Last active
May 27, 2017 11:12
-
-
Save linw1995/28d70604f8d430177fd0496ae15eb763 to your computer and use it in GitHub Desktop.
Monte Carlo method applied to approximating the value of π.
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
# coding:utf-8 | |
from __future__ import division | |
import random | |
import matplotlib.animation as animation | |
import matplotlib.pyplot as plt | |
import numpy as np | |
def dis(x, y): | |
return np.sqrt(x**2 + y**2) | |
def main(): | |
M = int(4e4) | |
inner_x = np.zeros(M) | |
inner_y = np.zeros(M) | |
outer_x = np.zeros(M) | |
outer_y = np.zeros(M) | |
count_inner = count_outer = 0 | |
fig = plt.figure() | |
ims =[] | |
count = 0 | |
for n in range(M): | |
x = random.random() | |
y = random.random() | |
if dis(x, y) < 1: | |
inner_x[count_inner] = x | |
inner_y[count_inner] = y | |
count_inner += 1 | |
else: | |
outer_x[count_outer] = x | |
outer_y[count_outer] = y | |
count_outer += 1 | |
if count >= M / 10 ** 2: | |
im = ( | |
plt.scatter( | |
inner_x[:count_inner], inner_y[:count_inner], | |
c='r', marker='.', s=1), | |
plt.scatter( | |
outer_x[:count_outer], outer_y[:count_outer], | |
c='b', marker='.', s=1), | |
plt.text(-0.2, 0, '$n=$%5d' % n), | |
plt.text( | |
-0.2, -0.035, '$\pi \\approx$%.7f' % (count_inner / n * 4))) | |
ims.append(im) | |
count = 0 | |
count += 1 | |
im_ani = animation.ArtistAnimation(fig, ims, interval=50, blit=True) | |
plt.axis('equal') | |
plt.show() | |
if __name__ == '__main__': | |
main() |
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
package main | |
import ( | |
"fmt" | |
"math" | |
"math/rand" | |
"github.com/gonum/plot" | |
"github.com/gonum/plot/plotter" | |
"github.com/gonum/plot/plotutil" | |
"github.com/gonum/plot/vg" | |
) | |
type points [][]float64 | |
type point []float64 | |
func appendIfMissing(slice []float64, i float64) []float64 { | |
for _, ele := range slice { | |
if ele == i { | |
return slice | |
} | |
} | |
return append(slice, i) | |
} | |
func unique(array []float64) (rv []float64) { | |
for _, v := range array { | |
rv = appendIfMissing(rv, v) | |
} | |
return | |
} | |
func randUniform(low, high float64) float64 { | |
return rand.Float64()*(high-low) + low | |
} | |
func randsign() (sign float64) { | |
if rand.Float64() > 0.5 { | |
sign = 1 | |
} else { | |
sign = -1 | |
} | |
return | |
} | |
func ranpt(low, high point) (pt point) { | |
length := len(low) | |
pt = make([]float64, length) | |
for i := 0; i < length; i++ { | |
pt[i] = randUniform(low[i], high[i]) | |
} | |
return | |
} | |
func meanAndVariance(array []float64) (mean, variance float64) { | |
length := len(array) | |
if length == 0 { | |
return | |
} | |
var a2sum, asum float64 | |
for i := 0; i < length; i++ { | |
asum += array[i] | |
a2sum += array[i] * array[i] | |
} | |
mean = asum / float64(length) | |
variance = a2sum/float64(length) - mean*mean | |
return | |
} | |
func miser(f func(vals ...float64) float64, regn [2]point, N int, dith float64) (pts points, average, variance float64, rvN int) { | |
const ( | |
MNBS = 60 | |
MNPT = 15 | |
PFAC = 0.1 | |
BIG = math.MaxFloat64 | |
TINY = 1 / math.MaxFloat64 | |
) | |
low := regn[0] | |
high := regn[1] | |
length := len(low) | |
Npre := int(math.Max(float64(MNPT), float64(N)*PFAC)) | |
V := high[0] - low[0] | |
for i := 1; i < length; i++ { | |
V *= high[i] - low[i] | |
} | |
if N < MNBS { | |
pts = make(points, N) | |
fvals := make([]float64, N) | |
for i := 0; i < N; i++ { | |
pt := ranpt(low, high) | |
pts[i] = pt | |
fvals[i] = f(pt...) | |
} | |
average, variance = meanAndVariance(fvals) | |
variance = math.Max(TINY, variance) | |
rvN = N | |
return pts, average, variance, rvN | |
} | |
//else | |
mid := make(point, length) | |
for i := 0; i < length; i++ { | |
s := randsign() * dith | |
mid[i] = (0.5+s)*low[i] + (0.5-s)*high[i] | |
} | |
fvalsL := make([][]float64, length) | |
fvalsR := make([][]float64, length) | |
for i := 0; i < Npre; i++ { | |
pt := ranpt(low, high) | |
fval := f(pt...) | |
for j := 0; j < length; j++ { | |
if pt[j] <= mid[j] { | |
fvalsL[j] = append(fvalsL[j], fval) | |
} else { | |
fvalsR[j] = append(fvalsR[j], fval) | |
} | |
} | |
} | |
var ( | |
sigmaL, sigmaR, sigmaLB, sigmaRB float64 | |
NL, NR int | |
) | |
sumB := BIG | |
iB := -1 | |
sSums := make([]float64, length) | |
variancesL := make([]float64, length) | |
variancesR := make([]float64, length) | |
for i := 0; i < length; i++ { | |
_, variancesL[i] = meanAndVariance(fvalsL[i]) | |
_, variancesR[i] = meanAndVariance(fvalsR[i]) | |
sigmaL = math.Max(TINY, math.Pow(variancesL[i], 2.0/3)) | |
sigmaR = math.Max(TINY, math.Pow(variancesR[i], 2.0/3)) | |
sSums[i] = sigmaL + sigmaR | |
if sSums[i] <= sumB { | |
sumB = sSums[i] | |
iB = i | |
sigmaLB = sigmaL | |
sigmaRB = sigmaR | |
} | |
} | |
if iB == -1 || len(unique(sSums)) == 1 { | |
iB = int(randUniform(0, float64(length))) | |
} | |
regnL := low[iB] | |
regnMid := mid[iB] | |
regnR := high[iB] | |
fracL := math.Abs((regnMid - regnL) / (regnR - regnL)) | |
fracR := 1.0 - fracL | |
if sigmaLB > 0 || sigmaRB > 0 { | |
L := fracL * sigmaLB | |
R := fracR * sigmaRB | |
NL = MNPT + int(float64(N-Npre-2*MNPT)*L/(L+R)) | |
} else { | |
NL = MNPT + int((N-Npre-2*MNPT)/2) | |
} | |
NR = N - Npre - NL | |
lowTmp := make(point, length) | |
highTmp := make(point, length) | |
copy(lowTmp, low) | |
copy(highTmp, high) | |
highTmp[iB] = mid[iB] | |
ptsL, aveL, varL, NL := miser(f, [2]point{lowTmp, highTmp}, NL, dith) | |
lowTmp[iB] = mid[iB] | |
highTmp[iB] = high[iB] | |
ptsR, aveR, varR, NR := miser(f, [2]point{lowTmp, highTmp}, NR, dith) | |
pts = append(ptsL, ptsR...) | |
average = fracL*aveL + fracR*aveR | |
variance = fracL*fracL*varL + fracR*fracR*varR | |
rvN = NL + NR | |
return pts, average, variance, rvN | |
} | |
func f(vals ...float64) float64 { | |
x, y := vals[0], vals[1] | |
if math.Sqrt(x*x+y*y) < 1 { | |
return 1 | |
} | |
return 0 | |
} | |
func main() { | |
// for i := 3; i < 10; i++ { | |
// N := int(math.Pow10(i)) | |
// _, average, variance, rvN := miser(f, [2]point{point{0, 0}, point{1, 1}}, N, 0.1) | |
// fmt.Printf("%d %.6f %.4e %.4e %d %d\n",i, average*4, 4*math.Sqrt(variance)/math.Sqrt(float64(N)), math.Sqrt(variance), N, rvN) | |
// } | |
N := int(1e7) | |
pts, average, variance, rvN := miser(f, [2]point{point{0, 0}, point{1, 1}}, N, 0.1) | |
fmt.Printf("%.6f %.4e %.4e %d %d\n", average*4, 4*math.Sqrt(variance)/math.Sqrt(float64(N)), math.Sqrt(variance), N, rvN) | |
plt, _ := plot.New() | |
inner := make(plotter.XYs, len(pts)) | |
outter := make(plotter.XYs, len(pts)) | |
var innerN, outterN int | |
for _, v := range pts { | |
if f(v...) > 0 { | |
inner[innerN].X = v[0] | |
inner[innerN].Y = v[1] | |
innerN++ | |
} else { | |
outter[outterN].X = v[0] | |
outter[outterN].Y = v[1] | |
outterN++ | |
} | |
} | |
s1, _ := plotter.NewScatter(inner[:innerN]) | |
s2, _ := plotter.NewScatter(outter[:outterN]) | |
s1.Shape = plotutil.Shape(4) | |
s2.Shape = plotutil.Shape(4) | |
s1.Color = plotutil.Color(0) | |
s2.Color = plotutil.Color(1) | |
plt.Title.Text = "MISER Monte Carlo" | |
plt.X.Label.Text = "X" | |
plt.Y.Label.Text = "Y" | |
plt.Add(s1, s2) | |
plt.Save(4*vg.Inch, 4*vg.Inch, "distributions.png") | |
} |
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
package main | |
import ( | |
"fmt" | |
"math" | |
"math/rand" | |
) | |
func f(x float64) float64 { | |
return math.Sqrt(1 - x*x) | |
} | |
func main() { | |
var pn, x, fn, fn2sum, fnsum, fsum, fnave, fn2ave float64 | |
for j := 1; j < 10; j++ { | |
N := math.Pow10(j) | |
for i := pn; i < N; i++ { | |
x = rand.Float64() | |
fsum += f(x) | |
fn = fsum / (i + 1) | |
fnsum += fn | |
fn2sum += fn * fn | |
} | |
fnave = fnsum / N | |
fn2ave = fn2sum / N | |
sigmaN := math.Sqrt(fn2ave - fnave*fnave) | |
sigma := 4 * sigmaN / math.Sqrt(N) | |
pn = N | |
fmt.Printf("%d,%.4e,%.4e,%.6f,%.6f\n", j, sigma, sigmaN, 4*fn, math.Abs(4*fn-math.Pi)) | |
} | |
} |
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
# coding:utf-8 | |
from __future__ import division | |
import random | |
import numpy as np | |
import matplotlib.pyplot as plt | |
def dis(x, y): | |
return np.sqrt(x**2 + y**2) | |
def main(): | |
M = int(4e4) | |
pi_n = np.zeros(M) | |
count_inner = 0 | |
for n in range(M): | |
x = random.random() | |
y = random.random() | |
if dis(x, y) < 1: | |
count_inner += 1 | |
pi_n[n] = count_inner / (n + 1) * 4 | |
error = pi_n - np.pi | |
error = np.abs(error) | |
plt.plot(error) | |
plt.ylabel('Error') | |
plt.xlabel('n') | |
plt.show() | |
if __name__ == '__main__': | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
蒙特卡洛方法及应用 | linw1995