############################################################################### ## News: to package RobExtremes ############################################################################### (first two numbers of package versions do not necessarily reflect package-individual development, but rather are chosen for the RobAStXXX family as a whole in order to ease updating "depends" information) ####################################### version 1.3 ####################################### user-visible CHANGES: + taking up a suggestion by Andreas.Scheidegger@eawag.ch, we introduced new argument propagate.names in our functionals controlling whether names obtained from parameter coordinates should be propagated to return values of specific S4 methods for functionals for Gumbel, GEV, GPD, Pareto + triggered by a mail by mekontsodimitri@gmail.com, we enhanced PickandsEstimator(); (a) it no longer requires the location parameter to be 0 and (b) it now also can be applied to GEVFamilyMuUnknown models. under the hood ####################################### version 1.2.2 ####################################### under the hood + fixed some internal .Rd files + moved from latin1 to UTF-8 encoding + removed the standardtexts (in comments) in our .Rd-files user-visible changes + added some more references + added dois to references where available + added ORCID references for the package authors + changed Description entry in DESCRIPTION -- now taking up references as well ####################################### version 1.2 ####################################### user-visible CHANGES: + return object of roptest (with interpolRisks) now contains information on its bias, so can, e.g., be used in confint with accounting for possible bias + CITATION file now uses bibentry bugfixes: + Bernhard discovered a bug in ".checkEstClassForParamFamily" for GEV (was GPD instead of GEV) + there were no classes [GPD/GEV]MDEstimate -> fixed now under the hood + moved quantile integration methods for expectation for Weibull and Gamma distribution to pkg distrEx (>= 2.8.0) + in asvarMedkMAD we now use distr::solve + made a helper function .qtlIntegrate out of existing code in RobExtremes 1.1.0 and moved it to distrEx where it is exported from version 2.8.0; it is reused in RobExtremes for the GEV methods + as with the interpolating - getStartIC methods in ROptEst, the makeIC-task is removed from the inner .modifyIC.0 function and delegated to the outer .modifyIC , so .getPsi, getPsi.wL, and .getPsi.P loose their argument withMakeIC + in the getStartIC methods for interpolRisks, we now produce slots modifyIC with argument withMakeIC (as before) and with "..." to pass on arguments to E() (e.g., when makeIC is called) + the timings are now about ~ 2s per estimator for GEV and GPD and check/makeIC are much faster + script updated + the makeIC methods for GPD/GEV... also gain an "..." argument + fixed minor issues in scripts/RobFitsAtRealData.R + expectation E() of Pareto, GPD, and GEV gain argument diagnostic and use dot-filtering (like in distrEx) + minor bugfixes in .getBetaXiGEW + new S4 classes "GPDML.ALEstimate", "GPDCvMMD.ALEstimate", and "GEVML.ALEstimate", "GEVCvMMD.ALEstimate" deleted classes "GPDMCALEstimate", "GEVMCALEstimate" as not every MCE is an ALE -> this gave misleading error messages + warning/caveat in the help to GEVFamily/GEVFamilyMuUnknown as to the accuracy of PickandsEstimator for GEV + introduced particular liesInSupport methods for Gumbel, Pareto, GPareto, and GEV ####################################### version 1.1 ####################################### user-visible CHANGES: + revised DESCRIPTION and staring help page + full robustness support now for ParetoFamily + moved "actuar" from require to imports in DESCRIPTION + NAMESPACE: export some formerly internal S4 classes + accessors and replacers for locscaleshapename; specify from which package (actuar,evd) to take gumbel p,d,q,r + smaller grid of xi values as starting values for GPD and GEVD for speed reasons + new accessor method scale for Pareto distribution + new checkIC and makeIC methods for GEV[U], GPD, and Pareto + new script RobFitsAtRealData + GEVFamily, GParetoFamily and GEVFamilyMuUnknown gain argument withMDE (by default TRUE) which controls usage of MDEs at finding startPars + gev/gpddiag and friends (i.e. interface to ismev methods) now apply to return values of roptest + new argument withMakeIC to control when to use makeIC to enhance accuracy + DESCRIPTION tag SVNRevision changed to VCS/SVNRevision minor changes: + new Rd files for now exported (formerly internal) intermediate classes + added timings for Sn + allow for versions < 3.0 when reading grid from sysdata.rda bug fixes: + fixed a sign error in FisherInfo.fct for GEV(U) + R CMD check did not like "usage" section in internal-methods.Rd, so we commented it out + in .define.tau.Dtau.withMu (only need theta[1:2]) + in getStartIC.R (in names risk type is without ".") + fixed modifyIC in "getStartIC",signature(model = "L2LocScaleShapeUnion") + some fixes in .checkEstClassForParamFamiliyMethods + class ParamWithLocAndScaleAndShapeFamParameter now contains ParamWithScaleAndShapeFamParameter + in ParetoFamily.R L2derivDistr was not attached to return value under the hood: + inserted new intermediate classes for return values of roptest, RMXEstimator, MBREstimator, OBREstimator, OMSEstimator, in case of GEV[U], GPD families (for ismev diagnostics) + Expectation E for DistributionsIntegratingByQuantiles to speed up things now uses stop.on.error = FALSE and for accuracy splits up the integration range in [0,0.98] and [0.98, upp] as soon as upp>0.98 + prepared everything for first release on CRAN + due to speed gains moved back diagnostic plot examples out of \donttest + added MakeIC calls in getStartIC to enhance accuracy + in order to avoid time-costly double calls to MakeIC, the respective modifyIC-slots gain attribute "hasMakeICin.modifyIC" once MakeIC is in the slot + added different route to get hand on L2Fam in gevgpddiag.R to make it applicable to return values of roptest + speeded up startEstGEV and startEstGPD + ParetoFamily now has interpolRisks + ParetoFamily distribution now has speed up for Sn + updated/prepared plotOutlyingness.R + fixed unit test suite for zero length + changed \dontrun in \donttest in examples + wherever possible also use q.l internally instead of q to provide functionality in IRKernel + in addition, now use slot locscaleshapename in generating function of GEVFamilyMuUnknown + new generics/methods for locationname, locscaleshapename(<-), scaleshapename, locscalename, shapename, scalename + use prefix evd:: to clarify which [p,d,q,r]gumbel to take + L2LocScaleShapeUnion gains slot locscaleshapename = "character" / looses slot scaleshapename = "character" + added some more specific links to other Rd files in distrMod + added code for the computation of LMs for ParetoFamily ####################################### version 1.0 ####################################### user-visible CHANGES: + title changed to title style / capitalization + GEV now has a robust starting estimator for mu unknown + GEVFamily and GEVFamilyMuUnknown now have changed default starting estimators realized in startEstGEV.R : a CvM-MDE with xi varying on a grid... + provide wrapper for ismev-diagnostics ie gev.diag, gev.prof, gev.profxi, gpd.diag, gpd.prof, gpd.profxi GENERAL ENHANCEMENTS: under the hood: + deleted some .bak file from R folder + risk measures VaR, CVaR, and EL for GPD, Weibull, Gamma, ... (ScaleShapeModels) + packed startEstGEV.R into a try-catch... + checking script to check GEV in scripts folder + prepared code for evaluation of LMs on xi grid for GEVFamilyMuUnknown family; + .pretreat.of.interest and .define.tau.Dtau are more accurate now; + moved some packages to Imports to avoid note on many packages in Depends, added Encoding due to warning about character 2292 in my locale + removed ::: internal dependencies (within distr&robast-Fam's of pkgs) by copying respective routines + enhanced imports in DESCRIPTION by explicating second order imports + gev.diag, gpd.diag now check for existence of se's and ascov BUGFIXES: + fixed issue with updateStep in kStepEstimator when using interpol risks (somehow modifyIC had not been attached in prior steps...) + added \value tag in getCVaR.Rd + fixed some issues with help of getVaR, getCVaR, ... + new print method for the results of these functions (and a corresponding S3class) + GParetoFamily now handles left endpoint correctly and catches xi < -1/2 + GEVFamily[MuUnknown] for xi>0 now handles left endpoint correctly and catches xi < -1/2 + warning for large xi is switched off in GEVFamily[MuUnknown] if called internally + double definition of ddigamma eliminated + bugfix in bounds / must be right for shape < 0 + eliminated some erroneous prints + in startEstGEV.R, the sigma search range could include negative values, now only let pass admissible starting estimators + bug in (population) variance of GEV xi=0 discovered + fixed the problem with inadmissible return values of MLE in startEstGEV.R + startEstGEV.R now works with soft bound 1+ xi (x-mu)/sigma > 0 (only to hold for lower quantile...) controlled by argument secLevel + fixed issue with check.validity as notified by B. Spangl, and extended starting estimator in GParetoFamily + fixed Matthias' error-issue in TeaserExample.R in cniperCont.R /plotWrapper.R -- used the wrong variance + fixed buglets in GEVFamilyMuUnknown which hindered evaluation of LagrangeMults, + cleaned small buglet in modifyPar + corrected link in Var.Rd + put code in \dontrun to reduce checking time + some issues fixed with gev.diag gpd.diag S4 methods ####################################### version 0.9 ####################################### Creation of this new package + moved Extreme value / scale shape infrastructure from distrEx and distrMod (i.e. Pareto, GPareto, GEV, Weibull, Gumbel) + Moved script GumbelLocationModel.R from pkg ROptEst + created generating functions for GEV, Weibull, Pareto families + PickandsEstimator (incl. as.var) + new L2ParamFamily GEVFamilyMuUnkown (with corresponding ES, EL, VaR) under the hood: + cleaned DESCRIPTION and NAMESPACE file as to Imports/Depends + RobASt-Pkgs: DESCRIPTION depends become stricter (requiring distrMod, distrEx, distr >=2.4 to be on the safe side) + added .Rbuildignore ++ kMad: + kMad.c now has Unix ending (LF instead of CR, CRLF + removed lazyload tag in DESCRIPTION + deleted occurrences of ll and lr in kMad.c (as proposed by Matthias) + deleted dll's and .d files from distrEx src folder + deleted stuff from "simple" estimators (LDE, Pickands, QuantileBCC) which is delegated to Estimator in distrMod() (i.e., asVar-calculations) + created methods for Sn and Weibull, Gammad and GEV + moved definition of S4 method for Sn and GPareto from interpolSn.R to SnQn.R + PickandsEstimator got a corrected asVar (due to the new computation for beta, increasing the bdp; + mentioned the increase of BDP in the Rd file. + bdpPickands committed + updated Metrika reference. + MedkMAD now has as.var information + interchanged order of ParamFamily and k (medkMAD) and alpha (PickandsEstimator); + implemented the Quantile estimator of Boudt Caliskan Croux for Weibull family as starting estimator (together with asvarQBCC..) + new LDEstimate class + LDEstimator now uses asvar + LDEstimators now may delay evaluation of variances and skip or delay evaluation of L2derivDistr + in LDEstimator.R changed ... any(is.na(es))... into ...any(is.na(estimate(es)))... + new S4-method .loc to center GParetoFamily and GEVFamily objects in LDEstimators + fixed an error in .LDMatch (forgot to substract loc0) + modularized generating functions GEVFamily, GParetoFamily, and WeibullFamily [ as to defining trafo, tau, Dtau ] + checked and computed FI in GEV model + explicit terms for the FI in the Weibull model + GEVFamily, ParetoFamily, WeibullFamily, GParetoFamily now have optional arguments * withCentL2 = FALSE (should L2deriv be numerically centered? -> formerly this was done automatically, and there was an error as the distribution under which centering was done was not the same (parameter) as the one to which L2deriv was calculated * withL2derivDistr: for GEVFamily, WeibullFamily, GParetoFamily this involves calls to imageDistr (hence random samples and calls to density()...) which was very costly now by default is skipped in slot call but not in direct calls * throw an error if argument startPar is no function. + GEVFamily, GParetoFamily, WeibullFamily * now set new slot .withMDE to FALSE (to avoid calling CvM-MDE); * .withEvalAsVar, .withEvalL2derivDistr equal to FALSE + getStartIC - method for interpolRisks (and hence internal-getpsi, too) moved from ROptEst to RobExtremes + getStartIC-method for interpolRisks and Sn methods now load grids from RobAStRDA namespace + added particular moveL2Fam2RefParam, moveICBackFromRefParam methods for scale shape models + considerably improved grid for Sn method for GPareto + finished interpolation grids (corrected a.o. Sn for GEV), + produced Weibull and GEV grids (in part; RMXE still missing; MBRE-GEV gave problems) + non-RobExtremes-specific interpolation infrastructure gets "imported" by something like .recomputeInterpolators <- ROptEst:::.recomputeInterpolators + minor changes in interpolation scripts + .mergeF and .recomputeInterpolators gain functionality to skip grids when merging/recomputing + write up to interpolation strategies + extract from rd file on internal interpolation infrastructure inserted into interpolation/WriteUp-Interpolators.txt + gave more detailed comments in interpolation/interpolationscripts.R + inserted scripts for generating interpol grids for GEVD Weibull Gamma + some diagnostic tools for interpolation grids (accessible through R-code) Rdir <- system.file("AddMaterial/interpolation", package= "RobExtremes") dir(Rdir) file.show(file.path(Rdir,"plotInterpol.R")) + forgot a "recursive = TRUE" in dir.create in interpolation script + shifted comments with auxiliary calculations to inst folder + TestInfrastructure for RobExtremes Tests for RobExtremes + to share some ideas: some test files in connection with RobExtremes + use of unit tests + wrapper functions use default rescaling controlled by S4-method rescaleFct dispatching on the respective model; particular methods for GEV, GPareto, GEVMuUnknown bugfixes: + forgot to initialize correctly slots L2derivSymm and L2derivDistrSymm in GParetoFamily.R; + minor error in CITATION file corrected + removed withPos relics in GParetoFamily + fixed error in initialize for Pareto-distribution (no lower endpoint was set) + some corrections in interpolation routines for GPD + some fixes in generating function to GParetoFamily + added class definition for Weibull family + minor correction in generating function for Weibull family + adapted title of the package to other RobASt-packages + fixed bugs in GEV-, GPareto-, and WeibullFamily + re-did the interpolation grids (for GPareto, so far) + functions where identical in grid..! + Corrected some errors in the scores function for the Weibull family; + fixed a minor issue in GEVFamily.R and suppressed warnings in call to imageDistr + bug in new S4 methods for Sn / fixed + found reason why GEV caused problems in finding optIC s for interpoloation: L2Fam call called GEV instead of GEVFamily ... + example to medSn had to be changed / lower upper bounds to be given; otherwise there was an error. + argh: forgot assignment of nE in getSnGrid... + .LDMatch produced names like scale.scale...