Statistics/Normal distribution: Difference between revisions
Content added Content deleted
m (→{{header|Perl 6}}: omit needless parens) |
(→{{header|PARI/GP}}: 2.4.3+) |
||
| Line 189: | Line 189: | ||
=={{header|PARI/GP}}== |
=={{header|PARI/GP}}== |
||
{{works with|PARI/GP|2.4.3 and above}} |
|||
<lang parigp>rnormal()={ |
<lang parigp>rnormal()={ |
||
my( |
my(u1=random(1.),u2=random(1.); |
||
sqrt(-2*log(u1))*cos(2*Pi*u1) |
sqrt(-2*log(u1))*cos(2*Pi*u1) |
||
\\ Could easily be extended with a second normal at very little cost. |
\\ Could easily be extended with a second normal at very little cost. |
||
| Line 212: | Line 213: | ||
}; |
}; |
||
show(10^4)</lang> |
show(10^4)</lang> |
||
For versions before 2.4.3, define |
|||
<lang parigp>rreal()={ |
|||
my(pr=32*ceil(default(realprecision)*log(10)/log(4294967296))); \\ Current precision |
|||
random(2^pr)*1.>>pr |
|||
};</lang> |
|||
and use <code>rreal()</code> in place of <code>random(1.)</code>. |
|||
=={{header|Perl 6}}== |
=={{header|Perl 6}}== |
||
<lang perl6>constant τ = 2 * pi; |
<lang perl6>constant τ = 2 * pi; |
||
Revision as of 16:06, 18 April 2012
Statistics/Normal distribution is a draft programming task. It is not yet considered ready to be promoted as a complete task, for reasons that should be found in its talk page.
The Normal (or Gaussian) distribution is a freqently used distribution in statistics. While most programming languages provide a uniformly distributed random number generator, one can derive normally distributed random numbers from a uniform generator.
- The task
- Take a uniform random number generator and create a large (you decide how large) set of numbers that follow a normal (Gaussian) distribution. Calculate the dataset's mean and stddev, and show the histogram here.
- Mention any native language support for the generation of normally distributed random numbers.
- Reference
- You may refer to code in Statistics/Basic if available.
Go
Box-Muller method shown here. Go has a normally distributed random function in the standard library, as shown in the Go Random numbers solution. It uses the ziggurat method. <lang go>package main
import (
"fmt" "math" "math/rand" "strings"
)
// Box-Muller func norm2() (s, c float64) {
r := math.Sqrt(-2 * math.Log(rand.Float64())) s, c = math.Sincos(2 * math.Pi * rand.Float64()) return s * r, c * r
}
func main() {
const (
n = 10000
bins = 12
sig = 3
scale = 100
)
var sum, sumSq float64
h := make([]int, bins)
for i, accum := 0, func(v float64) {
sum += v
sumSq += v * v
b := int((v + sig) * bins / sig / 2)
if b >= 0 && b < bins {
h[b]++
}
}; i < n/2; i++ {
v1, v2 := norm2()
accum(v1)
accum(v2)
}
m := sum / n
fmt.Println("mean:", m)
fmt.Println("stddev:", math.Sqrt(sumSq/float64(n)-m*m))
for _, p := range h {
fmt.Println(strings.Repeat("*", p/scale))
}
}</lang> Output:
mean: -0.0034970888831523488 stddev: 1.0040682925006286 * **** ********* *************** ******************* ****************** ************** ********* **** *
Mathematica
<lang Mathematica>x:= RandomReal[1] SampleNormal[n_] := (Print[#//Length, " numbers, Mean : ", #//Mean, ", StandardDeviation : ", #//StandardDeviation];
Histogram[#, BarOrigin -> Left,Axes -> False])& [(Table[(-2*Log[x])^0.5*Cos[2*Pi*x], {n} ]]
Invocation:
SampleNormal[ 10000 ]
->10000 numbers, Mean : -0.0122308, StandardDeviation : 1.00646
</lang>
MATLAB / Octave
<lang Matlab> N = 100000;
x = randn(N,1); mean(x) std(x) [nn,xx] = hist(x,100); bar(xx,nn);</lang>
Liberty BASIC
Uses LB Statistics/Basic <lang lb>call sample 100000
end
sub sample n
dim dat( n)
for i =1 to n
dat( i) =normalDist( 1, 0.2)
next i
'// show mean, standard deviation. Find max, min.
mx =-1000
mn = 1000
sum =0
sSq =0
for i =1 to n
d =dat( i)
mx =max( mx, d)
mn =min( mn, d)
sum =sum +d
sSq =sSq +d^2
next i
print n; " data terms used."
mean =sum / n print "Largest term was "; mx; " & smallest was "; mn range =mx -mn print "Mean ="; mean
print "Stddev ="; ( sSq /n -mean^2)^0.5
'// show histogram
nBins =50
dim bins( nBins)
for i =1 to n
z =int( ( dat( i) -mn) /range *nBins)
bins( z) =bins( z) +1
next i
for b =0 to nBins -1
for j =1 to int( nBins *bins( b)) /n *30)
print "#";
next j
print
next b
print
end sub
function normalDist( m, s) ' Box Muller method
u =rnd( 1) v =rnd( 1) normalDist =( -2 *log( u))^0.5 *cos( 2 *3.14159265 *v)
end function</lang>
100000 data terms used. Largest term was 4.12950792 & smallest was -4.37934139 Mean =-0.26785425e-2 Stddev =1.00097669
# ## ### ##### ######## ############ ################ ######################## ############################## ###################################### ############################################## ######################################################## ################################################################### ############################################################################## ####################################################################################### ################################################################################################ #################################################################################################### ######################################################################################################## ##################################################################################################### ############################################################################################## ######################################################################################### ################################################################################## ######################################################################### ############################################################## #################################################### ########################################## ################################# ########################## ################## ############# ######### ###### #### ## # #
PARI/GP
<lang parigp>rnormal()={ my(u1=random(1.),u2=random(1.); sqrt(-2*log(u1))*cos(2*Pi*u1) \\ Could easily be extended with a second normal at very little cost. }; mean(v)={
sum(i=1,#v,v[i])/#v
}; stdev(v,mu="")={
if(mu=="",mu=mean(v)); sqrt(sum(i=1,#v,(v[i]-mu)^2))/#v
}; histogram(v,bins=16,low=0,high=1)={
my(u=vector(bins),width=(high-low)/bins); for(i=1,#v,u[(v[i]-low)\width+1]++); u
}; show(n)={
my(v=vector(n,i,rnormal()),m=mean(v),s=stdev(v,m),h,sz=ceil(n/300));
h=histogram(v,,vecmin(v)-.1,vecmax(v)+.1);
for(i=1,#h,for(j=1,h[i]\sz,print1("#"));print());
}; show(10^4)</lang>
For versions before 2.4.3, define <lang parigp>rreal()={
my(pr=32*ceil(default(realprecision)*log(10)/log(4294967296))); \\ Current precision random(2^pr)*1.>>pr
};</lang>
and use rreal() in place of random(1.).
Perl 6
<lang perl6>constant τ = 2 * pi;
sub normdist ($m, $σ) {
gather loop {
my $r = sqrt -2 * log rand;
my $Θ = τ * rand;
take $r * $_($Θ) * $σ + $m for &cos, &sin;
}
}
sub MAIN ($size = 100000, $mean = 50, $stddev = 4) {
my @dataset = normdist($mean,$stddev)[^$size];
my $m = [+](@dataset) / $size; say (:$m);
my $σ = sqrt [+](@dataset X** 2) / $size - $mean**2; say (:$σ);
(my %hash){.round}++ for @dataset;
my $scale = 180 * $stddev / $size;
constant @subbar = < ⎸ ▏ ▎ ▍ ▌ ▋ ▊ ▉ █ >;
for %hash.keys».Int.minmax(+*) -> $i {
my $x = (%hash{$i} // 0) * $scale;
my $full = floor $x;
my $part = 8 * ($x - $full);
say $i, "\t", '█' x $full, @subbar[$part];
}
}</lang>
- Output:
"m" => 50.006107405837142e0 "σ" => 4.0814435639885254e0 33 ⎸ 34 ⎸ 35 ⎸ 36 ▏ 37 ▎ 38 ▊ 39 █▋ 40 ███⎸ 41 █████▊ 42 ██████████⎸ 43 ███████████████▋ 44 ███████████████████████▏ 45 ████████████████████████████████▌ 46 ███████████████████████████████████████████▍ 47 ██████████████████████████████████████████████████████▏ 48 ███████████████████████████████████████████████████████████████▏ 49 █████████████████████████████████████████████████████████████████████▋ 50 ███████████████████████████████████████████████████████████████████████▊ 51 █████████████████████████████████████████████████████████████████████▌ 52 ███████████████████████████████████████████████████████████████⎸ 53 ██████████████████████████████████████████████████████▎ 54 ███████████████████████████████████████████⎸ 55 ████████████████████████████████▌ 56 ███████████████████████▍ 57 ███████████████▉ 58 █████████▉ 59 █████▍ 60 ███▍ 61 █▋ 62 ▊ 63 ▍ 64 ▏ 65 ⎸ 66 ⎸ 67 ⎸
PureBasic
<lang purebasic>Procedure.f randomf(resolution = 2147483647)
ProcedureReturn Random(resolution) / resolution
EndProcedure
Procedure.f normalDist() ;Box Muller method
ProcedureReturn Sqr(-2 * Log(randomf())) * Cos(2 * #PI * randomf())
EndProcedure
Procedure sample(n, nBins = 50)
Protected i, maxBinValue, binNumber
Protected.f d, mean, sum, sumSq, mx, mn, range
Dim dat.f(n)
For i = 1 To n
dat(i) = normalDist()
Next
;show mean, standard deviation, find max & min.
mx = -1000
mn = 1000
sum = 0
sumSq = 0
For i = 1 To n
d = dat(i)
If d > mx: mx = d: EndIf
If d < mn: mn = d: EndIf
sum + d
sumSq + d * d
Next
PrintN(Str(n) + " data terms used.")
PrintN("Largest term was " + StrF(mx) + " & smallest was " + StrF(mn))
mean = sum / n
PrintN("Mean = " + StrF(mean))
PrintN("Stddev = " + StrF((sumSq / n) - Sqr(mean * mean)))
;show histogram
range = mx - mn
Dim bins(nBins)
For i = 1 To n
binNumber = Int(nBins * (dat(i) - mn) / range)
bins(binNumber) + 1
Next
maxBinValue = 1
For i = 0 To nBins
If bins(i) > maxBinValue
maxBinValue = bins(i)
EndIf
Next
#normalizedMaxValue = 70
For binNumber = 0 To nBins
tickMarks = Round(bins(binNumber) * #normalizedMaxValue / maxBinValue, #PB_Round_Nearest)
PrintN(ReplaceString(Space(tickMarks), " ", "#"))
Next
PrintN("")
EndProcedure
If OpenConsole()
sample(100000) Print(#CRLF$ + #CRLF$ + "Press ENTER to exit"): Input() CloseConsole()
EndIf</lang> Sample output:
100000 data terms used. Largest term was 4.5352029800 & smallest was -4.5405135155 Mean = 0.0012346541 Stddev = 0.9959455132 # ### ###### ########## ################## ############################ ######################################### ##################################################### ################################################################ ###################################################################### ###################################################################### ################################################################ ##################################################### ######################################### ############################# ################## ########## ###### ### #
SAS
<lang sas>data test; n=100000; twopi=2*constant('pi'); do i=1 to n; u=ranuni(0); v=ranuni(0); r=sqrt(-2*log(u)); x=r*cos(twopi*v); y=r*sin(twopi*v); z=rannor(0); output; end; keep x y z;
proc means mean stddev;
proc univariate; histogram /normal;
run;
/* Variable Mean Std Dev
x -0.0052720 0.9988467 y 0.000023995 1.0019996 z 0.0012857 1.0056536
- /</lang>
Tcl
<lang tcl>package require Tcl 8.5
- Uses the Box-Muller transform to compute a pair of normal random numbers
proc tcl::mathfunc::nrand {mean stddev} {
variable savednormalrandom
if {[info exists savednormalrandom]} {
return [expr {$savednormalrandom*$stddev + $mean}][unset savednormalrandom]
}
set r [expr {sqrt(-2*log(rand()))}]
set theta [expr {2*3.1415927*rand()}]
set savednormalrandom [expr {$r*sin($theta)}]
expr {$r*cos($theta)*$stddev + $mean}
} proc stats {size {slotfactor 10}} {
set sum 0.0
set sum2 0.0
for {set i 0} {$i < $size} {incr i} {
set r [expr { nrand(0.5, 0.2) }]
incr histo([expr {int(floor($r*$slotfactor))}]) set sum [expr {$sum + $r}] set sum2 [expr {$sum2 + $r**2}]
}
set mean [expr {$sum / $size}]
set stddev [expr {sqrt($sum2/$size - $mean**2)}]
puts "$size numbers"
puts "Mean: $mean"
puts "StdDev: $stddev"
foreach i [lsort -integer [array names histo]] {
puts [string repeat "*" [expr {$histo($i)*350/int($size)}]]
}
}
stats 100 puts "" stats 1000 puts "" stats 10000 puts "" stats 100000 20</lang> Sample output:
100 numbers Mean: 0.49355955990390254 StdDev: 0.19651396178121985 *** ******* ************** *********************************** ******************************************************** ****************************************************************** ************************************************************************* ****************************************** ************************************** ************** 1000 numbers Mean: 0.5066940614105869 StdDev: 0.2016794788065389 * ***** ************** **************************** ********************************************************** **************************************************************** ************************************************************* ****************************************************** *********************************** ************ ********* * 10000 numbers Mean: 0.49980964730768285 StdDev: 0.1968441612522318 * ***** *************** ******************************* ***************************************************** ****************************************************************** ******************************************************************* **************************************************** ********************************* *************** ***** * 100000 numbers Mean: 0.49960438950922254 StdDev: 0.20060211160998606 * ** *** ****** ********* ************** ****************** *********************** ***************************** ******************************** ********************************** ********************************** ******************************** **************************** *********************** ****************** ************* ********* ****** *** ** *
The blank lines in the output are where the number of samples is too small to even merit a single unit on the histogram.