A variety of functions to fit linear and nonlinear
regression with a large selection of distributions.
This package is intended to be the developmental version to the CRAN version of Jim Lindsey's gnlm. The .zip files listed on his homepage have been listed as version 1.0 since 2005. For the subsequent maintenance on this github and CRAN, we will start at 1.1.
Above this line will be News/Changes for gnlm only
Below this line corresponds to changes.txt, which was Jim Lindsey's file for
detailing changes across the v1.0 packages rmutil, repeated, gnlm, growth, event, stable on his homepage
elliptic: twins option added for model using covfn with covariance matrix
diagonal constant
28.11.10
elliptic: added an error check when covfn used
15.2.10 (rmutil)
changed s<0 to s<=0 in qlaplace & rlaplace (thanks to Peter Ehlers)
18.11.09 (repeated)
removed a redundant line in gausscop.r that now produced an error
(thanks to Patrick Lindsey)
7.4.09
removed extra } in biv.binom.Rd (thanks to Christopher Marcum)
20.10.08
discrete q functions: changed trunc to round (thanks to Frederic Gosselin)
3.7.08 (gnlm)
fit.dist: corrected check for negative values with Laplace, Cauchy,
and Student t plus error in counts (f -> ni) for Laplace
(thanks to Michael Anyadike-Danes)
24.10.07
fnenvir: changed way "=" is handled in gsub() because of error since R2.5.0
8.10.07 (event, gnlm, growth, repeated)
changed typsiz to typsize in nlm() throughout
11.7.07
romberg.c: added missing R.h (thanks to Olivia Lau)
8.2.07
print out name of response variable in elliptic, bnlr, gnlr, gnlr3,
gnlmm, gnlmm3, and fmr (thanks to Patrick Lindsey)
qsimplex: corrected search interval updates
27.9.06
qhjorth, qinvgauss, qginvgauss, qboxcox: changed lower limit of search
from 0.001 to .Machine$double.xmin (thanks to Franco Mendolia)
8.12.05 (rmutil, repeated, event)
minor modifications to Fortran for F95 compatibility (thanks to Jan de Leeuw)
22.11.05
finterp, objectrm: added na.action=NULL in calls to model.frame
(default is na.omit !!!!!) (thanks to Patrick Lindsey)
30.9.05
elliptic: corrected calculation of number of parameters for builtin
logistic function (thanks to Tom Van Dooren)
1.8.05
qbetabinom: changed trunc() to round() (thanks to Elias Krainski)
mprofile, iprofile: added check that times are available (thanks to
Patrick Lindsey)
6.7.05
ksurvb, ksurvg, kcountb: added break; after default: in C code to
satisfy some compilers (thanks to Patrick Lindsey)
30.6.05
finterp: correction for change in functioning of match()
gnlm family: added coef.gnlm() and vcov.gnlm() (thanks to Bendix Carstensen)
25.4.05 (stable)
rstable: eliminate production of NAs (thanks to Zhu Wang)
26.1.05
finterp: fixed a bug when >= or <= is used in a formula (thanks to
Eliot McIntire)
1.11.04
gnlmm3: generalized nonlinear mixed models for three-parameter distributions
28.9.04
catmiss: removed codes() from example in help (thanks to Kjetil Brinchmann)
21.9.04
finterp: fixed if test to remove occasional warning (thanks to Ken Knoblauch)
17.9.04
gnlmix: removed erroneous printing that distribution is censored for
binomial (thanks to Ken Knoblauch)
28.7.04
gnlmix, hnlmix: fixed printing of results when nonlinear function
contains a linear part (thanks to Ken Knoblauch)
2.7.04
tvctomat: fixed warning message on removing tm (thanks to Patrick Lindsey)
1.6.04
glmm: changed print.summary.glmm to work under R1.9 (thanks to Spencer Graves)
5.4.04
fnenvir: fixed obscure error when linear used in a function (thanks to
Ken Knoblauch)
help: corrected truncation of usage formula for certain functions
(thanks to Patrick Lindsey)
9.1.04
fitdist: fixed typo that stopped geometric distribution from working
6.1.04
ordglm: changed tapply to capply because of change to former (thanks
to Andrew Criswell)
9.12.03
corgram: start abscissa at 0 for PACF
fnenvir: fixed grep for checking redundant parameters
readdna, survkit, glmm, gnlmm, objectrm, readrm: removed obsolete
codes() function (thanks to Ken Knoblauch)
carma: give error when ccov=~1 used (thanks to Patrick Lindsey)
21.8.03
elliptic: corrected print function when the dispersion function depends
on the location function (thanks to Gabrielle Kelly)
31.7.03
hnlmix: corrected options for dnbinom (thanks to Jagat Sheth)
30.6.03
dftorep: corrected check for ordered times to allow for two levels of nesting
25.5.03
ordglm: added a data argument (thanks to Kosuke Imai)
13.5.03
ordglm: corrected order of printing of standard errors (thanks to Kosuke Imai)
25.4.03
gnlr, gnlr3, gnlmm, fmr, nordr: changed test for environment because
the value returned by parent.frame() has a class in R1.7
22.4.03
cphidden: a function to locate a changepoint in continuous time using
a two-state hidden Markov model
9.4.03
biv.binom: corrected degrees of freedom printed (thanks to Goran Brostrom)
12.2.03
restovec: fixed handling of delta when the response is a list
16.1.03
kalsurv: fixed typo in print.kalsurv (thanks to Anthony Gichangi)
4.12.02
int: changed eps
2.12.02
fit.dist: added Laplace distribution
1.12.02
glmm: added error message if (codes of) nesting variable not consecutively
numbered (thanks to Renaud Lancelot)
27.11.02
fit.dist: changed Weibull parametrisation so that mu refers to y and
not to y^alpha
22.11.02
fit.dist: added Cauchy and Student t distributions
use (log) density functions instead of writing formulae
18.11.02
fit.dist: added beta-binomial distribution
16.11.02
fit.dist: corrected error in calculation of log likelihood when censor=T
14.11.02
fit.dist: corrected error in calculation of fitted values for zeta distribution
31.10.02
int2: added default limits (thanks to Patrick Lindsey)
8.9.02 (repeated)
gar: corrected recursive fitted values when binomial (thanks to
Patrick Lindsey)
4.9.02
gausscop: exponential distribution now works (thanks to Patrick Lindsey)
30.8.02
restovec: modified checks for nesting in lists and allow nesting to be
supplied separately when a list is given (thanks to Patrick Lindsey)
gausscop: for positive-valued distributions, returned predicted values
without transforming by log link function (thanks to Patrick Lindsey)
18.7.02
ehr: addition checks of data supplied to this suite of functions
rs3: fixed typo
marg.hom: added checks on data
16.7.02
chidden.r, hidden.r: corrected negative binomial check so that 0
responses are allowed (thanks to Ben Cooper)
10.7.02
modified man pages for changed arguments to rgamma function
rmutil: created dist.h file
11.6.02
hnlmix: corrected AIC for penalty constraint (was too large by one)
changed calculation of multiplicative random effects
23.5.02
rmutil: added [pdqr]twosidedpower distribution
added log option to all density (d) functions
gar, gnlr, gnlmix, gnlmm, hnlmix: added two-sided power distribution
gnlr: user-supplied likelihood function works again (thanks to Martin Liermann)
finterp, fnenvir: added option to allow any response to be a covariate
9.5.02
hnlmix: recursive fitted values available
ordglm: fixed error that PearsRes not defined when individual data are
supplied (thanks to Kamal Desai)
6.5.02
gnlmix, hnlmix: added inverse gamma mixture distribution
gnlmix: handles censored data
gnlmm: finds nesting variable when repeated environment is specified
5.5.02
finterp: modified so that as.factor(times) works
30.4.02
hnlmix: nonlinear random effects models using a modified Lee and
Nelder h-likelihood
gnlr: modified check on location parameters for Levy distribution
added check that double Poisson, multiplicative Poisson, gamma
count, and logarithmic data are not censored
gnlmix: corrected typo in negative binomial distribution
23.4.02
carma, chidden, elliptic, hidden, kalseries: give error if censored
data supplied (thanks to Troels Ring)
22.4.02
elliptic: when two levels of nesting, calculate correctly first
recursive fitted value in each cluster (was plotted correctly
using iprofile) plus corresponding simplification of
plot.iprofile (thanks to Troels Ring)
17.4.02 (all packages)
gnlmix: corrected typo in inverse Gauss mixing distribution
print model methods: added option not to print correlations
15.3.02
restovec, tcctomat, tvctomat: added optional description slot to response,
tccov, and tvcov objects
13.3.02
glmm: convert repeated object to dataframe if supplied as data
tcctomat, tvctomat: corrected to detect contrast options when dataframe=F
12.3.02
tvctomat: corrected problem for list of factors when dataframe=F
(thanks to Patrick Lindsey)
finterp.default: give error if members of dataframe named using $
(thanks to Christof Bigler)
28.2.02
chidden, hidden: added check for correct number of initial estimates
when list of functions supplied (thanks to Patrick Lindsey)
22.2.02
corgram: added option for PACF
19.2.02
fmr: modified some discrete distributions to avoid overflow with large counts
17.2.02
elliptic: added as.double for y in call to C code because of change in
read.table
12.2.02
finterp: give error if offset used in W&R formula
31.1.02
%^%: power of a matrix
elliptic: corrected problem when common parameters in mean and
variance functions
20.1.02
plot.repeated: added selection of profiles to be plotted by using ccov
14.1.02
gar: added absolute value arch (names of others changed: additive -> square, multiplicative -> exponential)
volatility method for extracting values of nonconstant dispersion
parameter
Makefiles: removed . for SHLIB_EXT for R1.4.0
dist.c, kcountb.c, romberg.c, stable.c: changed malloc to R_alloc
10.1.02
(dhpqr)ggamma, fmr, gausscop, gnlmix, gnlmm, gnlr, gnlr3, hgamma:
changed argument of (dpqr)gamma for compatibility with R1.4.0
modified help to work with R1.4.0
18.12.01
contr.mean: provides correct labels on mean constraints (corrects
contr.sum)
4.12.01
chidden, hidden: corrected printing out family parameter with AR when
there is not one
28.11.01
qstable: corrected bug when tail<1 and skew=1 (thanks to Alec Stephenson)
23.11.01
corgram: handles NAs in the series
19.11.01
cprocess: fixed error in checking for list of events (thanks to Troels Ring)
stablereg: changed alpha to allow parameter to be (0,2) instead of (1,2)
18.11.01
chidden: added time-discretized Poisson process
17.11.01
chidden, hidden: added Student t distribution
changed Cauchy shape parameter to scale instead of scale^2
15.11.01
gar: added Student t distribution
added ARCH models
elliptic: when AR, take log of determinant returned by dpodi
(thanks to Gabrielle Kelly)
13.11.01
elliptic: when series of independent observations, calculate covariance
determinant as sum of log variances instead of log of product
(thanks to Gabrielle Kelly)
8.11.01
cmcre: corrected problems when a covariate is used (thanks to Anthony Gichangi)
6.11.01
print.response: do not print mean if nominal (but not binary) or ordinal
(thanks to Patrick Lindsey)
25.10.01
hidden.r: corrected check for fixed zeros in transition matrix
relaxed check for rows of transition matrix summing to one
chidden.r: relaxed check for rows of transition matrix summing to zero
(all thanks to Patrick Lindsey)
24.10.01
restovec: weights can be logical
14.10.01
gar: fixed output printing when shape is a function of location parameter
use dnbinom function
changed negative binomial shape parameter to be same as in gnlr
10.10.01
carma, chidden, gar, hidden, kalcount, kalseries: check for two levels of
nesting when serial dependence fitted
9.10.01
kalseries: corrected error when torder used with tvcov
8.10.01
hidden, chidden: added observed AR(1)
gnlr, gnlmm, gnlmix: changed parametrization of the shape parameter for the
beta distribution (thanks to Goran Arnoldsson)
binnest: duplicate variables in Fortran call
model functions using envir: check that response specified is one in
envir when only one present
3.10.01
plevy, qlevy: use pnorm and qnorm instead of integrating
26.9.01
elliptic: added second form of asymmetric multivariate Laplace
distribution with constant asymmetry parameter
25.9.01
elliptic: added asymmetric multivariate Laplace distribution
24.9.01
carma.r: removed unnecessary check that envir is a repeated object
(thanks to Troels Ring)
11.9.01
fit.dist: added checks that grouped frequency data are supplied
10.9.01
kalsurv: corrected output errors when environment is supplied
gar: use log option in dbinom, dpois
kalcount: set first recursive prediction in series to marginal prediction
6.9.01
gar: added loglog link for binomial data (corrected cloglog which was,
in fact, loglog)
20.8.01
gnlmix: set undefined sh3 to NULL for one parameter distributions
1.8.01
chidden, gar, gnlr3, hidden: added skew Laplace distribution
27.7.01
corgram: improved presentation of correlogram
25.7.01
d,h,p,q,rskewlaplace: probability functions for the skew Laplace distribution
24.7.01
autointensity.r: plots autointensity function of a point process
12.7.01
plot.repeated: fixed error of unknown type when plotted time-varying covariate (thanks to Patrick Lindsey)
carma: clearer error message when incorrect environment supplied
10.7.01
carma: will handle data objects with (one of) multivariate responses
chidden, hidden: handle Jacobian correctly with (one of) multivariate
responses
6.7.01
cprocess.r: recognizes data objects for events and not just for times
5.7.01
f2c.h included in library for toms614.c (missing in R1.3.0)
27.6.01
iprofile, mprofile: corrected links to other libraries for html help
plot.cum.pergram: corrected confidence interval
pergram: changed calculation of length for odd-lengthed series
22.6.01
gar.r: check that times are supplied and, if not, create if possible
19.6.01
fmr.r, gnlmm.r, gnlr.r, gnlr3.r: linear can be ~1 if mu not supplied
14.6.01
marg.hom.r: modified to handle factor variables with non-numeric levels
8.6.01
ordglm.r: corrected fitted values when weighted observations
(thanks to Troels Ring)
31.5.01
elliptic: changed check on initial variance function estimates
16.5.01
print.response, print.tvcov, print.repeated: added option to print
range of numbers of observations per individual instead of
vector of numbers (thanks to Markus Jantti)
dmultpois, etc: added additional check on dependence parameter
9.5.01
gar.r: corrected printout for generalized gamma parameter
26.4.01
changed F and T to FALSE and TRUE throughout
read.rep: removed col.names option because of changes in read.table
22.4.01
glmm: corrected typo when dataframe used with no offset
20.4.01
finterp: detects functions when given as arguments of other functions
19.4.01
finterp: formulae can be written on several lines, as several
instructions (e.g. to assign temporary variables)
11.4.01
dburr, pburr, qburr, hidden, chidden, gnlr3: changed parametrization
of Burr distribution (thanks to Patrick Lindsey)
28.3.01
chidden, hidden: corrected vector length problem in check for ordered
intercepts in ordinal models (thanks to Niko Speybroeck)
several p and d functions: changed check to y>=0 (thanks to Patrick Lindsey)
22.3.01
glmm: works again with weights and/or offset (thanks to Luc Duchateau)
gnlmix: changed to log dispersion for mixing distribution
21.3.01
int.c: corrected memory allocation problem
GExtVal: changed man page to agree with functions
(both thanks to Patrick Lindsey)
20.3.01
use log option in d and p functions for h functions
14.3.01
chidden, hidden: added further checks on ordering of intercepts for
ordinal data
13.3.01
gnlmix: changed dispersion parameter for normal mixing distribution
from standard deviation to variance
delta: returns a vector instead of a matrix if only for one variable
11.3.01
gnlmix: correction to work with unbalanced data
9.3.01
gar, gnlr3: added power variance function Poisson distribution
8.3.01
covariates: added expand option
dpvfpois, ppvfpois, qpvfpois, rpvfpois: functions for the
overdispersed power variance function Poisson distribution
kalcount: corrected for power variance function
7.3.01
plot.response: corrected indexing problem
1.3.01
kalcount, kalseries, kalsurv: removed constraints on family parameter
27.2.01
chidden, fmr, gar, gausscop, gnlmix, gnlmm, gnlr, gnlr3, hidden,
kalseries, kalsurv, nlr: relaxed type checks for continuous
and duration data
26.2.01
kalcount, kalseries, kalsurv: added two-parameter power variance family
mixture including gamma and inverse Gauss mixtures (family) for
serial dependence
23.2.01
response: if response is univariate, returns a vector instead of a matrix
covariates: if only one covariate requested, return as a vector
chidden, hidden: improved checks for ordered intercepts with ordinal response
improved calculation of ordinal probabilities
22.2.01
plot(iprofile()): works for models from kalsurv (thanks to Jacob Bowers)
19.2.01
chidden.r, hidden.r: corrected error in calculating individual profiles when
tvmu used (thanks to Jacob Bowers)
ordinal data can be used with multinomial option (thanks to
Patrick Lindsey)
work with ordinal data with a list of different formulae
(thanks to Niko Speybroeck)
31.1.01
glmm.r: works if response is a one-column matrix instead of a vector
(thanks to Luc Duchateau)
restovec: corrected manual so that arguments section appears (thanks
to Patrick Lindsey)
30.1.01
finterp, fnenvir: further correction to handle decimal numbers
(including scientific notation) correctly
finterp: replaced gsub by all.vars
25.1.01
name of response can be a character string when environment is supplied
(thanks to Patrick Lindsey)
hidden, chidden: added description of pintercept to man page
delta: works properly when name is supplied
plot functions: use 6 different line types instead of 4
gausscop: corrected mean for gamma margin
check that only one initial estimate when no dispersion function
18.1.01
transform.response: works when units is NULL
hidden, chidden: reversed order of categories for proportional odds
and continuation ratio
replaced dqrcf with dqrsl
nordr: bug fix to work with data objects
8.1.01
cutil.c, romberg.c, toms614.c: changed include for call_R
(thanks to Dennis Murphy)
7.1.01
model fitting functions check for correct type of response
dftorep, read.rep: modified to handle new "types" of responses
dftorep: now handles two column binomial response
ehr.r: rewrote to conform to other functions
4.1.01
restovec: option to add responses to an old response object and
types of responses returned in list instead of as a class
resptype: new method to return types of response variable(s)
finterp.repeated: check that binomial and censored responses are not
used as covariates
21.12.00
gar.r: corrected error in printing three-parameter distributions
19.12.00
finterp, fnenvir: methods for dataframes
gnlm functions: environment can be a dataframe
18.12.00
changed check for existence of response when environment supplied
bnlr, fmr, gnlmm, gnlr, gnlr3, nordr: fixed calculation of n for null function
17.12.00
various changes because of new R1.2.0 handling of formulae
finterp: check for + or - before ( (change to R1.2.0)
elliptic: removed check on tvcov, so can accept times and individuals
15.12.00
bnlr, fmr, gnlr, gnlr3, gnlmm, nordr: nonlinear formula need not
contain covariates
changes to cutil.c and coxre for R1.2.0
14.12.00
restovec, tcctomat, tvctomat: added slot for units of measurement
dftorep, read.rep: id can be a factor variable
carma, elliptic, gar, gausscop, kalcount, kalseries, kalsurv: test if name of
response exists when envir contains multivariate responses
stable: corrected man pages
3.12.00
gnlmix: generalized nonlinear models with one random parameter having
arbitrary mixing distribution
30.11.00
fitdist.r: calculate log probabilities and likelihoods to avoid underflow
(thanks to Goran Brostrom)
added d, p, q, and r functions for Consul generalized Poisson distribution
added PACKAGE option to .C and .Fortran calls
bnlr.r: added cloglog and loglog links
changed class of gnlr-type functions from gnlr to gnlm
gar.c: corrected calculation of censored Burr, Pareto, power
exponential distributions (thanks to Patrick Lindsey)
24.11.00
q and r functions: improved calculations and checks
23.11.00
bnlr, fmr, gnlmm, gnlr, gnlr3, nordr: nonlinear formulae can have a
linear part
qgweibull, qggamma, qgextval: corrected arguments to functions and
docs (thanks to Patrick Lindsey)
21.11.00
fmobj: find objects referred to in a formula
elliptic.r, fmr.r, gar.r, gausscop.r, gnlmm.r, gnlr.r, gnlr3.r: models with
common parameters in several regression functions can be specified
using formulae, not just functions
elliptic.r, gar.r, gausscop.r, gnlmm.r, gnlr.r: models with shape as a
function of location can be specified using formulae, not just
functions
20.11.00
finterp.r: formulae can have common parameters and/or depend on functions
16.11.00
hidden, chidden: added recursive predicted values
added q and r functions for distributions in rmutil
14.11.00
kalseries: corrected error in inverse Gaussian distribution (thanks to
Patrick Lindsey)
bnlr.r: added stable and mixture links
gnlr, gnlmm, gar: added beta and simplex distributions
rmutil: added psimplex and dsimplex
9.11.00
improved checking for multivariate response and choosing one response
when several present in a data object
6.11.00
fmr.r, printrm.r: corrected so that works with common parameters
(thanks to Laura Thompson)
29.10.00
gnlr3.r: corrected typo in normal and inverse Gauss distributions
19.10.00
gausscop: multivariate Gaussian copula with arbitrary marginals
residuals.elliptic: corrected error in calculation of raw residuals
31.12.99
objectrm.r: can select certain individuals with methods, covariates, delta,
nesting, times, weights
transform: handles NAs in response correctly
28.12.99
restovec: added name of response variable to list returned
objectrm.r: added as.data.frame and as.matrix methods for data objects
wr: works with my data objects
nordr: changed sign of coefficients for continuation ratio and
adjacent categories models so comparable with proportional odds
27.12.99
finterp: with W&R notation, design matrix no longer returned as attribute
when ~1 and .envir supplied, returns a function yielding a
vector of correct length
26.12.99
fit.dist: corrected exact fit of negative binomial
gnlr, gnlr3, fmr, gnlmm: improved speed
nordr, ordglm: ordinal categories numbered from 0 instead of 1
hidden, chidden: multinomial categories numbered from 0 instead of 1
handles ordinal models, thanks to Patrick Lindsey
gettvc: now handles NAs in response variable
25.12.99
improved documentation for methods to access data objects and functions
int: call C instead of Fortran for TOMS614
23.12.99
restovec: added additional checks that correct data are supplied
(thanks to Troels Ring)
mprofile.carma: corrected bug when no covariates
carma: corrected bug when delta is a scalar
carma, elliptic: added checks for incorrect formulae
21.12.99
hidden, chidden: improved printout and corrected error in checking
number of parameters in lists of formulae
20.12.99
lvna: creates a repeated object leaving NAs in
hidden, chidden: interactions between time-constant and time-varying
covariates allowed
17.12.99
hidden, chidden: improved printout
16.12.99
tvctomat: handles lists of factor variables correctly
restovec: value returned has class of type of response as well as "response"
added checks
hidden, chidden: can also use formulae if multinomial
12.12.99
hidden, chidden: can use formulae if not multinomial
7.12.99
cmcre: corrected memory leak
6.12.99
cmcre: continuous-time two-state Markov process with random effects
5.12.99
coxre: corrected several errors
1.12.99
stable: fixed plot arguments in help examples
29.11.99
finterp: fixed bug when multiple ('s or ^ before ( when detecting
function names
nobs: use method instead of direct access in all functions
provide default method
covind: provide default method
25.11.99
collapse: changed name to capply because of conflict in nlme
23.11.99
profile: changed to mprofile because of conflict in R0.90
22.11.99
finterp: properly distinguishes unknown parameters from functions
finterp and fnenvir: when no variables found, changed stop to warning
nobs: corrected for independent observations when length is one
18.11.99
stablereg: corrected bug when some parameters are not optimized
check for NAs in the hessian
17.11.99
plot.repeated, plot.response: added special call for ordinal responses
corrected plot for independent observations (thanks to Patrick Lindsey)
14.11.99
removed unneeded aliases in man pages
added aliases to plot.profile and plot.iprofile
11.11.99
added check for Inf (as well as NAs) in hessian to all functions using nlm
kalseries.r: added error message if times not available for Markov dependence
changed rep(1,n) to rep(1,nind) when mu function returns scalar
stable.r: moved call to C code into likelihood function for speed
int.r: limits can be specified as -Inf, Inf
4.11.99
kalcount.r, kalseries.r, kalsurv.r: with time-varying covariates in a
function or formula, initial estimates can be in preg or ptvc
and changed length(resp$response$y) to n for speed
31.10.99
gar.r: fixed undefined npt3 for autoregression parameter
finterp.r: fixed bug for : in W&R formulae
kalcount.r, kalseries.r, kalsurv.r: added error message when time-varying
covariates
22.10.99
covind: changed so that it works with carma, elliptic, gar, hidden,
kalcount, kalseries, and kalsurv objects
18.10.99
gar.r: corrected printing of parameter values for three-parameter distributions
gar.c: corrected calculation of lambda in three-parameter distributions
17.10.99
gar.r: corrected fitted values (due to changes on 12.10.99)
14.10.99
gar.r: corrected undefined variable, tm (due to changes on 12.10.99)
12.10.99
stable: stablereg with nonlinear regression replaces stableglm
finterp, fnenvir: check for factor variables instead of not being a
numerical or logical vector
gar: allow autoregression parameter to depend on covariates