Scribd
Upload
16 views

Bayesian Statistics: Lecture 9 MCMC: - Markov Chains - Gibbs Sampler

This document describes using Markov chain Monte Carlo (MCMC) methods, specifically the Gibbs sampler, to estimate parameters from data that may contain outliers. It generates samples for the mean (mu) and variance (sigma2) from their posterior distributions given the data and assigns each data point to being inlier (r=1) or outlier (r=2) in each MCMC iteration. It runs for many iterations and estimates the probability each data point is an outlier based on the samples.

Uploaded by

muralidharan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
16 views

Bayesian Statistics: Lecture 9 MCMC: - Markov Chains - Gibbs Sampler

This document describes using Markov chain Monte Carlo (MCMC) methods, specifically the Gibbs sampler, to estimate parameters from data that may contain outliers. It generates samples for the mean (mu) and variance (sigma2) from their posterior distributions given the data and assigns each data point to being inlier (r=1) or outlier (r=2) in each MCMC iteration. It runs for many iterations and estimates the probability each data point is an outlier based on the samples.

Uploaded by

muralidharan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 20

Bayesian statistics: lecture 9

MCMC:
- Markov chains
- Gibbs sampler

% Data
y=[28 26 33 24 34 -44 27 16 40 -2 ...
29 22 24 21 25 30 23 29 31 19 ...
24 20 36 32 36 28 25 21 28 29 ...
37 25 28 26 30 32 36 26 30 22 ...
36 23 27 27 28 27 31 27 26 33 ...
26 32 32 24 39 28 24 25 32 25 ...
29 27 28 29 16 23];
e=0.05; % prior probability of outlier
ns=1000; % number of update steps
k=[1 10];
n=length(y);
r=ones(size(y)); mu=mean(y); sigma2=var(y); % initial
values
rand('state',0); randn('state',0); % random generator
seeds
% allocate memory for the samples
mu=repmat(mu,1,ns);
sigma2=repmat(sigma2,1,ns);
ris2=zeros(1,n); % count occurrences of r(i)=2

% simulation loop
for is=2:ns
% draw mu from normal
i1=find(r==1); i2=find(r==2);
n1=length(i1); n2=length(i2);
wn=sigma2(is-1)/(n1+n2/10);
mn=(sum(y(i1))/k(1)+sum(y(i2))/k(2))/(n1/k(1)+n2/k(2));
mu(is)=normrnd(mn,sqrt(wn));
% draw sigma2 from invgamma
ns02=sum((y(i1)-mu(is)).^2)/k(1)+sum((y(i2)-mu(is)).^2)/k(2);
tau=gamrnd(n/2,2/ns02);
sigma2(is)=1/tau;
% loop through r's index in random order
nn=randperm(n);
for in=1:n
ii=nn(in);
logp(1)=log(1-e)-0.5*log(k(1))-(y(ii)-mu(is))^2/sigma2(is)/2/k(1);
logp(2)=log(e) -0.5*log(k(2))-(y(ii)-mu(is))^2/sigma2(is)/2/k(2);
p=exp(logp-max(logp));
p=p/norm(p,1);
[notused,r(ii)]=histc(rand,[0 cumsum(p)]); % draw r
end
ris2=ris2+(r==2);
end
% plot Prob(r(i)==2)
subplot(2,1,1)
stem(ris2/ns,'markerfacecolor','blue','markersize',4)
box off, title('Prob(r_i=2)'), xlabel('i')
axis([0 n 0 1])
mu_stats=[mean(mu) std(mu)]
sigma2_stats=[mean(sigma2) std(sigma2)]

You might also like