Jc. Kieffer, STRONGLY CONSISTENT CODE-BASED IDENTIFICATION AND ORDER ESTIMATION FOR CONSTRAINED FINITE-STATE MODEL CLASSES, IEEE transactions on information theory, 39(3), 1993, pp. 893-902
Observations are made of data generated by a stationary ergodic finite
-alphabet information source according to an unknown statistical model
. Two modeling problems, the identification problem and the order esti
mation problem, are considered. In the identification problem, one wis
hes to decide from the observed data whether the source model belongs
to a given model class. In the order estimation problem, one wishes to
decide from the observed data to which of infinitely many given model
classes the source model belongs. It is required that the given model
class in the identification problem and that each given model class i
n the order estimation problem be a constrained finite-state model cla
ss, which is a type of model class that includes many model classes of
information-theoretic interest. Strongly consistent decision rules ar
e exhibited in both the identification problem and the order estimatio
n problem. The decision rules are code-based in that a model class is
chosen based upon how well a certain code for that class encodes the o
bserved data. The code used for a model class is based upon the maximu
m likelihood code for that class, and asymptotic code performance is g
auged by means of a key property of divergence-rate distance.