NEWS.md

In alan-turing-institute/distr6: The Complete R6 Probability Distributions Interface

distr6 1.8.4

• Fix massive bottleneck in Matdist and Arrdist

distr6 1.8.3

• Add decorators argument to c.Matdist and c.Arrdist

distr6 1.8.1

• Add 'transformer' functions pdfcdf and cdfpdf, which use Rcpp to transform matrics/arrays/vectors between pdf->cdf and cdf->pdf respectively.

distr6 1.8.0

• Add Arrdist, which generalises the Matdist to a three-dimensional array, useful for Bayesian predictions where the third dimension is multiple distributions.

distr6 1.7.0

• Added dparse()

distr6 1.6.15

• Fix bug in plot.Matdist x limits

distr6 1.6.14

• Fix bug when extracting a single distribution with a logical vector from MatDist

distr6 1.6.13

• Fix reordering bug when extracting vector distributions

distr6 1.6.12

• Fix minor bug causing decorators not to be carried forward when using rep
• Fix bug causing vector distributions not to be able to be extracted

distr6 1.6.10

• Fixed bug in Matdist where d/p/q/r were transposed incorrectly
• Fixed bug in

distr6 1.6.9

• M1mac fix

distr6 1.6.8

• Added Matdist to simplify the process of creating a vector distribution for WeightedDiscrete distributions. Matdist acts as a univariate distribution whilst technically being multivariate - it contains a matrix of probabilities and provides methods for acting on those.
• Added methods: plot.Matdist to wrap matplot around Matdist, c.Matdist to combine multiple matrix distributions into one, [.Matdist for extracting WeightedDiscrete or Matdist distributions from Matdist, and as.Distribution.
• Added mixMatrix as a specialised mixing method for Matdist that creates a new Matdist instead of a MixtureDistribution
• Possibly breaking change - Changed default of as.Distribution.matrix to now coerce to a Matdist, a vector of WeightedDisrete is still possible.
• Slight speed improvements to EmpiricalMV
• Fixed r-devel linux warnings

distr6 1.6.7

• Fix for Apple M1

distr6 1.6.6

• WeightedDiscrete now correctly handles improper distributions by returning Inf and NA where appropriate
• Added constraint on WeightedDiscrete so cdf parameter must be increasing
• Add distribution method confidence to calculate confidence intervals
• Fix testing of suggested distributions

distr6 1.6.5

• Fixes error in c.VectorDistribution caused by R6 referencing

distr6 1.6.4

• Off-by-one error fixed in WeightedDiscrete distribution cdf function
• Speed-ups in pdf, cdf, rand functions in WeightedDiscrete

distr6 1.6.2

• Patch for testthat

distr6 1.6.1

• Bump param6 version to fix bug in HyperGeometric distribution

distr6 1.6.0

• Parameter set interface has been moved to param6. This should not cause breakages if parameters have only been accessed via distribution methods but may otherwise.
• R62S3 is no longer loaded by default, S3 methods must now be manually created with R62S3::R62Fun. This may cause breakages if scripts assume pre-loading but will work as before once R62Fun is called (see S3 and piping vignette)
• Kurtosis and skewness methods now vectorised
• Ids for vector distributions now forced to be alphanumeric
• Deprecated unused arguments from parameter methods
• print method now only shows active parameters
• Many other internal improvements to code quality and efficiency

distr6 1.5.6

• Fix for M1mac

distr6 1.5.5

• Minor internal change

distr6 1.5.4

• Added as.Distribution.matrix for converting matrices to a VectorDistribution of WeightedDiscrete distributions
• Added ... to dstrs to allow decorator argument to be passed through

distr6 1.5.3

• Internal changes only

distr6 1.5.2

• Fixed bug in extracting distributions from vector distributions

distr6 1.5.1

• Added ids active binding to more easily get ids of distributions in VectorDistribution

distr6 1.5.0

Minor Updates

Added Functionality

• Added length.VectorDistribution
• Added ids parameter to VectorDistribution to allow specifying names for wrapped distributions. Extended extract.VectorDistribution so these ids can be used for extraction.
• Added sugar function dstr to construct distributions more easily, e.g. dstr("Binom", size = 4) instead of Binomial$new(size = 4).
• Added sugar function dstrs to construct a VectorDistribution of distributions.
• ParameterSet method setParameterValue now includes resolveConflicts parameter for resolving conflicts between conflicting parameters (see below). This should primarily only be used internally and is useful for transforming composite distributions.

Edited Functionality

• The separator for composite (wrapped) distributions is now __ (before _). This may affect backwards compatibility in rare cases.
• Construction and setting of parameters is now changed. Previously this relied on knowing a confusing hierarchy in parameters which would take priority over one another. Now if conflicting parameters are supplied in construction or on setting an error is returned, e.g. Binomial$new(prob = 0.1, qprob = 0.2) would cause an error. This may affect bacwards compatibility however should not do so if distributions are correctly constructed/updated without conflicts.
• setParameterValue in ParameterSet no longer errors if a non-settable parameter is updated, instead a warning is given and the parameter is ignored.
• Multivariate distributions no longer error if constructed with a single variate, though this is still advised against.
• Internal changes to how SDistribution parameter defaults are set. Previously these were set in the constructor with function defaults, now they are set in the ParameterSet. This should not affect backwards compatibility. This should improve speed of distribution construction.

Deprecated/Removed Functionality

• Data parameter in WeightedDiscrete distribution has now been removed.

Bug fixes/Under the Hood Edits

• Minor bug fixes in some distribution properties.
• MixtureDistribution now uses extraDistr::rmnom for speed improvements.
• Slight improvement to error messages in ParameterSet but these will be improved further in a future update.
• Minor internal changes to storage method of SDistribution ParameterSet objects.
• Fixed typos in parameter sets

Documentation

• Documentation for SDistribution objects has been restructured for easier reading, the distribution default is now included.
• Fixed typos

distr6 1.4.8

• Fixes for R devel

distr6 1.4.7

• Patch for set6
• Add rep.Distribution for replicating distributions into vectors, mixtures, or products
• Kernels can now be used with vectors/mixtures/products

distr6 1.4.6

• Patch for solaris

distr6 1.4.5

• Added cdfSquared2Norm analytic methods to most kernels
• Minor internal patches

distr6 1.4.4

• Added coercions from Vector/Mixture/ProductDistribution to Mixture/ProductDistribution via as.MixtureDistribution and as.ProductDistribution
• Bugfix in MixtureDistribution parameter set
• Uneven vectors of WeightedDiscrete distributions are now truncated at the minimum length in the vector for pdf, cdf, quantile
• CoreStatistics methods now give the option to use cubature::cubintegrate and all associated parameters.
• Remove {pracma} requirement in ExoticStatistics

distr6 1.4.3

• Bugfix in WeightedDiscrete, cdf was previously positive when below distribution support, now zero.
• Added discrete p-norms in ExoticStatistics decorator
• Removed suppressMoments argument in Distribution and computation of movements in constructor, improving construction speeds.
• kurtosis and skewness type no longer included in properties. The kurtosis and skewness methods are still available as public methods, and the types are still in summary; to manually find these use exkurtosisType and skewType.
• Empty ParameterSet now allowed

distr6 1.4.2

• Critical patch for patched Solaris: Updated precision in C++ log

distr6 1.4.1

• Critical patches for R devel: workingSupport is now a public method and not an active binding

distr6 1.4.0

Minor Updates

Added/Edited Functionality

• Multivariate functions in VectorDistribution now return arrays
• Analytical median now available for SDistributions where closed form expressions are known
• Kernels now implemented in Rcpp with analytical log-pdf expressions. Analytical lower.tail and log.p will come in future versions.
• Adding plotting for bivariate distributions using {plotly}
• Added EmpiricalMV for empirical multivariate distributions
• All documentation now uses roxygen R6
• Distribution type is now required in construction custom Distributions as the previous method of 'guessing' was inefficient.
• Added ParameterSetCollection for Wrappers. Same functionality as before but much more efficient.
• Added $addDeps and $deps to ParameterSets to replace updateFunc. This allows faster and more precise control over parameter updates.
• Added $addChecks and $checks to ParameterSets to replace custom checks in distributions, and $addTrafos and $trafos to replace custom parameter transformations before setting. These allows clearner control over parameter setting.
• categories parameter in Categorical now called nCategories
• Constructor of Categorical now requires elements to be provided as a list to elements
• settable in ParameterSet now refers to the more intuitive flag that describes if a parameter can be set after construction
• Added length active binding to ParameterSet to return number of parameters in the set.
• Renamed squared2Norm in Kernels to pdfSquared2Norm and added x to arguments to allow for shifted norms.
• Added mixturiseVector for mixing vector distributions

Deprecated Functions/Fields/Methods

• Individual property and trait accessors deprecated: $support, $symmetry, $kurtosisType, $skewnessType, $valueSupport, $variateForm, $type
• Separated Loglogistic and ShiftedLoglogistic so only the latter has a location parameter. Additionally added an analytical rand expression to ShiftedLoglogistic, and Loglogistic now implemented from package actuar
• Removed deprecated classes: ArrayDistribution
• Removed deprecated functions: listSpecialSets
• WeighedDiscrete distribution now stores samples and probabilities as parameters.
• Removed verbose argument from distribution constructor
• Public update method in ParameterSet moved to private
• Removed squared2Norm from the ExoticStatistics decorator
• Deprecated updateFunc from ParameterSet
• Removed .getRefParams from SDistributions and added deps

Patches

• Bugfix in decorate function that was overwriting the decorators private field instead of appending
• Vastly improved speed of construction of SDistribution and Kernel
• Bugfix in $rand for Geometric distribution with trials == TRUE , previously was randomising between $[1, Inf)$ not $[0, Inf)$
• Changed default form parameter in NegativeBinomial, now errors if supplied argument is not in one of the four choices (previously reverted to "fbs")
• Fixed bug in MultivariateNormal rand() which was not adding the mean correctly
• Massive bottleneck in VectorDistribution for same distribution type removed in d/p/q/r functions as well as non-generating function methods
• Improved speed in VectorDistribution when only one row or column of data passed to d/p/q
• Dirichlet, DiscreteUniform, Frechet Gumbel, Pareto, Rayleigh, Triangular, Wald now implemented from package extraDistr
• Bug fix in Frechet $cdf
• Analytical log and lower.tail now available for all SDistributions and imputation possible for customs with CoreStatistics
• NegativeBinomial now uses Rcpp for d/p/q/r
• Distributions from stats are now vectorised across parameters in C
• Degenerate, ShiftedLoglogistic now Rcpp
• Removed redundant K parameter from MultivariateNormal
• Bugfix in Wald, symmetry was incorrectly flagged as "symmetric"
• Fixed bug in TruncatedDistribution support - previously support interval type was "[]" instead of "(]". An important implication is that even if truncating at or outside of the distribution limits, the support will still be changed if the distribution was left-closed to become left-open.
• Improved speed of assertions by using package {checkmate}
• Bugfix in listDistributions which was overriding filtering with simplify

distr6 1.3.7

• Minor internal Updates
• Next release will be 1.4.0, due to nature of R6 this will break dependencies.

distr6 1.3.6

• Bugfix in decorate function that was overwriting the decorators private field instead of appending
• Added plot.VectorDistribution for more efficient plotting of mutiple distributions at the same time
• Two bug fixes in MultivariateNormal$rand() . i) the distribution mean wasn't being added to the transformation correctly; ii) the Choleskey decomposition wasn't transposed causing incorrect simulations

distr6 1.3.5

• Added Erlang distribution
• Fixed mistake in documentation for Weibull, thanks to Paul Northrop (@paulnorthrop) for pointing this out.
• Fixed bug in plot preventing only quantile being plotted on its own
• Converted descriptive public methods to active bindings. Unfortunately this will break reverse dependencies as there's no way to soft deprecate these. Affected methods are: decorators, traits, valueSupport, variateForm, type, properties, support, symmetry, sup, inf, dmax, dmin, kurtosisType, skewnessType

distr6 1.3.4

• Bug fix in VectorDistribution that was causing decorators not to be added properly
• Bug fix in VectorDistribution that was causing an error when extracting decorated wrapped distributions
• Moved package pracma to suggests. Removed utils dependency.
• Added checks to assert all packages are installed when required by distributions
• Deprecated package field in distributions, now called packages
• packages lists all packages required to be installed for a distribution, if NULL only distr6 required
• Abstracted SetInterval and related classes to the set6 package
• Deprecated listSpecialSets, use set6::listSpecialSets instead
• Removed secondary checks on update of ParameterSet , any properly defined distributions will have this covered in primary parameter, thus increasing speed.

distr6 1.3.3

• as.data.table has been moved from a ParameterSet method to an S3 dispatch. Fixes an important bug of overloading as.data.table but may affect backwards compatibility.
• Added distrSimulate for convenient simulation from any distribution
• Updated plot to be able to handle distributions without quantile or rand. Optimised runtime by preventing automatic computation of pdf and cdf.

distr6 1.3.2

• Bug fix in VectorDistribution - missing which argument
• Bug fix in Bernoulli - rand was incorrectly calling dbinom not rbinom
• Bug fix in the mode of distributions
• Added mode to documentation

distr6 1.3.1

• Added wrapper for VectorDistribution for quick concatenation of constructed distributions, c.Distribution
• VectorDistribution print method more in line with base R vectors

distr6 1.3.0

Minor Updates

Added Functions and Classes

• plot function for plotting the pdf, cdf, survival, quantile, hazard, cumhazard or distr6 objects
• lines function for superimposing distr6 plots
• qqplot function for comparing distr6 distributions to each other or to other theoretical distributions
• Added Extract.VectorDistribution for extracted distributions from inside a VectorDistribution, see the big update below.

Updated Functions

• decorate now allows users to specify the Decorator as a character as well as supplying the object, this makes it simpler when using distr6 whilst unattached
• Big changes (and hopefully final) to the VectorDistribution. Now the VectorDistribution only constructs the internal wrapped distributions when they are extracted or when a function, such as d/p/q/r, are called. This massively reduces a bottleneck in constructing the distribution. Additionally added functions for extracting distributions from inside the VectorDistribution. The only difference that should affect backwards compatibility is that the distribution argument must now be a character and not an object. Custom (i.e. non-SDistribution) distributions should be used in conjunction with the distlist initializer.
• Changed the lower bound of positive Sets to .Machine$double.xmin as the previous value of 1.1e-15 was too restrictive
• Added skewness, kurtosis, entropy, mgf, cf, and pgf to WeightedDiscrete and Empirical
• Added support for custom distributions in VectorDistribution and for CoreStatistics functions as well as support for only one arguments passed to d/p/q/r for fast comparisons between wrapped distributions

Patches

• Bug fix in WeightedDiscrete distribution variance calculation
• Fixed bug in Empirical that was stopping the cdf of the first point in the distribution being evaluated
• Fixed bug that allowed invalid parameter values to be set for non-reference parameters
• Updated parameter error messages to be more informative
• Improved speed and efficiency in Distribution constructor for wrappers

distr6 1.2.0

Minor Updates

Added Functions and Classes

• Added non-central F, T, Chi-Squared, and Beta distributions
• Added WeightedDiscrete distribution. This allows users to supply a data.frame of samples and pdf/cdfs for more efficient usage with empirical discrete estimators (e.g. Kaplan-Meier)

Updated Functions

• Added the common survival parameterisation of Weibull distribution
• Updated the listX functions to make them quicker and functional when distr6 is not attached
• Added suppressMoments argument to Distribution constructor to allow for faster construction

Patches

• Added errormsg argument to assertions so a custom error message can be provided
• Fixed error in quantile function in Empirical that was causing results to be dropped
• Fixed bug in TruncatedDistribution that prevented multivariate distributions being truncated
• Fixed error in variance of Empirical that was calculating sample not population
• Moved rare imports to suggests

distr6 1.1.0

Minor Updates

Added Functions and Classes

• Empirical distribution for distributions arising from observed samples (e.g. from MC sampling methods)
• simulateEmpiricalDistribution: function for sampling without replacement from an Empirical distribution

Deprecated Functions

• ArrayDistribution - This wrapper is now merged with ProductDistribution by adding two new parameters to ProductDistribution that allow for construction either by a distribution list of by naming a distribution and providing a parameter list or data.table

Updated Functions

• Added quantile to Categorical distribution and updated its cdf efficiency
• cdf of discrete distributions evaluated between integers are now evaluated after rounding down and do not return 0, e.g. Binomial$new()$cdf(1.8) == Binomial$new()$cdf(1)
• Added two additional parameters to ProductDistribution: params and distribution, so that it can either function as before or as the ArrayDistribution wrapper which is now deprecated.
• Added two additional parameters to VectorDistribution: params and distribution, so that it can either function as before or as a generalisation to the ArrayDistribution wrapper which is now deprecated.
• Removed automated rounding of numerics when ParameterSet expected integer in setParameterValue
• Added parameter n to strprint and print to clean the print method for distributions (especially wrappers) with multiple parameters
• Added optional vectordist argument to MixtureDistribution for better compatibility with VectorDistribution
• Changed short_name of Uniform, Logistic, Normal, and Triangular kernels to remove Kern. The ClassNames (and thus constructors) remain the same
• listKernels and listSpecialSet now return data.tables with stringsAsFactors = FALSE
• Degenerate distribution re-classified as discrete

Patches

• Bug fix in quantile function of huberization wrapper
• Fixed the rand return of a Vector Distribution (transposed data.table and added column names)
• Fixed spelling mistakes in extension guidelines and removed calls to redundant parameters
• Removed redundant crayon dependency
• Fixed bug in MixtureDistribution that was causing rand to erroneously return integers
• Removed error in Categorical documentation
• Fixed bug in multi-modal Categorical distribution
• summary and print now return self invisibly for better piping

distr6 1.0.1

• Updated vignettes to remove redundant distr6 installation via GitHub that was causing a warning on Debian
• Redundant stopwarn argument removed from parameters() method in Distribution and ParameterSet classes
• Updated Kernel$new() error message to point user to Distribution class
• Fixed broken if/else statement in liesInSetInterval method of SetInterval class

distr6 1.0.0

• v1 API is ready to be shipped to CRAN!
• Again see here for the updated API lifecycle
• Tutorials, extension guidelines and other documentation (in appendices) are now on the website
• distr6 is a complete, unified, object-oriented interface to probability distributions in R. Complete with 36 probability distributions and another 11 kernels, distr6 also allows functionality for numerical imputation of methods and statistical functions.

distr6 0.1.0.9000

• Now public!
• distr6 API as whole is still very much experimental but analytic features are fairly stable, see here for details on feature lifecycles
• Development continues to be internal, through UCL and The Alan Turing Institute, but we welcome external users to test the API and report bugs, see our contributing guidelines for details including our code of conduct

alan-turing-institute/distr6 documentation built on Feb. 26, 2024, 11 a.m.