history

使用R语言的马尔科夫链蒙特卡洛模拟（MCMC）源代码程序。

//
// Changes and Bug Fixes
//

0.9-3 to 0.9-4
  * fixed issue with MCMCmetrop1R.R [thanks to Jim Albert]
  * fixed Linux compilation issue [thanks to Chris Lawrence]

0.9-2 to 0.9-3
  * added gcc 4.0 check in configure.ac and SystemRequirements
  * added verbose option to plotChangepoint()

0.9-1 to 0.9-2
  * fixed Makevars per Ripley's email [PKG_CXXFLAGS to PKG_CPPFLAGS]
  * fixed encoding on some documentation files [thanks to Kurt Hornik]
  * added other estimation algorithms for MCMCoprobit
  * added other estimation algorithms for MCMCmnl
  * added MCMCpoissonChangepoint with estimation in compiled C++
  * a variety of other minor fixes
  
0.8-2 to 0.9-1
  * first release of JStatSoft version
  * full port to Scythe 1.0.2 (nearly all C++ has been radically changed)
  * MCMCpanel() deprecated
  * a number of minor fixes [thanks to anonymous reviewers]

0.8-1 to 0.8-2
  * models with multivariate normal priors now check for symmetry and positive
    semi-definiteness of precision matrix
  * fixed bug in how models with marginal likelihood calculations check for
    prior propriety [thanks to Gary Rosner for spotting this]
  * fixed bug in how MCMCmetrop1R handled a singular Hessian 
    [thanks to Piers Dunstan for spotting this] 

0.7-4 to 0.8-1
  * added MCMCPoissonChangepoint() model (authored by Jong Hee Park)
  * added two plot methods for changepoint models:
       plotPostChangepoint() and plotPostState() (authored by Jong Hee Park)
  * package cleaned up prior to submission of piece to JStatSoft, including
    major edit of documentation
 
0.7-3 to 0.7-4
  * fixed minor bug in MCMCpoisson() that was causing the function to 
    not work on Windows machines.
  * changed 
      test <- grep("\.nonconst$", c.names)
    to 
      test <- grep("\\.nonconst$", c.names)
    and
      c.names <- sub("\.nonconst$", "", c.names)
    to 
    c.names <- sub("\\.nonconst$", "", c.names)
    in automate.R. [Thanks to Kurt Hornik for noticing this.]

0.7-2 to 0.7-3
  * following posting by Radford Neal at:  
      http://www.stat.columbia.edu/~cook/movabletype/archives/2006/03/ adaptive_metrop.html
    switched a < to a <= in the shrinkage procedures used in the various slice
    sampling implementations.
  * modified tomogplot() and dtomogplot() to handle situations in which 
    r1[i] == 0. [thanks to David Hugh-Jones for making this suggestion].
  * allowed for more user control of the initial call to optim in 
    MCMCmetrop1R [thanks to Luca La Rocca for this suggestion].
  * allowed users to pass the variance-covariance matrix of the Gaussian 
    proposal directly without resorting to a call to optim 
    [thanks to Luca La Rocca for this suggestion].
  * fixed minor bug in developer mode in automate.R [thanks to Ben Goodrich].
  * fixed a minor bug in the developer mode documentation template.
  * fixed minor bug in nonconst call in MCMCregress.



0.7-1 to 0.7-2
  * added procrustes() to NAMESPACE so that it can be seen (function and 
    documentation already there).
  * fixed the prior in the full conditional for theta_i in MCMCirtKdRob. 
    Previously was uniform on the unit circle which was not consistent with
    the documentation and could cause problems given how the starting values
    were chosen. 	
  * fixed prior for lambda_j in MCMCirtKdRob (similar to point above)	
  * fixed the the calculation of the full conditional for gamma[i] in 
    MCMCSVDreg()
  * Fixed a bug in MCMCbetabinomial() [thanks to John Wood].

0.6-6 to 0.7-1
  * Added robust k-dimensional IRT model MCMCirtKdRob().
  * Added SVD regression MCMCSVDreg().
  * Updated auto.Scythe.call() to have any number of non-constants args; this
    is necessary to return other things from the C++ code besides the
    posterior density sample (including log-marginal likelihoods,
    acceptance rates, etc.).
  * Updated form.mcmc.object() to add additional attributes to an
    mcmc object to hold other quantities of interest (such as
    log-marginal likelihoods, acceptance rates, data, etc.).
  * For linear regression model, added option to compute log-marginal
    likelihood via the method of Chib (1995) or via the Laplace Approximation.
  * For logistic regression model, added option to compute log-marginal
    likelihood via the Laplace Approximation.
  * For probit regression model, added option to compute log-marginal
    likelihood via the Laplace Approximation.
  * For Poisson regression model, added option to compute log-marginal
    likelihood via the Laplace Approximation.
  * Added methods to calculate Bayes factors and posterior probabilities 
    of models and handle log-marginal likelihoods. 
  * Added teaching model MCbinomialbeta().
  * Added teaching model MCpoissongamma().
  * Added teaching model MCnormalnormal().
  * Added teaching model MCmultinomdirichlet().
  * patched xpnd() function to provide more functionality.  Thanks to
    Gregor Gorjan for the patches.
  * added the function procrustes() that performs a Procrustes 
    transformation of a matrix.
  * made some minor changes to MCMCirt1d() to help conserve memory when
    dealing with large datasets (MORE OPTIMIZATION NEEDS TO BE DONE IN
    TERMS OF BOTH SPEED AND MEMORY)	

0.6-5 to 0.6-6
  * fixed the std::accumulate problem pointed out by Kurt Hornik. Thanks
    to Dan Pemstein for tracking the problem down and making the fix.
  * fixed up how the force.samp option works with MCMCmetrop1R-- previously
    if a diagonal element of the Hessian was 0 the correction wouldn't work. 
  * changed how additional arguments are passed to user-defined function
    in MCMCmetrop1R. This breaks old code but provides a more standard
    interface. 
  * when logfun==FALSE in MCMCmetrop1R the initial call to optim() now
    maximizes the log of the user-defined function
  * cleaned up the *.Rd files for the model fitting functions a bit 
    (more still needs to be done)
  * MCMClogit now optionally accepts a user-defined prior density

0.6-4 to 0.6-5
  * added a check so one can run MCMCirt1d without passing constraints
  * fixed the Senate dataset so there are no duplicate names in the
    member variables, and modified examples accordingly 
  * slightly edited DESCRIPTION file to get some important keywords
    in the first sentence
  * compute dinvgamma() on the log scale so it does not blow up
    for large shape and scale parameters
  * changed how MCMCmetrop1R handles a non-negative definite Hessian
    from optim--sampling can now proceed if the user flips a switch
  * fixed minor error in how std.mean was passed to MCMCmixfactanal
  * fixed the error in the beta full conditional in MCMCpanel that resulted
    from a typo in the Chib and Carlin paper
  * changed factor.score.eigen.start() in hidden.R to use squared 
    distances rather than distances so as to be compatible with 
    Poole, 2005. _Spatial Models of Parliamentary Voting_	
  * fixed a minor bug in how the verbose argument was handled in the C++
    code for MCMCmetrop1R and MCMCfactanal [thanks to Lingji Chen]
  * fixed a minor documentation bug in the MCMCirtKd model [thanks to
    Guillermo Rosas]
  * fixed the text echoed at start-up so it does not have to be updated 
    year to year

0.6-3 to 0.6-4
  * fixed bug with verbose in MCMCmetrop1R pointed out by a referee for 
    Rnews
  * added a little clarification to the lecuyer.h file about the dual-
    licensing scheme Pierre L'Ecuyer agreed to on 7 August 2004.

0.6-2 to 0.6-3
  * cleaned up the docs for MCMCirtKd
  * MCMCirtKd now fits a model with a difficulty parameter rather than
    a negative difficulty (easiness ???) parameter
  * MCMCirtKd and MCMCordfactanal now do a better job of dropping 
    variables that have 0 variance.
  * fixed Senate examples for MCMCirt1d and MCMCirtKd. EVENTUALLY THE 
    SENATE DATA SHOULD BE FIXED DIRECTLY SO THAT THERE ARE NO DUPLICATE 
    NAMES IN THE MEMBER VARIABLE.