From the archives

Perfect Simulation Scheme For Processes With Long Memory : a CIAFTP algorithm

Historical material. Affiliations, schedules and roles on this page describe their original context.

Presentation

A matlab implementation of the algorithm described my article Perfect Simulation Of Processes With Long Memory: a CIAFTP algorithm. This algorithm performs perfect simulation for stationary processes described by a transition kernel. These includes, in particular, variable length Markov Chains and Context Trees.

The term "CIAFTP" comes from [D. B. Wilson, How to couple from the past using a read-once source of randomness, Random Structures Algorithms 16 (2000), no. 1, 85–113].

Download

The package can be downloaded here. It uses object oriented programming, and thus requires a not-too-old version of matlab.

Files

README information file
demoPerfectSim.m demonstration file for the package
+Context/ the package in itself
+Context/dist2str.m internal: tweaked version of num2str specially dedicated to discrete distributions
+Context/Example1.m matlab class for the kernel of Section 7.1 (see the second example of demoPerfectSim.m)
+Context/Kernel.m matlab class for the transition kernels (contains function CIAFTP for perfect simulation)
+Context/mynum2str.m internal: personal version of num2str for integers between 0 and 9
+Context/part2str.m internal: modification of num2str specially dedicated to partitions of [0,1]
+Context/PCfunction.m matlab class for piecewise constant function on the set of semi-infinite sequences
+Context/Trie.m matlab class for trees with edges labelled by integers
+Context/VLMC.m matlab class for constructing a kernel via a VLMC representation (see the first example of demoPerfectSim.m)