annotate CHANGES @ 18:062d46712995 tip

Moved mc_global_info1 back to public folder
author samer
date Mon, 02 Apr 2012 21:50:43 +0100
parents 0e0f2805ef9c
children
rev   line source
samer@11 1 Changed transition matrix sampling procedure to try to avoid Markov chains
samer@11 2 with mupltiple stationary distributions (fixed points).
samer@11 3 To this end:
samer@11 4 * mc_fixpt has a tolerance parameter - all eigenvalues within this much of
samer@11 5 1 are considered fixed points.
samer@11 6 * mc_global_info1 (replaces mc_global_info) now checks that transition matrix
samer@11 7 has only one fixed point and returns a flag if this is the case
samer@11 8 * sampling logic is in mt_resample - transition matrices not passing uniqueness
samer@11 9 test are either resampled (method 1) or perturbed (method 2)
samer@11 10 * mt_init has new parameters, ErgMethod to select conditioning method and
samer@11 11 tolerance to pass to mc_fixpt.
samer@11 12
samer@11 13
samer@11 14 Other changes
samer@11 15 * Information measures are now displayed in bits, not nats.
samer@11 16 * Clicking on the title bar of a displayed transition matrix now displays
samer@11 17 information about its eigenvalues and eigenvectors (mt_show_transmat.m).