{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

L09_-_MCMC

# L09_-_MCMC - 6.046 Design and Analysis of Algorithms...

This preview shows pages 1–8. Sign up to view the full content.

6.046- Design’and’Analysis’of’ Algorithms Prof. Constan±nos Daskalakis Lecture’09’ Markov Chain Monte Carlo (supplementary material to these slides has also been posted)

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Menu Random walks on graphs Markov Chains Examples: – pagerank – card-shuffling – colorings
Menu Random walks on graphs Markov Chains Examples: – pagerank – card-shuffling – colorings

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Random Walks Given undirected graph G = ( V , E ) A squirrel stands at vertex v 0 : Squirrel ate fermented pumpkin so doesn’t know what he’s doing So jumps to random neighbor v 1 of v 0 Then jumps to random neighbor v 2 of v 1 etc Question: Where is squirrel after t steps? A: At some random location. OK, with what probability is squirrel at each vertex of the graph? Want to compute x t R n , where x t ( i ) : probability squirrel is at node i at time t . v t : random variable representing location at time t. v 0 v 1 v 2
x t x t + 1 ? Simplification: all nodes have same degree d, e.g. x 0 = (1, 0, 0, 0, 0) x 0 x 1 ? if u 1 , u 2 ,…, u d are the d neighbors of v 0 , then v 1 = u i with probability 1/ d so x 1 = (0, ½ , 0, 0, ½ ) x 2 = ( ½ , 0, ¼ , ¼ ,0) A = (adjacency matrix divided by d ) 1 2 3 4 5 ½ 0 0 0 ½ 0 ½ ½ 0 0 ½ 0 0 ½ 0 0 0 ½ 0 ½ 0 ½ 0 ½ 0 A ij : probability of jumping to j if squirrel is at i formally A ij = Pr[ v t +1 = j | v t = i ] x 1 = x 0 A x 2 = x 1 A = x 0 A 2 x 3 = x 2 A =x 0 A 3 x t +1 = x t A = x 0 A t +1 (1) “the transition matrix”

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
x t More general undirected graphs? Transition Matrix: A =adjacency matrix where row i is divided by the degree d i of i x t = x 0 A t Computing x t fast? – repeated squaring! – compute A A 2 A 4 A t (if t is a power of 2; if not see lecture 2) – then do vector-matrix product Limiting distribution x t as t →∞ ? e.g. what is x in 5-cycle? x = ( , , , , ) 1 2 3 4 5
Verifying x t ( , , , , ) Recall A = x 0 = [1 0 0 0 0 ] x 1 = [0 0.5000 0 0 0.5000] x 2 = [0.5000 0 0.2500 0.2500 0 ] x 3 = [0 0.3750 0.1250 0.1250 0.3750] x 4 = [0.3750 0.0625 0.2500 0.2500 0.0625] x 5 = [0.0625 0.3125 0.1562 0.1562 0.3125] x 6 = [0.3125 0.1094 0.2344 0.2344 0.1094] x 7 = [0.1094 0.2734 0.1719 0.1719 0.2734] x 8 = [0.2734 0.1406 0.2227 0.2227 0.1406] x 9 = [0.1406 0.2480 0.1816 0.1816 0.2480] x 10 =[0.2480 0.1611 0.2148 0.2148 0.1611] x 11 =[0.1611 0.2314 0.1880 0.1880 0.2314] x 12 =[0.2314 0.1746 0.2097 0.2097 0.1746] x 13 =[0.1746 0.2206 0.1921 0.1921 0.2206] x 14 =[0.2206 0.1833 0.2064 0.2064 0.1833] 1 2 3 4 5 ½ 0 0 0 ½ 0 ½ ½ 0 0 ½ 0 0 ½ 0 0 0 ½ 0 ½ 0 ½ 0 ½ 0 x 15 = [0.1833 0.2135 0.1949 0.1949 0.2135] x 16 = [0.2135 0.1891 0.2042 0.2042 0.1891] x 17 = [0.1891 0.2088 0.1966 0.1966 0.2088] x 18 = [0.2088 0.1929 0.2027 0.2027 0.1929] x 19 = [0.1929 0.2058 0.1978 0.1978 0.2058] x 20 = [0.2058 0.1953 0.2018 0.2018 0.1953] x 21 = [0.1953 0.2038 0.1986 0.1986 0.2038] x 22 = [0.2038 0.1969 0.2012 0.2012 0.1969] x 23 = [0.1969

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 24

L09_-_MCMC - 6.046 Design and Analysis of Algorithms...

This preview shows document pages 1 - 8. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online