The precision prior for damped state coefficients has changed from a fixed value of 1 to a dynamic value that depends on the s-th posterior sample of the estimated state. More specifically, an application of Zellner's g-prior is applied to a damped state's s-th posterior information matrix. Unlike a regression component where the design matrix is fixed, a damped state's design matrix is a function of the posterior sample. Thus, Zellner's g-prior in this case is adaptive.
Fixed a bug in posterior_dict() method. Specifically, the arrays associated with damped coefficients did not reflect the desired number of burned samples. This has been addressed.
A warning message has been added to sample() method to let the user know if the number of sampling iterations significantly exceeds the number of desired posterior samples. This is to caution against possible non-convergence.