Markov Chain Monte Carlo Simulation using C# and MathNet
Math.Net Numerics has capability to conduct Markov Chair Monte Carlo simulations, yet the document is very sparse. The only examples I found are in F# (see below). In this note, I attempt to port these examples into C# and hope others may find it useful in their research. Note that there are some errors in the original F# code, and this note corrected them. The ported code has been published as ASP.NET web services at x.ecoruse.org. Thus, one can easily copied over to any windows or web development projects.
Ported C# Code
using System;
using System.Collections.Generic;
using System.Web;
using System.Web.Services;
using MathNet.Numerics.Distributions;
using MathNet.Numerics.Statistics;
using MathNet.Numerics.Random;
using MathNet.Numerics.Statistics.Mcmc; [WebService(Namespace = "http://x.ecourse.org")]
[WebServiceBinding(ConformsTo = WsiProfiles.BasicProfile1_1)]
[System.Web.Script.Services.ScriptService]
public class MCMC : System.Web.Services.WebService { public MCMC () {} [WebMethod(Description = "Sampling Beta variable via rejection")]
public double[] BetaViaRejection(double a, double b, int N) {
var rnd = new MersenneTwister();
var beta = new Beta(a, b);
var uniform = new ContinuousUniform(0.0, 1.0, rnd);
var rs = new RejectionSampler(
(x => Math.Pow(x, beta.A - 1.0) * Math.Pow(1.0 - x, beta.B - 1.0)),
(x => 0.021), (() => uniform.Sample()));
var arr= rs.Sample(N);
return arr;
//string result = "Theoretical Mean:" + beta.Mean + " vs Sample Mean: "
+ Statistics.Mean(arr) + ";" + "Theoretical StarndardDeviation:"
+ beta.StdDev + " vs Sample StandardDeviation: "
+ Statistics.StandardDeviation(arr) + ";" + " Acceptance Rate:" + rs.AcceptanceRate;
//return result;
} [WebMethod(Description = "Sampling a normal variable via Metropolis")]
public double[] NormaViaMetropolis(double mean, double stdev, int N)
{
var rnd = new MersenneTwister();
var normal = new Normal(mean, stdev); var ms = new MetropolisHastingsSampler(0.1, x => Math.Log(normal.Density(x)),
(x,y)=>Normal.PDFLn(x,0.3,y),
x => Normal.Sample(rnd, x, 0.3), 20); var arr = ms.Sample(N);
return arr;
//string result = "Theoretical Mean:" + beta.Mean + " vs Sample Mean: "
+ Statistics.Mean(arr) + ";" + "Theoretical StarndardDeviation:"
+ beta.StdDev + " vs Sample StandardDeviation: "
+ Statistics.StandardDeviation(arr) + ";" + " Acceptance Rate:" + rs.AcceptanceRate;
//return result;
} [WebMethod(Description = "Sampling a normal variable via Metropolis symmetric proposal")]
public double[] NormaViaMetropolisSymmetricProposal(double mean, double stdev, int N)
{
var rnd = new MersenneTwister();
var normal = new Normal(mean, stdev); var ms = new MetropolisHastingsSampler(0.1, x => Math.Log(normal.Density(x)),
(x,y) => npdf(x,y,03),
x => Normal.Sample(rnd,x,0.3), 10); var arr = ms.Sample(N);
return arr;
//string result = "Theoretical Mean:" + beta.Mean + " vs Sample Mean: "
+ Statistics.Mean(arr) + ";" + "Theoretical StarndardDeviation:"
+ beta.StdDev + " vs Sample StandardDeviation: "
+ Statistics.StandardDeviation(arr) + ";" + " Acceptance Rate:" + rs.AcceptanceRate;
//return result;
} [WebMethod(Description = "Sampling a normal variable via Metropolis asymmetric proposal")]
public double[] NormaViaMetropolisAsymmetricProposal(double mean, double stdev, int N)
{
var rnd = new MersenneTwister();
var normal = new Normal(mean, stdev); var ms = new MetropolisHastingsSampler(0.1, x => Math.Log(normal.Density(x)),
(xnew, x) => Math.Log(0.5 * Math.Exp(npdf(xnew,x, 0.3))
+ 0.5 * Math.Exp(npdf(xnew, x+0.1, 0.3))),
x => MixSample(x), 10); var arr = ms.Sample(N);
return arr;
//string result = "Theoretical Mean:" + beta.Mean + " vs Sample Mean: "
+ Statistics.Mean(arr) + ";" + "Theoretical StarndardDeviation:"
+ beta.StdDev + " vs Sample StandardDeviation: "
+ Statistics.StandardDeviation(arr) + ";" + " Acceptance Rate:" + rs.AcceptanceRate;
//return result;
} [WebMethod(Description = "Slice sampling a normal distributed random variable")]
public double[] NormaViaSliceSampling(double mean, double stdev, int N)
{
var rnd = new MersenneTwister();
var normal = new Normal(mean, stdev); var ms = new UnivariateSliceSampler(0.1, x => npdfNoNormalized(x, mean, stdev), 5, 1.0);
var arr = ms.Sample(N);
return arr;
//string result = "Theoretical Mean:" + beta.Mean + " vs Sample Mean: "
+ Statistics.Mean(arr) + ";" + "Theoretical StarndardDeviation:"
+ beta.StdDev + " vs Sample StandardDeviation: "
+ Statistics.StandardDeviation(arr) + ";" + " Acceptance Rate:" + rs.AcceptanceRate;
//return result;
} public double npdf(double x, double m, double s)
{
return -0.5 * (x - m) * (x - m) / (s * s) - 0.5 * Math.Log(2.0 * System.Math.PI * s * s);
} public double npdfNoNormalized(double x, double m, double s)
{
return -0.5 * (x - m) * (x - m) / (s * s);
} public double MixSample(double x)
{
var rnd = new MersenneTwister();
if (Bernoulli.Sample(rnd, 0.5) == 1)
return Normal.Sample(rnd, x, 0.3);
else
return Normal.Sample(rnd, x + 0.1, 0.3);
}
}
Original F# Code
/
// Math.NET Numerics, part of the Math.NET Project
// http://numerics.mathdotnet.com
// http://github.com/mathnet/mathnet-numerics
// http://mathnetnumerics.codeplex.com
//
// Copyright (c) 2009-2013 Math.NET
//
// Permission is hereby granted, free of charge, to any person
// obtaining a copy of this software and associated documentation
// files (the "Software"), to deal in the Software without
// restriction, including without limitation the rights to use,
// copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the
// Software is furnished to do so, subject to the following
// conditions:
//
// The above copyright notice and this permission notice shall be
// included in all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
// OTHER DEALINGS IN THE SOFTWARE.
//
#r "../../out/lib/Net40/MathNet.Numerics.dll"
#r "../../out/lib/Net40/MathNet.Numerics.FSharp.dll"
open MathNet.Numerics
open MathNet.Numerics.Random
open MathNet.Numerics.Statistics
open MathNet.Numerics.Distributions
open MathNet.Numerics.Statistics.Mcmc
/// The number of samples to gather for each sampler.
let N = 10000
/// The random number generator we use for the examples.
let rnd = new MersenneTwister()
//
// Example 1: Sampling a Beta distributed variable through rejection sampling.
//
// Target Distribution: Beta(2.7, 6.3)
//
// -----------------------------------------------------------------------------
do
printfn "Rejection Sampling Example"
/// The target distribution.
let beta = new Beta(2.7, 6.3)
/// Samples uniform distributed variables.
let uniform = new ContinuousUniform(0.0, 1.0, RandomSource = rnd)
/// Implements the rejection sampling procedure.
let rs = new RejectionSampler( ( fun x -> x**(beta.A-1.0) * (1.0 - x)**(beta.B-1.0) ),
( fun x -> 0.021 ),
( fun () -> uniform.Sample()) )
/// An array of samples from the rejection sampler.
let arr = rs.Sample(N)
/// The true distribution.
printfn "\tEmpirical Mean = %f (should be %f)" (Statistics.Mean(arr)) beta.Mean
printfn "\tEmpirical StdDev = %f (should be %f)" (Statistics.StandardDeviation(arr)) beta.StdDev
printfn "\tAcceptance rate = %f" rs.AcceptanceRate
printfn ""
//
// Example 2: Sampling a normal distributed variable through Metropolis sampling.
//
// Target Distribution: Normal(1.0, 3.5)
//
// -----------------------------------------------------------------------------
do
printfn "Metropolis Sampling Example"
let mean, stddev = 1.0, 3.5
let normal = new Normal(mean, stddev)
/// Implements the rejection sampling procedure.
let ms = new MetropolisSampler( 0.1, (fun x -> log(normal.Density(x))),
(fun x -> Normal.Sample(rnd, x, 0.3)), 20,
RandomSource = rnd )
/// An array of samples from the rejection sampler.
let arr = ms.Sample(N)
/// The true distribution.
printfn "\tEmpirical Mean = %f (should be %f)" (Statistics.Mean(arr)) normal.Mean
printfn "\tEmpirical StdDev = %f (should be %f)" (Statistics.StandardDeviation(arr)) normal.StdDev
printfn "\tAcceptance rate = %f" ms.AcceptanceRate
printfn ""
//
// Example 3: Sampling a normal distributed variable through Metropolis-Hastings sampling
// with a symmetric proposal distribution.
//
// Target Distribution: Normal(1.0, 3.5)
//
// -----------------------------------------------------------------------------------------
do
printfn "Metropolis Hastings Sampling Example (Symmetric Proposal)"
let mean, stddev = 1.0, 3.5
let normal = new Normal(mean, stddev)
/// Evaluates the log normal distribution.
let npdf x m s = -0.5*(x-m)*(x-m)/(s*s) - 0.5 * log(Constants.Pi2 * s * s)
/// Implements the rejection sampling procedure.
let ms = new MetropolisHastingsSampler( 0.1, (fun x -> log(normal.Density(x))),
(fun x y -> npdf x y 0.3), (fun x -> Normal.Sample(rnd, x, 0.3)), 10,
RandomSource = rnd )
/// An array of samples from the rejection sampler.
let arr = ms.Sample(N)
/// The true distribution.
printfn "\tEmpirical Mean = %f (should be %f)" (Statistics.Mean(arr)) normal.Mean
printfn "\tEmpirical StdDev = %f (should be %f)" (Statistics.StandardDeviation(arr)) normal.StdDev
printfn "\tAcceptance rate = %f" ms.AcceptanceRate
printfn ""
//
// Example 4: Sampling a normal distributed variable through Metropolis-Hastings sampling
// with a asymmetric proposal distribution.
//
// Target Distribution: Normal(1.0, 3.5)
//
// -----------------------------------------------------------------------------------------
do
printfn "Metropolis Hastings Sampling Example (Assymetric Proposal)"
let mean, stddev = 1.0, 3.5
let normal = new Normal(mean, stddev)
/// Evaluates the logarithm of the normal distribution function.
let npdf x m s = -0.5*(x-m)*(x-m)/(s*s) - 0.5 * log(Constants.Pi2 * s * s)
/// Samples from a mixture that is biased towards samples larger than x.
let mixSample x =
if Bernoulli.Sample(rnd, 0.5) = 1 then
Normal.Sample(rnd, x, 0.3)
else
Normal.Sample(rnd, x + 0.1, 0.3)
/// The transition kernel for the proposal above.
let krnl xnew x = log (0.5 * exp(npdf xnew x 0.3) + 0.5 * exp(npdf xnew (x+0.1) 0.3))
/// Implements the rejection sampling procedure.
let ms = new MetropolisHastingsSampler( 0.1, (fun x -> log(normal.Density(x))),
(fun xnew x -> krnl xnew x), (fun x -> mixSample x), 10,
RandomSource = rnd )
/// An array of samples from the rejection sampler.
let arr = ms.Sample(N)
/// The true distribution.
printfn "\tEmpirical Mean = %f (should be %f)" (Statistics.Mean(arr)) normal.Mean
printfn "\tEmpirical StdDev = %f (should be %f)" (Statistics.StandardDeviation(arr)) normal.StdDev
printfn "\tAcceptance rate = %f" ms.AcceptanceRate
printfn ""
//
// Example 5: Slice sampling a normal distributed random variable.
//
// Target Distribution: Normal(1.0, 3.5)
//
// -----------------------------------------------------------------------------------------
do
printfn "Slice Sampling Example"
let mean, stddev = 1.0, 3.5
let normal = new Normal(mean, stddev)
/// Evaluates the unnormalized logarithm of the normal distribution function.
let npdf x m s = -0.5*(x-m)*(x-m)/(s*s)
/// Implements the rejection sampling procedure.
let ms = new UnivariateSliceSampler( 0.1, (fun x -> npdf x mean stddev), 5, 1.0, RandomSource = rnd )
/// An array of samples from the rejection sampler.
let arr = ms.Sample(N)
/// The true distribution.
printfn "\tEmpirical Mean = %f (should be %f)" (Statistics.Mean(arr)) normal.Mean
printfn "\tEmpirical StdDev = %f (should be %f)" (Statistics.StandardDeviation(arr)) normal.StdDev
printfn ""
Markov Chain Monte Carlo Simulation using C# and MathNet的更多相关文章
- PRML读书会第十一章 Sampling Methods(MCMC, Markov Chain Monte Carlo,细致平稳条件,Metropolis-Hastings,Gibbs Sampling,Slice Sampling,Hamiltonian MCMC)
主讲人 网络上的尼采 (新浪微博: @Nietzsche_复杂网络机器学习) 网络上的尼采(813394698) 9:05:00 今天的主要内容:Markov Chain Monte Carlo,M ...
- (转)Markov Chain Monte Carlo
Nice R Code Punning code better since 2013 RSS Blog Archives Guides Modules About Markov Chain Monte ...
- 马尔科夫链蒙特卡洛(Markov chain Monte Carlo)
(学习这部分内容大约需要1.3小时) 摘要 马尔科夫链蒙特卡洛(Markov chain Monte Carlo, MCMC) 是一类近似采样算法. 它通过一条拥有稳态分布 \(p\) 的马尔科夫链对 ...
- [Bayes] MCMC (Markov Chain Monte Carlo)
不错的文章:LDA-math-MCMC 和 Gibbs Sampling 可作为精进MCMC抽样方法的学习材料. 简单概率分布的模拟 Box-Muller变换原理详解 本质上来说,计算机只能生产符合均 ...
- 为什么要用Markov chain Monte Carlo (MCMC)
马尔科夫链的蒙特卡洛采样的核心思想是构造一个Markov chain,使得从任意一个状态采样开始,按该Markov chain转移,经过一段时间的采样,逼近平稳分布stationary distrib ...
- 蒙特卡洛模拟(Monte Carlo simulation)
1.蒙特卡罗模拟简介 蒙特卡罗模拟,也叫统计模拟,这个术语是二战时期美国物理学家Metropolis执行曼哈顿计划的过程中提出来的,其基本思想很早以前就被人们所发现和利用.早在17世纪,人们就知道用事 ...
- History of Monte Carlo Methods - Part 1
History of Monte Carlo Methods - Part 1 Some time ago in June 2013 I gave a lab tutorial on Monte Ca ...
- Monte Carlo Approximations
准备总结几篇关于 Markov Chain Monte Carlo 的笔记. 本系列笔记主要译自A Gentle Introduction to Markov Chain Monte Carlo (M ...
- Introduction To Monte Carlo Methods
Introduction To Monte Carlo Methods I’m going to keep this tutorial light on math, because the goal ...
随机推荐
- Spring Data Jpa (三)定义查询方法
本章详细讲解如何利用方法名定义查询方法(Defining Query Methods) (1)定义查询方法的配置方法 由于Spring JPA Repository的实现原理是采用动态代理的机制,所以 ...
- java复制项目中的补丁,完整的包路径
package com.bytter.audit.iface.util; import java.io.BufferedInputStream; import java.io.BufferedOutp ...
- Flume的断点续传解决
根据需求,首先定义以下3大要素 采集源,即source——监控文件内容更新 : exec ‘tail -F file’ 下沉目标,即sink——HDFS文件系统 : hdfs sink Sou ...
- SoapUI Pro官网原包百度云盘分享
SoapUI Pro下载是件很痛苦的事,经常断网,或者是下载时间过长,这里分享的是截止2019.01.01 最新的安装原包. 百度云盘资源:https://pan.baidu.com/s/1SXTFs ...
- 箭头函数详解()=>{}
摘要:箭头函数有几个使用注意点. (1)函数体内的this对象,就是定义时所在的对象,而不是使用时所在的对象,箭头函数继承而来的this指向永远不变. (2)不可以当作构造函数,也就是说,不可以使用n ...
- android 3.0 ationbar使用总结
1,ationbar的基本讲解 http://www.apkbus.com/forum.php?mod=viewthread&tid=125536 仅仅需要根据需求写出一个menu资源文件 2 ...
- Android NDK的生命周期JNI_OnLoad与JNI_OnUnload(转)
摘要 NDK的生命周期 //当动态库被加载时这个函数被系统调用 JNIEXPORT jint JNICALL JNI_OnLoad(JavaVM *vm, void *reserved) { LOGI ...
- oauth2.0协议原理
OAuth的授权不会使用第三方触及到用户的帐号信息(如用户密码),即第三方无需使用用户的用户名与密码就可以申请获得该用户资源的授权,因此OAuth是安全的. OAuth的作用:就是让“客户端”安全可控 ...
- Python基本语法_集合set/frozenset_内建方法详解
目录 目录 前言 软件环境 可变集合Set set函数创建集合 创建空集合 集合元素的唯一性 集合推导式 set类型对象的内置方法 add增加一个元素 remove删除一个元素 pop随机删除并返回一 ...
- 网页上预览pdf文件的几种方案
网页上查看pdf的方案: 1.使用adobe reader的插件 2.使用在线office控件 3.使用火狐开源项目pdf.js(浏览器需支持html5) 4.将pdf转换为swf文件 5.使用pdf ...