21.01.2015 Views

Metropolis-Hastings algorithm

Metropolis-Hastings algorithm

Metropolis-Hastings algorithm

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Efficiency, “burn-in”, and convergence<br />

• If proposal ldistribution ib i is narrow relative to target distribution ib i ,<br />

will take a long time for random walk to cover the distribution.<br />

– Algorithm will not be efficient: takes too many steps to accumulate<br />

a representative sample.<br />

– Particular problem if initial position of random walk is in region of<br />

target distribution that is flat and low.<br />

• Random walk moves only slowly away from starting position<br />

into denser region.<br />

– Unrepresentative starting gposition can lead to low efficiency even<br />

if proposal distribution isn’t narrow.<br />

• Solution: early steps of random walk are excluded from portion<br />

of Markov chain considered to be representative of target<br />

distribution.<br />

– Excluded initial steps: burn-in period.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!