Skip to content

Latest commit

 

History

History
298 lines (226 loc) · 13.8 KB

NEWS.md

File metadata and controls

298 lines (226 loc) · 13.8 KB
Raw

News

Version Updates

0.12.5

  • Issue #166: fix StackOverflowError with MvNormal().

0.12.4

  • Issue #168: Add Categorical() method to convert ArrayVariate{1} to Vector{Float64}.

0.12.3

  • Fix :DARBoot method of discretediag.
  • Fix overwritten MvNormal methods.

0.12.2

  • Fix jl_fill_argnames not found error.

0.12.1

  • Distributions compatibility fixes for Categorical, MvNormal, and MvNormalCanon constructors.
  • Julia 1.1 compatibility fixes for warn and indexin.

0.12.0

  • Initial release for julia 1.0.
  • Branched off of Mamba 0.11.2.

0.11.2

  • Disable package precompilation.
  • Fix for deprecated out_neighbors function in the LightGraphs package.
  • Fix for colorkey syntax deprecated in the Gadfly package.
  • Deprecate vectorization of logit and invlogit methods in favor of dot-syntax.

0.11.1

  • Accept functions as first arguments to Logical and Stochastic constructors.
  • Enable MCMC parallelization.
  • Add discrete convergence diagnostic.

0.11.0

  • Initial release for julia 0.6.
  • Branched off of Mamba 0.10.1.

0.10.1

  • Replace Graphs package dependency with LightGraphs..

0.10.0

  • Initial release for julia 0.5.
  • Branched off of Mamba 0.9.2.

0.9.2

  • Removed 0.9.1 deprecates.
  • Added logic to ABC sampler to skip distributional evaluations if candidate draws are not in the prior distribution support.
  • Updates for Distributions 0.10.0 package compatibility to fix MvNormal TypeError.
  • Added vector indexing to BMC3 and BMG samplers.
  • Implemented Binary Individual Adaptation (BIA) sampler.

0.9.1

  • Removed 0.8.1 and 0.9.0 deprecates.
  • Updates for Distributions 0.9.0 package compatibility to fix ambiguous new definition warnings and StackOverflowError.
  • Updated documentation links to new readthedocs.io subdomains.

0.9.0

  • Extended ABC sampler to allow specification of a decay rate for monotonically decreasing tolerances (epsilon) and perturbations of tolerance by random exponential variates.
  • Added GK distribution example for ABC sampling.
  • Revised stand-alone sampler interfaces (see documentation for full details).
    • Deprecated sampling functions amm!(), amwg!(), bhmc!(), bmc3!(), bmg!(), dgs!(), hmc!(), mala!(), nuts!(), rwm!(), slice!(), and slicesimplex!(); and replaced them with sample!().
    • Moved specifications of tuning parameters and target densities from sampling functions to SamplerVariate constructors.
    • Added target density fields to tuning parameter types.
    • Changed the following tuning parameter types/fields.
      • AMMTune.beta::Real -> Float64
      • AMMTune.scale::Real -> Float64 and redefined to be on the parameter scale instead of the covariance scale.
      • AMWGTune.target::Real -> Float64
      • AMMTune.SigmaF::Cholesky{Float64} -> SigmaL::LowerTriangular{Float64}
      • DGSTune -> DSTune
      • HMCTune.SigmaF::Cholesky{Float64} -> SigmaL::Union{UniformScaling{Int}, LowerTriangular{Float64}}
      • MALATune.scale::Float64 -> epsilon::Float64
      • MALATune.SigmaF::Cholesky{Float64} -> SigmaL::Union{UniformScaling{Int}, LowerTriangular{Float64}}
      • RWMTune.scale::Union{Real, Vector} -> Union{Float64, Vector{Float64}}
      • SliceTune.width::Union{Real, Vector} -> Union{Float64, Vector{Float64}}
  • Renamed Model method simulate!() to sample!().
  • Removed 0.7 deprecates.

0.8.2

  • Modified dgs!() to require parameter support in Matrix columns instead of rows to improve performance.

0.8.1

  • Added AbstractChains read and write methods.
  • Fixed restarting of multiple chains in the case of samplers with adaptively tuned parameters.
  • Renamed Model method tune() to gettune().
  • Simplified user-defined distributions examples.
  • Optimized performance of DGS, MISS, and SliceSimplex samplers and simulation engine.

0.8.0

  • Simplified sampler interfaces.
  • Implemented an approximate Bayesian computation (ABC) sampler.
  • Implemented a random walk Metropolis (RWM) sampler.

0.7.4

  • Simplified implementations of sampling function types and constructors.
  • Implemented SamplerTune and SamplerVariate types.
  • Extended contour plots from ModelChains to AbstractChains.

0.7.3

  • Parallelize logpdf() method for ModelChains.
  • Update BMG algorithm and interface.
  • Simplify BMC3 interface.
  • Rename sampler BMMG to BMC3.
  • Fix errant reinitialization of samplers when restarting chains with mcmc().

0.7.2

  • Deprecated model expression syntax, in favor of model function syntax, for the construction of Logical and Stochastic nodes and Sampler objects.
  • Implemented a Hamiltonian Monte Carlo (HMC) sampler.
  • Removed chain field from Model type.

0.7.1

  • Implemented Binary Hamiltonian Monte Carlo (BHMC) and Binary Metropolised Gibbs (BMG) samplers for binary model parameters.
  • Implemented pairwise posterior density contour plots.
  • Implemented the Metropolis-Adjusted Langevin Algorithm (MALA) sampler.
  • Added first(), step(), and last() methods for getting AbstractChains iteration information.
  • Implemented indexing of ModelChains by model node symbols.
  • Added logpdf() method for ModelChains.
  • Relaxed requirement that all sampled nodes be monitored for the calculation of DIC and for the simulation of draws from posterior predictive distributions.
  • Removed dependents field from Model type.

0.7.0

  • Implemented function syntax for specification of nodes and user-defined Sampler constructors.
  • Changed DSG sampler support field and arguments from Vector to Matrix.
  • Added PGF graphics format to Chains draw function.
  • Removed AbstractDependent linklength field.

0.6.3

  • Added support for model specification of stochastic nodes with Array{MultivariateDistribution} structures containing distributions of different lengths.
  • Added a Slice Simplex (SliceSimplex) sampler for parameters, like probability vectors, defined on simplexes.
  • Added AbstractDependent logpdf() methods for evaluating log-densities at specified values.
  • Renamed AbstractDependent methods link()/invlink() to unlist()/relist().
  • Implemented an AbstractStochastic rand() method for random sampling of node values.
  • Implemented Chains concatenation methods.
  • Implemented a Chains readcoda() method for importing CODA files.

0.6.2

  • Added support for Array{MultivariateDistribution} to the missing values (MISS) sampler.
  • Added a Binary Modified Metropolised Gibbs (BMMG) sampler for binary model parameters.
  • Added bar plots for the summary of AbstractChains that contain values simulated for discrete model parameters.
  • Compatibility updates for julia 0.4 release candidate 2.

0.6.1

  • Compatibility updates for julia 0.4 prerelease.

0.6.0

  • Stable release for julia 0.4.
  • Added support for the specification of MultivariateDistribution arrays in stochastic nodes.
  • Arrays in stochastic nodes must now be declared as a UnivariateDistribution[] or as a MultivariateDistribution[]. In previous package versions, arrays could be declared as a generic Distribution[] array. This is no longer allowable due to the need to distinguish between arrays of univariate and multivariate distributions.
  • The following changes were made to internal data structures and generally do not affect the user interface (i.e., model specification, sampling function calls, and sampler output diagnostics and summaries):
    • The VariateType, which aliases Float64, is deprecated and will be removed in a future version.
    • The abstract Variate type was separated into ScalarVariate and ArrayVariate types which are subtypes of Real and DenseArray, respectively.
    • AbstractVariate was defined as the Union(ScalarVariate,ArrayVariate).
    • The Logical type was separated into ScalarLogical and ArrayLogical types which are subtypes of ScalarVariate and ArrayVariate, respectively.
    • AbstractLogical was defined as the Union(ScalarLogical,ArrayLogical).
    • The Stochastic type was separated into ScalarStochastic and ArrayStochastic types which are subtypes of ScalarVariate and ArrayVariate, respectively.
    • AbstractStochastic was defined as the Union(ScalarStochastic,ArrayStochastic).
    • AbstractDependent was defined as the Union(AbstractLogical,AbstractStochastic).
    • The nlink field of logical and stochastic types was renamed to linklength.
    • An abstract AbstractChains type was implemented, the model field removed from the Chains type, and a new ModelChains type created to provide the model field.
  • Added an example of sampling different parameter blocks with different stand-alone samplers (amwg! and slice!).
  • Removed the insupport method for stochastic types.

0.5.2

  • Applied Mamba changes through 0.4.12.

0.5.1

  • Applied Mamba changes through 0.4.7 and updated compatibility with julia 0.4.0-dev.

0.5.0

  • Branched off of Mamba 0.4.4.
  • Initial release for the under-development version of julia 0.4. Mamba 0.5.x releases exist to incorporate changes being made in the nightly builds of julia, and should be considered unstable. They may contain compatibility issues or serious bugs. Most users are advised to use Mamba 0.4.x releases for julia 0.3 or to wait for stable Mamba 0.6.x releases for julia 0.4.

0.4.12

  • Updated documentation for user-defined distributions.

0.4.11

  • Implemented Chains method function changerate to calculate parameter state space change rates (per iteration).
  • Updated Truncated constructor for Flat distributions for compatibility with latest Distributions package.
  • Simplified documentation instructions for user-defined univariate distributions.

0.4.10

  • Optimized code and improved handling of sampler output in the Gelman convergence diagnostic.
  • Added ask argument to the draw plot method.

0.4.9

  • Added Heidelberger and Welch, and Raftery and Lewis convergence diagnostics.
  • Added documentation and illustrations for all included diagnostics.
  • Added documentation for creating user-defined distributions.

0.4.8

  • Fixed BoundsError() occurring with autocorrelation plots.

0.4.7

  • Improved formatting and outputting of posterior summary statistics
  • Improved efficiency of DOT formatting of Model graphs.
  • Exported graph2dot function to allow in-line processing of Model graphs with GraphViz package.

0.4.6

  • Added verbose argument to mcmc methods to suppress printing of sampler progress to the console.
  • Fixed calculation of effective sample size.

0.4.5

  • Replaced the Scale.discrete_color function deprecated in the Gadfly package with Scale.color_discrete.

0.4.4

  • Require julia 0.3.4 to remove version restriction on the Colors package.
  • Call new Distributions package methods to get InverseGamma shape and scale parameters in the tutorial example.

0.4.3

  • Fixed ERROR: too many parameters for type Truncated.

0.4.2

  • Added support for optional arguments in Chains plot method.
  • Implemented direct grid sampling for discrete univariate stochastic nodes with finite support.

0.4.1

  • Updated and documented predict function as an official part of the package.
  • Reorganized Chains methods documentation.

0.4.0

  • Added support for user add-on packages and functions to allow for their inclusion in Model specifications.
  • Added experimental predict (posterior prediction) function.
  • Required the Cairo package.
  • Removed deprecated MCMC* types and slicewg and SliceWG functions.
  • Fixed ERROR: GenericMvNormal not defined.
  • Distributions DiagNormal and IsoNormal removed and replaced by MvNormal.
  • Distributions DiagNormalCanon and IsoNormalCanon removed and replaced by MvNormalCanon.
  • Distributions DiagTDist and IsoTDist removed and replaced by MvTDist.

0.3.8

  • Updated to fix warning and work with the latest versions of the PDMat and Distributions packages.

0.3.7

  • Extend Chains draw method to allow automatic outputting of multiple plot grids to different files.
  • Add Chains plot method to accommodate vectors of plot types.
  • Fix variance calculation in gewekediag().

0.3.6

  • Fix for convert errors triggered by the Color package beginning with its version 0.3.9.

0.3.5

  • Documentation updates only - primarily the addition of results to examples.
  • No changes made to the source code.

0.3.4

  • Added distributions documentation.
  • Added jaws repeated measures analysis of variance example.

0.3.3

  • Fixed the rand method definition error (type DataType has no field body) that began occurring with late release candidates and final release of julia 0.3.

0.3.2

  • Fixed tuning parameter overwrites occurring with pmap() in single-processor mode.

0.3.1

  • Added chains field to Chains type for tracking purposes.
  • Fixed mcmc to accommodate restarting of chains subsetted by parameters and/or chains.
  • Fixed plot legends to properly reference the chains being displayed.
  • Added support for sampling of positive-definite matrices specified with Wishart or InverseWishart distributions.
  • Added a block-diagonal normal (BDiagNormal) distribution.

0.3.0

  • Implemented restarting of MCMC chains.
  • Deprecated slicewg and SliceWG. Replaced with :univar option to slice and Slice.

0.2.1

  • Updated documentation.
  • Simplified parallel code.

0.2.0

  • Automatically load Distributions package.
  • Implemented parallel execution of parallel chains on multi-processor systems.

0.1.0

  • Removed MCMC prefix from type names, and deprecated MCMC* types.

0.0.2

  • Renamed package from MCMCsim to Mamba.

0.0.1

  • Initial public release.