4
}
if y > x {
y = x
}
return 1 - simpson38(f, 0, y, int(y/h/math.Gamma(a)))
}
// this was used in my dissertation to generate equi-probable sample points
// for my disruption probability distribution.
func testSamplePoints(t *testing.T) {
fn := func(x float64) float64 {
k, theta, a := 1.5, 2.0, 1.0/600
return a / (math.Gamma(k) * math.Pow(theta, k)) * math.Sqrt(x*a) * math.Exp(-x*a/2)
}
x1, x2 := 0.0, 2400.0
xs := sampleUniformProb(fn, x1, x2, 10, 10000)
fmt.Println(xs)
fmt.Printf("x1-x0 = %v\n", xs[0])
for i, x := range xs[:len(xs)-1] {
fmt.Printf("x%v-x%v = %v\n", i+2, i+1, xs[i+1]-x)
}
}
func (self *JSDivFingerprint) calcSignificance(other *JSDivFingerprint) float64 {
p := self.histogram
q := self.histogram
n := len(p)
m := make([]float64, n)
for i := range p {
m[i] = 0.5 * (p[i] + q[i])
}
v := 0.5 * float64(n-1)
D := calcS(m) - (0.5*calcS(p) + 0.5*calcS(q))
inc := apporxIncompleteGamma(v, float64(n)*ln2*D)
gamma := math.Gamma(v)
return inc / gamma
}
func LowerGamma(a, x float64) (z float64) {
// x**a Γ(a) e**-x Σ{k=0..∞}x**k/Γ(a+k+1)
const ε = 1e-20
if x == 0 {
return 0 // γ(a, x) is an integral from 0 to x
}
d := math.Gamma(a)
m := math.Pow(x, a) * d * math.Exp(-x)
if m == 0 {
// overflow
return d // lim{x→∞}γ(a, x) = Γ(a)
}
s := 1 / (d * a) // x**0 / Γ(a+0+1)
z = s
for k := a + 1.0; s > ε; k += 1.0 {
s *= x / k
z += s
}
return m * z
}
// Survival Parity applies Survival to the parity of the random values.
func SurvivalParity(r rand.Source) float64 {
consec := 0
var parity int64 = -1 // prevent the first iteration from matching
counts := map[int]int{0: -1}
for i := 0; i < survivalParityN; i++ {
// software parity because I do not enjoy setting up asm to be used
// http://www-graphics.stanford.edu/~seander/bithacks.html#ParityMultiply
x := r.Int63()
x ^= x >> 1
x ^= x >> 2
x = (x & 0x1111111111111111) * 0x1111111111111111
x = x >> 60 & 1 // parity complete
if x == parity {
consec++
} else {
parity = x
counts[consec] = counts[consec] + 1
consec = 0
}
}
// copypasta
var maximum int
for i := range counts {
if i > maximum {
maximum = i
}
}
// If the source is truly random, then there should be half as many hits
// for counts[n] as there were for counts[n-1], with counts[0] being the
// maximum.
//TODO: E should be calculated from the median. This is causing crc64-ecma to NaN.
E := float64(counts[0])
var chi2 float64
for i := 1; i < maximum; i++ {
E /= 2
d := float64(counts[i]) - E
chi2 += d * d / E
}
k_2 := float64(maximum-2) / 2
return 1 - LowerGamma(k_2, chi2/2)/math.Gamma(k_2)
}
// The lagged survival test applies the survival test to an RNG, skipping N
// iterates between each sample.
func LaggedSurvival(r rand.Source, N int) float64 {
// hax teh copypasta
//TODO: prevent first iteration from matching
consec := [63]int{}
bits := [63]bool{}
counts := make(map[int]int)
for i := 0; i < laggedSurvN; i++ {
x := r.Int63()
for b := 0; b < 63; b++ {
if ((x & (1 << uint(b))) != 0) == bits[b] { // bit survived
consec[b]++
} else { // bit changed
bits[b] = !bits[b]
counts[consec[b]] = counts[consec[b]] + 1
consec[b] = 0
}
}
for n := 0; n < N; n++ {
r.Int63()
}
}
maximum := len(counts)
for i := range counts {
if i > maximum {
maximum = i
}
}
// If the source is truly random, then there should be half as many hits
// for counts[n] as there were for counts[n-1], with counts[0] being the
// maximum.
//TODO: E should be calculated from the median
E := float64(counts[0])
var chi2 float64
for i := 1; i < maximum; i++ {
E /= 2
d := float64(counts[i]) - E
chi2 += d * d / E
}
k_2 := float64(maximum-2) / 2
return 1 - LowerGamma(k_2, chi2/2)/math.Gamma(k_2)
}
func (self *JSDivFingerprint) calcSignificance(other *JSDivFingerprint) float64 {
p := self.histogram
q := other.histogram
m := make(histogram, len(p)+len(q))
min := self.minIndex
max := self.maxIndex
for i := range p {
if i < min {
min = i
}
if i > max {
max = i
}
m[i] = 0.5 * p[i]
}
for i := range q {
if i < min {
min = i
}
if i > max {
max = i
}
m[i] += 0.5 * q[i]
}
k := max - min
v := 0.5 * float64(k-1)
D := calcS(m) - (0.5*calcS(p) + 0.5*calcS(q))
inc := apporxIncompleteGamma(v, float64(self.count+other.count)*ln2*D)
gamma := math.Gamma(v)
return inc / gamma
}
func Test_frechet_03(tst *testing.T) {
//verbose()
chk.PrintTitle("dist_frechet_03")
μ := 10.0
σ := 5.0
δ := σ / μ
d := 1.0 + δ*δ
io.Pforan("μ=%v σ=%v δ=%v d=%v\n", μ, σ, δ, d)
if chk.Verbose {
plt.AxHline(d, "color='k'")
FrechetPlotCoef("/tmp/gosl", "fig_frechet_coef.eps", 3.0, 5.0)
}
k := 0.2441618
α := 1.0 / k
l := μ - math.Gamma(1.0-k)
io.Pfpink("l=%v α=%v\n", l, α)
l = 8.782275
α = 4.095645
var dist DistFrechet
dist.Init(&VarData{L: l, A: α})
io.Pforan("dist = %+#v\n", dist)
io.Pforan("mean = %v\n", dist.Mean())
io.Pforan("var = %v\n", dist.Variance())
io.Pforan("σ = %v\n", math.Sqrt(dist.Variance()))
if chk.Verbose {
plot_frechet(l, 1, α, 8, 16)
plt.SaveD("/tmp/gosl", "rnd_dist_frechet_03.eps")
}
}
// Variance returns the variance
func (o DistFrechet) Variance() float64 {
if o.A > 2.0 {
return o.C * o.C * (math.Gamma(1.0-2.0/o.A) - math.Pow(math.Gamma(1.0-1.0/o.A), 2.0))
}
return math.Inf(1)
}
// float32 version of math.Gamma
func Gamma(x float32) float32 {
return float32(math.Gamma(float64(x)))
}
// gammIPow is a shortcut for computing the gamma function to a power.
func (w Weibull) gammaIPow(i, pow float64) float64 {
return math.Pow(math.Gamma(1+i/w.K), pow)
}
func (pdf *PdfZ) ScaledValue(x, alpha, beta float64) (float64, error) {
if closeTo(alpha, 2, pdf.alpha_tol) {
// Gaussian case, for appropriately normalised levy distribution
return math.Exp(-0.25*x*x) / math.Sqrt(4.0*math.Pi), nil
} else if closeTo(alpha, 1, pdf.alpha_tol) && !closeTo(beta, 0, pdf.beta_tol) {
// This tends to suffer from small oscillations in the integrated distribution
// Need futher integration to sort this out
gamma := math.Exp(-0.5 * math.Pi * x / beta)
a := -0.5 * math.Pi
b := 0.5 * math.Pi
p, err := pdf.integrate(
func(theta float64) float64 {
return componentEq1(theta, beta)*gamma - 1.0
},
func(theta float64) float64 {
return integrandEq1(theta, beta, gamma)
},
a, b)
p *= 0.5 * gamma / math.Abs(beta)
return p, err
} else if closeTo(alpha, 1, pdf.alpha_tol) && closeTo(beta, 0, pdf.beta_tol) {
// Cauchy distribution
return 1.0 / ((1.0 + x*x) * math.Pi), nil
} else if !closeTo(alpha, 1, pdf.alpha_tol) {
zeta := -beta * math.Tan(0.5*math.Pi*alpha)
if x == zeta {
eps := math.Atan(-zeta) / alpha
p := math.Gamma(1+1/alpha) * math.Cos(eps) /
(math.Pi * math.Pow(1+zeta*zeta, 0.5/alpha))
return p, nil
} else if x > zeta {
eps := math.Atan(-zeta) / alpha
gamma := math.Pow(x-zeta, alpha/(alpha-1.0))
a := -eps
b := 0.5 * math.Pi
p, err := pdf.integrate(
func(theta float64) float64 {
return componentNeq1(theta, alpha, beta, eps)*gamma - 1.0
},
func(theta float64) float64 {
return integrandNeq1(theta, alpha, beta, eps, gamma)
},
a, b)
p *= alpha * math.Pow(x-zeta, 1.0/(alpha-1.0)) / (math.Pi * math.Abs(alpha-1))
return p, err
} else if x < zeta {
return pdf.ScaledValue(-x, alpha, -beta)
}
}
return 0, nil
}
// ExKurtosis returns the excess kurtosis of the distribution.
func (w Weibull) ExKurtosis() float64 {
return (-6*w.gammaIPow(1, 4) + 12*w.gammaIPow(1, 2)*math.Gamma(1+2/w.K) - 3*w.gammaIPow(2, 2) - 4*math.Gamma(1+1/w.K)*math.Gamma(1+3/w.K) + math.Gamma(1+4/w.K)) / math.Pow(math.Gamma(1+2/w.K)-w.gammaIPow(1, 2), 2)
}
// Variance returns the variance of the probability distribution.
func (w Weibull) Variance() float64 {
return math.Pow(w.Lambda, 2) * (math.Gamma(1+2/w.K) - w.gammaIPow(1, 2))
}
// Mean returns the mean of the probability distribution.
func (w Weibull) Mean() float64 {
return w.Lambda * math.Gamma(1+1/w.K)
}
func main() {
fmt.Println(" x math.Gamma Lanczos7")
for _, x := range []float64{-.5, .1, .5, 1, 1.5, 2, 3, 10, 140, 170} {
fmt.Printf("%5.1f %24.16g %24.16g\n", x, math.Gamma(x), lanczos7(x))
}
}
func nchoosek(n int, k int) float64 {
return math.Gamma(float64(n+1)) / (math.Gamma(float64(k+1)) * math.Gamma(float64(n-k+1)))
}
