Version 2.0-5
* Adapted to new rgl 1.0.1 interface
Version 2.0-4
* F77_ calling conventions adapted from fixed length strings to length parameter in BLAS calls
* fitDirichlet returned loglikehood fixed (by adding a "-" in front of n)
* as.matrix.rmult method returns now a matrix with the rmult class stripped; in this way, derivate functions of `lm()` work properly (e.g. `anova`, `summary.manova`, `MASS::lda`)
Version 2.0-3
* kernel density estimation on the simplex with Dirichlet kernels (kdeDirichlet)
Version 2.0-2.9000
* own implementation of dDirichlet
Version 2.0.2 and Version 2.0-1.9001
Minor changes / Bugfixes:
* functions that do not work with sticky classes are now wrapped
(e.g. `anova()`)
Version 2.0-1.9000
Major change: sticky classes!
* Now objects of compositional classes do not miss their class
when subsetting. For example, if `x` is an "acomp" object,
x[1:2,] will also be an "acomp" object. Selection of columns work
as well, but it is a bit more tricky. Check ?`[.acomp` for
details. You can get this behaviour skipped by setting
setStickyClassOption(FALSE).
* The dollar notation `$` also works for subsetting columns.
Added:
* Limited support for using compositions classes as formal S4
classes, for instance in slots of S4 classes expecting a
"data.frame" or a "structure" (matrix or array).
* self-invertibility: now "rmult" objects created by any
transformation (cdt, clr, cpt, idt, ilr, ipt, iit, ilt, alr,
apt) remember the original data and the transformation between
them, and can be back-transformed with `backtransform`.
* added cdt and idt methods for objects of class "factor" (returning
the contrasts and their ipt transformation), and for "data.frame"
objects (exploiting the attribute "origClass" if they have it).
* `pwlr()`: pairwise logratio transformation (and inverse: `pwlrInv()`)
* `pwlrPlot()`: pairwise logratio plots vs covariables, both as explained
and as explanatory variables. This makes use of the ability of plot(x~y)
to react to the nature of `x` and `y` to create scatterplots, boxplots
and spineplots.
* transformation functions between variation and clrvar: variation2clrvar
and clrvar2variation
* `split()` methods for compositional classes
* new panels for filling `pairs()` plots with boxplots, densities,
kde2d-densities, etc...: check ?vp.kde2dplot
Minor Changes:
* the function to fit linear models of coregionalisation is now called
`fit.lmc()`; use full name specification with it! `compositions::fit.lmc()`
Bugfixes:
* bug in subsetting of one-column rmult object corected
-----------------------------------------
Version 0.92-1 (R (>= 2.2.0), rgl, tensorA, robustBase)
Added: Robustness: see robustnessInCompositions (robust= parameter)
Outlier classification: see outliersInCompositions
advanced treatment of Missings: see missigsInCompositions
CodaDendrogram (Tolosana)
A replot functionallity, changing plot parameters on the fly
A onlyPanel Parameter to zoom individual panels of highdimensional
plots
ternaryAxis
A new coordinate subsystem
Changed:
*.row *.col and *.inv replaces by *Row, *Col, *Inv to
ensure compatibility wiht R>=2.7
added NAMESPACE: most gsi. functions are now hidden.
If you need them use attach(gsi)
Bugfixes: "s" and "m" warnings
wrong axis-labels in logarithmic pairs plots fixed
log option in boxplot.rplus now shows really logodds rather than
log relative portions.
Labeling in plots, missing treatment in plots.
------------------------------------
Version 0.91-6 ( R >= 2.2.0 , rgl )
Added: plot3d, biplot3d,
Basic treatment of missings
Version 0.9-11 ( R >= 2.2.0 beta)
Added:
read.geoeas, read.geoEAS (Tolosana)
isoPortionlines, isoProportionLines (Boogaart)
kingTetrahedron (Bren)
Version 0.9-10 (R > 2.0.0 )
Was the first complete version