NEWS.md

In retostauffer/Rfoehnix: Objective Foehn Diagnosis

TODO

• Regularized probability models (based on glmnet): implemented, test
• There is one import ":::" (cgaussian family) in one of the examples -> illegal!
• A separate windrose vignette might be nice. ... started a separate vignette for tsplot, implement separate vignettes for windrose and image.
• A Tyrolean and a Californian demo vignette? - draft added.
• summary(..., detailed = TRUE) does not show sigma coefficients.
• Seems I am having some problems with cgaussian/tgaussian/...
• Windrose: changed argument mfcol to ncol, adjusted handling of labels/legends.
• Updated a series of R documentation pages.
• See foehnix 0.1-6, tsplot() fix: Should be implemented everywhere?

foehnix 0.1-6 (June 2023; bugfixes)

• tsplot() for temperature difference can now deal with all-missing (shows NO DATA)

foehnix 0.1-5 (Mai 2023; bugfixes)

• Found an issue in the windrose.foehnix plotting routine which - by default - centers the sector around north by shifting the sector by one half of the interval selected (default is 10). Thus, the most northern sector internally got defined as [-interval/2, interval/2] but forgot to take into account the data 360 - interval / 2. Should be fixed now.

foehnix 0.1-5 (March 2023; bugfixes)

• Fix in demodata() when loading multiple stations.
• Series of small testthat fixes (new default color, matrix/array class).
• Fixing a bug in summary.foehnix() when no inflation happened (sprintf error)

foehnix 0.1-4 (August 2019; development)

• Additional tests added.
• Added generic function for IGN (ignorance) as alternative score.
• Windrose: now allows for inputs \code{main}, \code{legend.pos} ("right" or "left"), \code{legend.title} (add legend title), and \code{labels.angle} (where to plot the labels on density plots), and
• Updated structure of the pkgdown site: combined README.md (index.html and github readme), removed "getting started", added demo data sets and vignette "Import data".

foehnix 0.1-3 (August 2019)

• Additional functionality added to image.foehnix (custom xlim and ylim, allow for custom style files).
• Revamped/simplified internal structure of image.foehnix.
• Updated the demo data sets (Ellboegen and Sattelberg); quality-controlled updated extended data set (by Deborah Detka). Moved from csv to rda using usethis; reduces package size; added data/scripts in data-raw.
• Moved generic image function (image.foehnix) to R/image.R.
• image allows to set custom xlim, ylim (and zlim) limits. Decreasing xlim values allow to plot over the new years period.
• Added generic write.csv method to save estimated probabilities to a CSV file. Will be ued to test output (and compare R/python implementation).
• New vignettes (currently in draft stage): each plotting routine gets its own vignette.
• Allow for custom wind speed breaks when calling windrose.
• Allow for adding wind sectors when calling windrose.default.
• Added second demo data set "Viejas and Lucky Five Ranch (CA, USA)"; modified the demodata(...) function to be able to handle both (multiple) demo data sets.
• Data sets (ellboegen, sattelberg, viejas, luckyfive) now stored as binary time series objects (.rda; zoo).
• Fixed a bug in the p-value calculation in summary.crch (abs was missing).
• New test statistics in summary.foehnix (similar to flexmix).
• Additional plot type ("posterior", 5L) in plot.foehnix.
• Fixed an NA-handling bug in foehnix_filter. Only rows used have to be checked (controlled by new input argument \code{cols}).

foehnix 0.1-2 (August 2019)

• Mainly a development release.
• Revamped the structure of the time series plot (tsplot)
• Added new features to tsplot: wind sectors, styles, better customization options (tsplot.control).
• Fixed few tests.
• Fixed a but in the Hovmoeller routine (image.foehnix); further adjustments required.

foehnix 0.1-1 (April 2019)

• uv2ddff and ff2dduv now return a zoo object if the input is a single zoo object (uv2ddff(data) or ddff2uv(data) where data is of class zoo).
• Checking for constant values: new check on main variable (after applying the filter). The check for multiple constants for the concomitant model is now performed on the logitX model.matrix.
• Removed custom LASSO option (lambda) from the foehnix interface. iwls_logit would, technically, allow for iterative penalized estimates but is rather slow. Thus, an experimental control argument for regularized estimates based on the glmnet package has been implemented (see \code{\link{foehnix.control}}/\code{\link{glmnet.control}}).
• windrose.default allows for custom data filters (using the foehnix filter method) and custom variable names for wind direction and wind speed (var.dd/var.ff). Default var.dd = "dd" and var.ff = "ff". Custom filters only allowed with multivariate objects (zoo or data.frame).
• windrose now allows to highlight specific wind sectors.
• Fixed tsplot: observed temperature is now on top, fixed the issue with the gray boxes (periods with a probability >= 0.5).

foehnix 0.1-0 (January 2019)

• windrose allows to specify custom names.
• tsplot allows to specify custom variable names (rename defaults). Requires some more testing.
• foehnix objects (returned by foehnix) contain a new element nobs, the number of elements used for classification. Note that this is not the same information as provided by the good, the bad, and the ugly from the foehnix filter (the filter_obj only contains information based on the variables used with the filter, not on the covariates used for the foehnix model).

foehnix 0.0-9

• Added a warning which will be shown if the EM algorithm stops in the first iteration (returns initial values). Quite likely a misspecified model (non-suitable formula given the data).
• windrose manual extended, windrose also allows to plot a windrose of non-foehnix objects (e.g., univariate zoo objects of dd and ff or two numeric vectors). Examples included.
• Added inflation handling. foehnix tries to inflate the time series object provided by the user to create a strictly regular time series object. If the inflation ratio exceeds 2 (the data set would be inflated by a factor of 2 or more) the script will STOP! foehnix.control provides an option (force.inflate) to overrule this check, however, might cause memory issues.
• summary.foehnix shows "inflation" count.
• foehnix will stop if the data set is not regular (note: regular is not "strictly regular", but is one requirement to create a strictly regular time series (inflate feature).
• Fixed an issue with foehnix_filter where observations (rows) where not all elements have been NA have been treated as "outside wind sector" rather than "not all observations available" (FALSE has been returned instead of NA; now an NA will be returned by if multiple filters are used and at least one element is missing).
• Fixed a bug where N_inflated was missing when no inflation was needed.

foehnix 0.0-8

• Added vignette for the logistic regression model (IWLS).
• Added new references.
• In iwls_logit: renamed variable mu to prob.
• Added a Hovmoeller diagram method (image.foehnix) with some features. Has to be seen as "under development" at the moment!
• Removed some warnings/notes.
• Added inference to the summary output (asymptotic theory).
• Added vignette for the inference part.

foehnix 0.0-5

• Extended manual pages for Getting Started, Statistical Model, and References.
• Updated predict.foehnix, new phoenix_filter was not yet implemented.
• Updated/fixed some smaller issues inside the plot.foehnix method (colors/line types/labeling), added log = FALSE option to plot the paths on the EM iteration scale rather than the log iteration scale.
• If windfilter is used as an argument to the method foehnix the script will now shout at you! Depricated option (this check should be removed before release).
• Added functions to convert ddff2uv and uv2ddff (plotting).
• Was working on the windrose and probwindrose plotting functions, however, not yet production ready. TODO!
• Added windrose function(s), most interesting is the windrose function for the foehnix objects.

foehnix 0.0-4

• Initialization of prob ($\pi$) for the non-commitant mixture model is now mean(z) rather than 0.5.
• EM/IWLS: if the log-likelihood improvement for a specific iteration falls below the threshold: ignore this iteration and return the information/coefficients/iteration count from the previous iteration (`iter - 1).
• windfilter renamed to filter when calling foehnix!
• foehnix_filter has been extended. So far an integer vector has been returned with the observations within filter (the 'good' ones). This made it hard to find out which ones have to be set to 0 (where both, y was ont NA and all variables for the filter(s) were not NA). I made the return of the foehnix_filter slightly more complex. foehnix_filter now returns an object of class foehnix.filter, a list with three elements containing integer vectors which correspond to the row indizes of the data (input x, see foehnix_filter manual). Returns:
• good: observations/rows within filter
• bad: observations outside filter, but all values required for the filter have been available (not NA).
• ugly: if at least one of the variables used for the filter was missing.

foehnix 0.0-3

• Manual pages/documentation now based on roxygen2.
• Nicer implementation of the switch = TRUE option. Simply use z = f(y <= mean(y)) instead of z = f(y >= mean(y)) when initializing the component membership to flip the components.
• Removed some smaller uncritical bugs.
• Dependencies/Suggestions corretly resolved.
• R CMD check does not shout on me anymore.

foehnix 0.0-2

• Fist stable alpha version of the package. Some feature are still missing, some of the methods and functions might change in the near future!

retostauffer/Rfoehnix documentation built on June 5, 2023, 11:39 p.m.