TO DO LIST:
-----------
26/11/2010 TO DO:
fitted.GLMM_MCMC not yet implemented for situations when there are only fixed effects in the model
14/06/2010 TO DO:
In GLMM_Deviance.cpp
Commands: *marg_ll_iP += *w_k * AK_Basic::exp0_AK(loglik_k); (around line 239)
*marg_ll_iP = AK_Basic::log0_AK(*marg_ll_iP); (around line 252)
In the case of the poisson response, it would be desirable to add log(y!) to loglik_k before exp-it
and subtract it again after log the mixture sum. With higher counts, loglik_k is quite considerably negative
which leads to exp(-Inf) = 0 and then upon log again, we get -Inf log-likelihood where
-Inf is artificial due to imprecise calculation with numbers close to zero.
--> more or less solved on 03/12/2010
======================================================================================================================
01/09/2014 (version 3.7):
Bug in GLMM_MCMCdata corrected (it appeared when two or more discrete responses
of the same type, e.g., binomial, were included in the model).
CITATION file updated (methodological papers removed from there).
21/08/2014 (version 3.7):
NMixCluster function added (currently only method for class GLMM_MCMC).
06/08/2014 (version 3.6-1):
Update of index.html in /inst/doc directory, validation using http://validator.w3.org.
14/05/2014 (version 3.6):
Minor changes related to the final JSS submission.
22/09/2013 (version 3.5):
Gauss-Hermite quadrature inside the fitted.GLMM_MCMC function implemented using the
functions from the 'fastGHQuad' package. Arguments x2, z2 are no more needed.
16/08/2013 (version 3.4-2):
Function NMixRelabel.GLMM_MCMClist allows for a joint processing of two chains
(jointly = TRUE).
06/08/2013 (version 3.4-1):
Functions rWishart and dWishart renamed to rWISHART and dWISHART to avoid
conflicts with rWishart in package stats.
05/08/2013 (version 3.4):
Most dependencies moved from "Depends" to "Imports" in the DESCRIPTION file
and solved via importFrom in the NAMESPACE.
18/07/2013 (version 3.2-4):
Changes of names of some components inside the objects of classes NMixMCMC and GLMM_MCMC:
comp.prob2 --> comp.prob_b
comp.prob3 --> comp.prob
poster.comp.prob1 --> poster.comp.prob_u
poster.comp.prob2 --> poster.comp.prob_b
poster.comp.prob3 --> poster.comp.prob
quant.comp.prob1 --> quant.comp.prob_u
quant.comp.prob2 --> quant.comp.prob_b
quant.comp.prob3 --> quant.comp.prob
08/02/2013:
Support for parallel computation originally provided by snowfall/snow packages changed to parallel package.
01/05/2012:
All pdf vignettes moved to my personal webpage (to decrease the size of the package which must be stored on CRAN).
18/12/2011:
GLMM_MCMC creates by default two parallel chains and calculates also PED, support for parallel computation
on multicore CPUs using snowfall/snow packages added.
20/01/2011:
beta in GLMM_MCMC function (and all related) was renamed to alpha to be in agreement with papers.
29/11/2010:
Calculation of posterior quantiles for estimated component probabilities added to NMixRelabel functions
(both NMixMCMC and GLMM_MCMC version).
08/12/2009
Bug in computation of weights for the EB estimate of random effects in GLMM_longitPred_nmix_gauss.cpp fixed.
22/11/2009
Function NMixClust renamed to NMixPlugDA.
06/11/2009
Bugs in GLMM_updateRanEf.cpp found and fixed:
* eta_random was computed incorrectly as Z %*% bscaled and not as Z %*% b, where b = shift + diag(scale) %*% b
* pointers in the final part of the code did not shift correctly