* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Math 106 Lecture 19 Long Range Predictions with Markov Chains
Quadratic form wikipedia , lookup
Eigenvalues and eigenvectors wikipedia , lookup
Determinant wikipedia , lookup
Jordan normal form wikipedia , lookup
System of linear equations wikipedia , lookup
Matrix (mathematics) wikipedia , lookup
Four-vector wikipedia , lookup
Singular-value decomposition wikipedia , lookup
Non-negative matrix factorization wikipedia , lookup
Perron–Frobenius theorem wikipedia , lookup
Matrix calculus wikipedia , lookup
Math 106 Lecture 19 Long Range Predictions with Markov Chains 1 © m j winter, 2002 Transition Matrix; Transition Diagram Recall the cereal Jaz. Two states: buys Jaz (state 1), and buys not-Jaz (state 2) To 1 From 0.2 2 1 0.8 0.2 2 0.4 0.6 . from state 2 to state 1 1 0.8 2 0.4 0.6 2 1 Markov System (Chain) • A system that can be in one of several (numbered) states, and can pass from one state to another each time step according to fixed probabilities. • We use T for the transition matrix, and p for the probability matrix (row matrix). The entries in p represent the probabilities of finding the system in each of the states.The sum of these probabilities is 1. • We also regard p as describing the percentage distribution in each state of the system. 3 probability (distribution) after n steps If p is the initial probability matrix, then •pT is the probability matrix after one step •pT2 is the probability matrix after 2 steps •pTn is the probability matrix after n steps 4 2 (Jaz) probability (distribution) after n steps If p is the initial probability matrix, then In Jaz example pT is the probability matrix after one step pT = [ 0.5 0.5 ] pT2= [ 0.6 0.4 ] pT2 is the probability matrix after 2 steps p = [ 0.25 0.75 ] pT3= [ 0.64 0.36] pT4= [ 0.656 0.344] pTn is the probability matrix after n steps 5 Steady State It appears that, if we had used pT4= [ 0.656 0.344] as the initial probability matrix, the probability matrix after one step would be (about) the same. Define the equilibrium matrix, L, as the probability matrix which is the solution to LT = L. Example: find the equilibrium matrix for the cereal system 6 3 Systems with non-unique solutions Example 1 Example 2 in matrix form 3x + y = 5 AX = X 6x + 2y = 10 solution set is every point on the line y = 5 – 3x clearly X = O, the zero matrix is a solution 7 Singular Matrix Look at the system: 2 1 1 x 0 1 −2 2 y = 0 3 −1 3 z 0 One element of the solution set is x = 0, y = 0, z = 0 other solutions include x=4, y = – 3, z = – 5 x = 8, y = – 6, z = –10 If we try to find A–1 on the calculator, we get the message If the matrix of coefficients is singular, either the system has NO solutions or it has infinitely many 8 4 A closer look at this singular matrix Look at the system: with augmented matrix 2 1 1 x 0 1 −2 2 y = 0 3 −1 3 z 0 2 1 1 0 1 −2 2 0 3 −1 3 0 The third row/equation is the sum of the other two. It is not independent. We might be able to add another equation and still have a solution. 9 Equilibrium Matrix for Jaz Example Let L = [ x y ] ; know x + y = 1 Then LT=L becomes [x .8 .2 y] = [ x y] .4 .6 or 0.8x + 0.4y = x 0.2x + 0.6y = y (0.8 − 1) x + 0.4 y = 0 0.2 x + (0.6 − 1) y = 0 −0.2 x + 0.4 y = 0 0.2 x − 0.4 y = 0 This is really only one equation. Use the fact that x + y = 1 10 5 Solving for L The system L is sometimes called the (a) steady-state probability matrix. x+y=1 2x - 4y = 0 Technique: we used the fact that P is a probability matrix to replace one of the equations. becomes x+y=1 -6y = -2 or x = 2/3; y = 1/3 L = [ 2/3 1/3 ] = [ .6666… .3333…. ] 11 Bookbuying Problem Professor X is an avid reader of history books, biographies, and mysteries. She buys and reads a book every week. *If she is presently reading a history book, there is a 50% chance that she will switch to a mystery the next week. *She never reads two biographies in a row; after a week of reading a biography, there is a 75% chance that she will switch to a history book the next week. *She always reads history after reading a mystery. *She never reads two history books in a row. Assuming this trend continues indefinitely, what fraction of her library will consist of biographies 1,000 weeks from now? 12 6 Transition Diagram & Matrix States History 1 2 Biography 3 Mystery .75 0 0 1 0 .50 .50 .75 0 .25 0 0 1 2 .50 .50 1.0 .25 3 0 13 Solution - by solving LT = L L = [x y z] LT = L 0 .50 .50 [x y z] .75 0 .25 = [x y z] 0 0 1 Calculate solution: use calculator x = .47… History y = .23… Biography z = .29… Mystery -1x +.75y + 1z = 0 .50x - 1y + 0z = 0 .5x +.25y -1z = 0 x + y + z =1 14 7 Long Term Transition Matrix Assume (1) that some power of the transition matrix T has no zero entries We can use T to find the steady-state probability matrix. (2) The powers of T approach a fixed matrix T. T is called the steady-state transition matrix or the long-term transition matrix. 15 Show: For large values of n, Tn is, essentially, T Tn *T = Tn+1 is also essentially T. Write this equation in matrix form: [first row of T ] [second row of T ] T = : [last row of T ] TT=T [first row of T ] [second row of T ] : [last row of T ] Each row of T is the steady state probability matrix L where LT = L. 16 8 Solution- bookbuying - by taking powers of T screen showing A^8 and A^64 Here is A^64 stored as matrix B 17 9