NEWS

Generated header file, inst/include/extraDistr.h, to make C++ code callable from C++ code in other R packages (#23, contribution from Jon Fintzi, @fintzij).

Now zero-inflated negative-binomial (#12) and beta negative-binomial distributions (#14) do not have integer-only constraint on size parameter.
Fixed mistakes in the zero-inflated Poisson distribution documentation (#15) and negative hypergeometric distribution documentation.
Minor changes towards future Rcpp STRICT_R_HEADERS compatibility.
Fixed bug in vectorization code for pbbinom (#16). Additionally, this should make pbbinom and pbnbinom faster when working with large vectors.

Fixed bug in dinvgamma function
Updated the DESCRIPTION file by mentioning the packages used in optional unit tests for CRAN compatibility

Deprecated functions: dnst, pnst, qnst and rnst were removed
Fixed typos in documentation (thanks to #8 by philchalmers)
Converted a number of pdf and cdf functions to more numerically stable versions using logs (see #9 and #10)
Fixed bug in ppower: with lower.tail = FALSE it returned wrong values
Fixed bug in dgpd and pgpd: they assumed slightly wrong support
Improved the rgev and rgpd, now they give better random values since relying on exponential distribution random generator if possible
Documentation was improved and corrected in several places
Power distribution functions now check if alpha > 0 and beta > 0

Fixed bug in pinvgamma (lower.tail and log.p didn’t work)
Fixed underflow issues with rmnom and rdirmnom (#7)
The *nst functions are now deprecated and renamed to more informative abbrevation *lst

Now consistently with base R only the first elements of the logical arguments are used (thanks to #5)
Fixed bug in rtnorm (sampling from lower bound returned incorrect values)
When computation becomes slow, now pbnbinom, pbbinom, pgpois functions are easier to brake
Automatically registering native routines via Rcpp
Fixed bug in pinvgamma (it returned non-zero probabilities for q < 0)

Now rmnom and rdirmnom (issue #3) do not return NaN’s due to underflow issues
Fixed bug (issue #4) that resulted in hanging R if zero-length vectors were provided as input

Random generation from categorical distribution parametrized by log-probabilities rcatlp
Re-parametrized beta distribution is now more flexible since “prior” parameter was introduced (see documentation); this change is connected to switching to different prior equal to zero (instead of one as previously). Such parametrization is more commonly used in the literature. This change is documented in the ?PropBeta documentation
Improvements in documentation and examples
Fixed bugs in pbbinom, pbnbinom, pgpois, *nhyper that prevented compiling on RedHat Linux (#2)

Switched to C++11, underlying code was simplified and improved
Using memoization techniques for pbbinom, pbnbinom, pgpois and negative hypergeometric distribution that lead to major speed improvements
Improved and simplified code for beta-binomial, beta negative-binomial, and gamma-Poisson cdf; now recursive algorithm employing memoization techniques is used what noticeably improves performance
Discrete gamma, shifted Gompertz (d,p,r), negative hypergeometric and truncated binomial distributions (d,p,q,r) were added
Now lower.tail and log.p options for pbetapr work properly
Simplified code for multivariate hypergeometric, multinomial, Dirichlet-multinomial and categorical distributions
Truncated poisson distribution is now parametrized by lower and upper truncation points
Fixed bugs in dbvpois (it returned underestimated probabilities), dslash (there was discontinuity at x=0), pcat (randomly it broke if x was greater then the upper limit), and pdnorm.
Random generation functions throw warnings and produce NA’s on NA’s in parameters or incorrect parameters - as in base R
Order of parameters in discrete Laplace distribution was changed to location and scale (vs scale and location) for consistency with other distributions (e.g. continous Laplace)
Improved exception handling

Corrected and simplified documentation for *prop distribution
Categorical, multinomial, mixture of normals and mixture of Poisson distributions are now less restrictive about probability parameters and accept any non-negative values. Probability parameter vectors are normalized to sum to one (i.e. c(1,1,1) becomes c(1,1,1)/3)
NA’s and NaN’s in input now always lead to NA’s in output
Fixed bugs in rtnorm, now it properly handles sampling from extreme tails
Tests for multivariate distribution now done with tolerance 1e-2

Bug fixes in qtnorm - now it works properly for non-standard truncated normal
Bug fixes for rmnom and rdirmnom - now they correctly use the prob parameter values
Minor improvements in dmvhyper and dbvpois
Cleaned-up the documentation
Added discrete Laplace distribution
Faster RNG generator for Laplace distribution
Changes to using lower level RNG functions (unif_rand, norm_rand) when possible
More tests

Documentation clean-up
Minor improvements and simplifications in C++ code
For min == max discrete uniform distribution behaves now as degenerate distribution

Added Dirichlet-multinomial and beta prime distributions
Code clean-up for categorical distribution
Improvements in documentation and examples (e.g. bivariate Poisson, bivariate normal)
Other minor improvements in the code

Cleaning namespace - now mathematical functions are explicitly called from std library
“using namespace Rcpp” was removed from all the files
All the numerical values are now explicitly <double>’s, or casted to <double>; or <int>’s (for indexing)
Improvements in algorithms for discrete uniform, categorical, mixture of normal, mixture of Poisson distributions
Improvements in discrete uniform; now it accepts only integer valued parameters
Major C++ code clean-up

Removed erf, erfc, inv_erf that are not used at this moment and caused problems when compiling on Fedora and Solaris
Minor improvements in documentation
Added mixtures of normal and of Poisson distributions
Added truncated Poisson distribution

Minor changes and improvements in documention
Truncated normal returns normal for infinite truncation points
Added half-t, half-normal and half-Cauchy distributions
Changed naming of data-variable from x to q for CDF’s
Added inverse-CDF for discrete normal distribution, fixed random generation
Added tests
Multiple minor bug fixes (e.g. functions returning NaN instead of 0 for Inf values)
Added Birnbaum-Saunders (fatigue life) distribution
New algorithm for rtriang
Added quantile functions for: zero-inflated Poisson, zero-inflated binomial zero-inflated negative binomial, inverse gamma, inverse chi-squared distributions
Added Huber density

Exception handling in discrete uniform distribution functions
Bug fixes in rdunif
Clean-up in documentation and examples
Now dbvnorm and dbvpois work with matrixes
Renaming of parameters in bivariate normal to be consistent with base R normal distribution

Ranamed zero-inflated distributions to *zip and *zinb
Added zero-inflated binomial *zib
Code clean-up
Added pzipois and pzibinom
Added qlgser; changes in plgser and rlgser
Fixed bug in rlgser
Examples for most of the functions
Exception handling for dmvhyper and rmvhyper: values of x, n, and k are checked against being non-integers