- Fix typos in the vignette Using variables in the
designargument ofTVGEV.
- New vignette about the use of variables or parameters in the
designargument of theTSGEVfunction.
1
-
Experimental changes on
TVGEVto allow the use of Time Series covariates along with functions of time. CAUTION This may have created bugs!!! -
New vignette Timeseries Covariates with NSGEV. DRAFT version.
-
A warning is cast when the
designargument ofTVGEVis a call using variables not innames(data). These variables may or may not be found in the parent environment.
-
The generalised residuals computed with
"gumbel"were actually following the reversed Gumbel, not the Gumbel. New (experimental) choice"gev4"for the target distribution. -
When
TVGEVwas used withoptim = "nloptr", the choiceparTrack = TRUEdid not work as expected.
-
The (non-exported)
MLE.TVGEVfunction now has atraceargument which is set to the value oftraceargument ofTVGEV. Whenestim = "nloptr"and a value> 2is given the values of the parameter and of the objective are traced. -
The Hessian of the negalive log-likelihood is now exact and no longer computed with
optimHessian. -
The
.Rnwvignettes, such as the technical vignetteCompDet.Rnwshould no longer be built. Only the pdf will be shipped with the package tarball.
-
Experimental implementation of the ODE method for the determination of the profile likelihood confidence intervals on the quantile of the maximum. See the help of the
quanMax.TVGEVmethod, and the functionquantMaxPLODE. -
Edits in the vignette Introduction to NSGEV. Typos and clarification for the generalised residuals, thanks to Jesper Rydén (Uppsala University). New references related to the generalised residuals. The section on the quantile of the maximum a.k.a. design life level now includes an example with profile likelihood intervals, at the price of an increased computation time.
- In the
residmethod for the class"TVGEV"the generalised residuals now use the standard Gumbel target distribution rather than a standard exponential.
-
The
quantMaxmethod for the class"TVGEV"now computes the profile-likelihood intervals on the quantile of the maximum on an arbitrary "design-life" period. The computation is quite long for now hence is illustrated in adontrunpart of the examples forquanMax.TVGEVmethod. -
The
TVGEVfunction did not work when some (non temporal) covariates had missing values. -
The functions
qMax.TVGEVandpMax.TVGEVnow compute the gradient and the Hessian w.r.t. the parameters, opening the road to profile likelihood inference on the quantile of the maximum. The wew functiondMax.TVGEVcomputes the gradient w.r.t. the parameters. -
New tests dedicated to the
qMax.TVEV,pMax.TVEVanddMax.TVGEVfunctions. -
Fixes in the documentation.
-
New long-form documentation NSGEV: Computing Details, mainly dedicated to the the computation of the profile likelihood intervals. This is a Rweave document
.Rnwhence will not appear on thepkgdownsite. It could be removed from the vignettes if this generates installation problems for some users. -
The bibliography file for the vignettes
NSGEV.bibhas been cleaned and provides DOIs when these are available.
-
In
quantMax.TVGEVthe check onprobwas misleading. -
The
quantMax.TVGEVmethod did not work forTVGEVobject describing a stationary model.
-
For a
TVGEVobject, thenegLogLikFunfunction now optionally returns the Hessian. -
New method
quantMaxfor the"TVGEV"class and subsequent methodautoplotfor the relatedquantMax.TVGEVobjects. Using these methods is illustrated in the main vignette. -
New methods
quantMaxFunandcdfMaxFunfor the"TVGEV"class. The return functions (closures). -
An
autoplotmethod has been added for some classes:"TVGEV"and more. -
The vignettes of the NSGEVVal package are now included in NSGEV. The required technical functions from NSGEVVal are now included in NSGEV as "check" functions. The bibliography has been somewhat cleaned.
- The
NAMESPACEfile is now generated with roxygen2 making the package easier to maintain.
- The
TVGEVfunction did not work when a constant location was specified along with a non-constant scale and/or shape. Thanks to Jesper Rydén.
- The GEV probability functions
dGEV,pGEV,qGEVandrGEVhave been moved (with some changes) to the nieve package, available both on CRAN and on GitHub. So NSGEV now imports these functions from nieve. As a consequence the compiled code used by this functions has been discarded, making the package easier to install at least for Windows users.