CHANGES IN VERSION 3.4.1
- Fixed minor bug in localnulltest_fda

CHANGES IN VERSION 3.4.0
- Added localnulltest
- Added approximate posterior sampling method in LM when prior not
specifically implemented  
- Added check in modelSelection to detect NAs and infinite values in y or x
- Added localnulltest and localtest class
- Fixed bug in predict.msfit that could occur when colnames(object$xstd) was
null

CHANGES IN VERSION 3.3.1
- Added info on varnames and top model to the output of bestAIC, bestBIC, bestEBIC, bestIC
- Added methods for icfit objects: coef, confint, show, summary

CHANGES IN VERSION 3.3.0
- Fixed a bug in the calculation of BIC (it was off by log(n), and failed for
the null model with no covariates)
- Added bestAIC, bestBIC, bestEBIC, bestIC functions

CHANGES IN VERSION 3.2.0
- Adjusted cil so it has the same arguments as modelSelection. Added arguments priorCoef, center, scale, includevars and verbose. Removed arguments beta.prior and tau. Set the defaults of center and scale as in modelSelection (TRUE for all non-factor variables)
- Added GLM functionality to cil via arguments family, familyD which can now also be 'binomial' and 'poisson'
- Added plotprior method to cilfit
- cil internal parameter rho.min now set to 1/p rather than 1/p^2
- cil now returns BMA inference for all parameters in object$beta
- cil now returns BMA estimates, intervals and margpp in a single element object$cil.teff (before they were stored separately)
- cil now extracts BMA inference using coef function from msfit objects and allows for any priorCoef allowed by modelSelection
- Fixed incorrect BIC formula in CIL's lasso.bic for logistic regression
- cil now allows data frames as inputs
- Fixed bug in modelSelection where margpp==0 for variables with includevars==TRUE


CHANGES IN VERSION 3.1.4
- Removed pmomLM, pmomPM, emomLM, emomPM, zellnerLM, zellnerPM, pplProbit

CHANGES IN VERSION 3.1.3
- Fixed NOTE from comparing class() to string in cil.R

CHANGES IN version 3.1.2
- Added cil
- Fixed gcc warnings for comparison of pointer addition to NULL

CHANGES IN VERSION 3.1.1
- Added bicprior to linear models

CHANGES IN VERSION 3.1.0
- Fixed modelSelection to work with GLMs with bicprior() and group constraints
- Added initpar option to modelSelection with 'MLE', 'MLE-aisgd', 'L1',
'L2-aisgd' to improve ALA accuracy and increase speed of Laplace approx

CHANGES IN VERSION 3.0.7
- Fixed bug in predict.msfit where covariates in newdata were not scaled appropriately

CHANGES IN VERSION 3.0.6
- Added posterior sampling under Normal approximation for GLMs, in the
presence of groups

CHANGES IN VERSION 3.0.5
- Added getBIC and getEBIC methods

CHANGES IN VERSION 3.0.4
- Added bicprior() to modelSelection
- Fixed memory error in pmomLM Rd examples file

CHANGES IN VERSION 3.0.3
- Added step size decrease in Newton-Raphson optimization (modselfunction.cpp)
- Increased default max step size decreases for GLMs from 5 to 10
- Added optim_maxit option to modelSelection and nlpMarginal 

CHANGES IN VERSION 3.0.2
- Fixed numerical overflow issue in postProb

CHANGES IN VERSION 3.0.1
- Added logistic and Poisson regression to modelSelection
- Added ALA for GLMs

CHANGES IN VERSION 2.3.2
- Added method='ALA' to modelSelection for AFT survival models
- Implemented asymetric modelbinomprior
- Fix calibration of groupmomprior
- Add taustd prior to set dispersion in terms of unit information prior

CHANGES IN VERSION 2.3.1
- Added option to specify multiple nknots in modelSelection
- Implement groupzellner prior combinations with MOM, Zellner and normalid
priors
- Implement groupmom prior with MOM
- Fix some inconsistencies between tau and taugroup usage

CHANGES IN VERSION 2.3.0
- Fixed error in coefByModel (typo in calling postProb)
- Fixed unnecessary warnings returned by hastPostSamples
- Fixed error in modelSelection Gibbs when using includevars

CHANGES IN VERSION 2.2.9
- Fixed NOTE due to specifying non-canonical URL for mombf package

CHANGES IN VERSION 2.2.8
- Fixed error in Windows checks related to a test in test-modelSelection-Gibbs.R

CHANGES IN VERSION 2.2.7
- Fixed error caused by class(smoothterms) returning multiple values

CHANGES IN VERSION 2.2.6
- Fixed error caused by class(y) where y is "matrix" returning multiple values
- Added coefByModel, and hidden function modelSelectionGLM with preliminary GLM support
- Added nomalidprior
- Added support for groups and constraints to nlpMarginal

CHANGES IN VERSION 2.2.5
- Fixed bug that prevented modelSelection to run for survival outcomes where no formula is specified

CHANGES IN VERSION 2.2.4
- Fixed bug in C++ code for checkConstraints
- Implemented Newton-Raphson optim in modselFunction.cpp
- Implemented MOM orthogonal approx in SurvMarg
- greedyVarSel now does unconstrained search, the result is subsequently
modified to satisfy constraints
- Improved modelSelection parameter initialization via Taylor expansion at 0

CHANGES IN VERSION 2.2.3
- Removed cleanup script that wrongly deleted the Makevars file

CHANGES IN VERSION 2.2.2
- Improved speed of modelSelection for AFT models (improved Mill's ratio
evaluation and avoided unnecessary calls to pnorm)

CHANGES IN VERSION 2.2.1
- Fixed bug in coef and predict methods for msfit objects
- Fixed compiler warnings in Windows

CHANGES IN VERSION 2.2.0
- Fixed bug in modelSelection when computing Binomial model prior probabilities
- Added Normal Accelerated Failure Time models
- Added coef method to extract BMA estimates and intervals from msfit objects
- Added predict method to extract BMA predictions and intervals from msfit objects
- Added additive models to modelSelection
- Added support for formulas to modelSelection
- Added RcppArmadillo
- Added option to group variables and impose hierarchical restrictions to
modelSelection
- Added rnlp method that only requires msfit
- msfit objects now store priorCoef, priorDelta, priorVar and standardized versions of (y,x)
- Added baseDensity='laplace' to dmom
- Added dmomigmarg, pmomigmarg

CHANGES IN VERSION 2.1.2
- Added complexity prior to modelSelection and bms_ortho
- Added option includeModels to postModeOrtho.
- Fixed bug in postModeOrtho under Zellner's prior that caused it to return wrong posterior model probabilities
- rnlp now accepts y formatted as a 1-column matrix
- pMOM integrated likelihood now defaults to orthogonal approximation, under method=='auto'

CHANGES IN VERSION 2.1.1
- Fixed R_registerRoutines and R_useDynamicSymbols NOTE when running R CMD check

CHANGES IN VERSION 2.1.0
- Added Normal - Inverse Wishart conjugate analysis
- Added Normal mixtures
- Removed dependency on package actuar

CHANGES IN VERSION 1.9.6
- Removed register statements causing a warning in new C++ compiler

CHANGES IN VERSION 1.9.5
- Added option includevars to modelSelection
- Added eMOM and iMOM priors for two-piece Normal and two-piece Laplace errors
- Fixed a bug in integrated likelihood Laplace approximation for the MOM prior and Normal errors
- Fixed valgrind memory warnings

CHANGES IN VERSION 1.8.3
- Fixed spectral clustering on cor(x)^2 instead of cor(x)

CHANGES IN VERSION 1.8.2
- Added blocksearch and spectral clustering

CHANGES IN VERSION 1.8.1
- Fixed use of infinity that was creating trouble in some C compilers

CHANGES IN VERSION 1.8.0
- Major vignette update
- Added model enumeration option to modelSelection
- Added integrated likelihood for product eMOM prior in linear models
- Added general nlpMarginal function
- Added linear regression with two-piece normal, laplace and two-piece laplace residuals
- Fixed numerical stability issues in orthogonal and block-diagonal regression

CHANGES IN VERSION 1.7.1
- Fixed issue in block-diagonal variable selection that caused it to skip some model sizes

CHANGES IN VERSION 1.7.0
- Added block-orthogonal and block-diagonal variable selection

CHANGES IN VERSION 1.6.1
- Fixed maximum number of variables <=n in pmomLM, pmomLM (this also solves a numerical overflow issue when p>n)

CHANGES IN VERSION 1.6.0
- Added pimomCoxMarginalR
- Added method=='plugin' to modelSelection, pimomMarginalK, pimomMarginalU

CHANGES IN VERSION 1.5.9
- Fixed warning in logPL about not found mclapply
- Fixed R functions computing pMOM, piMOM, and peMOM marginal likelihood when input model has no covariates

CHANGES IN VERSION 1.5.8
- Fixed bug in emomLM when sampling residual variance under model with no covariates
- Added rnlp method for objects of type 'Surv'
- Added rnlp method for general d(theta) * N(theta;m,V)
- Added pmomCoxMarginalR
- Added function eprod

CHANGES IN VERSION 1.5.7
- Fixed memory access error in pmomLM

CHANGES IN VERSION 1.5.6
- Fixed apparently harmless invalid memory read at modselIntegrals.cpp

CHANGES IN VERSION 1.5.5
- Implemented faster Laplace approx for piMOM marginal likelihood (now uses gradient algorithm to find posterior mode)
- Implemented p>n case in modelSelection
- Changed default to method=='auto' in modelSelection
- Changed default priorCoef, priorModel, priorVar in modelSelection to recommendations in Johnson & Rossell (2010)
- Fixed bug in modelSelection when method=='auto' and a prior different from MOM was used.

CHANGES IN VERSION 1.5.4
- Fixed R CMD CHECK warnings

CHANGES IN VERSION 1.5.3
- Fixed compiler warnings

CHANGES IN VERSION 1.5.0
- Added function rnlp to sample from MOM, eMOM and iMOM posteriors under Normal linear models
- Added class msfit and several basic methods
- Added momprior, imomprior, emomprior, modelunifprior, modelbinomprior, modelbbprior, igprior functions to facilitate creating objects of class msPriorSpec
- Prevented numerical overflow in computing iMOM integrated likelihood with method=='Hybrid' and method=='MC'. Improves pimomMarginalU and modelSelection for both these methods.

CHANGES IN VERSION 1.4.1
- Fixed bug in modelSelection that caused to report the wrong model (only) at the first Gibbs iteration
- Increased precision of Laplace approximation to compute MOM and iMOM integrated likelihood

CHANGES IN VERSION 1.4.0
- Fixed C++ bug in Solaris caused by overloading of 'log' and 'sqrt' functions

CHANGES IN VERSION 1.3.9
- Fixed C++ bug in Solaris caused by overloading of 'pow' function
- Limited number of saved models by modelSelection Gibbs scheme when p>16, to avoid running out of memory

CHANGES IN VERSION 1.3.8
- modelSelection Gibbs scheme now uses saved log(integrated likelihood) + log(prior) for previously considered models. Observed a 5-10 fold increase in speed.
- modelSelection with Beta-Binomial prior now marginalizes over Binomial success prob, which greatly improves mixing of the Gibbs sampler.

CHANGES IN VERSION 1.3.7
- Fixed warning due to visibility of mclapply

CHANGES IN VERSION 1.3.6
- Added truncated multivariate normal random number generation routines

CHANGES IN VERSION 1.3.5
- Increased precision in pimom marginal Laplace approximation, which could cause crashing due to non-positive definitiness

CHANGES IN VERSION 1.3.4
- Fixed code chunks in vignette to enhance their visibility

CHANGES IN VERSION 1.3.2
- Added check to T dmom to ensure that prior is proper
- Added greedy initialization algorithm for probit models
- Added multivariate version of heavy-tailed pMOM to dmom
- Modified heavy-tailed eMOM implemented in dmom so that it is consistent with heavy-tailed pMOM. Now both are defined via hyper-priors on tau.
- Added pplPM function
- Added emomLM, emomPM function

CHANGES IN VERSION 1.3.1
- modelSelection now returns the estimated regression coefficients for the most probable model

CHANGES IN VERSION 1.3.0
- Added pmomPM function
- Added pmomLM function

CHANGES IN VERSION 1.2.1
- Added dmom prior for unspecified tau. Added parameters a.tau and b.tau to dmom.

CHANGES IN VERSION 1.2
- Added demom and pemom functions.

CHANGES IN VERSION 1.1.3
- Improved vignette model selection section
- Implemented new Laplace approximation for product iMOM marginals which is much more efficient computationally than the previous one

CHANGES IN VERSION 1.1.2
- Removed makefile from inst/doc
- Added hierarchical Beta-Binomial prior for model space indicator in modelSelection

CHANGES IN VERSION 1.1.1
- Fixed some C warnings due to uninitialized variables

CHANGES IN VERSION 1.1.0
- Product MOM and iMOM priors now implemented
- Linear model variable selection for MOM and iMOM priors implemented

CHANGES IN VERSION 1.0.5
- Implemented product iMOM prior density in dpimom. Renamed quadratic iMOM density from dimom to dqimom. Added wrapper dimom to access either dpimom or dqimom.

CHANGES IN VERSION 1.0.4
- Implemented tMom Bayes factors as an option to mombf, mode2g and g2mode.
- Implemented tMom prior density as an option to dmom

CHANGES IN VERSION 1.0.3
- A couple minor changes in documentation files .Rd to avoid warnings

CHANGES IN VERSION 1.0.2
- Fixed bug in the multivariate calculation of MOM Bayes Factors (affects mombf, momknown)

CHANGES IN VERSION 1.0.1
- Fixed bug in the multivariate calculation of dmom

CHANGES IN VERSION 1.0.0
- Added option 'method' to functions 'imombf', 'momknown' and 'imomunknown' to allow users to choose integration method. Default integration method changed from Monte Carlo to integration based on routine 'integrate' from package 'stats'.
- Functions 'imombf', 'imomknown' and 'imomunknown' now accept parameter 'g' to be a vector instead of a single value.
- Added pmom and pimom functions
- Removed g2mode.univ and mode2g.univ functions. The same results can now be obtained via g2mode and mode2g.
- Corrected error in dimom evaluation
- Corrected error in zbfknown evaluation
- Changed order of dmom and dimom arguments: theta0 is not 1st argument anymore
- Added contents to vignette illustrating the new functions, and changed the name of the vignette from "mombf_manual" to "mombf"

CHANGES IN VERSION 0.0.1
- Changed word "library" in Vignette title to "package"
- Indicated GPL(>=2) in the DESCRIPTION file


