corpus.useR.2008.abstracts was removed as suggested package and the dataset added to the package to ensure that the vignette code can be successfully executed even if the package is not available for installation.
Performance tweaks for random number generation from the vMF distribution.
Update URLs.
Model driver added to extend package flexmix.
Package HSAUR3 now used instead of HSAUR2 for
dataset household
.
Minor vignette improvements.
A bug in function solve_kappa_uniroot
is now corrected.
Minor vignette improvements.
The information on publication of the vignette in JSS is now also mentioned in the vignette.
The package vignette has been published in the Journal of
Statistical Software, Volume 58, Issue 10
(doi:10.18637/jss.v058.i10), and the paper should be
used as citation for the package, run citation("movMF")
for
details.
In the calculation of 0F1 it is now ensured that summation is continued as long as the terms are increasing and only then numerical convergence is checked as stopping criterion.
For simulating from the vMF distribution the necesssary quantities in the rejection sampling are now calculated in a numerically more stable way. Thanks to Mike Allerhand for pointing out that problems did arise for large values of the precision parameter for the previous implementation.
Package corpus.useR.2008.abstracts is now listed in the DESCRIPTION file together with the information that is available from the additional repository http://datacube.wu.ac.at.
An improved approximation is used to calculate the normalizing constants of the log-likelihoods. See the package vignette for more information.
For determining the concentration parameters the following
additonal methods were implemented: "Halley"
, "hybrid"
and "Newton_Fourier"
. The default method is now
"Newton_Fourier"
.
Package SnowballC is now suggested instead of Snowball.
Package now uses registration for native (C) routines.
For determining the concentration parameters the additional
method "Song_et_al_2012"
was added which performs 2 Halley
steps.
A vignette documenting the use of the package and some implementation issues is added.
Function A()
and the functions of its first and second
derivative now use the Taylor series expansion for small values of
kappa.
Function A()
modified to have now three methods
("PCF"
: Perron continued fraction, "GCF"
: Gauss
continued fraction, "RH"
: using a ratio of generalized
hypergeometric series). By default, method "PCF"
is used.
The random initializations for multiple runs are now determined within the loop separately for each run.