Metropolis-Hastings algorithm
Metropolis-Hastings algorithm
Metropolis-Hastings algorithm
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Metropolis</strong>-<strong>Hastings</strong> <strong>algorithm</strong><br />
• Note: target distribution ib i does not need to be normalized.<br />
– Based on ratio: P(θ proposed ) / P(θ current ) .<br />
– Useful when target distribution P(θ) is a posterior distribution<br />
p D| p .<br />
proportional to <br />
pD|<br />
p<br />
<br />
Bayes’ rule:<br />
p<br />
<br />
|<br />
D<br />
<br />
p D<br />
<br />
p | D<br />
pD|<br />
p<br />
<br />
– Need only evaluate the product p D| p , not the separate<br />
likelihoods and priors.<br />
– By evaluating p D<br />
| <br />
p<br />
<br />
,<br />
can generate random representative<br />
values from the posterior distribution.<br />
– Don’t need to evaluate evidence, pD.<br />
• Can do Baysian inference with rich and complex models.