NEWS.md

v2.0.2

• deseasonalize dependency (not available anymore) to allow installing the package again
• Fix warnings in try-error checks. Some objects have more than a single class/return a string list when provided to the class() function. All checks whether the class of an error is "try-error" is now wrapped in any() to ensure it both works and does not produce any warnings
• Fix bootstrap unit tests. There seems to have been some changes in the sampling algorithms or random number generation of R invalidating the previous bootstrap error estimates. Well, since there was an update to a new major release, this is not too unexpected. Also, the errors of the same order and still reproducible. I updated the values.

v2.0.1

• Updating the vignette. Removing the part describing the removed input function source.data() and the customization of the no longer available climex.path global option.
• Removing the cran-comments.md file.
• Removing GitHub part of the installation instruction in the README.md
• Updating the temp.potsdam data set. It now covers two more years.
• Updating all tests to cope with the new temp.potsdam data.
• Moving res/ folder to inst/res/ to match the R conventions and removing the obsolete glyphicons-199-ok-circle.png file
• Removing the deprecated ... input argument of the return.level function.
• Using the devtools::check() function in the GitLab CI and Travis checks instead of the R CMD check
• Incorporating the upper.limit function to calculate the upper limit for the GEV and GP distribution for shape parameters smaller than zero and the corresponding fitting errors. For the latter the Maximum Likelihood, Monte Carlo, as well as bootstrap method are available

v2.0.0

The functionality of the 1. climex core package, 2. the shiny-based GUI, and 3. import of the data of the German weather service (DWD) will be split in three different entities. This way, it is more simple regarding housekeeping. And I might finally succeed in getting a reply from one of the CRAN fellows by just submitting the core package.

• Rewriting the source.data() function to search for .RData in a specified folder recursively, print all results well-formatted in the command line and prompt the user for a selection. The option for a default data set is now deprecated.
• Adding S3 function checking.completeness. For input of class xts it returns TRUE or FALSE whether or not the time series has at least the supplied number of complete years. If the input is of class list, the function trims the list to only those objects fulfilling the requirement.
• Rewriting the functions fit.gev, fit.gpd, return.level, aic, bic, remove.seasonality, anomalies, block, remove.incomplete.years, decluster, and threshold as S3 objects. They now accept both the basic elements they already were able to accept beforehand and a list of those objects. In the latter case the output will be a list of the output of the individual calls.
• Streamlining the documentation and formatting Updating the formatting of the code and the documentation. All lambda functions now have running variables composed of two times the same character. Function arguments are marked with the \strong{} highlighting in the documentation and packages are referenced using \pkg{}.
• Allowing for parallel execution of S3 functions The functions anomalies, block, decluster, threshold, check.completeness, fit.gev, fit.gpd, remove.incomplete.years, remove.seasonality, and return.level can now harness the support of the parallel package using the mclapply function. Per default their additional input argument mc.cores is set to NULL. If set to a numerical value instead, this value is used as the number of cores and the parallel execution is triggered. The functions aic and bic, while being S3 generic, have been excluded since they are calculated in no time.
• The separation.mode argument of the block function was removed, since it was overwritten internally anyway.
• The default block.number argument of the block function was set to NULL. This way its internal handling using the missing function was changed to a check using is.null to ensure it will work when called by another function (the fit.gev one).
• The threshold function now hands the cluster.distance argument to the decluster function.
• fit.gev and fit.gpd are now capable for preprocessing the time series on their own. Using the arguments blocking and thresholding the functions can call block or threshold internally and hand over all the required arguments. This makes the use of those functions more convenient to the user.
• The extreme.type argument was introduced to the fit.gev, fit.gpd, block, threshold, and return.level function to be also able to fit the block minima or all extreme points below a certain low threshold. This argument replaced the block.mode one of the block function.
• Reducing the default value of the monte.carlo.sample.size argument of the return.level function form 1000 to 100.
• Bug fixes and additional test functions for the error estimation in the return.level, fit.gev, and fit.gpd function.
• Using seeds instead of tolerances in the test functions.
• Bug in ttplot prevented correct x axis for all base time units instead of days.
• The positions of the returned list elements x and return.level of the fit.gev function was switched.
• Both the fit.gev and fit.gpd function do now have a fallback method, which is able to handle numerical data. It contains a warning message suggesting the usage of xts-class objects instead. The resulting object will most probably be incompatible with a lot of functionalities.
• The fit.gev and fit.gpd functions do now feature a new input argument called debug. If set to TRUE, it displays both warning messages and intermediate results from the outer loop of the constrained optimization routine. In addition, the silent argument now properly mutes all output of the fitting functions.
• Fixing the correct linking and exporting of the C++-based function likelihood_GEV (only used within the likelihood.plot function since it yields a significant speedup in there.
• The convert.date function was some heritage from times the climex package was still performing the download and formatting of the data of the German weather service (DWD). Since this is no longer the case, the function was removed.
• The source.data function is now removed as well. It way more plausible to have it in the dwd2r package and a very bad idea to have two copies.
• The climex.path global variable is no longer part of the climex package. Since dwd2r uses its own directory for storing the downloads, ~/R/dwd_data in the global option dwd2r.download.path, the climex package does not directly use or access the climex.path folder anymore. The variable will be instead defined by the climexUI package.

v1.2.1

• Only perform the test of the API of the DWD on Unix-based systems. The winbuilder of CRAN just throws non-reproducible error messages. In some way the server is not able to properly download the .zip file. This doesn't seem to be an issue of this package at all.
• The test of the API of the FTP server of the DWD is now only performed in case the computer running the tests has a working connection to the internet.
• Using Travis to run the tests when pushing the package to Github.
• Registered C++ routines as suggest in RcppCore/Rcpp#636.
• Instead of toggling the attachment of the DWD data in the source.data function to either the global environment or the one used to call the function, the desired environment can now be provided as an input argument.
• Fixing the handling of global variables in the shiny app by using a custom environment climex.environment instead of true global variables.
• The vignettes in the //vignette/ folder are no longer used as the online documentation. There were some problems with them loading images in R. Instead they are stripped of all figure links and a copy of them is introduced in the //res/ folder. These ones still contain the figure links and are saved as proper .md files (and ignored during build). In the end the //res/ folder was included again, since win-builder tries to build the README.md using pandoc.
• Fixing a bug in the formula calculating the run.length in the extremal.index for a threshold distance then 2.
• Updating the documentation, namespacing, file layout, build ignore, DESCRIPTION file
• Using proper UTF-8 encoding
• With bootstrapa third option of calculating the error estimates for the GEV/GP parameters, as well as the return level is now available.
• The likelihood.initials now uses a proper scaling when searching for an initial parameter combination using a Markov random walk.
• In the fit.gev function the monte.carlo.sample.size variable was not handed over properly.

v1.2.0

The return.level function now returns a list containing both the estimates and the approximated errors, regardless of whether an error estimation was selected or not. The error estimate for the GP return level can now be obtained without supplying the total.length variable. Its value will be estimated. The fit.gev function is able to directly fit a Gumbel distribution by supplying an initial parameter combination with a shape parameter equal to zero. In the same fashion the exponential distribution can be fitted via the fit.gpd function.

v1.1.0

All the optimization is now relying on the augmented Lagrangian method implemented in the 'alabama' package. Therefore the 'dfoptim' dependence, all traces of rerunning the algorithm and the third tab of the Climex app were removed. Additional functions for the augmented likelihood and its gradient were added. Changes of the API of the DWD were incorporated.

v1.0.3

The GenSA function is not that robust. Therefore I added the stats::optim( method = "SANN" ) implementation as a fallback function in case the former one does not work. In addition I updated the documentation to the tile control panel in the leaflet tab of the climex app and fixed a bug in the likelihood.gradient function.

v1.0.2

The likelihood gradient now works for the shape parameter equals zero too (for both GEV and GP). In addition the a bug in the internal function likelihood.plot was fixed and the climex app now can choose between the OpenStreetMaps and OpenTopoMaps tiles via a control panel.

v1.0.1

Improved heuristics for the estimation of the initial GEV and GP parameters used during the optimization. The likelihood.initials function now uses a more fine grained precision tree for deciding about the initial value of the shape parameter for both the GEV and GP distribution. Beforehand this function also featured a stochastic component drawing random shape parameter around the estimate and searching for the best possible. This was replaced by a sequence function to make the fit deterministic and hence reproducible. In addition a number of fixes and improvement in the documentation and the vignettes were performed.

v1.0.0

Major changes in the climex app. Among others modularization, numerous bug fixes, and disabling of features not implemented yet (e.g. the animation of GP fits). Removal of the global CLIMEX.PATH variable in favor of a option called "climex.path". Fixing of the likelihood and return level calculation of the shape parameter == 0. Introduction of tests. I will note down new features more properly in the future. Up to now I (theGreatWhiteShark) was the only one using this app.

v0.8.2

With new version of the climex app

Now featuring a map to choose the individual stations and to show some summary statistics.

v0.8.1

updating namespacing

Previously I wrote the NAMESPACE by hand. But since there was a problem with the exporting of the box() function from the shinydashboard package breaking my app I decided to redo stuff. There seems to be a new version of Hadley's R pkg book. Or at least I do not remember much of it. Now everything is clean up with using the :: operator in function and @import and @export in the ROxygen documentation

.default fallback methods for functions.

In the very beginning of the package (when I still had the years, stations and bulk S3 classes) it was more or less handy to have a fallback method handling just plain numerical input. Since I just use xts class input since quite a while its now time to drop the .defaults and to clean up the package and it's exports. But this might break some code which is directly referring to e.g. block.xts()

stripped the ttplot.list method

Since always use multiplot for it and its more a tool to inspect a time series.