2012-11-15 Martin Maechler
* DESCRIPTION (Version): 0.999-4
* tests/rstable-ex.R: add a check for U[0,1] margins for rCopula(.)
of "all" our copulas.
* R/stable.R (rPosStableS): is a *scaled* version of rstable(), and
it was always *correct* to omit the factor cos(.)^(1/alpha) factor,
but the docu (which implicitly said gamma=1) was wrong.
* R/gumbelCopula.R (rgumbelCopula): needs rstable1(..., gamma = GG)
where GG is *not* == 1.
2012-08-13 Martin Maechler
* DESCRIPTION (Version): 0.999-1, r 835, released to CRAN.
2012-07-02 Martin Maechler
* R/fitCopula.R: t copulas now can be fit with and without fixed
'df'. With varying df ('df.fixed=FALSE'), it no longer needs an
initial parameter estimate 'start'.
2012-06-20 Martin Maechler
* DESCRIPTION (Version): 0.99-4 after fixing the gnacopula() bugs.
2012-02-21 Martin Maechler
* DESCRIPTION (Description): big merge of 'copula' + 'nacopula'
2012-02-06 Martin Maechler
* DESCRIPTION (Version): 0.8-0, released to CRAN
2012-01-20 Martin Maechler
* R/AllClass.R (nesdepth): nesdepth() gives the nesting depth.
2011-12-27 Martin Maechler
* R/aux-acopula.R (coeffG): 'method' now contains all dsumSibuya()
methods; former "dsumSibuya" is deprecated in favor of "dsSib.log".
2011-11-24 Ivan Kojadinovic
* Added the function exchTest (resp. exchEVTest) that can be used
to test whether data come from an exchangeable bivariate copula
(resp. an exchangeable bivariate extreme-value copula).
* Added the hideWarnings flag in fitCopula. Also, maxit set to
1000 by default when calling optim.
* Corrected a few bugs. Many thanks for all the reports.
2011-08-31 Martin Maechler
* DESCRIPTION (Version): 0.7-9, ready for CRAN; released as 0.7-9-1, 2011-09-23
2011-08-30 Martin Maechler
* R/special-func.R (log1mexpm): more careful with NAs.
2011-06-26 Martin Maechler
* R/special-func.R (Bernoulli, Bernoulli.all): first "experiments":
need accuracy even more than speed.
2011-04-28 Martin Maechler
* R/special-func.R (polylog): introduce 'is.minus.log' so we can
get much more accurate Frank density for high correlation.
2010-07-01 Martin Maechler
* DESCRIPTION (Date): ready for CRAN submission
2010-06-26 Martin Maechler
* R/nacopula.R (onacopula): allow 'family' be a family object as
well (not only a family *name*) ===> can use it with opower()-cops!
(onacopulaL): new function, working recursively with *list*
argument.
2010-06-17 Martin Maechler
* man/tauAMH.Rd, NAMESPACE: add help file and export
2010-06-16 Martin Maechler
* R/aux-acopula.R (tauAMH): nicely working for theta -> 0
2010-06-10 Martin Maechler
* R/aux-acopula.R (retstableC): maybe use *both* methods, when V0
is vector; new printAcopula()
(printNacopula): as well: now show() is nice.
2010-06-09 Martin Maechler
* R/safeUroot.R: add 'check.conv' argument.
2010-05-27 Ivan Kojadinovic and Jun Yan and Jun Yan
* R/rstable1.R (rstable1C): use rstable1 := rstable1C
and look at it --> do *NOT* pass gamma to C as it is sometimes a vector.
* src/retstable.c (rstable_c, ...): corresponding changes
2010-05-04 Martin Maechler
* src/retstable.c (retstable_c): with 'method', instead of
retstable_MH_c() etc.
* tests/retstable-ex.R (histSt): add first comparisons "MH" <--> "LD"
LD is much faster but looks wrong.
* DESCRIPTION (Version): 0.3-0
* R/aux-ACopula.R (retstableC): finally switch to retstableC()
which now works; new 'method' argument, in tests only using
retstable_MH at the moment.
* tests/NAC-experi.R (print.chiSqChk_cop3d): need to allow P-value < 0.05
* src/init.c (updates)
2010-05-03 Marius Hofert
* src/retstable.c (retstable_LD): Luc Devroye's algorithm for
tilted stable added; fixed buglets; some renaming (after meeting
with MM).
* src/nacopula.h, src/init.c: renaming
2010-04-28 Marius Hofert and Martin Maechler
* src/retstable.c: new retstable() versions; but buggy, according
to our distribution tests in tests/NAC-experi.R
2010-04-27 Ivan Kojadinovic and Jun Yan
* Improved evTestC.
* Added evTestK.
* Added calibKendallsTau and calibSpearmansRho for the FGM copula.
* Improved documentation and changed CITATION file.
2010-04-10 Martin Maechler
* DESCRIPTION (Version): 0.2-2
* R/nacopula.R (onACopula), man/onACopula.Rd: a - correct -
constructor for "outer_nACopula".
* tests/NAC-experi.R: using onACopula() consequently.
* ChangeLog: update it hugely (using svn2cl and much editing).
2010-04-09 Martin Maechler
* inst/doc/Makefile:
add Makefile for my convenience
* tests/NAC-experi.R, tests/retstable-ex.R, tests/retstable_Nstat.rda:
investigate #{calls} to rstable1(1, *)
* R/aux-ACopula.R (retstable), src/retstable.c,..: retstable() C
version: correct interface, wrong result.
2010-04-08 Martin Maechler
* src/init.c, src/nacopula.h, src/retstable.c, *:
retstable using eval(); using registration and nacopula.h try
fixing mess of rfjoe/rFJoe by reverting to original name
2010-04-07 Marius Hofert
* src/retstable.c: first C version; does not work (...)
2010-04-07 Martin Maechler
* NAMESPACE, R/nacopula.R, src/rfjoe.c (rFJoe): port to C
2010-04-05 Martin Maechler
* R/*.R, inst/doc/nacopula.Rnw, tests/*.R:
ensure copula function()s also work with NULL instead of
numeric(0); improve testing output; cosmetics
2010-04-04 Marius Hofert
* tests/NAC-experi.R:
now, NAC-experi.R works! (except for Gumbel, see FIXME)
* R/nacopula.R (rn): solved the problem!
* tests/NAC-experi.R: add Chi-Square Test. --> problems found
2010-04-03 Martin Maechler
* inst/doc/nacopula.Rnw:
solve Sweave.sty/jss.cls issue; use {bm} instead of {bbm}
for indicator
* inst/doc/jss.bst, ...: add jss files
* R/cop_objects.R,...: cosmetic (style) improvements.
2010-04-02 Marius Hofert
* R/cop_objects.R: fixed error in rFJoe, ....
2010-04-01 Marius Hofert
* R/cop_objects.R: much smaller default 'noTerms' for tau in Joe.
Leads to a uniformly bounded error (over all theta) < 10^-5.
* tests/NAC-experi.R: taueps adjusted in order to pass ...
* R/aux-ACopula.R (retstable: simplified
* tests/copula-play.R: cleaned up; works flawlessly
* tests/NAC-experi.R: clean up; added large-dim. example
2010-04-01 Martin Maechler
* R/cop_objects.R (rlog): nomore extraneous runif(); slight
simplification.
2010-04-01 Marius Hofert
* R/cop_objects.R: slight improvement for rejFFrank
* R/aux-ACopula.R (retstable): vectorized: reduces runtime for
Clayton by 60%.
2010-03-31 Marius Hofert
* R/cop_objects.R (copGumbel): Vectorized V01().
* R/cop_objects.R (rnchild): fixed error
* R/nacopula.R (rn): found and fixed sorting error.
* tests/NAC-experi.R: improvement; ....
* inst/doc/nacopula.Rnw: finished theory part
2010-03-30 Martin Maechler
* inst/doc/nacopula.Rnw: first version, 99.9% from Marius.
2010-03-29 Martin Maechler
* DESCRIPTION (Version): 0.2-0 {up from 0.1-1};
(Imports): gsl [no longer in 'Depends']
* NAMESPACE: newly added namespace.
* R/nacopula.R (rn, rnchild): fix another missing lapply().
* man/rn.Rd, man/rnchild.Rd, man/value-methods.Rd: rudimentary
2010-03-29 Marius Hofert
* man/*.Rd, R/*.R: renaming class to "outer_nACopula"
2010-03-22 Marius Hofert
* R/nacopula.R: fixed error
* tests/NAC-experi.R: added tests of rn().
(almost) implemented "mean example".
* R/nacopula.R (rn): first trial implementation.
2010-03-18 Marius Hofert
* R/cop_objects.R: improved Frank and Joe.
2010-03-15 Marius Hofert , Ivan Kojadinovic , and Jun Yan
* R/cop_objects.R (Frank, Joe): new versions for V0() and V01().
* Corrected a bug in rcopula for claytonCopula for dim >= 3.
Thanks to Zhongyi Yuan .
* Added mvtnorm in NAMESPACE to make the package work if these
functions are overwritten by other packages:
importFrom(mvtnorm, dmvnorm, pmvnorm, dmvt, pmvt)
2010-02-02 Marius Hofert
* R/cop_objects.R (Frank, Joe): Improved versions for sampling.
2010-01-14 Marius Hofert
* R/cop_objects.R (AMH): Sampling V01() now directly via rnbinom()
2009-12-15 Marius Hofert
* R/cop_objects.R (Joe): new idea for sampling.
* tests/copula-play.R (tstCop): included "Joe" again.
2009-12-14 Marius Hofert
* tests/copula-play.R (tstCop): Added tests for psi(), psiInv()
+ boundaries ==> all work
2009-12-10 Martin Maechler
* tests/copula-play.R (tstCop): improve, measuring CPU time used.
* R/safeUroot.R (safeUroot): use instead of just uniroot().
2009-12-09 Marius Hofert
* tests/NAC-experi.R: compute "by hand" and compare.
2009-12-08 Martin Maechler
* DESCRIPTION, R/cop_objects.R (Frank): use gsl::debye_1() for tau().
* R/*.R, tests/NAC-experi.R: value() fix and first test.
2009-12-07 Marius Hofert
* tests/copula-play.R: testfunction added
* R/nacopula.R (value): written; not yet tested
* R/cop_objects.R: vectorized tau()
2009-11-24 Marius Hofert
* R/cop_objects.R (Frank): found & fixed error from
Devroye(1986) which is wrong!
2009-11-23 Martin Maechler
* R/MH-Acopula.r, tests/*.R: fix, extend, ...
2009-11-23 Marius Hofert
* R/AllClass.R: ..
* R/cop_objects.R: fixes
2009-11-17 Marius Hofert
* R/cop_objects.R: new: moved functions from MH-Acopula.r
2009-11-17 Martin Maechler
* DESCRIPTION, R/AllClass.R (dim): change to validity for nAC*.
2009-11-16 Martin Maechler
* R/AllClass.R (initialize): method for ACopula which
auto-constructs the "paraConstr" function from the "paraInterval"
specification; allow to set parameter "theta" of ACopula
* R/rstable1.R: tweaks
* R/Auxiliaries.R, man/interval-class.Rd: add "interval" class and
interval() constructor
2009-11-16 Marius Hofert
* R/cop_objects.R: parameter interval for copAMH
2009-11-14 Martin Maechler
* R/rstable1.R, man/rstable1.Rd: fix rstable(alpha = 1, beta != 0)
2009-11-10 Martin Maechler
* R/MH-Acopula.r: original code of Marius Hofert
2009-10-07 Ivan Kojadinovic and Jun Yan
* Removed dependence on package adapt, which has been withdrawn from CRAN
due to license issues.
2009-09-22 Ivan Kojadinovic and Jun Yan
* Added the function evTestC that can be used to test whether data come from
an extreme-value copula.
2009-07-28 Ivan Kojadinovic and Jun Yan
* Fixed a bug in rcopula for claytonCopula when dependence is negative.
Thanks to Dominik Sznajder for
reporting it.
2009-07-23 Ivan Kojadinovic and Jun Yan
* Fixed a bug in rcopula for frankCopula when dependence is negative.
* Default df for tCopula and tevCopula are set at 4.
2009-06-04 Ivan Kojadinovic and Jun Yan
* Significantly improved the multiplier version of gofCopula.
There are no more repeatability problems and the test is slightly
faster. Documentation updated.
* Fixed various warning messages.
2009-01-22 Jun Yan and Ivan Kojadinovic
* Fixed a bug in the random number generation for the
Husler-Reiss family.
* Changed the default optimization method in fitCopula to
"Nelder-Mead".
* Added "optim.method" parameter to "gofCopula".
* The functions "fitCopula" and "gofCopula" should now work
for the Husler-Reiss and the Galambos families. The derivatives
w.r.t. parameters and arguments are also implemented.
* Added conditions in "fitCopula" so that the variance is not
estimated if the appropriate functions are not defined.
* Switched order of argument "copula" and "data" in fitCopula
to be consistent with gofCopula.
* Changed "UseMethod" to "setGeneric" for defining generic
functions.
* Documention for "gofCopula" improved.
2008-11-18 Ivan Kojadinovic
* Jun has modified the structure of "fitCopula". Inversion
of Spearman's rho and Kendall's tau can now be used to estimate
the dependence parameters. The large-sample standard errors of
the estimates can also be computed in almost all situations.
* The argument list of function "fitCopula" has changed from
the last version. A newly added "method" argument precedes
the "start" argument, because "start" may not be needed for
inverting Kendall's tau or Spearman's rho. The returned object
of "fitCopula" also has component "est" renamed to "estimate"
to be more informative.
* The function "gofCopula" has been modified to reflect the
changes in "fitCopula". The multiplier method has also been
improved.
* Warning messages from evaluating likelihood at invalid
parameter values is suppressed when calling optim. It is not
the optimal way, but let's experiment with it for a while.
* Corresponding documentation improved.
2008-10-16 Roger Koenker
* Added a logical variable to mvdc objects "marginsIdentical"
intended to permit the use to force identical marginals. When
this is specified as TRUE, then fitMvdc and loglikMvdc take a
restricted vector of parameters and proceed accordingly.
* Made a few minor alterations in the documentation to clarify
points that I thought were unclear, notably the ordering of the
parameter vector for mvdc fitting.
2008-09-17 Jun Yan
* Fixed numerical rounding problem for rcopula method for
bivariate Frank copula with alpha >= 35 using its symmetry,
since it works for alpha <= -35.
* Added "indepCopula" class.
* Changed default method to "parametric.bootstrap" for
gofCopula; the multiplier needs to be fixed for its
considerable variation between replications.
2008-08-29 Jun Yan
* Bug fixed in the random number generation for the
Clayton copula.
* empcop*.test functions renamed to *indepTest.
* Documentation improved.
2008-07-17 Jun Yan
* Ivan has implemented and tested multiplier CLT
goodness-of-fit tests (bivariate and multivariate) for
certain copulas.
* Restructured R files under directory R.
* Expressions for derivatives of cdf/pdf are now stored
as opposed to computed on the fly.
* Numerical approximation functions for tau and rho
and its derivatives for special cases are now stored
in sysdata.rda under directory R.
2008-01-22 Ivan Kojadinovic
* The structure of the empcop*.test has been changed.
All the statistics are computed using the same code.
The only difference comes from the array J which changes
according to whether * in {u,m,s,sm}. The computation
is much faster.
2008-01-14 Jun Yan
* Restructured empcopsm for better performance.
Arrays W, K, and L are stored for reference in each permutation.
The computation is about 3 times faster.
* Fixed a bug in random permutation:
(i + 1) * runif() instead of i * runif().
The bug is seen for the case n = 2, where no permutation
would happen.
* When compute p-values of Tippett, use obs <= sim as oppose to
obs < sim because this statistic is discrete.
2008-01-02 Ivan Kojadinovic
* The tests of independence based on the empirical copula
have been renamed to empcopu.test (univariate) and
empcopm.test (multivariate).
* Some computations have been improved.
* Serial analogs of these tests have been implemented. They
are called empcops.test and empcopsm.test and can be used to
test serial independence in univariate and multivariate
continuous time series.
2007-12-11 Jun Yan
* Fixed a bug in fitCopula (thanks to Rodrigo Dupleich
for reporting).
2007-12-07 Ivan Kojadinovic
* Added the function empcop.rv.test, a test of independence
among continuous random vectors based on the empirical copula.
* Improved empcop.test.
* Fixed a bug in fgm.c.
2007-10-16 Jun Yan
* Added \encoding{latin1} in empcop.test.Rd.
* Fixed warnings and notes issued by R (2.6.0) CMD check.
2007-08-26 Jun Yan
* Added try-error handler for loglikCopula and loglikMvdc.
This will allow the optimizer to keep searching when NaN is returned.
* Changed the way to generate function calls to evaluate
[dpqr] for each margin, thank to
Martin Maechler .
The package can interact now with package nor1mix.
2007-06-06 Jun Yan
* Fixed random number generator of amhCopula.
The formula in Johnson (1988) has undefined quantities.
2007-06-02 Ivan Kojadinovic
* Farlie-Gumbel-Morgenstern class implemented with
distribution, density and random number generation.
Class needs to be properly tested, especially random
number generation.
2007-06-01 Jun Yan
* Merged with package copulab by Ivan Kojadinovic
, who
provides the multivariate independence test of
Genest and Rémillard (2004).
2007-05-18 Jun Yan
* Association measures are exported into the namespace:
kendallsTau, spearmansRho, and tailIndex. Calibration
functions are implemented for Kendall's tau and Spearman's rho.
* Extreme value copula class is implemented. This class
includes Galambos, Husler-Reiss.
* Added Archimedean copula Ali-Mikhail-Haq.
* Added Plackett copula.
2007-04-28 Jun Yan
* The three Archemedean copulas (clayton, frank, and gumbel) now
have their density expressions imported from mathematica, after
some symbolic simplification, which helps to eliminate some numerical
precisions problems on the boundary of the unit square.
Frank copula has the most complicated expressions. On 4GB memory
machine it ran out of memory for dim = 10. So for frank copula, the
maximum dimension implemented is dim = 6.
The symbolic expressions are processed in R with function deriv
to generate efficient algorithmic expressions.