R-Forge Project: distr - S4 classes for distributions - News

distr release 2.8

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Wed, 10 Apr 2019 10:15:27 GMT

================================================================================
### distr release 2.8 ###
================================================================================

Updates for the packages of the distr family are now avaialable on CRAN in
version >= 2.8.0.

The most important changes are enhancements as to
- implementation of distributions
- generic integration/expectation
- plotting
- minimum criterion estimators

For some details see points D-1 to D-4 below, for more details please see the
NEWS files in the packages, available as NEWS("<pkgname>").

--------------------------------------------------------------------------------
Some details
--------------------------------------------------------------------------------
================================================================================
### distr release 2.8 ###
================================================================================
%---
D-1. enhanced implementation of distributions:
%---
+ Function liesInSupport() gains an argument checkFin which in case of
DiscreteDistributions, implements information on the finiteness of
the support.
+ We revised the code determine the quantile function when there are gaps in
the support of a distribution with non-trivial (abs.) continuous part.
This fixes an issue recently raised at the introduction of staged
installation.
%---
D-2. enhanced integration/expectation:
%---
+ The return values of distrExIntegrate and all E()-methods gain an optional
attribute "diagnostic" for information on integration accuracy etc; this
comes with functionality to inspect/access this information through functions
showDiagnostic, getDiagnostic and the S3method for print for S3class
DiagnosticClass
+ We introduce filter functions to warrant some safety when digesting "..."
args for distrExIntegrate(), E(), integrate(), GLIntegrate(), quantiles
and IQR
+ We enhance method .qtlIntegrate to integrate w.r.t. the quantile
function for expectations (e.g. for Weibull, GPD, GEVD, Gamma)
%---
D-3. enhanced plotting:
%---
+ Our plot methods gain a return object of S3 class c("plotInfo","DiagnInfo"),
containing the information needed to (re)produce the respective plot, which
at a later stage could be used by different graphic engines (like, e.g.
ggplot) to produce the plot in a different framework. A more detailed
description will follow in a subsequent version.
+ For diagnostic plots to be displayed in several windows, we enhance our helper
function devNew() to react on the opening of two many devices: in interactive
mode it now asks the user to shut some devices after reaching a certain
threshold of opened devices; in non-interactive mode, when this threshold is
reached, it shuts the first 15 open devices; see ?devNew
%---
D-4. enhanced minimum criterion estimators
%---
+ cleverer parsing of "..." args in M[L,C,D]Estimator:
(where: ML = MaxLikelihood, MC = MinCriterion, MD = MinDistance)
We actively filter out arguments causing clashes with formals in
optim/optimize and calling arguments; from the remaining ones only [exactly]
_named_ arguments of the optimizer (optim/optimize) and those matching either exactly
arguments of the criterion or all remaining ones (if "..." is a formal of the
criterion) are passed on.
+ For diagnostic purposes, MCEstimate-class gains a slot 'optimReturn' which is
filled by the return value of the optimizer in "mceCalc" -- it has a
corresponding accessor
+ For Cramer von Mises MD-Estimator, we distinguish 3 cases
selected by argument muDatOrMod = c("Dat","Mod", "Other"):
* in case "Dat", we integrate w.r.t the emp. cdf (default),
* in case "Mod", we integrate w.r.t. the current best fit model distribution,
* and in case "Other" one has to supply an integration probability mu.
The provided asymptotic covariances reflect this choice.
+ We add some theory/references to help file to MD estimators.

distr release 2.7

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Mon, 20 Aug 2018 07:25:35 GMT

Updates for the packages of the distr family are now avaialable on CRAN in
version >= 2.7.0.

The most important changes are:
- we switch from SweaveListingUtils (no longer maintained as announced
prior to 2015) to knitr in vignettes
- graphics (plot-methods) are now more stringently vectorized in their arguments
- internally, we use accessor q.l instead of q for the quantiles which
makes our packages available for use with Jupyter IRKernel and easier
with RStudio (both catch calls to q() and treat them differently to
standard R)

For details please see the NEWS files in the packages, available as
NEWS("<pkgname>").

Versions 2.7 of the distr family on CRAN

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Tue, 24 Jul 2018 23:30:28 GMT

Up to pkg 'distrMod' all packages in the distr family of packages are on CRAN in version >= 2.7.0 now. 'distrMod' is to follow soon, together with the 1.1-release of the RobASt family
of packages.

Versions 2.6 of the distr family on CRAN

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Mon, 25 Apr 2016 19:18:12 GMT

With the usual mirror delay, versions 2.6 of the packages in the distr family are available on CRAN in the next few days.

Version 2.4 of distr packages on CRAN soon

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Fri, 08 Feb 2013 19:03:10 GMT

New versions 2.4 of our distr-family of packages are submitted to CRAN and will soon be available on all mirrors

(i.e.; startupmsg, SweaveListingUtils,
distr, distrEx, distrDoc, distrEllipse,
distrMod, distrSim, distrTEst, distrTeach)

and a new package distrRmetrics with distr support for
(some) distributions provided in the Rmetrics family of packages.

For more details see the corresponding NEWS files
in the pkgs, e.g. news(package = "distr") or
NEWS("distr") [the latter using fct NEWS from pkg
startupmsg].

Version 2.3 of distr packages on CRAN soon

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Thu, 20 Jan 2011 09:48:58 GMT

New versions 2.3 of our distr-family of packages will be available on CRAN soon
[we have just uploaded them...]

Most importantly, we have included:

+ a quasi-MC trick by Nataliya Horbenko to better produce
image distributions under complicated not necessarily
monotone transformations
+ enhanced function qqplot
+ (enhanced) support for GEV distribution
+ new functional kMad
+ as well as several bug fixes

For more details see the corresponding NEWS files
(e.g. news(package = "distr")
or using function NEWS from package startupmsg
i.e. NEWS("distr")).

Best
Peter
Matthias
Nataliya

Version 2.2 of distr-family on CRAN

stamats@users.r-forge.r-project.org (Matthias Kohl) — Mon, 09 Nov 2009 14:43:16 GMT

We recently submitted version 2.2 of the distr-family of packages to CRAN where the new package distrEllipse, which includes S4 classes and methods for elliptical symmetric distributions, was added.
Peter
Matthias

Versions 2.1 of the distrXXX family of packages available

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Fri, 27 Mar 2009 13:01:06 GMT

After having merged devel branch 2.1 of the distrXXX family of packages back into the trunc, these versions will be available for download/installation on R-Forge after nightly builds.

They will also be released to CRAN in the next few days.

For a detailed overview for this family of packages, consider (large) vignette "distr"
in package distrDoc herein.

As to the new versions, there will be a more detailed announcement on R-pkgs after the release on CRAN.

Best, Peter & Matthias

New (patched) versions to be released on CRAN as of Oct 11

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Tue, 14 Oct 2008 10:17:44 GMT

After some changes in the S4 system by John Chambers, traced by http://developer.r-project.org/blosxom.cgi/R-devel to:

+ treatment of "class" "matrix", Mon, 06 Oct 2008,
+ treatment of "setIs" relations Fri, 19 Sep 2008,

we had to make some minor modifications to our code; this should do now.

New versions released on CRAN

ruckdeschel@users.r-forge.r-project.org (Peter Ruckdeschel) — Fri, 12 Sep 2008 18:20:50 GMT

-------------------------------------------------------------------------------
New versions released for the distrXXX family of package
-------------------------------------------------------------------------------

We would like to announce the availability on CRAN (with possibly a minor delay
until on every mirror) of new versions of our packages in the "distrXXX"-family
(version 2.0), i.e.; "distr", "distrEx", "distrSim", "distrTEst", and
"distrDoc", as well as of the new packages "distrMod" and "distrTeach".

[all of them require R >= 2.2.0]

For details, please also refer to the NEWS file of the corresponding package,
accessible via NEWS("<pkgname>") [after require(<pkgname>)]

-------------------------------------------------------------------------------
Devel versions on r-forge
-------------------------------------------------------------------------------

Please note that from this version on, we have moved development of these
packages under r-forge project /distr/:

http://r-forge.r-project.org/projects/distr/
http://distr.r-forge.r-project.org/

If you find this project interesting and would like to collaborate, you are
warmly welcome. You may find more information on how to collaborate under

http://distr.r-forge.r-project.org/HOWTO-collaborate.txt

We look forward to receiving questions, comments and suggestions

Peter Ruckdeschel
Matthias Kohl

-------------------------------------------------------------------------------
Major news in package "distr" :: extended arithmetics
-------------------------------------------------------------------------------

We have extended the arithmetics of (univariate) distributions: besides
convolution and affine linear transformations (with constant coefficients),
now also multiplication, division, exponentiation understood as binary
operations acting on distributions, as well as min-- and maximum, truncation
and Huberization of distributions are realized analytically.
Moreover, we introduce new S4 classes for univariate mixing as well as Lebesgue
decomposition of distributions; confer class?UnivarMixingDistribution,
class?UnivarLebDecDistribution and the example sections therein.

-------------------------------------------------------------------------------
New package "distrMod"
-------------------------------------------------------------------------------

This new package is to provide S4 class infrastructure for parametric models.

For the parameter of these families we introduce class 'ParamFamParameter'
which allows for a partition of the parameter into a main, fixed and nuisance
part as well as for (smooth) transformations of the parameter; for details
confer the help pages and the vignette in (updated) package "distrDoc".

For estimation in smooth parametric models, we introduce class 'L2ParamFamily'
which [most importantly] has slots for the distribution of the observations,
for the parameter, for the scores function and the Fisher information and,
to be able to ``move model P_theta from theta to another parameter value
theta', we have functional slots realizing maps
theta -> Scores_theta, theta -> Fisher_Info_theta,
theta -> observation_distribution_theta.

For 'L2ParamFamily' objects, we define 'Minimum Criterium Estimators' (MCEs)
(with corresponding S4 class 'MCEstimate'), i.e. estimators which are defined
as minimizers of a certain criterium. Particular cases are Maximum Likelihood
and Minimum Distance Estimators (the latter are available for Kolmogorov,
Hellinger, total variation, and Cramér von Mises distance)

We also have corresponding 'confint' methods to produce asymptotic confidence
intervals.

Also there is a coercion method to class 'mle' of package 'stats4',
so methods available for class 'mle' are also available for class
'MCEstimate'.

+++ beyond 'fitdistr' and 'mle'

The implementation of our S4-class approach goes beyond 'fitdistr' of
Venables/Ripley's MASS package and 'mle' of package 'stats4' in the following
sense:
MCEs can (but do not need to) use method dispatch to decide on runtime by
which way to produce the corresponding estimate; this way particular
methods beyond numerical optimization (like 'mean' in Gaussian location)
can be used. More importantly, alternative methods may also be defined
later, in other packages and by other people without interfering with
existing MCE code or even without just to signal their mere existence
to the package maintainer of package "distrMod".

This is not possible for 'mle' and 'fitdistr': 'mle' only allows
for calls to 'optim' and 'optimize', while 'fitdistr' only allows for
a restricted number of alternatives to branch for, which is "hard-coded"
within 'fitdistr'-code, so extensions could not be made from outside of
this function; the implemented alternatives include Gaussian location,
though.

For examples how easy one can define new classes / new particular methods
confer the scripts/demos in this package.

-------------------------------------------------------------------------------
New package "distrTeach"
-------------------------------------------------------------------------------

For the use of R in high school, we have initiated a new package "distrTeach".
So far we have moved illustrations illustCLT, illustCLT_tcl and illustLLN
from package "distrEx" to this new package.
Later on, it is to contain the results of the graduate thesis project by
Eleonare Feist and Anja Hueller, in particular some S4-class infrastructure
for small teaching units in Stochastics and Statistics.