bayesImageS 0.5-0

New features

  • improvements to Bayesian indirect likelihood (BIL) algorithm (mh$alg="aux"):
    • MH hyperparameter Ecrit for 2nd order phase transition when k > 4
    • MH hyperparameter factor to inflate the variance of the surrogate model
    • separate MH hyperparameters Vmax1,Vmax2 for $\beta < \beta_{crit}$ and $\beta > \beta_{crit}$, respectively
  • option mh$sort to impose an ordering constraint on the component means

Bug fixes

  • removed OpenMP options from Makevars to fix problems with RcppArmadillo. Users should edit ~/.R/Makevars to enabel multithreading.

  • no longer wait for mh$auxiliary iterations before beginning to update $\beta$ using the exchange algorithm or ABC. For $\beta > \beta_{crit}$, this increased the propensity of the Gibbs sampler to become stuck in a suboptimal local mode, far from the true parameter value.

  • added jss.bst and jss.cls to .Rbuildignore

bayesImageS 0.4-1

  • removed the deprecated slices parameter from mcmcPotts()

  • some minor edits to the Roxygen documentation

  • added ORCID to DESCRIPTION

  • removed the R package PottsUtils from Suggested packages; added mcmcse

  • moved README-chunkname.png images from root directory to inst/images

bayesImageS 0.4-0

New features

  • Functions getNeighbors(), getEdges(), getBlocks() contributed by Dai Feng. These are currently exact copies of the equivalent functions in the R package PottsUtils. The reason for duplicating this code is to avoid picking up dependencies on miscF, R2jags, BRugs, etc.
  • Added new option to swNoData and mcmcPottsNoData to initialise the labels using either random or deterministic values. This can be useful when comparing the convergence of the Swendsen-Wang and Gibbs sampling algorithms to known distributions (e.g. when $\beta$=0 or $\infty$)

  • Additional documentation in README.md with an example of usage

Bug fixes

bayesImageS 0.3-3

  • First version released on CRAN

bayesImageS 0.1-21