Dotnet/FastRng
Dotnet/FastRng
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases 5 Wiki Activity

Refactored to use a shape fitter

Browse Source
This commit is contained in:
Thorsten Sommer 2020-09-29 20:30:16 +02:00
parent f69330d79e
commit 25c773b411
1 changed files with 20 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
FastRng/Double/Distributions/Normal.cs
View File

@ -1,4 +1,5 @@
using System;
using System.Runtime.CompilerServices;
using System.Threading;
using System.Threading.Tasks;
@ -6,42 +7,38 @@ namespace FastRng.Double.Distributions
{
public sealed class Normal : IDistribution
{
private double standardDeviation = 1.0;
private const double SQRT_2PI = 2.506628275;
private const double STDDEV = 0.4;
private const double MEAN = 0.5;
public IRandom Random { get; set; }
private ShapeFitter fitter;
private IRandom random;
public double Mean { get; set; } = 0.0;
public double StandardDeviation
public Normal()
{
get => this.standardDeviation;
}
public IRandom Random
{
get => this.random;
set
{
if(value <= 0.0)
throw new ArgumentOutOfRangeException(message: "Standard deviation must be greater than 0", null);
this.standardDeviation = value;
this.random = value;
this.fitter = new ShapeFitter(ShapeFunction, this.random, 100, 0.99);
}
}
private static double ShapeFunction(double x)
{
return 1.0 / (STDDEV * SQRT_2PI) * Math.Exp(-Math.Pow((x - MEAN) / STDDEV, 2.0));
}
public async ValueTask<double> GetDistributedValue(CancellationToken token = default)
{
if (this.Random == null)
return double.NaN;
//
// Previously:
//
// var u1 = await this.Random.GetUniform(token);
// var u2 = await this.Random.GetUniform(token);
// var r = Math.Sqrt(-2.0 * Math.Log(u1));
// var theta = 2.0 * Math.PI * u2;
// var value = r * Math.Sin(theta);
// return this.Mean + this.StandardDeviation * value;
const double SQRT_2PI = 2.506628275;
var x = await this.Random.GetUniform(token); // BUG: It seems, that uniform is not uniform (enough) or RunningStatistics had specific issues. Test, if Math.NET's uniform is better.
return 1.0 / (this.StandardDeviation * SQRT_2PI) * Math.Exp(-0.5 * Math.Pow((x - this.Mean) / this.StandardDeviation, 2.0));
return await this.fitter.NextNumber(); // TODO: Add token!
}
}
}