Version 2.3

Version 2.3

Bug fixes

• rtnorm now checks arguments length and throw an error rather than try to recycle mean and standard deviation vectors (issue #6).
• rtnorm recycles arguments when ltrunc or rtrunc are length 1.
• *tmvt and *tmvnorm now check that the scale matrix sigma is symmetric, positive-definite and non-degenerate (issue #8) by default.
• rmvnorm returns a vector rather than an 1 by n matrix for the unidimensional setting.
• Degrees of freedom larger than 350 are not treated as infinite, as the code suffers from overflow and returns missing values otherwise.

Changes

• The package uses tinytest rather than testthat for unit tests.
• Since scrambling is disabled in randtoolbox (throwing warning messages), the package now relies on the qrng package and the Owen scrambling method from spacefillr package for quasi Monte Carlo.
• Added alias in package documentation, as requested by Kurt Hornik.

v2.2

Fix error in sanity checks

v2.1

• Change back to scrambled Sobol via randtoolbox
• Fix segfault bug #2
• Copy back probit example to documentation following the publication Durante (2019), Conjugate Bayes for probit regression via unified skew-normal distributions, Biometrika

v2.0

Version 2.0

• Can specify mean vector directly in the functions
• Functions for multivariate Student simulation ported from Matlab
• Change to the interface
• Many functions now internal (documented)
• Cholesky matrix permutation (GGB and GE reordering now supported)
• Selected code rewritten in Rcpp
• New vignette
• Removed random seed
• Change to the Sobol sequence because previously used package is unmaintained (no more scrambling for QMC)
• Throw error if NaN in bounds
• Handle univariate cases in mvrandt and mvrandn
• Use nleq solver to solve convex program (also check KKT conditions)