
Release notes for hierarchicalDS  May 13 2013
Paul Conn paul.conn@noaa.gov 

v2.0 Release 9/28/12

-Species misclassification models now implemented
-Removes dependency on ggplot2 
-Input data are now required to be in data.frame form; "Levels" as an input is deprecated
-Resolves a few bugs when trying to analyze data from >1 species
-Resolves a few bugs when trying to analyze data when the # of observers varies between transects
-Number of spatial strata now defined by "Mapping" instead of "Adj" input
-Resolved a bug that led to incorrect calculations when there was greater than one transect in a spatial grid cell
-Resolved a bug with translation of MCMC output to a coda object when there is only one grid cell (i.e. for data pooled over space)
-Resolves issues with covariate updates when the latent # of groups in a transect equals 0
-Incorporates a greater # of options for point independence; in particular, full independence can be last bin or first bin,
 and linear change on log scale can be positive or negative (with options to fix the sign)
-Implemented posterior predictive loss criterion for model discrimination (sensu Gelfand and Ghosh 1998) in function "post_loss"
-Incorporates additional Control variable "iter.fix.N" which specifies the number of iterations to estimate parameters while
 keeping abundance fixed.  This can help stabilize estimation when data are sparse
-The prior distribution for habitat beta parameters is now assumed to be Normal(0,(tau_beta*X'X)^(-1)) where tau_beta is provided by the user
 This scaling helps mixing and provides structure on beta parameters; previous implementation had a bug that the prior sd could be specified
 for beta parameters but in effect an improper prior was always used during estimation.

v2.1 Release - ??/??/??

-Zero-inflation implemented; when spatial autocorrelation is modeled separate spatial models are used for the
Poisson process and for the zero inflation process.  This makes the models very data hungry and limited testing suggests
that results will only be reliable using this approach if sample sizes (and spatial coverage) is high
-Corrects a bug in how posterior predictions of abundance were calculated during MCMC
-Removes reference to MH.beta in list of control inputs (A Gibbs step is used to update these parameters)
-Corrects major bug in posterior loss calculations
-Corrects a bug that wouldn't allow program to run with n.obs.cov=0
-Corrects bug in which Area.hab wasn't accounted for in Metropolis-Hastings updates of nu
-Added rect_adj and rect_adj_RW2 functions for spatial neighborhood construction
-fixed documentation for Hab_pois_formula to indicate it needs to be a list
-Added vignette
-fixed problems with small detection probabilities and/or high detection correlation causing numerical errors with sparse data
 (max correlation now has bounds [-0.95,0.95]; detection probability lower bound 10^-20)