VEGAN DEVEL VERSIONS at https://github.com/vegandevs/vegan
ChangeLog provided a detailed development history of vegan, but it
is not updated after September 11, 2014. Vegan moved to git source
code management and linear ChangeLog is poorly suited for
branching git development. Use git log to track the history in
your current branch. The main upstream repository of vegan is
currently https://github.com/vegandevs/vegan.
ChangeLog gave technical details and was mainly provided for vegan
developers. Most important user-visible changes are listed in the
NEWS of the current release and its patched version.
Version 2.1-43 (opened September 11, 2014)
* cca, rda, capscale: remove u.eig, v.eig and wa.eig items or
scores scaled by eigenvectors. We have had a warning of their
eventual removal since vegan 1.6-8, and now we finally did
so. Scaling of scores happens when scores() function access
normalized score items of the results.
* commsimulator: commented out deprecation warning at the moment,
because this triggered warnings in CRAN packages bipartite and
metacom. Contacted the maintainers of these packages, and will put
back the warning in 2.2-1.
* getPermuteMatrix: works also when 'strata' are
missing. Triggered an error in CRAN package mpmcorrelogram.
* permustats: new function to extract permutation results from
vegan objects, with support functions summary, density,
densityplot, qqnorm and qqmath. Deprecated previous density and
densityplot methods directly accessing the same results.
* ordirgl, ordiplot3d: 3D functions using rgl or scatterplot3d
packages were removed from vegan and moved to a new CRAN package
vegan3d (released Oct 7, 2014).
* scores.hclust: combined documentation with other hclust methods.
* scoverage: removed from vegan.
* dispweight: veganified so that uses nullmodels. Added
gdispweight(): a new generalized method taking any glm() formula
instead of one grouping.
* betadisper Permutation tests via the `permutest()` method
are now parallelised. The number of cores to parallelise over
is specified by argument `parallel` in common with the
implementation for `cca()`.
Fix a couple of bugs in the examples; number of permutations
was specified using `control` which is not an argument to the
`permutest()` method for `betadisper()`.
* text.cca: was incorrectly testing if factor constraints were
in the biplot scores when user-suppiled lables for factors were
given.
* ccanova: removed from vegan. These were backup functions of old
anova.cca, and were completely rewritten for new vegan.
* metaMDSrotate: made defunct (replaced with MDSrotate).
Version 2.1-42 (closed September 11, 2014)
* Opened a new version to prepare release 2.2-0.
* Moved main development from R-Forge to GitHub.com.
* NAMESPACE, DESCRIPTION: adapted to current R CMD check that has
more stringent tests on attaching and importing external packages
and using their functions.
* getPermuteMatrix: routines have a common API based on the new
function.
Version 2.1-41 (closed September 4, 2014)
* ordiellipse, ordihull, ordispider: can now handle (omit) NA
cases in 'groups'. They were able to omit to NA cases in scores,
but having NA in 'groups' triggered really cryptic error
messages.
* adipart, multipart, hiersimu: it is now an error to provide
non-nested sampling hierarchy (used to be a warning).
* new version opened with the CRAN release of vegan_2.0-10 on
December 12, 2013.
* anosim, mantel, mantel.partial: R CMD check told that "... may
be used in an incorrect context". The dots were added in r2765,
and now removed.
* ordistep: add1.cca reported P=0 for redundant terms with 0 Df,
and this caused an error in ordistep.
* as.hclust.spantree: a new function to cast a "spantree" result
object to an "hclust" tree.
* hclust: add reorder() and rev() methods for standard "hclust"
trees of R. I have no clue why base R does not have these
methods, but I provide them now in vegan. An additional reason for
providing these methods is that reorder(, wts,
agglo.FUN = mean) will use unweighted mean of merged groups even
when these are of very unequal sizes. The reorder method provided
here will use group sizes as weights and the value of the group
will be the mean of its leaves (terminal nodes).
Add scores() method to extract coordinates of internal nodes or
leaves from a plotted hclust() tree. The function is whimsical and
may be removed before release (it is documented separately to make
this easier).
* biplot.rda: failed in axis scaling with negative 'scaling'
values when some species had zero variance (and hence species
scores was 0/0 = NaN).
* cascadeKM: Calinski index for one group will now be NA instead
of randomly chosen Inf, -Inf or NaN which can cause confusion (see
http://stackoverflow.com/questions/21022848/r-produces-different-result-after-io-on-file
* ordiellipse: failed if all points were on a line. Now handles
these cases by drawing a line trhough the points, and issuing a
warning from chol(): "the matrix is either rank-deficient or
indefinite". Earlier we required at least three points for an
ellipse, but these could still be on a line and fail. Now we
accept two points, and draw the line with a warning. The problem
with three points was reported by Paul Bacquet (Louvain,
Belgium).
* radfit: plotting of radfit frames of several communuties failed
if there were one-species or no-species (empty) rows. Part of this
was fitted.radfit that now returns sensible and consistent output
for these cases as well, and radfit.data.frame() completely
removes empty rows from the data (with a warning).
* RsquareAdj: return list(NA, NA) for capscale objects with
imaginary component, and use rda method if there is no imaginary
component.
* tabasco: "hclust" objects (use, sp.ind) are reordered using
weighted means. This is a better method than the unweighted means
used for reordering of dendrograms. Earlier "hclust" objects were
changed to dendrograms, but now we provide reorder.hclust() and
rev.hclust() in vegan, and can use improved method of ordering the
table. Dendrogram and hclust labels must match the community data
names, and now the community matrix is internally reordered to
match the dendrograms. This requires that the clusterings do have
labels attributes.
* treedive, treedist: default is now match.force = TRUE, and
treedive() gained new argument 'verbose' to turn of most messages
-- which is practical in oecosimu().
* BCI: names checked after http://www.theplantlist.org, but kept
the old (alphabetic) order of species. The changes are: Abarema
macradenium -> A. macradenia, Apeiba aspera -> A. glabra,
Aspidosperma cruenta -> A. desmanthum, Cassipourea ellipitica ->
C. guianensis, Chlorophora tinctoria -> Maclura t., Coccoloba
manzanillensis -> C. manzinellensis, Cupania sylvatica ->
C. seemannii, Dipteryx panamensis -> D. oleifera, Eugenia
coloradensis -> E. florida, Eugenia oerstedeana -> E. oerstediana,
Inga marginata -> I. semialata, Lonchocarpus latifolius ->
L. heptaphyllus (this is ambiguous, since Hebestichma cubense is
another alternative), Maquira costaricana -> Maquira guianensis
var. costaricana, Phoebe cinnamomifolia -> Cinnamomum triplinerve,
Swartzia simplex var. ochnaceae -> var. continentalis, Tabebuia
guayacan -> Handroanthus g. Guarea is ambiguous: data have three
taxa (fuzzy, grandifolia and guidonia), but theplantlist.org says
grandifolia is an ill. synonym of guidonia. This change allows
matching 206 of 225 BCI species with
http://datadryad.org/resource/doi:10.5061/dryad.63q27. In
addition, there are two vars of Swartzia simplex in BCI which
could be matched at species level.
* dune data sets: use 4+4 letter CEP names for species instead of
old 3+3 names. Botanical nomenclature was updated: Leontodon
autumnalis -> Scorzoneroides (Leoaut -> Scorautu), Potentilla
palustris -> Comarum (Potpal -> Comapalu). Rows are now arranged
numerically, and species names alphabetically, vascular plants
first, and then the two bryophytes.
Added data 'dune.phylodis' which are ages of dune species
extracted from
http://datadryad.org/resource/doi:10.5061/dryad.63q27 (Zanne AE et
al. 2014, Nature doi:10.1038/nature12872, published online Dec 22,
2013).
dune.taxon was updated to APG III.
* varespec: use 4+4 letter CEP names instead of 3+dot+3.
Nomenclature was cautiously fixed. Most important changes: Ledum
palustre -> Rhododendron tomentosum (Led.pal -> Rhodtome),
Cetraria nivalis -> Flavocetraria nivalis (Cet.niv ->
Flavniva). In addition, Dip.mon was corrected to Diphcomp
(Diphasiastrum complanatum, should perhaps be Lycopodium c.), and
Barbilophozia lycopodioides to B. hatcheri.
* plot.envfit: when `envfit()` is called as
`envfit(ord, foo[, "bar"])`, there are no useful rownames on the
objects returned. Hence the logic in `plot.envfit()` when called
with argument `labels` was failing.
* rarecurve: line colour and type for each sample can now be
specified through formal arguments `col` and `lty`. Incidental
wish of http://stackoverflow.com/q/22714775/429846.
* simper: now doesn't fail with obscure error when groups have a
single member.
Version 2.1-40 (closed December 12, 2013)
* anova.cca: Function is now based on the new code, but the old is
avaialable in a function called ccanova. The API changes:
arguments 'step' and 'perm.max' are gone and replaced with
'permutations' that accepts a how() object defining permutations,
or a single number like previously or a permutation matrix. The
new anova.cca adds an option of analysing a sequence of ordination
models. This also means that '...' is now the second argument and
the names of all arguments must be written in full. This change
can be expected to cause trouble outside vegan. Currently the
tests in examples are passed, but tests fail (looks like being
caused by the change in API).
* add1.cca, drop1.cca, ordistep, ordiR2step: functions were
adapted to new anova.cca and this brought along similar changes in
API.
* envfit: function assumed that environmental variables are either
factors or numeric, and choked if they were neither but, say,
character strings. Now the function tries to coerce all
non-numeric variables into factors, including character strings
and logical. This is one possible reason for problems reported in
https://stat.ethz.ch/pipermail/r-sig-ecology/2013-December/004217.html,
although the reported case is irreproducible and we cannot be
sure. Also fixed handling of tied values in assessing the P-values
in vectorfit.
* nestednodf: vegan 2.1-36 (release 2.0-10) changed the function
so that weighted analysis of binary data was equal to unweighted
binary analysis, but this broke consinstency with the original
software and publication by Almeida-Neto & Ulrich. The fix was now
made optional, and the default is to follow published method by
adding argument 'wbinary' (defaults FALSE). Based on the
suggestion by Matt Barbour in GitHub.
* ordispider: can now use spatial medians as centres instead of
the default centroids. The kind of centre is defined by new
argument 'spiders'.
* tests for cca/rda/capscale: commented out test that failed with
the anova.cca. Known issue was that 'by = "term"' and 'by =
"axis"' stop with error with missing data. This was designed and
and can be re-designed to handle missing data by listwise deletion
so that models will have the same number of observation for every
variable. An unknown issue was that 'by = "margin"' failed in
capscale(). This may be a scoping issue an needs inspection. We
keep the old vegan-tests.Rout.save file so that we are reminded of
the problems at every check.
Version 2.1-39 (closed December 3, 2013)
* anova.cca: started to rewrite the anova.cca family of functions
for permute package. At the first stage, a temporary development
function anovacca was created. The user interface was changed, and
the function no more adapts the number of iterations for the
P-value, and arguments 'step' and 'perm.max' were removed.
Instead, permute package is used to create a permutation matrix
used in all cases with fixed number of permutations.
In addition to the overall test, the function allows now testing
a sequence of models (anova.ccalist). Specific tests provided are
by = "term" (anovacca.byterm) which is fitted as a sequence of
models in anova.ccalist. Case by = "margin" directly calls
permutest.cca and gets the significances from differences of
residual variation similarly as anova.ccalist. Case by = "axis" is
is implemented as a marginal model.
Simple permutations give identical results for "term" and "axis"
cases, but by = "margin" is different. Marginal models were
implemented as partial models with other parameters partialled out
in vegan 2.0 and earlier, but the current implementation is an
anova.ccalist model where the parameter in question was removed
and model compared against the complete model.
* commsim: documentation (commsim.Rd) was restructured so that
nullmodels were collected under separate sections with a brief
introductory text and shorter specific text of the
algorithm. Hopefully this makes easier for an outsider to grasp
the width of the choices.
* oecosimu: change printed quantiles to match the direction of the
test as changed in r2495.
* permutest.betadisper: updated to the new permute API and operates
similarly to permutest.cca in respect of how the permutation test
can be defined. It currently doesn't support the parallel processing
of the cca method, however.
* tests: vegan examples and vegan-tests have been out of sync for
a long time. These have not been updated because most of the
changes seem to be triggered by switching to R 3.0-x, and we have
not had time to analyse the reasons. The differences also seem to
be platform specific, and Linux and MacOS give slightly different
results. In particular, there seem to be differences in
permutations, constrained ordination, in particular in capscale()
and rounding of output. We have also introduced some changes in
output that were not yet synced.
* janitorial: vegan has been dependent on R >= 2.14.0 since
version 2.1-34 (r2597 Wed 28-Aug-2013). Now superfluous references
and tests for the older R version were removed in R code and
documentation. FIXME: cca() returns residuals.zombie item that was
supposed to be needed in R < 2.13.0. This was not yet removed: we
must first check that this can be safely done.
vegan also now depends on permute version 0.7-8 or later.
Version 2.1-38 (closed November 10, 2013)
* DESCRIPTION: depends on permute >= 0.7-5, where the
permute::how() result object can be updated.
* bioenv: returns now the number of the best model ('whichbest'),
the standardized environmental data ('x') and the distance
function used for the environmental data ('distfun'). New function
bioenvdist() uses these to re-calculate the environmental
distances for the best model, or any other model selected by its
number.
* permutest.cca: permutation test uses 'permute' package. The old
interface was retained, and 'permutations' can be a single number
or a permutation matrix (as previously), but now it can also be a
how() object of the 'permute' package. The argument 'strata' was
also retained, but it is planned to be deprecated in the future,
and it is recommended that users switch to defining 'blocks' in
how().
* renyiaccum: plot() works also when only one index ('scales') was
used. The function gained new argument 'collector = FALSE' to
accumulate sites in the order they are in the data set in addition
to the summary statistics of permutations. This can be used to
analyse the randomness of the particular order of sites (looks
like my student would need this).
Version 2.1-37 (closed November 5, 2013)
* anova.cca: added new function anova.ccalist() to compare a
sequence of models. The function is still experimental ("proof of
the concept") and unexported. If this stays in vegan, it should
eventually be called from anova.cca(). This would bring along a
change of API to anova.cca(object, ..., alpha=...): the dots must
follow the first argument which turns of positional and partial
matching of arguments so that the function can collect the "cca"
models. We must decide whether the new function is worth such a
change that can make life harder for ordinary users.
One potential advantage is that the code in anova.ccaby* functions
could be simplifed to a anova.ccabylist() calls.
The function is based on calling permutest.cca for each model with
identical permutations. We can then compare the change in model
for each permutation and collect the test statistics for
differences. This requires that the models really are nested so
that residual deviance certainly decreases in bigger model
(testing theory requires nesting, but this is commonly violated by
users: here nesting is necessary).
* parellel: default cluster defined by setDefaultCluster() is no
longer used in functions with parallel processing. Using default
cluster would need querying an unexported environment
parellel:::.reg, and this gives a NOTE in R CMD check.
* specaccum, renyiaccum, tsallisaccum: gained argument to select a
'subset' of sites (looks like my student would need them).
Version 2.1-36 (closed October 14, 2013)
* opened with the release of vegan 2.0-9.
* decostand(..., "normalize") uses now .Machine$double.eps to
replace zero sum of squares instead of matrix minimum.
* envfit: if a variable is constant, no fitted vector can be
calculated and results will be given as NA (with warnings).
Plotting of all vectors will fail in such cases because no finite
scale was found for arrows. Fixed on ordiArrowMul.
* envfit: try to produce something sensible if fitted variable has
constant values (invariable vector, one-level factor). Report
arrow heads as all zero, and R2=0 for vectors, and the centroid to
the data centroid and R2=0 for factors. Tied values are now
treated differently in factorfit: now they support null
hypothesis, previously they decreased the P-values.
* fisherfit: completely rewritten and estimates of standard error
removed: I could not find no justification for these. Actually, it
seems that the value of Fisher alpha as estimated in the function
was independent of the abundance distribution of species, but will
be defined by the number of species (S) and number of individuals
(N). Now the Fisher alpha is estimated from the relationship S =
alpha*(1 + log(N/alpha)) using function uniroot(). Because of
this, standard errors cannot be estimated and they were
removed. In addition, functions confint.fisherfit,
profile.fisherfit and plot.profile.fisherfit were removed. The
estimation of standard errors was also removed in function
fisher.alpha (that only calls fisherfit).
* nestednodf: matrix fill was wrongly calculated in weighted
analysis. The nominator was length of 'comm', and if input was a
data frame that was the number of columns instead of the number of
cells. The fill was correct in non-weighted analysis because there
data were transformed to a matrix, and the length of a matrix is
the number of cells (unlike in data frames).
* nestednodf: weighted analysis gave all statistics as zero if
binary data were supplied. Some ">" comparisons were changed to
">=" and now weighted analysis of binary data gives same results
as non-weighted analysis. However, this can change results of
weighted analysis of quantitative data. The change needs
endorsement by the function author Gustavo Carvalho.
* oecosimu: warns user if the specified nullmodel 'method' changes
quantitative input data to binary data under cover. Some people
have not noticed this.
Version 2.1-35 (closed September 25, 2013)
* ordilabel: colour arguments ('col', 'fill', 'borderä) can now be
vectors and will be recycled if needed. May need care if used with
'select', but should work with 'priority'. This was needed for a
function under development and testing: when labelling over lines,
it may be nice if the border has the same colour as the line
covered.
* MDSrotate: can now rotate a solution to more than one vector,
provided that the number of dimensions is higher than the number
of vectors. Because fitted vectors usually are correlated, only
the first vector is aligned to dimension 1, but second vector and
further are oblique to the corresponding axis. In any case they
will have zero correlation to all subsequent dimensions.
* simulate.rda, simulate.cca: gained argument 'correlated' for
using covariances of species in generating correlated multivariate
normal residuals in parametric simulations. The covariances are
estimated from the residual ordination of species. The argument
defaults FALSE which implements the old parametric simulation
where each species is simulated independently. The argument has
no effect in capscale() which stores no information on species.
Version 2.1-34 (closed September 5, 2013)
* DESCRIPTION: dependent on R >= 2.14.0.
* DESCRIPTION, vignettes: R 3.0.2-to-be checks with --as-cran
requires (with a NOTE) the vignettes source files (Rnw) to be in
vignettes/ directory. Because Makefile is not executed in
inst/doc if vignettes/ directory is present, all other sources had
to be moved to vignettes/ as well with vignettes/.install_extras
to move those to inst/doc after building. This also made
/.Rinstignore unnecessary. The vignettes directory was introduced
and made recommended in R 2.14.0 so that the this version of vegan
depends on R >= 2.14.0.
* tweaks for R < 2.12.1 removed: code to fix buggy naming of
columns in qr.X before 2.12.1 (as.mlm.cca/rda, intersetcor), and
change in the cmdscale() output (eigenvals.default).
* tweak for R < 2.13.0 removed: change in cmdscale() output
(capscale).
* vegandocs: does not use unexported tools:::httpdPort. Now only
remaining ':::' case is querying default cluster in the
environment parallel:::.reg.
Version 2.1-33 (closed August 28, 2013)
* DESCRIPTION: new dependence on lattice. Passes new strict R
checks with NOTE on ':::' calls to tools:::httpdPort (to launch
reading vegan NEWS.html in existing browser window) and
parallel:::.reg to acces the defaultCluster in an unexported
environment within 'parallel' Namespace.
Version 2.1-32 (closed August 19, 2013)
* opened a new version with the CRAN release of vegan 2.0-8.
* merged Eduard Szöcs's code on dispersion weighting of
overdispersed species following Clarke, K. R., M. G. Chapman,
P. J. Somerfield, and H. R. Needham. 2006. Dispersion-based
Weighting of Species Counts in Assemblage Analyses. _Marine
Ecology Progress Series_, 320, 11–27. The basic development was
made in github.com and merged to R-Forge.
* nullmodel: replaced internal indshuffle() function in
make.commsim() with much faster stats::rmultinom(). The
rmultinom() function takes argument 'n' for the number of random
vectors, and using this could be still faster, but we only
generate one vector in time. Using 'n' would require better
analysis of individual nullmodels. The commit changes random
sequences, but passes tests.
* oecosimu: Gained argument 'batchsize' to set the maximum size of
simulated nullmodels (in Mb). If a larger object would be
produced, the analysis is broken into several batches to avoid
exceeding the maximum size. This avoids exhausting memory which
can make whole R unresponsive and analysis very, very slow. In
general, the argument is needed with large data sets and/or large
number of simulations.
* orditkplot: bmp has been available in unix-alike OSes since
2008, or a moth after writing orditkplot. Thanks to Brian Ripley
for informing us.
* vignettes: vignettes use now standard article style instead of
hacked jss style. Decision and diversity vignettes are in two
columns, but intro in one (R output did not fit in one
column). The common packages, macros and definitions were moved to
new vegan.sty which is written so that it should work both with
amsart and article, and with one and two columns. Figures are now
in standard figure environment, but intro redefines this to use
sidecaption figures.
Version 2.1-31 (closed July 10, 2013)
* Dependencies: Vegan now depends on a version equal to 0.7-4 or
later.
* betadisper, permutest.betadisper: Modified to use the new
permute package API (from version 0.7-3 onwards). The `permutest`
method gains a new argument `permutations`, which takes a matrix
of permutations to be used.
* protest: modified to use the new permute API. Gains argument
`control` which describes the design. As a result, `strata`
argument has been removed and `permutations` argument can only
be used to supply your own matrix of permutations. The number of
permutations and other features of the design are set via `control`
and function `how()` from permute.
* ordipointlabel: now uses `ordiArgAbsorber()` to stop the warnings
about non-graphical paramters being passed to plotting functions.
* ordisurf: can now pass in a line width for the contours via
argument `lwd.cl`.
* ordiArrowTextXY: New (internal) support function that finds
coordinates of text box at the point of the arrow so that the
arrow point just touches the text. This should improve
positioning of the arrow labels and avoid writing labels over the
arrows.
* plot.envfit, plot.cca, text.cca: use ordiArrowTextXY() for arrow
labels instead of expanding arrow heads by 10%.
* plot.envfit: plot(..., add = FALSE) estimated string (text)
dimensions after plot.new() but before plot.window(). Since
plot.new() sets xlim, ylim to c(0,1) and then plot.window resets
the limits to the data values, string dimensions in user units
were poorly estimated. This became evident with new positioning of
arrow text based on string dimensions.
* plot.ordipointlabel: gains a plot method that is very similar to
`plot.orditkplot()` but which does not mess with graphical
parameters. This allows it to fit more naturally into a standard
R workflow (it plays nicely with `layout()` for example.
* stressplot: metaMDS and monoMDS stressplot() use now
expression(R^2) instead of ascii R2. The stressplot.monoMDS()
function returns the plotting structures in the original input
order.
* vignettes: Brian Ripley urged as to remove
\usepackage{inconsolota} as this package is on the way to be
removed from CTAN.
Version 2.1-30 (closed June 12, 2013)
* bioenv: can now use Mahalanobis, Manhattan and Gower distances
for environmental variables. The Mahalanobis distances are based
on orthogonalized data, Manhattan distances give the direct sum of
differences of environmental variables, and Gower distances can
also handle factor variables. This involves adding internal
function veganMahatrans() for Mahalanobis transformation. The
change was triggered by a recent email by Lydia Beaudrot (UC
Davis) to implement Mahalanobis distances, and in the same I also
implemented Robby Marotte's suggestion of using Gower distances
(vegan Forum item in R-Forge in July 2012). The output is changed
to show the 'metric' and the name of the 'method' is fully
expanded. No more fails if 'upto' is too large. Passes "..." to
cor() like documented, but never done.
* rankindex: can now use Mahalanobis or Manhattan
distances. Scales automatically gradient variables to unit
variance (Euclidean) or range (Manhattan), and the scaling is
already inbuilt in Mahalanobis and Gower.
* vegdist: added Mahalanobis distance.
Version 2.1-29 (closed April 19, 2013)
* ordisurf: significant changes were made to this function:
- The default for `method` and `select` were changed to `"REML"`
and `TRUE` respectivelt.
- Argument `thinplate` is deprecated in favour of `isotropic`. A
warning is now issued if `thinplate` is used.
- The spline basis for the smoother can now be specified from a
subset of those implemented in the mgcv package. This is achieved
via the `bs` argument, which defaults to `"tp"` for thin plate
regression splines.
- Argument `knots` and `bs` can now be a vector of length two, one
per ordination dimension considered. This is only of use with
anisotropic surfaces with `isotropic = FALSE`.
- New argument `fx`; indicates whether the smoothers are fixed
degrees of freedom regression splines (`fx = FALSE`) or a
penalised regression spline (`fx = TRUE`). Can be a vector of
length 2 for anisotropic surfaces (`isotropic = FALSE`).
- The number of locations in each ordination dimension at which the
fitted surface is evaluated can now be specified via new argument
`npoints`.
- The formula passed to `gam` is now built in greater detail. When
the model is printed the user can see exactly how the smoother was
constructed.
* oecosimu: the interpretation argument "alternative" was really
twisted. We now changed the test direction and have a much clearer
explanatory text in printed output. The issue was raised by Juan
Manual Barreneche (jumanbar) in GitHub (issue #14). NB. The
direction of the test changed from previous versions.
* raupcrick: adapted to the change in oecosimu() to define the
direction of the test.
Version 2.1-28 (closed April 19, 2013)
* betadisper: failed with type = "centroid" when there was only
one group (i.e., in estimating the overall beta diversity in the
data). Reported by Pierre Legendre.
Now correctly reports distance to "medoid" in the print method
when type = "median". Reported by Pierre Legendre. The print
method also now shows only the first 8 eigenvalues.
* eigenvals: new method for class "betadisper".
* rda: eigenvalues are now regarded as zero if they are very small
compared to the first eigenvalue. Earlier we used fixed limit of
1e-4, but now the limit is first eigenvalues * 1e-5. Similar
change was not made in cca, since there the theoretical maximum of
the eigenvalue is 1, and comparison to an absolute minimum
threshold of 1e-4 makes sense.
Version 2.1-27 (closed March 17, 2013)
* cca/rda/capscale: removed references to 'u.eig' and 'v.eig'
items in support functions. These now use 'u' and 'v' items and
scale these with 'eig' (or diag(sqrt(eig))) when needed. This
prepares for removing items 'u.eig', 'v.eig' and 'wa.eig' in vegan
2.3 and 2.4. Concerns fitted.capscale, goodness.cca and
goodness.rda. The documentation has warned about removing these
*.eig items for years, and most vegan functions already avoided
using them. We have no idea if any external packages depend on
these and therefore we delay the removal till vegan 2.3 and 2.4
release. However, we warn about this in NEWS for 2.0-7 and also
say this clearly in cca.object documentation.
* nestedtemp: function failed if the matrix fill was < 0.38%,
because the fill line parameter was outside the original
estimation bracket. Now bracket is moved up if the estimation
fails. The problem was reported by Carsten Dormann (Univ Freiburg)
and Benjamin A. Sikes (Lincoln University, NZ).
* specaccum: gained argument 'w' for weights to give the sampling
effort. This resurrects a feature that was introduced in r1505 (22
Feb 2011) and 1507, and removed in r1606 (26 May 2011). It seems
that BiodiversityR::balanced.specaccum() provides the same feature
but we still try it here. The feature was resurrected after a
user query by Bastien Mérigot (Univ Montpellier II). The working
of the new argument is still untested with specaccum() support
functions.
* fitspecaccum: model = "asymp" was actually fitting logistic
regression (the same as model = "logis").
* text.cca, points.cca: gained argument 'axis.bp' (defaults TRUE)
to suppress drawing axes for scaled biplot arrows. Only effective
if 'bp' scores were requested.
* wcmdscale: setting only weights 'w' does not force full
"wcmdscale" output but returns only the matrix of coordinates of
real axes. Similarly, setting 'add = TRUE' would not force
"wcmdscale", but 'add' is not implemented.
Version 2.1-26 (opened February 11, 2013)
* New version opened with the release of vegan_2.0-6 on February
11, 2013.
* metaMDS: exposed argument 'maxit' in metaMDSiter() so that users
can set the number of iterations in monoMDS() or isoMDS().
metaMDS(..., trace = 2) show the stopping criterion used in with
engine = "monoMDS". After wishes by Jon Bakker, Univ Washington
(U.S.A.).
* tabasco: a sister function of vegemite() to display a compact
community table using heatmap(). Both vegemite() and tabasco() can
handle cluster::agnes() trees.
* wcmdscale: return a full "wcmdscale" object if any argument is
set to non-default value. This also implies that if weights 'w'
are set, the result will be full "wcmdscale" object with a
"weights" item, instead of a simple matrix with no information
about weights used.
Version 2.1-25 (closed February 11, 2013)
* FAQ: new entry on impossibility of using random effects in cca,
rda, capscale & adonis, and telling how to do this
approximately. Please comment and correct.
* oecosimu: changed the structure of the oecosimu() result
object. It now returns a list of two items: "statistic" is the
observed statistic, and "oecosimu" which contains the simulation
records. In previous version the function returned the statistic
and added item "oecosimu" there. Previously, the result object was
of type c("oecosimu", class(statistic)) as the original
"statistic" object was returned amended with an "oecosimu" item,
but now it is only of type "oecosimu" with the original
"statistic" as a separate item. So "statistic" is now one level
deeper instead of being the main object.
* predict.cca, predict.rda: In rev2412, 2413 we studied including
partial (pCCA) component in predict(..., type="response") and
predict(..., type="working") models, but then adopted a policy of
never having partial component, but always returning only the
component requested for. Therefore warning messages of ignoring
pCCA component were removed. This log entry was made to emphasize
that there is a policy decision, and an alternative policy can be
tracked in the repository.
* fitted.[cca,rda,capscale]: can now return partial component with
argument 'model = "pCCA"'.
* simulate: the simulate functions for "rda" and "cca" return
objects with original row and column names. Function capscale()
already did so, and simulate.nullmodel() was changed so in vegan
2.1-24 (r2396).
Version 2.1-24 (closed February 2, 2013)
* simulate.nullmodel: output array inherits dimnames after the
input data matrix, this is often required by oecosimu.
Dimension names are now only stored once for all nsim
simulations, so it is an improvement over the implementation
in commsimulator.
* oecosimu: the 'oecosimu' list cannot be added to the result if
the nestfun() returns a data frame. In that case, the 'oecosimu'
list is treated like a variable, and this gives either an error of
wrong length or if the length matches, a mess. Now data frame is
silently turned into a list which also means that it will not be
printed with the output. The name of the 'statistic' is not used
for unnamed vector output where it would only name the first item
(like would be the case if the 'statistic' was extracted from a
data frame). It is now checked that only one 'statistic' is given
and a comprehensible error message is issued instead of the
current confusing one. The changes were trickered when testing
picante::pd().
* protest: huge speed-up. Instead of calling procrustes() in every
permutation step, we only calculate the goodness of fit statistic
in svd(). This avoids a huge overhead of procrustes(). In a test
with a 160 x 12 matrix (RDA scores from Bryce Canyon data
'bryceveg') with 9999 permutations, the time went down from 12 sec
to 0.8 sec in my desktop. The analysis prints now also the 'ss'
term (residual sum of squares) which for symmetric analysis is
equal to squared m12.
* procrustes: marginally faster way of getting sum of squares of a
matrix. This should not influence the results, but one metaMDS()
trace result has a small difference in nearly-zero rmse (was
1.094382e-06, is 1.09439e-06) in my desktop.
* treedive, treedist: treedive() did not correctly match data and
tree when the tree contained species that did not occur in the
data. Function treedist() tries to match tree and data when their
sizes differ, and argument 'match.force' was added to force
matching even when sizes do not differ.
Version 2.1-23 (closed January 25, 2013)
* clamtest: Richard Telford reported an extreme case
{rbind(a=c(1,0,5,10,4),b=c(0,10,5,2,1))} where
clamtest failed due to inadequately setting up minimum
abundance thresholds for rare species. The issue is solved
by hard coding the minimum values to be 1 when no suitable
solution is found. Also, clamtest failed when the community matrix
had no column names, this is now fixed.
* capscale: It was wrongly assumed that eigenvalues could be used
in normalization of species scores, but this worked only with
Euclidean distances. Now normalization is done explicitly with
decostand() function. This change means that scaling of species
scores will change, and graphs can look different than
previously. All analyses should be redone. Function now displays
the value of the additive constant with 'add = TRUE'.
* stressplot: added stressplot() methods for wcmdscale(),
capscale(), cca(), rda(), prcomp() and princomp() results. These
also work with constrained ordination. These methods display the
ordination distances in given number of dimensions (defaults 'k =
2') against original observed distances. These original distances
are found from the full space solution, and in capscale() and
wcmdscale() they are correct for the imaginary axes. The weights
are used in wcmdscale() and cca() so that their distances differ
from plotted ordinations, but agree with eigenvalues. Partial
models (p-dbRDA, pRDA, pCCA) add the partial component both to the
original dissimilarities and the fit.
The row scores (u) alone will not correctly estimate original
dissimilarities in constrained (or partial) ordination. In
unconstrained ordination we can get the distances as dist(u %*%
diag(sqrt(eig))), but in constrained ordination this will not give
the observed dissimilarities with all axes. Currently we get the
ordination distances from a (low-rank) approximation of the data
as dist(u %*% diag(sqrt(eig)) %*% t(v)). However, it is not sure
that this the right thing to do, but perhaps we should acknowledge
the fact row ordination with constraints does not approximate
distances. So this may change.
* wcmdscale: added method functions print(), plot() and
scores(). Now class "wcmdscale" results also retun the function
call and dimensions have names.
* ordilabel: was missing ordiArgAbsorber() on the plotting calls
to text() and polygon(). Thus lots of warnings were raised in use.
* orditorp: added argument select, to choose which of the rows of
scores are plotted, so matches with ordilabel() and
ordipointlabel().
Version 2.1-22 (closed January 8, 2013)
* multipart: argument global was printed as TRUE regardless of the
actual argument value. This did not affect calculations. Reported
by Valerie Coudrain.
* monoMDS, metaMDS: Default convergence criteria were changed in
monoMDS. Most importantly, now scale factor of the gradient is
sfgrmin = 1e-7. The former limit 1e-5 was much too slack with
large data sets and iterations stopped too early without getting
close to the solution. In addition, scores() ignore now requests
to scores beyond those calculated instead of failing, and
scores.metaMDS() does not drop dimensions.
* metaMDS: Iteration sometimes finds a false convergence, or an
identical solution to the previous best although a better solution
can exist. These are undetected except sometimes in parallel
processing. Therefore we first label convergence false when we
find a new best solution, and then study if it really
converged. This is a rare effect, and can only be seen with
parallel processing. An example is
set.seed(7)
metaMDS(BCI, k=4, parallel=2)
which will converge in run 7 with the old code although run 8
would be better, but converges in run 18 with the new code.
* nestednodf: added plot() method modelled after
plot.nestedtemp().
* ordiR2step: gained argument 'R2scope' (default TRUE) which can
be used to turn off the criterion of stopping when the adj-R2 of
the current model exceeds that of scope. This option allows model
building when the 'scope' would be overdetermined (number of
predictors higher than number of observations). Pierre Legendre
needed this option for some checks with huge AEM/PCNM scopes.
* envfit, plot.envfit: Plotting an object fitted by envfit() would fail
if p.max was used and there were un-used levels for one or more factor
constraints. The un-used levels could result from deletion of
observations with missing values or simply the result of supplying
a subset of a larger data set to envfit(). Both cases are now handled
through the use of droplevels().
Version 2.1-21 (closed November 19, 2012)
* New version opened with the CRAN release of vegan 2.0-5 on Oct
8, 2012.
* dispindmorisita: output gained a new column for Chi-squared
based probabilities that the null hypothesis (random distribution)
is true.
* clamtest: output was wrong when some of the possible
species groups were missing (bug report submitted by R Telford).
* procrustes: plot() of two-dimensional solutions often draw
original axes in a wrong angle. The problem was reported by
Elizabeth Ottesen (MIT).
* msoplot: gained legend argument for positioning the legend
according to user needs.
Version 2.1-20 (closed October 8, 2012)
* anova.cca: Dr Sven Neulinger (Christian Albrecht University,
Kiel, Germany) reported several problems with anova.cca cases. All
these were problems in scoping. Two problems solved with this
commit were: (1) anova.cca(..., by = "axis") always failed in
partial analysis (with conditions), (2) anova.cca(..., by =
"term") failed in partial models when there was no 'data='
argument, but the variables were in the global workspace. The
first fix also seems to allow anova(, by = "axis")
that used to fail. In addition, there is one unsolved problem with
search order: stats function C() is found instead of variable 'C'
and this gives an error message "object is not a matrix".
* radfit: The methods are now more consistent over different
levels of radfit models (radline, radfit, radfit.frame). The
common methods to all include now AIC(), coef, deviance(),
logLik(), predict() and fitted(). The radfit() objects gained
points() and lines() methods. The data frame method also works
with matrices. The predict() method gained new argument to change
the expected 'total' size of communities. The fitted() returns now
named vectors or matrices. It seems that radlattice() never
displayed BIC contrary to documentation and labelling in graph.
This has been wrong since the introduction of radlattice in r551
(2008-11-09). Now radlattice also check that it gets the "radfit"
object it can handle.
* Rd: documentation files upgraded to the third English edition
(2012) of Legendre & Legendre.
Version 2.1-19 (closed September 16, 2012)
* adonis: Small changes in calculations. Simplified calculations
of matrix G as centred distance matrix, and does not keep n x n
matrices that are not needed (A) or used only once (identity
matrix I). These can make calculations marginally faster and
reduce the memory usage, but probably there are no observable
effects in most data sets. The new centring is probably more
accurate than older, and therefore the last significant digits can
slightly change (magnitude 1e-12 in tests).
* betadisper: An effective R stats .C function was used for double
centring, but it was removed from the API in r60360 | ripley |
2012-08-22 07:59:00 UTC (Wed, 22 Aug 2012). This removal stopped
betadisper() with error. Now we have a less efficient R code for
the same purpose. However, the effects in timing should be
negligible.
* density methods: all vegan functions that return simulated or
permuted statistics have now density() methods that directly
access the returned statistic. The functions return an object of
class "vegandensity" that inherits from class "density". The
object is identical to class "density", but it is amended with
item "observed" that contains the observed statistic. The observed
statistic is also put among permuted values when estimating the
density. This can cause a pimple in density lines when the
observed statistic is very different from simulated values, but it
is consistent with the permutation tests.
The function has a plot.vegandensity() function that is similar to
plot.density(), but it also draws a vertical line for the observed
statistic.
The density methods were made available for adonis, anosim,
mantel & partial.mantel, mrpp, permutest.cca and procrustes. The
anova.cca function does not return permutated statistics.
All density methods handle only one statistic. Function adonis()
can return a matrix of permuted F-values for each term, and it
gained a densityplot method (lattice package) that can handle all
these simultaneously.
Functions adipart, hiersimu and multipart were made
oecosimu-compliant earlier and simultaneously they also gained the
density and densityplot methods.
* mantel, mantel.partial: Gained argument na.rm (defaults FALSE)
to remove missing values from dissimilarities. To implement this,
there were some internal changes in functions (that should not
influence the results): The functions mixed cor.test() and cor(),
but now only use cor(). Function cor.test() was only used to get the
textual presentation of the correlation 'method', but this is now
found internally.
* protest: do not return the observed statistic as one of the
permuted values but separately.
* radfit: gained a predict method which works for single models
('radline'), radfit, and radfit.frame. All predict functions
accept 'newdata' which need not be integer, but extrapolation may
fail for some models. Needs still documentation. The function was
provided due to a user request.
Version 2.1-18 (closed August 20, 2012)
* cIndexKM: internal function count() triggered a warning in R CMD
check with R-devel because the funtion used .C() call to an
unloaded packages "cclust". count() was never called in the
current cIndexKM() and was removed together with the following
currently unused functions: withinss(), varwithinss(),
maxmindist(), vargss(). These deletions should have absolutely no
visible effects.
* envfit: the plot() method gained argument 'labels' to change the
default labels. The default labels are displayed with the new
labels() function.
* vignettes: building vignettes failed in CRAN and R-Forge. These
sites use TeXLive 2012 distribution which was not yet used by any
vegan developer when vegan 2.0-4 was released. Ubuntu Linux still
stocks TeXLive 2009, and new version is promised first for the
12.10 release, and many other Linuxes are just as old. TeXLive
2012 for MacOS was released in July 2012 and with that we could
pin down the problem. There is hardly any user visible changes
except that building vegan succeeds with vignettes.
Version 2.1-17 (closed July 30, 2012)
* New version opened with the CRAN release of vegan 2.0-4 on June
18, 2012.
* biplot.rda: bug in specification of `type` argument if not supplied
by the user; should have been a vector of length == 2.
* adipart, hiersimu, multipart: default and formula methods of
these functions were identical (also for the calculations) except
in interpreting the input. Now the formula method only interprets
the formula and calls the default method for the actual
calculations without replicating its code. The "call" attribute of
these functions now returns the generic function name without
".default", ".formula" suffix.
Functions use now print.oecosimu() for displaying results and
their specific print.*() functions were deleted. This involved
changes in attributes: the printed attributes are now in
object$oecosimu$simulated instead of object.
* oecosimu: returns "call" attribute similarly as adipart(),
hiersimu() and multipart(). The print.oecosimu() output changed,
and shows the call. print.oecosimu() is able to display adipart(),
hiersimu() and multipart() results, but does not show all
informations that those dedicated functions showed about options.
* Formula methods for adipart/multipart/hiersimu functions
use a new internal (hierParseFormula) to interpret the formula.
Version 2.1-16 (closed June 18, 2012)
* envfit: plot() gained new argument 'bg' that triggers labelling
with ordilabel() using the colour given in 'bg' as the background.
* simper: added parallel processing for permutation tests in
accordance with other vegan functions and with similar user
interface. The code was developed by Eduard Szöcs in
http://github.com.
* predict.cca: number of rows must match in the original data and
'newdata' of cca() result, because original row weights are used
in scaling the results. Now the match is checked, and
non-matching 'newdata' is ignored with warning. Earlier this gave
an error. Reported by Glenn De'ath.
* betadisper: the method is biased with small, unequal group
sizes. Bias corrected version can now be used with new argument
'bias.adjust' (defaults 'FALSE'). The problem was analysed and
bias correction developed by Adrian Stier and Ben Bolker.
* .checkSelect: standardise those plotting functions that have a
'select' argument that controls which rows of the scores are
plotted. All these functions now use .checkSelect() to check and
apply 'select' as appropriate.
* ordipointlabel: gains argument 'select' which allows some rows
of the plotted scores to be skipped in the same manner as for
text.cca(). This only applies when a single set of scores is
plotted. Otherwise it is ignored and a warning issued.
* ordihull, ordiellipse: defaults to use semitransparent fill
colour with 'draw = "polygon"', and gain argument 'alpha' to set
the transparency.
* ordihull: gained explicit 'col' argument and adds labels after
drawing convex hulls so that filled hulls (with 'draw = "polygon")
do not cover labels. With these changes, the behaviour of
ordihull() is similar to ordiellipse(). The labels are centred
more correctly.
* metaMDS: A warning is issued on too good stress (zero or nearly
zero). This is often a symptom of insufficient data. In general,
you need n > 2*k + 1 points for k dimensions, and Kruskal's advice
is to have n > 4*k + 1. With low number of points there can be
several complete (zero stress) but different results, and no two
convergent solution can be found. The warning is issued also when
convergence was obtained, and information on dimensions is
printed. FAQ gained an entry on the issue.
* bioenv: accepts now dissimilarities or a square matrix that can
interpred as dissimilarities instead of a community data frame.
This allows using other dissimilarities than those in vegdist().
* update.nullmodel: explicit assignment is required to update
the input object (nm <- update(nm, ...)).
* ordiplot3d: the returned envfit.convert() function did not
recognize 'choices'.
Version 2.1-15 (closed May 11, 2012)
* rrarefy, drarefy: check the data are integers -- the functions
do not give sensible results with real values.
* monoMDS: checks now that there is a sufficient number of non-NA
dissimilarities for the analysis. People really try to use NMDS
with too small data sets. The change was triggered by a user who
had tried to find a six-dimensional solution for seven points
(21 dissimilarities, 42 scores) using 100,000 random starts in
metaMDS. With over-defined models there is an infinite number of
different solutions with nearly zero stress, and no convergence is
found.
* ordiplot3d: function returns the projected coordinates of the
origin, and function envfit.convert() that can project a
three-dimensional envfit() result to the current plot. Unlike
originally assumed, the function will not set equal aspect ratio
for all axes. We try to compensate this by setting equal scaling
to all axes.
Version 2.1-14 (opened March 9, 2012)
* Opened a new version with the CRAN release of vegan 2.0-3 on
March 3, 2012.
* Warton, Wright & Wang (Methods Ecol Evol 3, 89-101; 2012) had a
paper where the analysed the confusion of location (differences
between groups) and dispersion (variability within groups) in
dissimilarity-based analyses. We have warned on this for long in
vegan, but now we can get support from this paper. Explicit
arnings added to anosim, mrpp, simper (where this is worst) and
adonis (where this was already analysed by Marti Anderson when
introducing the method).
* adipart, multipart, hiersimu: permutation tests assumed constant
full gamma diversity in all simulations even when the null model
could produce variable gamma diversities. The default method
("r2dtable") had constant gamma diversity.
* adipart, multipart, hiersimu: these are now generic functions
with default and formula methods. The formula method is identical
to the previous function, the default method can take two matrices
as input, but the second argument describing the hierarchy can be
missing. In this case a trivial two-level hierarchy will be
assumed (each row is a seperate group, all rows are in same
group).
* anova of prc() objects by "axis", "terms" or "margin" failed due
to NAMESPACE issues. Reported as issue #7 by Eduard Szöcs in
github.com.
* clamtest: wrongly used frequencies instead of the counts
when calculating sample coverage to test x < coverage.limit.
No detectable differences were produced when rerunning
examples from Chazdon et al. 2011 (Ecology, 92, 1332--1343)
and vegan help page.
* envfit: failed if some of the environmental variables were
factors with unused factor levels. Fixed in centroids.cca.
Reported as issue #8 in github.com by Eduard Szöcs.
* msoplot: expose 'ylim' as an argument. Previously, 'ylim' was
set internally and user could not change it setting, although
y-axis maximum was sometimes so low that standard errors were
outside the plot, and legend covered lines. The default setting of
'ylim' was improved, and the function follows R idiom more
closely. Reported in a private email to J.O. by Ricardo Pita.
* scoverage: new function for sample coverage based correction for
calculation of relative frequencies in count community matrices
(Good 1953, Biometrika 40, 237--264).
Version 2.1-13 (closed March 9, 2012)
* cca/rda/capscale: names of levels could be dropped if a
two-class factor was used as the only constraint. Noticed in an
email of Sascha Kirchner (Univ Helsinki, Finland).
* scores: expand description of the default method to avoid
confusion with specific vegan methods.
* scores.monoMDS: did not know 'choices' and hence plot() was also
unable to choose dimensions.
* vegdist: Added Cao dissimilarity (CYd). Thanks to Yong Cao for
consultation.
Version 2.1-12 (closed February 23, 2012)
* scores.default: failed if users asked scores of non-existing
axes. This was reported as an error in ordiplot() when the user
tried to plot 2-dim graph of 1-dim solution in
https://stat.ethz.ch/pipermail/r-sig-ecology/2012-February/002764.html
* simper: new function to implement "similarity percentages" of
Clarke (Austral. J. Ecol. 18, 117-143; 1993) contributed by Eduard
Szöcs (Uni Landau, Germany).
Version 2.1-11 (closed February 9, 2012)
* indpower: now can handle input objects without dimnames. This
caused problems with oecosimu, because nullmodel objects have no
dimnames to save memory. Extended example on indpower help page
shows the p-value and heterogeneity calculations suggested in
Halme et al. 2009.
* adonis, anosim, mantel, mantel.partial, mrpp, permutest.cca: do
not need clusterEvalQ(parallel, library(vegan)) for socket
clusters.
* adonis: added missing 'mc.cores=' for multicore parallel
processing.
* bioenv: implemented parallel processing.
* metaMDS: implemented parallel processing which runs iterations
(tries) in batches of 'parallel' iterations. However, it seems
that this does not work completely with monoMDS: basic results are
OK, but 'diss' and 'dist' vectors are scrambled (tested in Linux &
R 2.15.0) which is evident if you try to run stressplot() on the
result. The problematic behaviour can also be generated directly
with monoMDS():
mods <- mclapply(1:8, function(i) monoMDS(d), mc.cores=2)
stressplot(mods[[1]])
The 'diss' and 'dist' seem to contain random rubbish as soon as
mc.cores > 1. On the other hand, there are no similar problems
with isoMDS() -- but it does not directly return 'diss' and
'dist'. If this cannot be solved, the parallel processing (r2069)
will be reverted.
* nesteddisc: new argument 'niter' to give the number of
iterations to reorder tied columns.
* renyi.Rd: fixed a broken link reported by Arne Erpenbach (Uni
Frankfurt, Germany).
Version 2.1-10 (closed February 5, 2012)
* adonis: print info that terms are added sequentially -- this
seems to confuse users. Make this change in adonis() instead of
print.adonis to be prepared to add other policies.
* adonis, anosim, mantel, mantel.partial, mrpp: implemented
parallel processing.
* RsquareAdj: implemented adjusted R2 for partial RDA results.
The adjusted R2 of model rda(Y ~ X1 + Condition(X2)) is defined so
that it is the same as component '[a] = X1|X2' in
varpart(). Removed some dead code from RsquareAdj.cca().
* varpart: do not scale constraints to unit sd -- this makes
constant columns (like all zero) into NaN and causes an error in
simpleRDA2. Not scaling may help in problems like that reported
in "[vegan-help][5477] Nested factors in function "varpart"?" by
Katie Shelef on 26 Jan 2012.
* use inconsolata fonts in vignettes.
* added .Rinstignore file to list inst/doc files that should not
be installed (Makefile, tex, bib, sty). Background: R 2.13.0
mandated to put vignettes to their specific vignettes/ directory,
but this is incompatible with R 2.12.* where vegan also should
work. The inst/doc directory can still be used (it isnow
'deprecated'), but system files (such as Makefile) and
intermediate files (such as tex) are silently copied to the
installation. R 2.15.0 (under development) added R CMD check test
for these extra files. File .Rinstignore allows maintaining R
2.12.* compatibility and silences R 2.15.0 tests.
Version 2.1-9 (closed January 22, 2012)
* public launch of parallel processing in vegan. First step was to
explain the implementation in decision-vegan.Rnw.
* DESCRIPTION: vegan suggests 'parallel'. The 'parallel' package
was released with R 2.14.0. If you need to check or use vegan with
older R, you should set environmental variable
_R_CHECK_FORCE_SUGGESTS_=FALSE (see, e.g., discussion
https://stat.ethz.ch/pipermail/r-devel/2011-December/062827.html).
* oecosimu, permutest.cca: new parallel block which honours
setDefaultCluster() in R-devel (becoming R 2.15.0) and
automatically uses parallel processing with socket clusters if
setDefaultCluster was defined. Tested in R (unstable) (2012-01-16
r58122) with full features, in R 2.14.1 without setDefaultCluster,
and in R 2.13.1 (2011-07-08) without parallel processing
* anova.ccabymargin failed if none of the terms was analysed (all
were aliased) and no permutations were performed. This would
happen with, e.g.,
A <- dune.env$Management
anova(rda(dune ~ Management + A, dune.env), by = "margin")
or in general if all marginal effects were aliased.
* capscale: plotting failed for mod <- capscale(dune ~
Condition(Management) + A1, dune.env). The centroids for
Management were completely removed leaving a zero-row matrix of
centroids, and this caused an error in plot() and would probably
fail elsewhere. The same problem can appear with completely
aliased classes, or when class centroids are nearly zero.
* ordistep: handle cases where the marginal effects for
adding/dropping are completely aliased. This should avoid problems
like that reported in
https://stat.ethz.ch/pipermail/r-help/2012-January/300167.html
Version 2.1-8 (closed January 8, 2012)
* betadisper: failed with an error in internal function
betadisper() if there were empty levels. This could happen when
'groups' was a factor with empty levels, and was reported in
https://stat.ethz.ch/pipermail/r-sig-ecology/2011-November/002525.html
The behaviour is now corrected in ordimedian() which will return NA
for empty factor levels.
* nestedbetasor, nestedbetajac: New functions that implement
decomposition of Sorensen and Jaccard beta diversities into
components of turnover and nestedness following Baselga (Global
Ecology and Biogeography 19, 134-143; 2010). These are documented
with nestedness indices and are ready to be used with oecosimu().
* oecosimu: more informative text on 'alternative' hypotheses in
the printed output. Return also the mean of simulations and show
that in the printed output.
* plot.cca: works with degenerate solutions where constraints are
aliased and 'biplot' scores have zero rows. Stops with a
comprehensible error message if a user requests non-existing
scores (such as "bp" scores for unconstrained ordination).
* rarecurve: new function to draw rarefaction curves for each
plot, optionally with sample size vertical and corresponding
horizontal richness lines. Soil microbiologists with sequencing
data seem to want these, and I have seen them used elsewhere as
well.
Version 2.1-7 (closed November 19, 2011)
* adonis: speed up implemented in r1636 was not passing the
transposed matrices to internal f.test function. Reported by
Nicholas Lewin-Koh.
* metaMDS: arguments 'noshare = 0' and 'noshare = FALSE' are now
different: zero is taken as the numeric threshold and always
triggers stepacross(), whereas FALSE is logical and never triggers
stepacross.
* vegan 2.0-2 was released on November 15, based on this version.
Version 2.1-6 (closed November 12, 2011)
* FAQ: correct mark-up of hyper links (@uref{} instead of @url{}),
add new entries on RDA scaling, scaling of NMDS stress and scaling
of environmental arrows in cca/rda/capscale/envfit, plus some
minor updates of old entries.
* ordiarrows, ordisegments: gained argument 'order.by' that can be
used to order the points within groups before drawing the arrows
or segments. This message in R-sig-ecology seems to need this:
https://stat.ethz.ch/pipermail/r-sig-ecology/2011-November/002464.html
* ordispider: returns invisibly the coordinates to which each
point is connected. Triggered by an email query of this
functionality.
* ordiplot3d: expanded example to show how to use xyz.convert() to
add points as per
https://stat.ethz.ch/pipermail/r-help/2011-October/293955.html
* oecosimu: new proposition for implementing parallel processing
following suggestions of Peter Solymos. The only relevant
argument is now 'parallel' which can be either the number of
parallel processes (defaults getOption("mc.cores", 1)) or a
pre-defined socket cluster or NULL in which case it is taken as
the default cluster defined by setDefaultCluster (this last option
only works in R-to-be-2.15.0 unstable, and using NULL is a user
error in R 2.14.0 and hence undocumented). The 'parallel' defaults
to 1 (no parallel processing), but if the user sets the "mc.cores"
option, all parallel processing functions will automatically use
that number of parallel processes. (The "mc.cores" argument is
used by the 'parallel' package, but it is normally unset.) If
'parallel' is a socket cluster or there is a default cluster
('parallel = NULL'), this will be used without setting up and
closing the cluster. This (1) saves time, (2) allows using of
other packages than 'vegan' if user has given command
'clusterEvalQ(library(foo))', and (3) makes unix-like OS
(incl. MacOS X and Linux) to use the socket processing instead of
forking.
* permutest.cca: parallel processing modelled after oecosimu().
Version 2.1-5 (closed October 30, 2011)
* opened a new version with the release of vegan 2.0-1 on Oct 20,
2011.
* metaMDSdist: the default value of "noshare" was inconsistent
with metaMDS(), and therefore stressplot() could fail for
engine="isoMDS". Usually metaMDS() sets the 'noshare' depending on
the engine, but metaMDSdist() is called directly from
capscale(..., metaMDSdist = TRUE) and metaMDSredist (for
stressplot), and now these default to extended dissimilarities.
The problem was reported by Falk Hildebrand
* capscale: could fail if constrained component had zero rank,
typically in partial models where constrained component was
completely aliased. This was observed when checking an R-News
query of October 19, 2011
(https://stat.ethz.ch/pipermail/r-help/2011-October/293077.html)
Version 2.1-4 (opened October 20, 2011)
* adonis, anosim, CCorA, envfit (factorfit, vectorfit), mantel,
mantel.partial, mrpp, protest: user interface changed and
'permutations' can now be a matrix where each row gives permuted
indices. Internally first find a permutation matrix or use the
given permutation matrix, and then find the statistics with single
{ls}apply. Functions adonis and mrpp already worked like this, but
they gained the option of matrix input. This makes the functions
ready both for the 'permute' package and for parallelization
(replace {ls}apply with mclapply, par{SL}apply). Function
envfit() was much simplified by generating a common permutation
matrix in envfit.default() and using that as the input to
vectorfit() and factorfit(). The anova.cca* cases should also be
made to use a single generated permutation matrix, as
permutest.cca() allows this.
Version 2.1-3 (closed October 16, 2011)
* added plot.preston, lines.preston and plot.fisher (that also can
add points and lines). These are similar as plot.prestonfit and
plot.fisherfit, but without the fitted model. Among other things,
they can be used to add alternative models to fisherfit and
prestonfit models.
* sd() function for matrix or data.frame columns was deprecated in
R r57185 (R-to-be-2.15.0). The reason seems to be that users were
confused when median() did not work on data.frames, and as a
solution the R developers decided to take care that mean() or sd()
will not work either (it would be nice to understand how these
people think). Fixed in rda.default, capscale and
simulate.rda/cca/capscale. It seems that this was also implemented
in soon released R 2.14.0 as r57209 | maechler | 2011-10-10
19:28:33 +0300 (Mon, 10 Oct 2011), but as message() instead of a
warning().
* nesteddisc: use only max 200 tries to reorder columns: tracing
showed that in most cases an improved ordering is found rather
quickly, and trying up to 1000 times takes awfully long. Now
faster, and usually as good as earlier, but slacker.
* simulate.rda/cca: implemented 'nsim' or an option to generate an
array of simulated matrices inheriting from "simmat" object and
using print.simmat() for a compact display. If 'nsim = 1', similar
2-dim matrix is returned as before so that cca(simulate(mod))
still works. For 'nsim > 1', the 'indx' argument should have
'nsim' rows, but if 'nsim' is missing, number of rows in 'indx'
will give the number of simulations, and 'indx' can be made with
permute::shuffleSet(). Implemented for rda() and cca() results,
but not for capscale() where simulate returns a "dist" object
which is nasty to pack into an array.
* oecosimu: An attempt to set 'parallel' processing in evaluating
the statistic, and only evaluating the statistic -- the simulation
of null models is not influenced. Both "multicore" (fork) and
"snow" (socket) style parallelization are implemented.
* permutest.cca: implemented 'parallel' processing in
permutest.cca. The parallelization only works in R 2.14.0 (alpha)
and later with the 'parallel' package. Function permutest.cca gets
a new arguments 'parallel' (defaults 1) that gives the number of
parallel process, and 'kind' that selects the parallelization
style which is either "snow" (large overhead, but works in al
OS's) and "multicore" (faster, but only works in unix-like systems
like Linux and MacOS X). The arguments are silently ignored if the
system is not capable of parallel processing. The functionality
cannot be included cleanly: it depends on the package 'parallel',
but suggesting 'parallel' fails R CMD check in the current R
release (2.13.2) which does not yet have 'parallel'. So we get
warnings: 'library' or 'require' "call not declared from:
parallel", and "permutest.cca: no visible global function
definition for ‘mclapply". However, with these warnings,
the function passes tests in R 2.13.2.
* permutest.cca: the user interface changed so that argument
'permutations' can be either the number permutations (like
previosly), or a matrix of permutations like produced by
permute::shuffleSet(). This was done to move RNG outside
parallelized code. This will also allow much simpler and
anova.cca* code. Currently, the 'strata' argument will not work,
but this will be fixed "real soon now".
Version 2.1-2 (opened October 4, 2011)
* permutest.cca could not be update()d, because "permutest.cca"
was not exported from NAMESPACE -- only "permutest" was
exported. Another buglet (and this calls for checking other 'call'
items that return non-exported calls).
* metaMDS did not reset 'tries' when the analysis was started from
'previous.best' of a different model (except when no. of dims 'k'
changed). I think this was a bug(let). Fixed in metaMDSiter.R.
* commsimulator is going to be deprecated: it is no longer used in
oecosimu() ore elsewhere in other functions. Currently, functions
make.commsim(), nullmodel() and simulate.nullmodel() do the same,
and more. As the first step, its documentation in oecosimu.Rd is
moved to vegan-deprecated.Rd.
* examples: cut donw some excessively time consuming examples.
Profiling of all vegan examples showed that 25% of total time was
spent in anova.cca, and 12.6% in ordistep, but they probably are
sufficiently documented more quickly.
Version 2.1-1 (opened September 20, 2011)
* oecosimu: the 'comm' argument can be either 1) community data,
2) a nullmodel object or 3) a simmat object. If 'comm' is a
nullmodel, simulation method is found from the nullmodel object,
and if 'comm' is a simmat object, its matrices are analysed
without simulations within oecosimu(), and different statistics
can be swept out based on the same set of simulated matrices.
* permatfull/swap is using the new simulate.nullmodel(...)
infrastructure. permatfull1 and permatswap1 removed from vegan
devel.
* Value of 'mode' is set to "double" for the following
null model algorithms in make.commsim: abuswap_r, abuswap_c,
r00_samp, r0_samp, c0_samp. These can take any nonnegative real
valued matrix as input. nullmodel function now handles storage
mode reliably.
* meandist bug fix: tapply() function used to find mean group x
group dissimilarities could reorder the class levels and return a
confused matrix. This could happen in particular when the
'grouping' was a vector of integers which then were ordered
alphabetically so that "1" < "10" < "2". Now uses internally more
stable way of applying tapply() which should have the danger of
reordering the levels. Incidentally, this also seems to be
faster. The problem was found by Dr Miguel Alvarez (Univ Bonn).
* nestedness.c: changed interface in "swapcount" and "rswapcount"
which now require integer data matrix. The first argument in .C
call should now be defined as as.integer() instead of old
as.double().
* str.nullmodel: new function to display the *str*ucture of the
"nullmodel". The "nullmodel" is an environment and therefore does
not show in usual str() although its items can be accessed with
$-notation. No documentation, but alias in nullmodel.Rd.
* New functions: commsim is used to define Null Model Algorithms
via a function that returns n x m x nsim array of simulated
matrices based on structural constraints. make.commsim contains
Null Model Algorithms already defined in vegan from commsimulator
and permat* functions (and some more). The nullmodel function
creates an environment, where statistics of the input matrix are
stored. The environment also stores updated status of sequential
algorithms and current number of iterations. The update and
simulate methods are used to update the nullmodel (for sequential
algorithms) or simulate random matrices, respectively. The
simulate method returns the n x m x nsim array (simmat class).
Efficiency gains are sometimes high (because marginal statistics
are calculate only once by nullmodel), but not significant in most
cases. Most advantageously, this implementation can unite the
commsimulator and permat* branches and can serve as basis for
further extensions. Current intent is to investigate how this low
level infrastructure can be used within oecosimu and permat*
functions without breaking current vegan functionality.
Version 2.1-0 (closed September 20, 2011)
* New major version opened with the release of vegan_2.0-0 on
September 8, 2011.
* nestedness.c: isDiag* uses now switch(sX) where sX is the number
of non-empty cells. The "swapcount" method uses new isDiagFill
which finds the largest swappable element that does not change the
fill, and "rswapcount" uses isDiag which finds both the largest
swappable element the change in fill. The swap and trialswap also
find first the fill of the 2x2 submatrix, and continue only if
fill == 2. The measurable effects are small again (perhaps 1%).
* tests: added tests for commsimulator, permatswap1 and
permatfull1 before starting the adventures with nestedness.c.
* commsimulator: a bit less overhead -- all attributes set
simultaneously instead of setting separately dim, rownames and
colnames. The results should be identical(), running should be
marginally faster or neutral.
* permatswap: there was a bug in internal C routine, and therefore
not all permissible swaps were performed. Based on limited
analysis, the effects of this bug seem to be negligible. It is,
however, recommended to re-run all analyses. The C code was made
faster by getting quickly out from isDiag* if there are only 0 or
1 filled items, because there is nothing to swap. Tests show that
the C code indeed is ca 10% faster in permatswap(BCI,
method="swa", thin=1000, times=999), but only a 20-25% of time was
spent in C, and the new permatswap/permatswap1 spends 2.2x longer
in other parts. With faster C code the net slowdown is 1.7x.
* various attemps of speed-up (often in vain): The speed-up
started with permatfull/permatswap which appeared to be by far the
slowest functions in R CMD check --timings. Later this proved to
be a feature of checking these functions in MacBook Air/MacOS X:
the functions were not slow in Linux, nor in MacBook when the very
same tests were run outside R CMD check. However, several changes
were made:
- permatfull/permatswap avoid data.frame/matrix casting, and save
results in matrices. This was based on misleading test statistics
in MacOS, and it is not sure what are the real effects.
- commsimulator: smaller overhead.
- sample.int was used in place of sample in commsimulator,
nesteddisc, permuted.index and poolaccum (the effects may be
measurable for permuted.index, but not large).
- centroids.cca: much faster. This was perhaps real -- profiling
(not timing) showed that cca.formula/rda.formula have a large
overhead over cca.default/rda.default, and centroids.cca was
responsible for a lot of that -- most is due to ordiParseFormula.
- permutest.cca uses La.svd, but the effects are non-measurable
and only concern case first = TRUE.
- MDSrotate example: envfit does not do permutations.
* New functions: permatfull1 and permatswap1. Both functions
return a single permuted matrix. These functions are now called
repeatedly by the corresponding permatfull and permatswap
functions.
* New function: clamtest (with summary and plot methods). The
method uses a multinomial model based on estimated species
relative abundance in two habitats, it minimizes bias due to
differences in sampling intensities between two habitat types as
well as bias due to insufficient sampling within each habitat. The
method permits a robust statistical classification of habitat
specialists and generalists, without excluding rare species a
priori. Based on Chazdon et al. 2011 (Ecology, 92, 1332--1343).
* raupcrick: new function to implement Raup-Crick (dissimilarity/
probability) index with unequal sampling probabilities of species.
Brian Inouye informed about their paper (Chase et al., Exosphere
2:art24 [doi:10.1890/ES10-00117.1]; 2011) where they showed that
Raup & Crick said that we should use sampling probabilities
proportional to species frequencies in assessing their index, but
vegdist(x, "raup") uses equal probabilities. Unequal sampling
probabilities cannot be directly implemented in vegan, but the
Chase et al. method can be implemented as oecosimu(x, function(x)
designdist(x, "J"), method="r1"). Basically, the current function
uses this, but with boosted code that is much faster than
designdist().
Version 2.0-0 (released September 8, 2011)
* opened the release candidate of vegan_2.0-0 on September 3,
2011.
* some old functions used attributes(x)$which instead of more
correct attr(x, "which"), and in addition postMDS() used
attributes(x)$names instead of names(). Concerns anosim(),
initMDS(), postMDS(), mantel(), mantel.partial() and
mso(). initMDS() was one of the two functions that was unchanged
since the first vegan release on 6/9/01, and it was changed day
before it turned ten-years-old. After this, wisconsin() is the
only unchanged function from the first release.
* vegan-defunct: put definitions of removed "new" permutation
functions to vegan-defunct and have a vegan-defunct.Rd for them.
Version 1.92-2 (closed September 3, 2011)
* monoMDS: saves and displays info on the dissimilarities used,
and displays the number of points. Checks that input is
dissimilarities, and warns on negative ones.
* MDSrotate: metaMDSrotate() was renamed to MDSrotate(), and
metaMDSrotate() was .Deprecated.
* meandist: warns if some dissimilarities are negative (small
negative values are tolerated).
* vegandocs: copied branches/1.17/inst/NEWS as ONEWS (old news),
and adapted vegandocs. Expanded documentation of vegandocs().
Version 1.92-1 (closed August 28, 2011)
* adonis, betadisper, mrpp: check that dissimilarities are
non-negative (small negative values are tolerated). The test was
not added to meandist: there may be legitimate usages for
correlation-like objects although summary.meandist() is
invalid. The addition was triggered by a user who tried to analyse
Chase et al. Raup-Crick values in the range -1 .. +1 in
betadisper().
* anosim: warn if some dissimilarities are negative. Function can
handle negative dissimilarities because it is based on ranks, but
probably this is a user error (but can be intentional and
correct).
* metaMDSrotate: can rotate monoMDS() results.
* monoMDS: gained argument 'pc' (defaults TRUE) to rotate the
final scores to principal components.
* specaccum, fitspecaccum: gained predict() methods. With
specaccum() predict(..., newdata) can estimate values for
'newdata' with linear or spline interpolation if necessary.
Without 'newdata' the function returns the the fitted expected
richness. With fitspecaccum(), the predict.nls() is performed to
each fitted model. Function fitspecaccum() is written so that
default fitted(), residuals() and coef() work.
* specpool: failed if 'pool' had NA values or empty factor
levels. Checks that 'pool' length matches nrow of data.
* NAMESPACE: imports nobs from permute_0.5-3 in R 2.12.2 and older
which do not yet have nobs() generic. Do not export orderingKM &
ordimedian. Remove aliases to non-exported and non-documented
print and print.summary methods in Rd files (but leave documented
ones).
* vegandocs: shows the permutation vignette of the 'permute'
package.
Version 1.92-0 (closed August 23, 2011)
* open new version with the release of vegan_1.17-12 on August 17,
2011.
* fitspecaccum: a typo and a condition was never true.
* monoMDS & metaMDS: always print types of stress and ties.
* NAMESPACE: export pasteCall for analogue.
* NEWS: added file NEWS.Rd for vegan release 2.0 news in Rd format.
Keeping neat formatting of this file may be challenging to many
text editors, but it is worth trying. This file should list the
important items for regular vegan users, whereas this ChangeLog
should explain the changes to other vegan developers. If
getOption("help_type") == "html", should start a browser to show
the NEWS, and use text console otherwise ("html" tested on MacOS
R.App GUI, text mode in Linux and MacOS bash shells, no testing on
Windows yet). Although the browser knows how to display NEWS.Rd,
it does not format the page properly (\code{foo} shows as ordinary
text etc). As a kluge we now make a NEWS.html file in the inst/doc
directory. NEWS.html cannot be made in the inst/ directory since
browser only agrees to show NEWS (plain) in the main directory or
files in doc directory.
Version 1.91-0 (closed August 18, 2011)
* NAMESPACE: added namespace to vegan. Currently hides all that
can be hidden, and exports only main functions: Some method
functions will be probably exported later. Superfluous aliases
were removed with the registration of S3 methods, and scoping was
changed a bit in anova.ccabyterm, model.matrix.cca and
model.frame.cca. as.mcmc is not yet registred: it need changes in
Imports field of DESCRIPTION (same with vif.cca: vif is defined in
car, but we could have our private vif generic here?).
* Vignettes: use now jss.cls shipped with R instead of amsart.cls
for better consistency with R and permute style.
* plot.envfit: When `add == FALSE`, make sure we leave enough room
around the plotting locations to contain the arrow and centroid
labels.
* plot.isomap: will always plot points or text above net so that
they are visible. Gained argument 'type'.
* ordiplot: expose arg 'cex' so that user can change its value.
Age shows in this function, and it is badly in need of complete
re-writing to improve its flexibility.
Version 1.90-2 (closed August 6, 2011)
* ordilabel: gained argument 'select'.
* orditorp: works with NA scores.
* make.cepnames: ignores duplicated and trailing dots. It is
assumed that the names to be changed came via R make.names()
mechanism which changes all blanks and non-accepted characters to
dots, and this may give extra dots in names. Gained argument to
take the second item of the name instead of the last which may
help in names like "Vaccinium vitis-idaea" (which becomes
"Vaccinium.vitis.idaea" -> "Vaccidae") or names which have authors
or other rubbish at the end.
Version 1.90-1 (closed July 23, 2011)
* cca/rda/capscale support functions: cca/rda/capscale were
changed to return zero components instead of NULL for completely
aliased constraints (CCA) or saturated models (CA), but several
support functions still checked for is.null() only, although they
also should check if rank == 0. Some functions failed, and these
are fixed here: anova.cca, bstick.cca, goodness.cca/rda,
predict.cca/rda, screeplot.cca. The following functions did not
fail, but were fixed: anova.ccanull, calibrate.cca, ordirgl,
ordiresids, ordiplot3d, deviance.cca/rda (simplified),
ordixyplot.
* swan: gained argument 'maxit' that can be used to restrict the
number of beals() passes on zeros. The default is 'maxit = Inf'
which is the old behaviour: beals() passes on zero entries
continue until there are no zeros or the number of zeros does not
change. Probably only 'maxit = 1' makes any sense (in addition to
the default 'maxit = Inf'), but all values are allowed.
* metaMDSrotate: results could be a bit off with NA in
environmental vector and >2D solution. The starting scores must be
orthogonal, but the initial orthogonalizing PCA was run without
removing NA cases.
* ordiellipse, ordihull, ordispider: will remove NA scores that
can appear with na.action in cca/rda/capscale (and perhaps in
other cases).
* betadiver: argument 'index' was renamed to 'method' for
compatibility with dist() objects. Argument 'index' is
still recognized with a warning.
* metaMDS: Documentation and interface adapted to monoMDS().
Argument 'noshare' defaults to FALSE with monoMDS(), since tie
breaking seem to handle tied maximum dissimilarities even better
than stepacross(). For 'engine = "monoMDS"' 'noshare' defaults to
TRUE. The 'noshare' argument now accepts logical arguments in
addition to old numeric (FALSE worked previously because it was
interpreted as numeric 0, but TRUE has now meaningful
interpretation). Remove argument 'old.wa' introduced in rev280 on
24/3/08 for backward compatibility when calculating WA scores
started to use the same transformed data as analysed.
* ordiplot: Now handles non-par arguments to plot.default without
warnings. Things like `axes = FALSE` in the call now just work!
Version 1.90-0 (closed July 3, 2011)
* vegan now depends on the permute package for new permutation
designs. This initially version strips out all the old "new"
permutation code from the R and Rd files, and patches up
permutest.betadisper to use shuffle() to generate permutations.
The 1.90-0 devel series will lead the way to vegan 2.0-0. There
may well be build problems as we iron out permute / vegan
meshing.
* capscale: could not be embedded in another function after fixing
the update() in 1.18-31 -- problems were anticipated in ChangeLog,
and they were realized.
* monoMDS: text for a proper help (Rd file). Other help files
updated for monoMDS. Stress type 2 is printed both in monoMDS and
metaMDS. stressplot() prints now R2 for monoMDS() results, and
docs tell that "linear fit" of stressplot() is related to "stress
2".
Version 1.18-33 (closed June 21, 2011)
* adonis: made faster (much faster in tests) by polishing the
f.test() in the innermost loop.
Version 1.18-32 (closed June 14, 2011)
* closed with the release of vegan_1.17-11.
* anova.prc: new function that casts prc() result to a
corresponding rda() and submits this to anova.cca(). With this is
it is possible to analyse anova(..., by = "axis"). Other cases are
possible as well, but not very useful as "margin" and "term" only
analyse all constraints as a single matrix. Now some functions
that were made refuse prc() results in 1.18-30 work again, and
anova.ccabyaxis() even makes sense. This fixes a problem Jae
Pasari raised in his email.
* vegdist docs: Brian Inouye informed us that we are not
calculating the Raup-Crick index like originally suggested: we use
equal probabilities for all species, but the species probabilities
should vary with their frequencies. Doing this would break
vegdist.c design, and require Fisher's non-central hypergeometric
distribution, and that is not available in base R. However, it
seems that a simulation version could be implemented in
oecosimu(). Now we document the problems in vegdist.Rd, but do not
correct the behaviour. The key paper is available over
http://www.esajournals.org/doi/full/10.1890/ES10-00117.1
* rda & capscale: similar changes as in cca in r1517 (version
1.18-24). CA component is always returned, even with zero rank,
and CCA and pCCA components are NULL only if they were not
specified originally, and if they become zero rank because of
aliasing etc, they are returned as zero components (instead of
NULL).
* print of cca/rda/capscale: do not "Proportion" column in
unconstrained ordination which only has one component. However,
this will be printed also with zero pCCA or CCA (but not with
NULL).
Version 1.18-31 (closed May 30, 2011)
* ordiR2step: more informative about rejecting 'scope': partial
models are not (currently) accepted due to the design of
RsquareAdj.rda().
* specnumber: gained argument 'groups' to find total richness
within each group. An example shows how to use this to find one
index that Whittaker suggested to estimate beta diversity.
* capscale: species scores will be NA if input data were
dissimilarities and no 'comm' was given. update(capscale_result)
failed because LHS data were evaluated in the
environment(formula). Now removed the explicit environment: I ran
some tests and found no problems, but let's see what the real life
brings along. The changed point is commented so that this change
can be reconsidered if necessary.
* oecosimu: reverted r1557 of version 1.18-26, or splitting tied
values. This was introduced because Chase et al. (Ecosphere 2011,
http://www.esajournals.org/doi/pdf/10.1890/ES10-00117.1) used this
in their Raup-Crick index. This is against established practice in
order statistics and therefore we cancelled this.
* specaccum: reverted r1505 & 1507, or introduction of weighted
species accumulation model in version 1.18-23. The method was
introduced because of user request, but these persons lost their
interest and therefore there is no reason to see the trouble that
finalizing these functions would require (see comments on version
1.18-23 about some of the work needed, and there is more).
Version 1.18-30 (closed May 26, 2011)
* opened with the release of vegan 1.17-10 on April 27, 2011.
* nobs: documented nobs() methods for vegan. These seem to be
needed in R 2.13.0 and should be released soon (there is no need
to add R >= 2.13.0 dependence as long as we do not use these
functions ourselves within released vegan code).
* treeheight: dramatic simplification and speed-up of the code.
* scores.cca/rda: failed if user requested only non-existing
scores. Now return an empty list of scores instead (with possible
attributes).
* prc: several support methods did not work with 'prc' results,
and these now stop with an informative error instead of an
uniformative error. Concerns anova.ccabymargin, anova.ccabyaxis,
drop1.cca, add1.cca, model.frame.cca, model.matrix.cca. This was
found out after a private user query about having signficance test
for the sencond PRC axis.
Version 1.18-29 (closed April 27, 2011)
* capscale: could fail with add = TRUE in R 2.13.0. A kluge was
necessary in cmdscale() pre 2.13.0, but the feature was fixed in R
2.13.0 and unnecessary fix failed. Now the kluge is used only in
older versions of R and capscale should be working both in 2.13.0
and older R.
* adipart.Rd, multipart.Rd: cancelled some excessive speed-ups in
example code made in r1204. Some of these speed-ups never made to
the release version, but were only made in the devel branch. The
sped-up examples were also merged to the release (adipart.Rd,
kendall.global.Rd, multipart.Rd, mantel.correlog.Rd,
pertmatfull.Rd, radfit.Rd). Also reverted r1224 so that
Guillaume's original r1206 (49 permutations instead of 99) is used
again. Probably we should speed up some other examples. The five
worst cases (timing in MacBook Air) are: ordistep (19.0 sec),
add1.cca (7.2 sec), tsallis (7.0 sec), prc (5.6 sec),
vegan-package (5.2 sec).
* vegan-package.Rd, ordistep.Rd: speed up examples.
Version 1.18-28 (closed April 11, 2011)
* Peter Minchin joined the vegan team.
* monoMDS: a new function with Peter Minchin's Fortran90 code for
NMDS. The full Fortran90 code has global, local, linear and hybrid
MDS, configurable and valid tie treatment, handles missing values,
and allows adding new points to existing ordinations. The Fortran
code is highly tuned, and much faster than other alternatives in
R. The R interfaces allows access to most features of the Fortan90
code, except adding new points to the ordination (which will need
a separate function). The missing values in dissimilarities are
also handled in the R interface. The function will eventually
replace isoMDS() of the MASS package as the main NMDS engine in
metaMDS().
* metaMDS: uses monoMDS() as its default engine (and gained
argument 'engine' to select either monoMDS() or isoMDS()).
* stressplot: is generic with a specific method for "monoMDS"
objects, and the old function as the default method.
* goodness: got a new method for "monoMDS" objects. The function
handles all monoMDS() models, but does not take into account
possible missing values in input dissimilarities.
* metaMDSrotate: gained argument 'na.rm' (defaults FALSE) to
remove missing data from the continuous vector.
Version 1.18-27 (closed April 1, 2011)
* orderingKM: Kurt Hornik found a problem when inspecting the
Fortran subroutines with gcc 4.6 tools when checking vegan release
1.17-9.
Version 1.18-26 (closed March 31, 2011)
* version closed with the release of vegan 1.17-9.
* oecosimu: added argument 'tiesplit' (defaults FALSE). If TRUE,
simulated values tied with the observed statistic are split so
that half of them are regarded as satisfying the condition, and
removed from those cases which are greater or less (as given by
'alternative'). This is usually *not* done, but implementing
Raup-Crick dissimilarity in oecosimu() following
http://www.esajournals.org/doi/pdf/10.1890/ES10-00117.1 (Chase et
al., Ecosphere 2011) uses this option. However, cross-checking
the "r0" null model against the vegdist(..., "raup") indicates
that the options should be set FALSE (and this would mean that the
change is reverted).
* anova.ccabyaxis: I (JO) noticed in an ORDNEWS message that
anova(..., by = "axis") ignores the original Conditions in partial
models and so gives wrong results in partial models. The fix finds
the conditioning matrix as qr.X(object$pCCA$QR), and uses this as
a Condition in updated models. Added a regression test that checks
that statistics and residual df match.
* tolerance: new function to compute species tolerances and sample
heterogeneities as Canoco does. Includes a method for objects of
class "cca".
* plot.contribdiv: plot now draws axes and box *after* doing the
drawing - stops polygons overplotting the axis and box.
Version 1.18-25 (closed March 23, 2011)
* ordilabel: gained argument 'xpd' to draw labels outside the plot
region.
* ordisurf: got a formula interface as an alternative to define
the model. Also now accepts `gam()` argument `select` to add an
extra penalty to smooths so they can be penalized to zero (i.e.
removed from the model). This is a form of model selection.
Estimation of smoothing parameters can now be determined via
argument `method`. Both `select` and `method` have been set to
defaults that were implied in previous versions of the function.
If smooth is penalized to (effectively) 0, countour plotting is
suppressed. New argument `gamma` can be used to increase the cost
of each degree of freedom used by a model in the GCV or UBRE/AIC
scores used in smoothness selection. New argument `plot` can be
used to stop ordisurf producing any graphical output.
A plot method is now provided that can draw contour or perspective
plots of ordisurf results, or plot the fitted GAM surface.
* metaMDSrotate: can now rotate > 2-dim solutions so that the
first axis is parallel to a given vector and all others are
orthogonal to the vector (and rotated to PC if they were
originally). Earlier had argument 'choices' which was supposed to
select only two axes to be rotated (but failed before rev 1533),
but now lost 'choices' and always uses all dimensions, and
orthogonalies dim 2 and beyond against the given vector. There was
a query in R-sig-ecology on having two vectors in 3-dim ordination
which triggered the current changes, but this does not still
answer to the question of
https://stat.ethz.ch/pipermail/r-sig-ecology/2011-March/001947.html.
It would be possible to have this with recursive metaMDSrotate for
1..N for first 'vec', then 2..N for the second 'vec', but the
later 'vec's could be correlated with previous axes.
* metaMDS: 'previous.best' can now be a user-supplied
configuration, or a result of isoMDS(). If 'previous.best' is an
isoMDS() or a metaMDS() result, it can have different number of
dimensions than the current 'k': extra dimensions are dropped off,
or noisy extra dimensions are added as needed. This may help in
running metaMDS() with a range of 'k'. Not yet documented.
Version 1.18-24 (closed March 10, 2011)
* cca: will return NULL item for CCA only if constraints were not
given. If the CCA component has zero rank (= constraints were
aliased or were orthogonal to the data), will still return a CCA
item with zero eigenvalue, rank, scores etc plus info on
"alias". The residual component will never be NULL, but similar
zero-containing component is returned even when there is no
residual variation. Usually these changes only manifest in
arbitrary data sets, but we have had email queries in R-sig-eco on
completely aliased constrainsts when users assumed that model
cca(y ~ A + Condition(A)) would be similar to a random effects
model and A could be analysed both as a Condition and as a
Constraint instead of being aliased. Naturally, overfitted models
with empty residual components do appear often. Currently the
printed output includes these zero items with Inertia and Rank 0,
whereas NULL components are not displayed. The changes so far only
concern cca(); rda() and capscale() are only changed after (and
if) this change appears sensible and correct in cca().
* drarefy: a new function to return the probabilities that a
species occurs in a rarefied sample (= sampled without
replacement) of a given size, (or given sizes specified
separately for each row if 'sample' is a vector)..
* diversity, rarefy, rrarefy, specnumber: accept vector input. All
functions documented in ?diversity now accept vector input.
Function fisher.alpha and drararefy already accepted vectors.
Version 1.18-23 (closed February 28, 2011)
* opened with the release of 1.17-7 based on version 1.18-22 on
February 16, 2011.
* does not use unnecessary return() plus other stylistic
twitches. This touches 24 functions, but users should see no
difference.
* procrustes: checks and reports different number of rows instead
of failing in crossprod() with incompatible arguments.
* prestonfit, as.preston: default now 'tiesplit = TRUE'.
* as.preston (prestonfit, prestondistr): uppermost octave could be
NA with 'tiesplit = TRUE' (a bug).
* decorana: Cajo ter Braak notified about false expansion of
computed zero eigenvalues in rescaling which then are used in
estimating eigenvalues. They are now zeroed. Cajo's example was a
6x5 petrie matrix:
petrie <- matrix(0, 6, 5)
diag(petrie) <- 1
petrie[row(petrie) - 1 == col(petrie)] <- 1
decorana(petrie)
Gavin Simpson had another failing case:
dummy <- matrix(c(1,3,1,0,0,0,0,0,0,
0,0,0,1,3,1,0,0,0,
0,0,0,0,0,0,1,3,1), nrow = 3, byrow = TRUE)
which also failed in orthogonal CA ('ira = 1'). Now the zeroing is
done after estimating each axis, and the results are similar as in
Canoco. However, the results of orthogonal CA with 'dummy' are
wrong in both: there should be two eigenvalues of 1, but only one
is reported. This is a fundamental problem in the
orthogonalization algorithm in the Fortran code presumably shared
between vegan:decorana and Canoco.
* specaccum: first step in implementing weights on
method="random". Not yet documented nor supported. Issues that
must be solved are plotting (all lines? summaries?), summarizing
(use approx to get average predictions?) and the very natural link
to fitspecaccum models.
* nobs: R 2.13.0 will introduce new generic nobs() that will
return the number of observations. This is at least needed in
functions to evaluate BIC and in step() and functions it
calls. Implemented nobs() for several vegan classes -- some of
these may be unnecessary. Not yet documented but aliases in
vegan-internal.Rd.
* FAQ updates: using 'select' in ordination text() and points(),
using 'xlim' to flip axes or zoom into ordination plot(), the
non-effects of 'strata' in adonis(). All these indeed are
frequently asked.
* scores.default: clarified documentation (PR#1300).
* vegdist: rephrased Chao index following PL.
Version 1.18-22 (closed February 16, 2011)
* nestednodf: did not go to 'weighted = FALSE' with quantitative
data if 'order = FALSE'. Reported by Daniel Spitale.
* eigenvals: learnt to know 'dudi' objects of the ade4 package,
and 'pco' an 'pca' obejct of the labdsv package.
* betadisper: streamlined code somewhat.
* prestonfit: implemented splitting "tied" counts (1, 2, 4, 8 etc)
between octaves following Williamson & Gaston (J Anim Ecol 43,
381-399; 2005) with argument 'tiesplit = TRUE'
* specaccum: fixed typo -- 'individuals' instead of
'invidividuals'. plot() gained argument to use "individuals" as
x-axis instead of "sites" with method = "rarefaction".
Version 1.18-21 (closed January 19, 2011)
* Fixes bug report #945 at R-Forge: anova.cca should now work with
subset, missing data, functions of constraints (like log(x)),
constraints or subset variables in data=, attached data frames or
in working environments.
* capscale: return additive constant ('ac') when 'add = TRUE'.
* fitted.capscale, predict.rda: additive constant ('ac')
subtracted from the estimated distances when 'add = TRUE' was used
in capscale().
* predict.rda: type = "working" for capscale returns now scaled
scores that produce the dissimilarities as Euclidean distances --
like was claimed in the help page.
* simulate.capscale: implemented simulate() for capscale. This
uses internal metric scaling mapping to simulate data and add
error, but returns Euclidean distances adjusted to the original
scale of input dissimilarities. The function uses only the real
axes, and imaginary dimensions are ignored.
* anova.cca: by = "term" and by = "axis" work now both with the
data= statement and with variables in the GlobalEnv and subset=
both in cca/rda. It seems that 'subset' is not implemented for
capscale(), but otherwise models work.
* cca/rda: added item 'subset' that is the subset argument
evaluated to a logical vector.
* capscale: implemented 'subset' (and this also works with
anova.cca). It looks like all anova.cca tests pass also with
capscale (but more thorough test will be run before closeing the
anova.cca bug in vegan-forge). It seems that subset and na.action
also work.
* anova.cca: most cases work now when terms are given in
attach()ed data frame after changes to anova.ccabyterm() and
ordiParseFormula().
* anova.cca: works (mostly) when there are missing values. The
exception found was 'by = "term"' when the data frame is not given
as a 'data=' argument, but as an attached data.frame. In this case
'by = "term"' detects a change in the number of rows and stops
with a message.
Version 1.18-20 (closed January 9, 2011)
* base of vegan release 1.17-6 (released on Jan 10, 2011).
* capscale: the total inertia was defined as the sum of absolute
values of inertia so that negative eigenvalues of imaginary
dimensions increase inertia (changes in 1.18-17 and release
1.17-5). Torsten Hauffe reported this as a bug in vegan
open-discussion forum in R-Forge on Dec 22, 2010. I (JO) answered
on Jan 1, 2011 and justified the change, and said this is a design
decision, and not a bug. Now I have reconsidered this design
decision, and it really seems to me that it was wrong. Now the
negative eigenvalues again reduce the total inertia, and are
similar as in vegan 1.17-4 and earlier. The capscale() output got
a new item of "Real Total" that gives the sum of eigenvalues
without the imaginary component, and the proportions are given
only within this real component (if there are negative
eigenvalues). The eigenvalues of negative eigenvalues are not
added to the result, because they are not strictly for the
unconstrained component only like previously indicated. However,
summary.eigenvals() estimates the proportions explained using
absolute values of eigenvalues since this seems to make sense
there (the proportion explained is now based on absolute
eigenvalues to give more sensible output).
Version 1.18-19 (closed January 3, 2011)
* new version opened with the release of vegan 1.17-5 (submitted
Dec 17, 2010, released at CRAN on on Dec 18, 2010).
* CCorA: Pierre Legendre's minor fixes, including scaling of
scores Cy an Cx, and removing requirement of MASS.
* wascores: checks input -- weights (species abundances) must be
non-negative and not all zero.
* metaMDS: function was written for community data with all
entries non-negative, but it is used increasingly often with other
data types with a risk of failure (especially in wascores()). Now
data are checked for negative entries, and arguments wascores,
autotransform and noshare are set to FALSE with negative data
entries. Triggered by email queries and problem reports.
Version 1.18-18 (closed December 17, 2010)
* mantel.correlog: proper treatment of ties in permutation
statistics. Function uses mantel() to evaluate p-values, and this
uses one-sided test. If the statistic is negative, the test
direction is reversed, but then p-value cannot be directly
complemented if there are ties: the observed statistic must be
switchec to another side of the tied block. Therefore
mantel.correlog() now evaluates the p-value directly in reversed
test.
* nestedtemp: updated the literature reference to Almeida-Neto &
Ulrich (2011).
* vegdist.Rd: fixed legends to Gower indices and checked binary
formualations (ok).
* vegan-package.Rd: added a package intro file (to be merged to
the release version 1.17-5).
* sipoo.rda: got island names, and decision-vegan.Rnw vignette was
changed accordingly.
* ordistep: Pierre Legendre's edits of the help page.
* prc: got back the formula interface and with that the na.action
lost in r1356 (9 Nov, 2010).
* plot.procrustes: user can select the direction of the arrow with
argument 'to.target', and the 'type' of the plot. With 'type =
"text"', ordilabel() is used to mark the tails of the arrows.
* plot.betadisper: would fail when plotting anything other than
axes 1 and 2. Now fixed and an example included in the Rd file to
make sure this doesn't happen again. Reported by Sarah Goslee.
Version 1.18-17 (closed December 1, 2010)
* predict.procrustes: gained 'truemean' so that new points can be
added to the plot.procrustes() graph.
* alpha release version: preparing release (1.17-5) of
vegan. Mainly clean up and checking of functions intended for the
release.
* mantel.correlog: vegan mantel() was used standard order
statistics or (hits+1)/(nperm+1) since May 2009, but
mantel.correlog() did not notice the change and still tried to
correct the mantel() result.
* as.mlm.cca/rda and intersector: should get the pivoting of names
in qr.X() correct both in the current buggy R and hopefully fixed
next version 2.12.1.
* vif.cca: aliased terms must be removed when calculating
VIFs. Test case vif.cca(cca(dune ~ Manure + Management, dune.env))
should give decent values all below <100 (used to give magnitude
10^31). VIF of aliased terms is NA. Pierre Legendre noticed this.
* print.eigenvals uses zapsmall.
Version 1.18-16 (closed November 17, 2010)
* procrustes: fixed centring of fitted(). Added predict() method
that can be used to add new rotated coordinates to the target. The
function cannot be used for symmetric analysis with 'newdata'.
* vignette on design decision: updated to changes in 'const' in
scores.rda() in 1.18-15 and to Canoco 4. Explains now 'const'
more thoroughly.
* pcnm: gained argument 'dist.ret' to return the distance matrix
on which PCNMs were based.
* cca/rda/capscale: JO yielded to the pressure, and added
proportions of inertia components in the printed output.
* capscale: failed with na.action -- fix in ordiParseFormula().
na.action was introduced in rev1002 (Sept 11, 2009), and then
ruined with a change in ordiParseFormula() in rev1007 (Sept 15,
2009) so it lived four days and was broken when released. The
problem was reported by Nevil Amos.
* SSgleason: a new SS-species-area-model.
* SSlomolino: improved starting values for 'xmid' (and
'Asym'). Now fitspecaccum(..., "lomolino") works in several cases,
including BCI and bryceveg (but fails in <1% of cases). Now 'Asym'
and 'xmid' are estimated from Arrhenius at log-log scale: 'Asym'
is the predicted value at max(x), and 'xmid' is the value of x
giving 'Asym/2'.
* capscale: defines total inertia as the sum of absolute values of
eigenvalues to be consistent with cmdscale(), wcmdscale(),
eigenvals(), Gower and Mardia, Kent & Bibby.
* eigenvals: added a method for "pcnm" and docs discuss handling
negative eigenvalues in summary.eigenvals(). Function is ready for
the new version of cmdscale() in R 2.12.1.
* prc: Cajo ter Braak wrote new code for more direct
implementation of PRC. This drops formula interface and directly
finds model matrices avoiding aliased terms.
Version 1.18-15 (closed November 9, 2010)
* procrustes: 'translation' needs to take into account 'scale',
although Mardia et al. omit it. Reported, analysed and fix
suggested by Christian Dudel (Bochum).
* fitspecaccum: uses now new SS-models for species-area
relationship and several of standard SS-models for
nls(). Placeholders for documentation added. Has a plot()
function.
* Self-starting nls() models for species-area relationship: added
Gitay and Lomolino models from Dengler 2009 (J Biogeog 36,
728-744). Some more should be added, but these were among the best
in Dengler's simulation. The starting values of the Lomolino model
(SSlomolino) need work, as do several other candidate models. The
purpose is to use these also in fitspecaccum().
* rda: the scaling 'const' can now be a vector of two items: the
first is used for species and the second for sites. This allows
compatibility with Canoco 3.x (but Canoco 4 changed scaling): if
'mod' is an rda() result, and 'nr' is the number of rows in data,
then "scaling = -2, const = c(sqrt(nr-1), sqrt(nr))" in scores(),
summary() or plot() will use default scaling of Canoco 3.x. Help
page now points to decision vignette explaining scaling and
'const' (and the vignette needs updating).
* prc: coefficients folded in a wrong way. Diagnosed, reported and
fixed by Cajo ter Braak. CtB also fixed the scaling of species
scores and coefficients to be more consistent with Canoco and
common standards. The function now ignores user settings of
contrasts and will always use treatment contrasts. Cajo ter Braak
is now recognized as a co-author of the function.
* nestednodf: breaks ties in column and row totals by row sums and
column sums in quantitative data ('weighted = TRUE'). Version
submitted by Gustavo Carvalho in a follow-up to Feature Request
#1097.
Version 1.18-14 (closed November 1, 2010)
* treedist: works now with zero-height trees (for instance, only
one item in a tree). Function gained a new argument 'relative'
(defaults TRUE): if FALSE, finds raw dissimilarities of tree
heights. Help page now tells that relative tree distances are in
range 0..2 instead of 0..1, since combining two trees may add a
new common root.
* isomap: returns only axes associated with positive eigenvalues.
* wcmdscale: give two goodness of fit statistics for 'k' axes: one
is for the real part and based on the sum of positive eigenvalues,
and second is for all axes and based for the sum of absolute
values of all eigenvalues.
* summary.eigenvals: handle now possible negative eigenvalues from
capscale() or wcmdscale(): cumulative sum and proportions are now
based on the sum of absolute values of eigenvalues, and all real
eigenvalues explain less than 100% if there are negative
eigenvalues.
Version 1.18-13 (closed October 15, 2010)
* anova.cca(..., by = "axis"): now really uses marginal tests for
every axis. Significance of axis k with LC scores lc is from model
update(object, . ~ lc[,k] + Condition(lc[,-k])). That is, axis k
is used as constraint and all other axes as conditions, when axis
k is analysed. This also solves the failure of anova(..., by =
"ax") of rda(dune ~ poly(pH, 2) + Baresoil, varechem) of PR#945,
but no other problems of the same report.
* treedive: handles trivial cases of zero (diversity NA) or one
species (diversity 0).
* oecosimu: handles NA values as na.rm = TRUE in simulations for
cases where these are produced by the function to evaluate
statistics.
* oecosimu: takes care that the statistic is evaluated with binary
data when null models are binary. This was a one-year old bug
introduced in rev 967 on Aug 31, 2009. Previous time this bug was
corrected in rev 438 on June 28, 2008.
* nestednodf: Gustavo Carvalho sent a new upgraded version which
also implements a new quantitative method of Almeida-Neto & Ulrich
(Env Mod Software, in press; 2010). Based on feature request #1097
in R-Forge.
* ordistep: name clash if the fitted model contained item or term
called 'mod'. Reported by Richard Telford (Bergen, Norway).
* fitspecaccum: Stephen Sefick suggested adding Michaelis-Menten
fit to random species accumulation and sent his model function.
These ideas were used to add a new more general function. In
addition, there is an option of fitting Arrhenius model S = c*A^z
with a new self starting function SSarrhenius() for nls(). No
documentation yet: needs work.
* decorana: gives a warning if residual is bigger than tolerance.
Version 1.18-12 (closed September 26, 2010)
* swan: no eternal loop with disconnected data where zeros will
remain in the matrix.
* ordilabel: gained argument 'col' to set the text colour of the
labels separately from 'border' (default action is to use 'border'
colour).
* ordiellipse: gained argument 'col'. For draw = "lines" this
works similarly as previously when 'col' was passed on. For draw =
"polygon" 'col' sets only the polygon fill, and the text of the
labels will use the colour of the border or of the foreground.
* ordiR2step: gained argument 'direction = c("both", "response")',
and with "both" gives also adjusted R2 when dropping terms. I have
not yet seen a case where a term is dropped. Abandoned dangerous
practice of getting the R2.all directly from a fitted model of the
'scope' since this may have different lhs.
* anova.ccalist: removed. Revisions were 1273, 1274, 1276. Could
not make this work consistently with other anova models, except
with 'model = "direct"'. The function would also change the API so
that "..." come as the second argument breaking partial matching
of other arguments.
Version 1.18-11 (closed September 17, 2010)
* diversity: example says that rarefy(x, 2) - 1 is the same as
unbiased Simpson of a true random sample of Hurlbert (1971,
eq. 5). Added after several wishes to have this function.
* metaMDS: more verbose about connectivity. Use na.rm = TRUE in
metaMDSdist().
* FAQ: entry on ade4:cca name clash triggered by an invalid bug
report #1066 and queries in r-sig-ecology.
Version 1.18-10 (closed August 31, 2010)
* New version opened with the release of vegan_1.17-4 on August
20, 2010.
* rankindex: can take a list of dissimilarity functions as an
argument.
* deviance.cca/rda: return 0 (instead of NULL) if there is no
residual variation.
* vegdist: help page gives equations for binary variants of the
indices.
* biplot.CCorA: resets par that it sets.
* ordiR2step: can go to the end and include all variables of the
scope (and checks there is a scope). Additional stopping
criterion of exceeding P-value. Looks like being up to specs and
left for approval.
* permutest.cca: defaults to 99 permutations instead of 100 (to
be consistent). Item 'call' now stores the call to permutest()
instead of the call of the ordination which is now saved in
'testcall'. This means that update() works now.
* anova.cca: use correct degrees of freedom when there is no
constrained or no residual component.
* anova.ccalist: Started to work on the concept of permutation
test comparison of several cca/ rda/ capscale models. The basic
function was added to the R-Forge (undocumented!). First idea is
to run permutest.cca with the same RNG seed and collect the
permutations from the differences of these models. This won't be
consistent with other anova.cca* because permuted residuals differ
from the corresponding partial model (except with model =
"direct"). Another option that I have on my mind is that I use
model.matrix.cca to reconstruct a matrix of RHS of each model, and
then run a series of partial model ANOVAs. Here the problem is
that the statistics will be sequential, i.e., the residual of each
model is from the next model and not from the final model, and the
models must be in increasing order or ordered programmatically;
the current test will study differences of any sequence.
Naturally, we have to figure out how the permutation should be
done. Probably a note to R-Devel to open up discussion on the
issue. NOTE: there is no automatic delegation to this function. If
you try this, you must explicitly write "anova.ccalist(model1,
model2, model3). NOTE2: There are no tests of the sanity of the
input: be careful. NOTE3: Does not work with saturated model with
null residual variance.
Version 1.18-9 (closed August 20, 2010)
* mrpp & meandist: John Van Sickle notified us that his
Classification Strength (CS) uses 'weight.type = 1' (or n)
insteaad of 'weight.type = 3' (or n(n-1)/2). Calculation of CS
was dropped from mrpp(), because with this weighting it no longer
has an exact relation to the corresponding MRPP statistic and the
function mrpp() cannot provide a significance test for both
statistics together. CS is kept in meandist(), where its
calculation is corrected for the correct weight type.
* vegdist: Anderson et al. (Ecol Lett 9, 683-693; 2006) defined
their "alternative Gower" without range standardization of
columns. Reported as bug #1002 in http://r-forge.r-project.org/
by Sergio Garcia. Also some small edits of vegdist man page.
* msoplot: legend mixed symbols and labels.
* Rd (help) files: removed some superfluous \-characters that were
visible in rendering text (accented letters, Latex notation in
wrong places).
* predict.rda/cca: document 'newdata' with 'type = "response"' or
'type = "working"'. The feature was added in vegan
1.18-4 (rev1180, April 15, 2010).
Version 1.18-8 (closed August 12, 2010)
* DESCRIPTION: does not suggest package 'ellipse'.
* FAQ: new entry on negative or non-numeric data, and updates of
variance explained by ordination.
* intro: ordispider uses label.
* adonis: uses print.anova for cleaner output.
* plot.MOStest: uses veganCovEllipse instead of
ellipse:::ellipse.glm.
* ordiellipse: failed to plot labels with filled polygons, since
the same colour was used for polygon filling and labels. Now uses
ordilabel() with draw == "polygon" (but still text() with draw ==
"lines").
* decorana: change order of data checks so that non-numeric data
is detected (by rowSums()) before checking negative data entries.
* ordigrid: arguments lty, col and lwd to use different lines for
levels and replicate of the grid.
Version 1.18-7 (closed July 19, 2010)
* new version opened with the release of vegan_1.17-3.
* veganCovEllipse: new internal function for drawing covariance
ellipses (to be used both with ordiellipse() and plot.MOStest()).
* ordiellipse: uses veganCovEllipse.
Version 1.18-6 (closed June 16, 2010)
* Version 1.18-6 closed simultaneously with the release of
vegan_1.17-3 on June 16, 2010).
* mantel, mantel.partial: considerable speed up by cleaning
innermost loop and replacing as.dist() with direct extraction of
lower diagonal.
* treedist: documenting tree dissimilarity function that has been
in vegan devel since Aug 17, 2009 (rev928).
* betadisper: 'type = "median"', the default, was not computing
the spatial median on the real and imaginary axes separately.
Reported by Marek Omelka.
* permutations: First attempt to document within Vegan the general
workings of permutation tests. See ?permutations for details.
Version 1.18-5 (closed May 31, 2010)
* cca, rda: plot() failed if Condition() had factors, but
constraints had no factors. An example of failure:
m <- cca(dune ~ A1 + Condition(Management), dune.env)
plot(m) ## failed
* ordiarrows: gained 'label' argument which uses ordilabel() on
arrow starts.
* ordisegments: gained 'label' argument which uses ordilabel() on
both ends of the segment.
* ordispider: gained 'label' argument (uses ordilabel()). Can now
handle ordihull() result objects: connects points in the line to
the hull centroid. Extended examples.
* ordiR2step: new function to perform forward model selection
following Blanchet, Legendre & Borcard (Ecology 89, 2623-2632;
2008) or by adjusted R2 until the adjusted R2 of the full model
(scope) is exceeded or adjusted R2 starts to decrease.
* RsquareAdj.default: handles vector arguments.
* ordiellipse: works only with 2D data, and now uses only first
two dimensions of input if more given. For instance, the following
works now, but gave very cryptic error message previously:
m <- decorana(dune)
plot(m)
with(dune.env, ordiellipse(m, Management))
* ordiresids: removes now the row weights from Residuals and
Fitted in CA. Now shows the same values that are shuffled or
resampled in simulate.cca() where they are again weighted after
being added to the new fitted value.
* mantel.correlog: changes the Sturges rule to conform with the
standard R nclass.Sturges ('ceiling' instead of 'round'). Extends
and corrects documentation and messages.
Version 1.18-4 (closed May 21, 2010)
* envfit: defaults now to 999 permutations. Used to be 0 or no
permutations.
* metaMDS: can handle NA values in dissimilarities (but I do not
know if isoMDS can handle them correctly).
* mantel.correlog: accepts now user supplied breakpoints or number
of classes like intended.
* RsquareAdj.rda: uses now correct model df in rank deficit RDA
models.
* adonis: implemented tie handling in permutations.
* predict.cca/rda: implemented 'newdata' for type =
"response". For CCA, 'newdata' must be constraints (environmental
data), and for CA they must be community data. Test version, not
yet documented (except here).
Version 1.18-3 (closed April 15, 2010)
* new version opened with the release of vegan 1.17-2 (March 8,
2010).
* DESCRIPTION: Corrected the name of Bob O'Hara (also in release
1.17-2).
* scores.cca/scores.rda: scores will always have names even if the
input data have no names. This fixes some display issues that
Pierre Legendre had with biplot.rda(), and makes redundant the
name checking code in summary.cca (the redundant code was not yet
removed, but I see that this fix here was sound).
* permutest.cca: re-weighting of permuted CCA data matrix is now
done with compiled code using existing C function 'wcentre' (in
goffactor.c). This may give huge speed-up (>50% time save in first
tests) both in CCA and pCCA. The print function of permutest puts
now the observed statistic among permutations (the P-value is only
evaluated at printing so that this does not influence any
anova.cca function using permutest.cca).
Version 1.18-2 (closed March 8, 2010)
* ordistep: documented the 'anova' component added to the result
object.
* densityplot.oecosimu: gives no warning if there is only one
statistic and hence only one panel.
* nesteddisc: failed if highest species frequencies were tied. The
error message was "Error: dims [prodcut 15] do not match the
length of object [0]". Found by Eric Fuchs in the help forum of
vegan.r-forge.r-project.org. The function does not try as hard to
optimize the index, and hence it is much faster (but slacker).
* screeplot: New argument 'legend' for all methods (except
'decorana') to draw a legend if the observed and broken stick
distribution are both plotted. Now also respects 'pch' for the
broken stick line.
* permutest.betadisper: print method got ordering of permutation
p-values wrong in displaying the matrix of pairwise tests. The
computed values were correct. The formatting for printing contained
the bug. Reported by Dan O'Shea.
Version 1.18-1 (closed February 18, 2010)
* closed with the release of vegan 1.17-1. The release contains
bug fixes and multipart of 1.18-1, but excludes MOStest and 1.18-1
upgrades to ordistep.
* ordistep: saves model changes in the 'anova' item similarly as
standard step(). Experimental feature: not yet documented.
* envfit: did not pass weights to factorfit or to vectorfit when
the environmental variable was a single vector. Reported for
vectors by Richard Telford. Broken in rev1004 (Sep 12, 2009).
* CCorA: Fixed bug in presentation of variables in plots. Adds new
biplot types. General improvement in checking exceptional cases
improve stability.
* predict.cca, predict.rda: match 'newdata' by row names or column
names in type = "wa" and type = "sp". This is similar as
predict.prcomp/princomp. Gained choice type = "working" for
working residuals (as fitted.cca/rda).
* simulate.rda, simulate.cca: gained argument rank passed to
predict.rda() or predict.rda(). This also allows simulation under
null model by setting rank = 0.
* permatfull.Rd: fixed braces in the help file + mark-up.
Version 1.18-0 (closed January 11, 2010 with release 1.17-0)
* Documented na.action in cca.object.
* Opened concurrently with copying 1.16-34 (at rev 1087) to
branches/1.17 for a new release.
* betadisper: 'type = "median"' is now the default type of
analysis.
* fieller.MOStest: removed erroneous multiplication of variances
by overdispersion, because this was already done in cov.scaled,
and therefore was doubled and gave much too wide (up to Infinite)
limits.
* MOStest uses now ANODEV of free model and model where the
hump/pit is forced to a given location instead of the test based
on the Wald statistic (coefficient/sd) of the coefficient of the
linear term. The ANODEV and Wald statistic are equivalent in
Gaussian models, but in with other error families the ANODEV is
more correct. This fix removes the conflict between the basic
MOStest() and its profile() and confint() methods in non-Gaussian
families.
Version 1.16-34 (closed January 2, 2009) -- (codename: Lucia)
* twostagechao: deleted, because question regarding the
notation in the original article remained unanswered.
* permat* cleanup: permat.control deleted, not needed any longer
due to changes in oecosimu, documentation for plot.permat merged
with permat* help page.
* goodness.cca: was not using correct weights. Broken in r992 (Sep
6, 2009).
* vegdist: added UI for the alternative Gower index ("altGower")
that skips double zeros. The C code has been in vegan since Oct 24
(r1051), but no R interface. Together with decostand(..., "log")
of r1052 this closes Feature Request #473 submitted by Etienne
Laliberte in R-Forge.
* anova.cca: tied values could be randomly broken due to
numerical precision, and this gave false P-values, for instance in
anova(cca(dune ~ Management, dune.env),
strata=dune.env$Management). Now rounds statistics to 12 decimal
places in permutest.cca and in anova.ccabyterm.
* permutest.cca: uses pasteCall() in print.
* betadisper: added spatial median code written by Jari to betadisper.
All associated methods work OK, and this should be the default method
but needs a little checking first. This requires the new function
'ordimedian', which is currently undocumented in ?betadisper.
Version 1.16-33 (closed December 13, 2009)
* DESCRIPTION: Guillaume Blanchet added to the Authors.
* spandepth: New function to find the depths of nodes in a
spanning tree from spantree() function. Per request of
W. E. Sharp.
* bstick: refuses to analyse capscale() models with imaginary
components because I have no idea how to do it. Reported by
Richard Telford, Bergen (who probably wanted another
solution...).
* predict.rda: works now with rank = 0 -- returns matrix of column
means. (Not yet done for predict.cca.)
* simulate.rda: can now add permuted residuals to the fitted
values. Experimental, low level interface that requires giving the
index of right length. The permutation index can contain
duplicates such as from sample(nrow(data), replace = TRUE) so that
bootstrapping is possible unlike in anova.cca/permutest.cca. Works
now with partial model: adds partial fit to the constrained fit.
* simulate.cca: implemented by cloning simulate.rda. Function
works with internal Chi-square standardized form and
back-transforms the result by marginal totals and matrix grand
total at the end. The marginal totals may still be negative. Not
explicitly documented, and needs a connoisseur user.
Version 1.16-32 (closed November 13, 2009)
* alias.cca: gained argument 'names.only' (defaults FALSE) to only
return the names of the aliased variables insted of their defining
equations.
* rarefy: failed with one site and many sample sizes like
rarefy(rpois(10, 2), sample=2:3).
* msoplot: uses standard legend() to display info instead of hand
crafted positioning of the legend information.
* ordisurf: plotting main title was broken with NA handling in
1.16-31 (r1049). Also increased grid density and adjusted pushing
out limits of convex hull.
* pcnm: new function for PCNM (Principal Coordinates of
Neighbourhood Matrix) with option for row weighs allowing PCNM for
cca. Based on Stéphane Dray's PCNM function in his (unreleased)
SpacemakeR package. Imported with history from sedarVegan in
http://sedar.r-forge.r-project.org/. Has scores.pcnm to select all
(default) or some vectors of 'choice'.
* decostand: implemented Marti Anderson's log scaling of type
log(x, base = logbase) + 1 as a part of Feature Request #473. The
coding follows closely Etienne Laliberte's suggestion, and also
handles non-negative non-integer data.
* vegdist: C code for an alternative version of Gower Index that
skips double zeros. No R level interface yet. If this function is
to be used from vegdist.R, it should be added as item 14 in
'method' list, and the R level preprocessing should be done
similarly as for "gower" (method choice 6). This was implemented
for inspecting the implementation of second part of Feature
Request #473, but may be removed if that part of the request is
not implemented.
Version 1.16-31 (closed October 24, 2009)
* ordisurf: removes observation with NA in scores (as result of
na.action) or in predicted variable.
* specaccum: "accumulations" in one-site collections are handled
with a message() instead of stop()ping with error. Per request of
Roeland Kindt.
* constrained ordination (cca, rda, capscale) could not handle all
anova(.., by= "something") cases when the data were not defined by
data=, but were in the working environment or given using
with(). The attach'ed data frames that were later detach'ed won't
work. This means some potentially unstabilizing changes in scope:
please report all problems.
* envfit: could drop data frame to a vector with na.action (broken
in 1.16-28).
* model.frame.cca, model.matrix.cca: new support functions to
reconstitute the model frame and model matrix (or a list of
condition and constrain model matrices) from a cca/rda/capscale
result object. Documented separately.
* constrained ordination (cca, rda, capscale) did not analyse
constrained component in model like cca(varespec ~ Al:P +
Condition(Al + P)) or in constrained interaction terms if the
corresponding main effects were partialled out.
* constrained ordination (cca, rda, capscale) failed if they only
had partial term Condition() without constraints.
* ordistep: new function for automatic model choice in the cca
family of methods. This is similar to step(), but uses permutation
P-values instead of AIC (that does not exist for these methods,
and sometimes fails). This is similar to forward.sel() in Stephane
Dray's 'packfor' package at R-Forge. Better things here are that
the function has both drop and add instead of only forward
selection, handles correctly factors, and works with cca(), rda()
and capscale(). On the other hand, it is much slower than
forward.sel(). The function has similar user interface as step():
it knows 'scope' and 'direction'.
* ordisurf: pushes out limits of convex hull so that contours
better cover corners and margins of points.
* plot.decorana & plot.cca: use linestack() if only one dimension
was chosen.
* scores.lda: new function to extract row scores and predictor
coefficients from "lda" objects (MASS package). Magically,
ordiplot() started to work with "lda" results. Not yet documented,
but alias in scores.default.Rd.
* ordiplot: uses linestack() to plot 1-col results, and now really
is able to identify if there were no species scores (tried
earlier, but failed because species had different names than
sites).
* scores.default: returns vector scores as 1-col matrix.
* linestack: passes "..." to plot(), handles missing names.
Version 1.16-30 (closed October 3, 2009)
* manual pages (Rd files) were run through aspell. With this, an
error in an argument choice was corrected also in
plot.nestedtemp function.
* meandist: plot method has a new alternative of drawing
histograms (van Sickle 1997) as an alternative to dendrograms. A
histogram connects groups at theri within-cluster dissimilarity
level to the average between-cluster dissimilarity.
* oecosimu: gained new keyword 'alternative' for "two.sided",
"less" and "greater" one-sided tests.
* oecosimu: tied values in simulations and observed statistics
treated more correctly.
* oecosimu: if all simulations were constant, their sd became
zero, and z became NaN. Now z = 0 for these cases. A response to
the query of Martin Kopecky at the vegan-help forum in R-Forge.
* densitplot.oecosimu: panels keep the order of statistics. The
panels were ordered alphabetically, but the vertical line for the
observed statistic was in the original order (and thus often in
the wrong panel).
* calibrate: is now a a generic function and has a new method for
the 'ordisurf' object in addition to the old 'cca' method.
* FIXED BUG in cca, rda & capscale interpretation of
formulae. This was introduced in rev 972 (Sep 1, 2009), and fixed
in rev 1011 (Sep 18, 2009). Basically, all evaluation of 'subset'
and 'na.action' is done after extracting the model frames with
proven methods so that the first half of 'ordiParseFormula' is
more similar to pre-972 versions. Function 'ordiParseFormula'
could be re-written more elegantly, but deliberately breaking it
twice instead of breaking it once is just too much.
* cca/rda/capscale: cca(dune ~ poly(A1, 2), dune.env)
failed. Internal changes in 'ordiTerminfo' which now always looks
at the model frame returned by 'ordiParseFormula' instead of
looking at the 'data' (and the 'data' arg is now unused in
'ordiTerminfo').
* FIXED (minor): cca(dune ~ Management, dune.env, subset =
Management == "SF") failed in 'contrasts()' called from
'model.matrix()' because there is only one level in Management.
Now 'ordiParseFormula' adds a level with an unlikely name of
.ThisVarHasOnly1Level, so that model.matrix works, but
predict.cca(..., newdata=, type = "lc") rejects unknown factor
levels.
Version 1.16-29 (closed September 18, 2009)
* meandist: plot got an 'ylim' argument. This is practical when
users want to display several plots side by side, and requested by
Heli Suurkuukka.
* WARNING OF A BUG in cca, rda & capscale: all versions have
failed since rev 972 (Sep 1, 2009) if an argument in a model was a
matrix. Now these things work so that tests pass, but results are
wrong , because get_all_vars() does not get names right for matrix
columns in the model frame. This is manifest in the example of
varpart() that uses matrix items (such as mite.pcnm). It can be
also seen with this:
p <- poly(dune.env$A1, 3)
cca(dune ~ p + Management, dune.env)
The following seems to work:
cca(dune ~ poly(A1, 3) + Management, dune.env)
* cca, rda etc. drop unused levels of factors. This became actual
with 'subset' and 'na.action'.
* cca & rda (but not (yet?) capscale) gained argument 'subset' to
select rows for analysis. A 'subset' can be a logical vector (TRUE
for kept rows), or a logical expression which can contain
variables in 'data' or species (variable) names of the dependent
data. Some valid examples are:
cca(dune ~ A1 + Moisture, dune.env,
subset = Management %in% c("HF","BF"))
cca(dune ~ Moisture, dune.env,
subset = Lolper > 0 & Management != "SF")
S <- specnumber(dune)
cca(dune ~ 1, subset = S > 9)
Version 1.16-28 (closed September 15, 2009)
* drop1.cca: 'scope' given as a formula was not recognized and
permutation tests were not run. Only character vectors of names
worked.
* envfit: handles missing values in ordination scores or in
environmental variables using na.rm argument. This also changes
the user interface so that 'vectorfit' and 'factorfit' clearly
become low level functions which only accept scores (no ordination
objects), and all processing is done in envfit.default.
* wcmdscale: scores (points) with zero weights (w = 0) and other
non-finite entries will be NA.
* capscale: implemented na.action. The WA scores are always NA for
missing observations in capscale.
Version 1.16-27 (closed September 10, 2009)
* cca & print.cca: no warning on removed species in cca.default,
but print information on their number.
* predict.rda: failed when rda was based on correlations (scale =
TRUE), and there were missing data (0/0 division in scaling).
* cca & rda: New way of handling na.action. Function ordiNAexclude
adds items CCA$wa.excluded or CA$u.excluded for NA rows into
ordination result, and function ordiNApredict called in scores.cca
and scores.rda pads the results with these excluded data. In this
way the internal structure of the cca object remains unchanged,
but functions accessing the result with scores will get the NA
data.
* anova.cca(..., by = "term") and step() did not work correctly
with na.exclude or na.omit. These fitted sequential models, and
due to missing value removal the data set could change in the
sequence. The problem case was step() which relies on standard
stats functions add1.default and drop1.default. They detect the
change in data by length(object$residuals). Therefore
ordiNAexclude adds item 'residuals.zombie' to the ordination
result so that step() works. The same mechanism is also used in
anova.ccabyterm (not all NA deletions change the numbers of
observations).
* weights.cca and weights.rda know na.action.
* Fixing cca/rda functions for changes in weights(). The rule is
that you can mix scores() and weights(), but if you have internal
cca object stuctures (like x$CCA$u), you should have internal
weights x$rowsum. Concerns as.mlm.cca, goodness.cca,
permutest.cca, spenvcor, inertcomp and intersetcor. (The inertcomp
solution is a kluge.)
* plot.cca: works when there are NA values in scores.
* anova.cca(.., by = "axis") works when there are omitted or
excluded observations due to missing values.
Version 1.16-26 (closed September 1, 2009)
* cca and rda: can now handle missing values in constraints or
conditions. With 'na.action = na.omit', the rows with missing
values are removed, and with 'na.action = na.exclude' all rows are
retained but results that cannot be computed are NA. The LC scores
cannot be computed with missing values, but WA scores (usually)
can be computed. The results with 'na.exclude' probably fail with
most support functions because these do not expect missing values
in the results. These functions are fixed with time. capscale does
not (yet) handle missing data.
* oecosimu: does not depend on permatfull/permatswap, but method
can be a function. The r2dtable method implemented also
to be used with adipart and related functions.
* bioenv: uses standard R function combn() and removes
ripley.subs() and ripley.subsets() from vegan. The result should
be unchanged.
* ordiNAexclude: internal infrastructure to implement na.action =
na.exclude for cca and rda (and capscale?). The option returns
scores where the excluded sites get NA results and are not
omitted. In constrained ordination, LC scores will be NA, but WA
scores are found from the community data, and the current function
uses predict.cca/predict.rda with 'newdata' of removed
observations to estimate WA scores of excluded observation.
* ordiParseFormula: handles missing values with na.action.
Version 1.16-25 (closed August 31, 2009)
* mantel.correlog: updated to the version 1.4 of Pierre Legendre's
package.
* RsquareAdj: documented.
* ordiellipse & ordihull: argument 'draw' got a new alternative
"none". With this, nothing is drawn but the invisible plotting
structure is returned. This was added so that it is easier to
develop permutation tests for hulls and ellipses.
Version 1.16-24 (opened August 17, 2009)
* ordiellipse & ordihull: gained new argument 'label' to put the
name of the group in the centre of the ellipse or the hull. The
functions now return invisibly data on their plots for each
plotted item: ordiellipse() the covariance matrix (cov.wt()
result), and ordihull() the coordinates of the hull vertices.
These inivisible results have a summary() method that displays the
centres of the hulls or ellipses and their surface areas.
* ordiellipse: does not depend on the 'ellipse' package, but finds
the coordinates of ellipse using the equation taken from the 'car'
package. However, this devel version of 'vegan' still must SUGGEST
'ellipse', because 'plot.MOStest' requires 'ellipse'. The current
release version does not include 'plot.MOStest' and could be
liberated from the 'ellipse' dependence.
* spantree: first argument 'd' like in hclust()instead of 'dis'.
Got a 'call' item so that update() works.
* treedive & treeheight: accept 'spantree' objects. EXPERIMENTAL
CHANGE THAT MAY BE CANCELLED.
* treedist: average distance of species trees for two
sites. Generalization of treedive(): builds a species tree of two
sites and sees how much of the tree is shared between two sites,
and how much is unique to the combination. The measure is similar
to Jaccard index: the proportion of unique (unshared) tree of the
combined tree. In principle, this is similar to UniFrac metric
(C. Lozupone & R. Knight, Appl. Environ. Microbiol. 71:8225-8235;
2005), but numerically and computationally different (and the
current implementation may specify a new unpublished method). AN
EXPERIMENTAL CHANGE THAT MAY BE CANCELLED.
Version 1.16-23 (closed August 17, 2009)
* poolaccum: new function to find the extrapolated richness
values (or specpool() estimates) for random accumulation of
sites (with summary and plot methods).
* specaccum, specpool, specpool2vect and friends: index names are
now all lower-case.
* estaccumR: new function similar to 'poolaccum', but uses
'estimateR' for individual (count) based estimates of extrapolated
richness. Inherits from 'poolaccum' and shares its print, summary
and plot methods.
Version 1.16-22 (closed August 11, 2009)
* metaMDS: accepts now user supplied dissimilarities either as a
"dist" structure or as symmetric square matrix (Susan Letcher's
wish). As a side effect, dissimilarity matrices are now correctly
detected and will not be wrongly treated as raw data.
* summary.cca: fix in 1.16-20 failed if all variables were
continuous and there were no centroids of factor levels.
* indpower: new function to calculate indicator power for a
species (Halme et al. 2009, Cons. Biol. 23: 1008-1016)
* mantel.correlog, print.mantel.correlog & plot.mantel.correlog:
new functions to construct multivariate mantel correlograms ---
use in ecology: to describe the spatial structure of species
assemblages. See Legendre and Legendre, Numerical ecology; 1998,
section 13.1.5. Written by Pierre Legendre.
Version 1.16-21 (opened July 4, 2009)
* capscale: has now 'fitted' method (and hence, 'residuals'), and
'predict' works with 'type = "response"'. These give
dissimilarities that produce same ordination as original data, or
their low-rank approximations. Function 'fitted' has 'model =
"Imaginary"' that returns correction distances of complex
eigenvectors with negative eigenvalues. The dissimilarities are
not additive, and the "Imaginary" component is not used in
estimating the unconstrained component.
* capscale: really works now either with argument 'dfun' or
'metaMDSdist' (but not with both!).
* wcmdscale: returns negative eigevalues scaled by
sqrt(abs(eigenvalues)) if 'k' is not given or 'k' would include
any negative eigenvalue.
Version 1.16-20 (closed July 4, 2009)
* new version opened concurrently with the release of vegan 1.15-3
on June 17, 2009.
* orditorp: works now when axes are reversed like xlim = c(1,-1).
* rrarefy: a new function to generate randomly rarefied
communities.
* betadisper: fix bug in removal of (effectively) zero Eigenvalues.
* summary.cca: takes care that all scores have rownames --
head/tail fail without names. No ugly printing for constrained
eigenvalues when these do not exist in unconstrained models.
* biplot.rda: honour args to scores ('scaling', 'const').
* points/text for cca: accept 'const' argument for rda scores.
* ordilabel: no superfluous warnigns when args ('scaling',
'const') are given to scores.
* ordiplot3d: honour args to scores ('scaling', 'const') and
without warnings.
* ordixyplot: honour ags to scores ('scaling', 'const').
* spantree: plot honours args to scores ('scaling', 'const').
Version 1.16-19 (closed June 17, 2009)
* capscale: handles now negative eigenvalues with semimetric
indices. These cause no more warnings. The MDS engine is now
'wcmdscale' instead of 'cmdscale' unless 'add = TRUE' was
specified (and then negative eigenvalues are not produced). The
negative eigenvalues are listed after positive unconstrained
eigenvalues, but no scores are given for them. The total inertia
is now the sum of all eigenvalues, including the negative ones,
and inertia components adds item 'Imaginary' for negative
eigenvalues after 'Uncostrained'. New argument 'sqrt.dist' makes
square root transformation of dissimilarities, producing metric
indices in many cases (e.g., Bray-Curtis, Jaccard). Help page and
examples now show all these, and therefore the entry on negative
eigenvalues was dropped from FAQ-vegan.texi. The closed (max 1)
indices are now correctly adjusted for 'rda' and hence the
eigenvalues and inertia components are now sqrt(n-1) smaller and
similar as in 'cmdscale' and 'wcmdscale'.
* wcmdscale: remove zero eigenvalues, keep negative, and do not
assume that the last eigenvalue is zero.
* RsquareAdj: 'cca' returns R2, but no adjusted R2 (which is
NA). The adjusted R2 is always NA for partial models.
* kendall.global: fixed a bug that caused incorrect counts of ties
in large data sets.
* tsallis: division by zero occurred with hill=TRUE, scale=1,
this bug has been fixes to give exp(x) as scales goes to 1.
* multipart: a sequel to adipart, but implements multiplicative
diversity partitioning based on Hill numbers (Jost 2007,
Ecology, 88, 2427-2439.)
* eigenvals: new function to extract eigenvalues of cca, rda,
capscale (constrained & unconstrained), wcmdscale, prcomp,
princomp, eigen and svd. For svd returns squares of singular
values, and for prcomp and princomp squares of 'sdev'. The
'summary' of eigenvals() retrurn eigenvalues with the proportion
explained and cumulatitive proportion explained.
* summary.cca: uses summary.eigenvals to display eigenvalues.
* protest: observed value of test statistic is now considered as
one of the permutations. Default N.perm decreased to 999 to account
for this.
* envfit (vectorfit, factorfit): observed value of test statistic
is now considered as one of the permutations.
* mantel, mantel.partial: observed value of test statistic is now
considered as one of the permutations.
* anosim: observed value of test statistic is now considered as
one of the permutatons.
* mrpp: argument defaults and printing adated for having test
statistic among permutations (like the function always had).
Version 1.16-18 (closed May 14, 2009)
* tsallis: got new argument 'hill' similar to that of renyi.
* twostagechao: function that calculates multiple-community
similarity based on Chao et al. 2008 (Biometrics 64, 1178-86).
Some debugging is needed since it cannot reproduce the numbers
in Table 3 exactly (lower estimates and SEs). The Rd file
also needs more details.
* anova.cca: there was a name clash and failure in
anova.ccabymargin, anova.ccabyaxis and anova.ccabyterm if data
were indexed with 'i' in the fitted model. Now the internal
indexing uses less likely name (.ITRM) to avoid name clashes.
* orditkplot: copes with missing or NaN scores (omits them).
* ordixyplot: failed in constrained ordination (cca, rda,
capscale). Was broken in rev108 on Nov 7, 2007(!). The biplot was
rotated 90 degrees: it seems that lattice calls x y and y x.
* ordixyplot & ordisplom: Gained a new type = "arrows" that is a
Lattice version of ordiarrows.
* RsquareAdj: defined as generic function. The old RsquareAdj is
now the "default" method, and there are specific methods for
"rda", "cca" (= NA), "lm" and "glm" objects. Not yet documented,
but functions are listed as aliases in varpart.Rd. On Pierre
Legendre's request.
Version 1.16-17 (closed April 15, 2009)
* closed concurrently with the release of vegan 1.15-2.
* spantree.Rd: example plots use type = "t" to demonstrate the use
of ordilabel().
* nestedtemp: docs tell that Carsten Dormann's bipartite package
has a direct port of BINMATNEST programme of Rodriquez-Girones &
Santamaria.
* anosim: uses much less memory allowing analysis of larger
problems (number of rows, number of iterations). The change
concerns only internal data structures, and the permutation
results should be identical with old and new code when the same
random number seed is used. The Rd file got an example of using
'strata' in permutations.
* meandist: a new sister function for mrpp. Function meandist
calculates a matrix of mean within and between group
dissimilarities. Its summary function returns the overall averages
of those distances, and all three variants of MRPP A statistic,
and classification strength. The plot method draws a dendrogram
based on the mean dissimilarity matrix, with leaves hanging to
within-group dissimilarity.
* betadisper: Fix bug corrects the way distance to centroid is
computed when only one group is present in the data. betadisper
can now also detect if 'd' is Euclidean, zapping zero Eigenvalues.
Bug report and suggested improvements provided by Etienne
Laliberté.
betadisper now also handles missing observations in either the
dissimilarity matrix or the grouping variable.
* decostand: added dots to the argument list to pass (ignored)
arguments to decostand(). This was done to to fix
stressplot(metaMDS(x, dist="gower", trymax=40)), which failed
because metaMDSdist/vegdist were passing 'trymax' to decostand()
while evaluating "gower" dissimilarity.
* ordisurf: will fit a linear trend surface if knots = 0 or knots
= 1, and a quadratic trend surface if knots = 2. All these are
below the lowest permissible values of 'gam' smoothers and free
for creative reuse.
Version 1.16-16 (closed April 2, 2009)
* metaMDS: scales ordination scores to the same range as input
dissimilarities if halfchange = FALSE. Better heuristics to decide
whether to use halfchange scaling: the old heuristics failed if
Euclidean or other open-ended dissimilarities were small (max <
1), like after wisconsin() in metaMDSdist. The heuristics are now
in postMDS instead of metaMDS wrapper, and this changed the
default of internal argument 'halfchage' in postMDS, and now the
user can set the 'halfchange' argument in metaMDS. The
undocumented internal 'maxdis' argument is changed to logical in
metaMDSdist (TRUE if maximum dissimilarity = 1 for a pair of sites
with nothin shared).
* mrpp: reports now "classification strength" when weight.type = 3
or weights n*(n-1) are used. Returns mean dissimilarities and
counts for classes. Expected delta is now the mean of original
dissimilarities instead of mean of permutation deltas.
Version 1.16-15 (closed Mar 25, 2009)
* envfit help page: explains the automatic scaling and 'arrow.mul'
argument.
* add1/drop1 help: better example after a query of usage in R
mailing list.
* oecosimu & permatswap: much faster 'burnin' implemented through
'thin', and hence using a single C loop instead of multiple calls
and assignments to an R object.
* as.mcmc.permat: provides same functionality for permat objects
as as.mcmc.oecosimu for oecosimu objects.
* permatswap: same default values of 'burnin' and 'thin' (0 and 1,
respectively) are set for permatswap and oecosimu functions.
Previous defaults for permatswap were in conflict with oecosimu
when called by permat.control.
* hiersimu: new function to evaluate a statistic value at
different levels of a nested hierarchical sampling design. Based
on oecosimu and null model testing, but with evaluation process
similar to adipart.
* metaMDS: automatic stepacross works with Euclidean, Manhattan
and other distances with no fixed upper limit. However, this may
be meaningless, and a warning is issued. Fixed after a query to
R-sig-ecology by Manuel Spinola.
Version 1.16-14 (closed Mar 10, 2009)
* as.mcmc.oecosimu: a new function to translate sequential null
models to "mcmc" object of the coda package. The coda package
provides a wealthy source of the analysis of things like
stationarity, adequate sample size, autocorrelation, length of
needed burnin-in etc.
* density and densityplot: new methods for "oecosimu" objects.
* as.ts.oecosimu: new function for applying time series methods
for sequential null models ("swap", "tswap", "abuswap").
* metaMDSrotate: a new function to rotate metaMDS configuration so
that the first axis is parallel with a continuous site variable.
* as.ts.permat: extracts summary statistics from 'permat' objects
as time series, to use for time series diagnostics.
* permatswap: two new methods (under "abuswap") added based on the
paper of Hardy 2008 (J Ecol 96, 914-926). These preserve row *and*
colunm occurrences, and row *or* column sums at the same
time. Thus these complement the methods available in vegan. Uses C
code (not thoroughly tested yet). Needs some more work, and
documentation (examples and reference in Rd file).
Version 1.16-13 (closed Feb 24, 2009)
* anova.cca: documented the 'cutoff' argument for by = "axis".
* text.procrustes: added text() function for procrustes() result plots.
* rad.null: failed with family = gaussian and Gamma.
* simulate.rda: new method to simulate responses with random error
for an rda() result object. The function uses Normal error, and
estimates the sd separately for each species from their rda
residuals. Normal error is justified by least squares fitting in
rda. This could be made to use permutations of residuals. Not yet
implemented for cca or capscale results. For cca we -- at least --
need to handle weights, and the capscale result would be rda
result, since the method cannot be directly implemented for
dissimilarities. The function was triggered by Ben Bolker's email
to R-devel mainling list
https://stat.ethz.ch/pipermail/r-devel/2009-February/052116.html
Version 1.16-12 (closed Feb 15, 2009)
* anova.ccabyaxis: gained keyword 'cutoff' (defaults 1) to break
from permutation tests after exceeding the P-value given in the
argument. The keyword was introduced because rda and cca are
occasionally used with a huge number of constraints (like PCNM
vectors) where tests take a very long time, although only a couple
of first axes are significant. The keyword is still undocumented,
and will remain so because this change is still experimental.
* contribdiv: new functions for contribution diversity as an
alternative to simple additive diversity following Lu, Wagner &
Chen, Basic and Applied Ecology 8, 1-12 (2007).
Version 1.16-11 (closed Feb 9, 2009)
* anova.cca: used different random number sequence for the second
term with by = "margin".
* ordirgl & orgltext: removed work arounds for very old versions
(pre-0.65) of the rgl package.
Version 1.16-10 (closed Jan 28, 2009)
* commsimulator: failed if input data (x) were a data.frame.
* metaMDS: really do not evaluate no.shared() if set
noshare=0. Triggered by a problem report with missing data.
* vegdist doc: Morisita and Horn-Morisita were given as
similarities in the help.
* ordirgl: fixed wrong warnings in ordirgl(..., type="t").
* kendall.post, kendall.global: knows now all P value adjustment
methods of 'p.adjust()' plus "sidak".
Version 1.16-9 (closed Jan 20, 2009)
* kendall.post: added one line that was accidentally lost in
editing.
* Checked with R version 2.9.0 Under development (unstable)
(2009-01-13 r47593) having experimental strict Rd parser, and
corrected the reported problems in Rd files so that vegan is R
2.9.0-ready.
* adonis: uses now canonical (hits+1)/(tries+1) for P-values
instead of hits/tries. Default number of permutations increased
from 5 to 999. Added terms component so that things like update()
and drop.scope() magically started to work. Renamed design.matrix
into model.matrix (and documented the item).
* permatswap: new algorithm "swsh" is added to keep row/column
incidences (but not sums) constant. The name refers to the two
steps: (1) binary *sw*aps are made then (2) individuals in
non-zero cells are *sh*uffled.
* permat methods: calculations of the Chi-squared statistic is
added to summary, and the type argument can be used in the plot
"bray" or "chisq" values.
* nestednodf: gained argument 'order = TRUE' allowing tests for
other orderings than using marginal frequencies (default)
following suggestion by Almeida-Neto et al. (2008).
* summary.cca: cumulative "accounted for" statistics failed in
unconstrained models.
Version 1.16-8 (closed Jan 7, 2009)
* Fixed unescaped $-signs in documentation files after Ripley told
to do so ("[Rd] Misuse of $$ in Rd files",
https://stat.ethz.ch/pipermail/r-devel/2008-December/051634.html)
* kendall.global & kendall.post: new functions to analyse the
Kendall's coefficient of concordance --- use in ecology: to
identify significant species associations. See P. Legendre,
J. Agric. Biol. Envir. Statistics 10, 226-245; 2005. Written by
Guillaume Blanchet and Pierre Legendre.
* nestedness: added nestedness index based on overlap and
decreasing fill (Almeida-Neto et al., Oikos 117, 1227-1239; 2008).
Not yet all properties of the Oikos paper are implemented, and the
UI needs work (print, plot methods). Coding by Gustavo Carvalho
as a part of R-Forge Feature Request #265. Documented with
nestedtemp.
Version 1.16-7 (closed Dec 17, 2008)
* permat*: functions were rationalised, strata argument is used
instead of reg and hab, and returned object got several new
arguments. The tide of changes affected methods as well, and some
other function (adipart, oecosimu). Stratification was tested and
corrected.
Version 1.16-6 (closed Dec 7, 2008)
* adipart: got a formula interface, and aggregate() was replaced
by matrix multiplication. Now it is 10 times faster. The formula
interface has some consequences on the specification of the
sampling design.
* permatfull: Jari Oksanen made the C port for the quantitative
quasiswap algorithm. So the permat* null model family now can be
used for a wide array of null model analyses and is quick enough
to make reliable testing. The permatswap function and help page
were modified accordingly.
* plot.rad: gained argument log = "y", allowing log = "xy" so that
Zipf model is a straight line, or log = "" with arithmetic
scales. Not (yet) implemented for Lattice alternatives.
Version 1.16-5 (closeed Dec 4, 2008)
* adipart: new implementation of adipart is now ready to be
released (needs some testing). Summary and plot methods have been
removed. The new implementation is based on oecosimu and
quantitative null model settings via permat.control. It contains
only traditional diversity indices. More features will be added
later, gradually.
* plot.nestedtemp: label argument accepts a vector of length 2
with elements for row and column labels, respectively.
* mite.rda: Oribatid mites got species names (from Pierre
Legendre's web pages).
* text.decorana: failed (bug introduced in rev344, on May 9,
2008).
* commsimulator: simulated null models have same column and row
names as the original input (as.)matrix. There was a problem
report by a user who tried to use column names during evaluation
of the statistic. Done by setting attributes (and it would be
possible to replace dim() setting by attributes() setting in the
code).
* ordisurf: gains argument 'bubble' to use bubble plots for
points, and 'cex' to set the symbol size (or the maximum size with
'bubble = TRUE').
Version 1.16-4 (closed November 25, 2008)
* oecosimu: got a new control argument for quantitative null model
analyses; this is available if method = "permat". The print
method and help file modified accordingly.
* vignettes: Figures with narrower margins. Vignette
"diversity-vegan" adds functional diversity ('treedive'), 'beals'
and 'radfit' graphics updated to current vegan. Still missing
pieces: nestedness & null models, quantitative null models,
additive diversity partioninng, Tsallis entropy.
* radlattice: a new function to display each radfit model in a
separate panel together with their AIC or BIC values.
* plot.cca: takes care that plotted items have non-NULL names that
plot neatly. This was already take care of with formula
interface, but now it should work also in other cases in
plots. Reported by Daniel Borcard.
* permatwsap: quasiswap method is implemented for count data based
on the idea of Carsten Dormann in the bipartite::swap.web
function. This takes a random matrix first with same row and
column sums as the original (via r2dtable), and then restores the
original matrix fill through steps of swapping 2x2
submatrices. The current R code is slow.
* permatfull: when both cells and individuals are shuffled, the
matrix fill remains constant.
* nesteddisc: replaced with a new function that orders tied
columns to minimize the discrepancy statistic. If there are <=7
tied columns, all upto 5040 permutations are inspected, and if
there are >7 tied columns, 5000 random permutations are
inspected. The ties are inspected only if they could change order
of columns within range of row sums (species richness). A warning
with results is only printed if there were >7 tied columns that
could influence the statistic, because in other cases the search
for solution was exhaustive. The new function may be slow, and a
more clever solution should be found (or nesteddisc dropped from
vegan). With this code, the discrepancy statistic of the 'sipoo'
data is 50 (cf. notes on 1.16-3) instead of the original 55.
* betadisper: now works for cases where there is only a single
group. 'scores', 'plot' and 'boxplot' methods updated to work in
such cases. 'anova' and 'permutest' methods stop with an error if
used for such cases as they implement tests that do not make sense
for a single group.
Version 1.16-3 (closed November 1, 2008)
* spantree: saves labels of points, and 'cophenetic' and 'plot'
use these labels. Function 'plot' uses 'ordilabel' for text
labels.
* orditkplot: added option to copy the current graph to a TIFF
file. The capabilities("tiff") check was added in R-2.8.0, but the
test works in R < 2.8.0 as well.
* nesteddisc: Carsten Dormann (UFZ, Leipzig, Germany) turned my
attention to the sensitivity of discrepancy to ties in column
frequencies. These are not handled by any way (yet), but the users
are warned now about ties. For instance, the example matrix 'A' on
page 259 of Brualdi & Sanderson (Oecologia 119, 256-264; 1999) can
be ordered to give discrepancy value = 2 instead of value = 3
reported in the paper with a different ordering of columns.
Different random reorderings of species in the 'sipoo' data in
vegan can give discrepancy values 51 to 58 (and the original
arbitrary ordering gives 55). This means that discrepancy values
are not very reliable, but ties should be handled better or the
function withdrawn from vegan.
* ordisurf: returns x, y and z used by contour() following a query
"[R-sig-eco] output from ordisurf" on Oct 23, 2008.
Version 1.16-2 (closed October 27, 2008) --- for Annukka
* beals: replaced old beals.R with the new code by Miquel de
Caceres with new methods described in De Caceres & Legendre
(Oecologia 156, 657-669; 2008). This also implements the
cross-validatory Beals of Munzbergova & Herben (Oikos 104,
408-414; 2004) plus many new options of unbiasing estimates.
* permatfull: bug of row/col margins is fixed, shuffle argument is
added to enable individual and sample based randomisation of count
data (or both) when not all the margins are fixed.
* anosim: API and documentation made similar to mrpp to combine
their documentation. Now anosim can take data frame input and find
the dissimilarities internally.
Version 1.16-1 (closed October 10, 2008)
* dispindmorisita: function to calculate Morisita's index of
dispersion, and its standardized version. Uses the correct way of
standardizing, see Note section in help.
* radfit: Should work with empty sites (no species) or when the
number of species is less or equal the number of parameters
estimated (like may happen in sweeping analysis of data frames or
simulations). Takes care that input data frames have row names
which are necessary for displaying results.
* head.summary.cca & tail.summary.cca: shortcuts to
print(summary(x, ...), head, tail) for nicer Sweave tutorials.
Version 1.16-0 (closed on September 30, 2008)
* New devel version after branching the release version
1.15-0. Identical to the release version, but includes MOStest and
adipart functions that were removed from 1.15-0.
Version 1.14-12 (closed September 26, 2008)
* add1.cca, drop1.cca: new functions that also implement
permutation tests with argument test = "permutation", but only use
default methods with test = "none". Function drop1.cca uses
anova.cca(..., by = "margin"), and add1.cca implements a new type of
analysis for single term additions. The immediate reason for
implementing these tests was that I found out that step(..., test
= "Chisq") worked(!) producing parametric likelihood ratio tests,
and this hole had to be closed before users notice this.
* anova.cca: improved treatment of models where constrained or
unconstrained models are NULL so that permutation tests are
impossible. These used to stop with error, but now they return a
zeroed result so that things like anova(..., by = "term") and
drop1() work. Added function anova.ccanull to handle these cases.
* anova.cca(..., by = "margin") was handling wrongly 'x' in ~
Condition(x) + x + z, or model formulae where same variables were
used both as Conditions and (aliased) constraints.
* anova.cca(..., by = "margin") or function anova.ccabymargin
gained a 'scope' argument which is used in drop.scope().
* orditkplot & ordipointlabel: improved interface allows passing
font type from ordipointlabel to orditkplot.
* copied to version 1.15-0 at rev506.
Version 1.14-11 (closed September 10, 2008)
* anova.cca/permutest.cca: default permutation model changed from
"direct" to "reduced" after Pierre Legendre's simulation test
demonstrated slightly better Type I error for "reduced" models in
by = "axis" and other partial models. This is also consistent with
CANOCO (but permutation of CCA residuals differs from CANOCO). For
compatibility with old versions of vegan, you must set model =
"direct" explicitly.
* ordiplot: did not use partial match for "sites" and "species".
Version 1.14-10 (closed September 5, 2008)
* treedive: tries to match 'comm' and 'tree' names if forced (by
an arg), or when sizes do not match. Output vector has names. Docs
discuss estimating signifcance using oecosimu().
* oecosimu: accepts now a vector of statistics and plain
statistics instead of the list item statistics, and the user can
define the name of the statistic in the function call. This means
that things like functional diversity (treedive) for sites can be
analysed with approprite simulation model.
* nestedtemp & oecosimu documentations split.
* ordiParseFormula: formula of the partial terms has the same
environment as the general formula. Cures some tricky issues when
the partial term of cca/rda/capscale model is updated within a
function. (The environment() drives me crazy, says J.O.)
* ordiParseFormula gains new argument 'envdepth' to define the
depth where the community data (dependent data) is evaluated. Most
functions evaluate community data in the parent environment of the
calling function, but capscale() creates new dependent data and
needs evaluate the community within calling function. (The
environment() drives me absolutely crazy, says J.O.)
* specaccum: empty (all zeros) species removed from input
data. These confused method = "exact" which gave sd = NA.
* permutest.cca: tells whether the test was for all eigenvalues or
for the first (this is implicit in anova.cca).
* vegan-internal.Rd: brief documentation of internal functions for
developers with some guidelines of good practices.
Version 1.14-9 (closed August 20, 2008)
* closed concurrently with the release of vegan 1.13-2: changes up
to this version may have been there.
* adipart: new function to perform additive diversity partitioning
with corresponding S3 methods.
* tsallis: new function to calculate Tsallis-entropy, a scalable
diversity index family
* treedive: new functions for estimating functional diversity
defined as the height of a dendrogram of species properties for a
site (Petchey & Gaston). Still very preliminary.
* plot.procrustes: could fail if two solutions were *exactly*
*exactly* identical (noticed by Peter Solymos with wcmdscale
example).
Version 1.14-8 (closed August 10, 2008)
* wcmdscale: new function to perform metric scaling
(a.k.a. principal coordinates analysis) with weights for points.
* ordiArgAbsorber: now knows about 'choices' and 'display' args
of scores(). Reported by Michael Denslow.
* orditkplot: plotting text labels was broken in the default case.
Zooming works again. Plotting character (pch) is now better
centred on the canvas, and the code was simplified.
Version 1.14-7 (closed July 30, 2008)
* orditkplot: imitates now plotting character (argument
pch). Zooming maintains graphical parameters (such as mar). Label
selection shown by a rectangle, since label may already be
red. Label family, size and font can be vectors.
* permat.R, swapcount.R: summary method was modified according to
standard R ways, ylab argument was deleted from plot method. The
R version of swapcount was deleted and replaced by the C
version. Help file was modified accordingly.
* ordilabel: new function for cluttered ordination plots -- text
is written on a non-transparent label. Similar to s.label()
function in ade4.
* ordipointlabel: new function for cluttered ordination plots --
points are in fixed positions, but their text label is located to
avoid overlap. The optimization is based on optim(..., method =
"SANN"). Similar to pointLabel function in maptools. Returns an
"orditkplot" object, but orditkplot cannot yet completely handle
this (and this feature is undocumented).
* permutations: permuted.index2 and associated functions now allow
for restricted permutations of strata (i.e. restricted shuffling
of the blocks). This changes the acceptable 'type' options and adds
a new argument 'permute.strata' to permControl(), to control how
and what is permuted.
Version 1.14-6 (closed July 5, 2008)
* permatswap (nestedness.c): translated Peter Solymos's
swapcount.R to C. This is still experimental code, and the user
interface is undocumented, except here: use method = "Cswap" in
permatswap.
* permatswap: was not updating swap matrix but using the same
starting matrix after burnin for every swap.
* swapcount.R: was advancing 'thin' counter even when a swap was
rejected so that fewer than requesed 'thins' were done.
* permat.R, swapcount.R: genereal cleanup, most importantly
replacing vector operators & and | with logical operators && and
||.
* commsimulator: "quasiswap" written in C and *much* faster
now. Times are for MacBook Intel 1.86 GHz and 100 matrices:
"sipoo" from 7 min to 4 sec, "BCI" from 2+ hrs to 45
sec. Actually, quasiswap is now much faster than ordinary swap
which also should be written in C. "Backtracking" is so much more
complicated code that it could probably never reach "quasiswap"
even if written in C, and it may be dropped in the future.
* commsimulator: "swap" and "tswap" written in C, and clearly
faster now. In MacBook, 100 swaps with "sipoo" went from 8 sec to
0.5 sec, and thinning hardly influenced timing (it used to
thin-fold times).
* oecosimu: takes care that original statistic is found from
binary data.
Version 1.14-5 (closed June 19, 2008)
* scores.rda: scaling = 0 returns now unmodified scores from the
object (like documented) without multiplying by the scaling
constant. Gains argument 'const' for user-settable general scaling
constant of the scores. The default 'const' is found internally to
giv a biplot scaling so that species and site scores together
approximate the original data. The value of 'const' is returned as
an attribute of 'scores'.
* summary.cca: prints the value of general scaling constant used
in 'rda' or 'capscale'.
* summary.cca: prints 'capscale' like 'rda' (used to have some
'cca' style output, like calling site scores "weighted averages"
instead of "weigthed sums").
* permutest.cca: estimates now residual df as number-of-rows -
rank-of-constraints-and-conditions - 1 so that df and pseudo-F are
equal to those reported by lm() for single response variables in
rda. anova(..., by = "ax") needs rethinking though: it only had
one df. Because the test was based on permutation and order
statistics the change of df does not change the significances:
the observed pseudo-F and permuted pseudo-F's are influenced
similarly by the change of df, and order statistic does not
change.
* varpart: long formulae were cut off in printoutwith a cryptic
warning.
Version 1.14-4 (closed June 12, 2008)
* permatfull, permatswap and swapcount: functions to generate
unrestricted and restricted null model community data matrices
under diferent constraints (preserving row/columnsums,
or incidence pattern), with print, plot and summary methods.
(submitted by PS)
* adonis: speed-up and doc updates.
Version 1.14-3 (closed June 9, 2008)
* rda.default, cca.default: vegan naively used only the rank of
the ordination, but indeed, there are three cases of ranks: rank
of the ordination or number of axes (returned as rank like
before), rank of the constraints after conditions (now returned as
qrank), rank of conditions + constraints (returned in QR$rank).
When only 'rank' was used, vegan for instance aliased terms beyond
rank of ordination axes, extractAIC did not get df right when
going beyond rank of ordination, anova(..., by="t") gave zero df
for constraints beyond rank of ordination, and calibrate.cca
ignored those extra variables (now stops). In addition predict.cca
and predict.rda(..., type="lc", newdata=somedata) gave wrong
results in partial model. The following changes were made if
necessary to cca/rda support functions:
* alias.cca: QR$rank
* anova.ccabyaxis: OK
* bstick.cca: OK
* calibrate.cca: stop with error if qrank > rank
* extractAIC: qrank
* mso: OK (HW confirms)
* msoplot: OK (HW cofirms)
* permutest.cca: qrank, also fixes anova.ccabyterm
* predict.cca/rda: QR$rank with type = "lc" & newdata
* summary.cca/rda: OK
* scores.cca/rda: OK
Version 1.14-2 (closed June 5, 2008)
* Helene Wagner joined the vegan team.
* mso: automatically collapses distance classes larger than half
the maximum distance into a single distance class.
* mso: plot.mso renamed to msoplot so that plot.cca also works,
and the example uses now Oribatid mite data set.
* mso: gained a print method that also shows the variogram
(Helene, please check this!), and mso knows about its own call.
* lines.spantree: did not pass graphical arguments to lines
(broken in 1.12-15).
* ade2vegancca: a new function which tries to change an ade4:::cca
object to vegan:::cca object (and manages partially). This is used
internally in print.cca, scores.cca and summary.cca so that vegan
can display some information from ade4:::cca (with a warning).
Since scores.cca works for ade4, so do plot, orditkplot etc which
rely on scores.
Version 1.14-1 (closed May 29, 2008)
* plot.mso: passes extra arguments (...) to graphical functions.
* mite.xy: new data set with spatial coordinates for the Oribatid
mite data. From Pierre Legendre's web page.
* weights failed in several ordination utilities, because
weights.default in R 2.7.0 (stats) could not cope with objects
that do not have weights. R 2.6.2 still returned NULL for those
objects, but now weights.default gives an error. As a workaround
several vegan functions define internally
weights.default <- function(object, ...) NULL
to get back the old behaviour. Concerns orglspider, ordicluster,
ordispider, ordiellipse, ordisurf, factorfit, vectorfit.
* inertcomp and intersetcor only work with objects inheriting from
"cca", but they did not check that the object was of the correct
type and so could end up with calling weights.default or give
obscure error messages.
* permutest.betadisper: Could give the wrong pair-wise permutation
p-values. Reported by Daniel O'Shea.
Version 1.14-0 (closed May 21, 2008)
* The base of the release version 1.13-0.
Version 1.12-15 (closed May 14, 2008)
* closed the revision and made it as the base of release branch
1.13-0 at rev353.
* capscale: prints info on possible metaMDS transformations, and
uses metaMDSdist transformed data for getting species scores
(similarly as metaMDS recently does).
* ordiTerminfo: does not give superfluous warnings if constrained
ordination called without constraints and without data=, like
capscale(x ~ 1).
* graphical functions: Graphical functions that also pass '...' to
non-graphical functions (such as scores) now accept non-graphical
arguments without issuing an warning. These warnings were harmless
but it is neater to stop them happening in the first place. This
is facilitated by the new internal function ordiArgAbsorber().
* MOStest: added fieller.MOStest for approximate confidence
intervals of the location of the hump or pit. The method is based
on Fieller's theorem following ter Braak & Looman (Vegatatio 65,
3-11; 1986) and profile based condifence limits following Oksanen
et al. (Ecology 82, 1191-1197; 2001). Both are based on the code
from the ESA Ecological Archives accompanying Oksanen et
al. (Ecology 82, 1191-1197; 2001) and published in package optgrad
(http://www.esapubs.org/archive/ecol/E082/015/default.htm), but
profile methods heavily borrow from the MASS package.
Version 1.12-14 (closed May 9, 2008)
* mso: added Helene Wagner's mso function for direct multi-scale
ordination or spatial partioning of 'cca' and 'rda'
results (Ecology 85, 342-351; 2004). Thanks to Helene Wagner for
allowing the inclusion of the code.
* MOStest: new function to implement Mitchell-Olds & Shaw test for
the location of quadratic extreme in a defined interval.
* capscale: accepts now other dissimilarity function than vegdist,
and optionally uses metaMDSdist to manipulate dissimilarities
similarly as metaMDS. This provides now a one-shot unconstrained
principal coordinates analysis ("metaPCoA"), optionally with
extended (flexible shortest path) dissimilarities. Incidentally
this also implements a "new" possibly unpublished ordination
method: extended distance based redundancy analysis.
* metaMDSdist: argument 'add' gained new alternative "ignore" to
do nothing with zero distances. This can only and should be used
with metric scaling (cmdscale(), capscale()).
* ordispantree: removed after being deprecated and defunct for a
long time. Use lines.spantree instead.
Version 1.12-13 (closed May 7, 2008)
* orditkplot: labels can be edited with double clicking, and
shift-mouse allows zooming into smaller are in the plot.
Version 1.12-12 (closed May 1, 2008)
* adonis: added calculation of coefficients both for the species
(when data matrix was given) and for sites (both with the data
matrices and dissimilarities).
* betadisper: was not calculating distance to centroid correctly
for observations where the imaginary distance to centroid was
greater than the real distance (resulting in negative distance)
which resulted in NaN when we took the square root. betadisper()
now takes the absolute value of the combined distance before taking
the square root. This is in-line with Marti Anderson's PERMDISP2.
Version 1.12-11 (closed April 23, 2008)
* adonis: permutation tests fixed. The permutation statistics were
wrongly calculated. The results will now be different than in
older version, including the example(adonis). Basically, the
reported statitistic used SS improvement compared to the residual
of the complete model, but permutations used SS accumulation
compared to the residual after the current model.
* specaccum: returns item "individuals" or the average number of
accumulated individuals with method = "rarefaction". A user
request.
* rarefy: accepts now a vector of sample sizes, and then returns a
matrix with a column for each sample size. If also 'se = TRUE',
returns two rows for each case, one for rarefied richness, and
after than one for its standard error. This was a recurrent user
request.
* zzz: vegan got package startup message.
Version 1.12-10 (closed April 19, 2008)
* adonis: the fix for the unused factor levels (r312) revealed
that adonis did not handle properly aliased variables and
pivoting. Results need checking.
* designdist: gained argument 'abcd = FALSE' to use the 2x2
contingency table notation instead of the more tangible default
notation.
Version 1.12-9 (closed April 13, 2008)
* ordiArrowMul: added function 'ordiArrowMul' to improve automatic
scaling of biplot arrows (in plot.cca, text.cca, points.cca) and
fitted vectors (in plot.envfit). Reversed axes 'xlim=c(1,-1)' work
now, and shifted origin 'at = c(1,1)' taken into account in
plot.envfit.
* plot.cca: scaling used for biplot arrows is returned as an
attribute to the 'biplot' item in the 'ordiplot' object (but this
attribute is not used for anything useful yet).
* summary.cca: contributions of eigenvalues accumulated for
constrained and unconstrained axes, and separately (new) for
constrained axes.
* allPerms: Now accepts a vector from which number of observations
can be deduced. This was the intended behaviour but allPerms failed
in such cases.
* adonis: Was not dropping empty levels of factors, which was
leading to incorrect results (d.f. was inflated). Reported by
Tyler Smith.
Version 1.12-8 (closed April 6, 2008)
* permute: New high-level untility function for facilitating
the production of permutation tests using the new permutation
designs allowed by permuted.index2(). An example of the new
API is shown in ?permute.
* permutest.betadisper: Pairwise t test function made more
computationally stable.
Version 1.12-7 (closed Mar 30, 2008)
* permuplot: New function, produces a graphical representation
of a permutation design given a number of observations and a
object returned by permControl(). The function handles all the
permutation designs currently handled by permuted.index2().
* permuted.index2: was not returning correct sample indices
for grid designs within strata. Also, was not consitently
mirroring series and grid designs when 'constant = TRUE' (i.e.
mirroring was not applied the same way within each level of
strata).
* permCheck: Now includes several sanity checks for balance
in permutation designs ('grid' & 'strata') and when using
same permutation in each level of 'strata' (constant = TRUE).
Also works if object is a (numeric or integer) vector of length
1, by expanding it to seq(from = 1, to = object). This is now
in-line with the way numPerms works.
New argument 'make.all' in permCheck() allows greater control
over the checking process. Allows user to check a permutation
design without generating the matrix of all possible permutations.
Used in allPerms() to allow that function to check permutation
designs without getting stuck in infinite recursion, as permCheck()
calls allPerms() if complete = TRUE. This change allows allPerms()
to be called by a user without having to make sure the supplied
design makes sense - the design is checked using
'make.all = FALSE'.
* screeplot: functions return now invisibly the xy.coords of
bars or points for eigenvalues. They used to return
invisibly the input data ('x'). Concerns screeplot methods for
'decorana', 'cca', 'princomp' and 'prcomp' objects in vegan.
* varpart4: result could be wrong if 'varpart' was bypassed and
'varpart4' was called directly, because 'Y' was not centred before
calculating sum of squares in 'varpart4' ('varpart2' and
'varpart3' were OK). Reported by Guillaume Blanchet, Uni Alberta.
* permutest: Now a generic function. permutest.cca is now the
'cca' method.
* permDisper: This function has been renamed to permutest.betadisper
and documentation updated.
* TukeyHSD.betadisper: Function is now documented with betadisper.
Version 1.12-6 (closed Mar 25, 2008)
* metaMDS: WA scores for species are now based on the same
transformations and standardization (like wisconsin(sqrt())) as
the dissimilarities. They used to be based on the original data
even when the data were transformed for the ordination. The change
is minimal: data are evaluated similarly as in metaMDSdist if
wascores are requested. An alternative would have been to split
data manipulation function into an independent function from
metaMDSdist, but I try this first. For compatibility reasons, the
previous behaviour can be re-established using argument old.wa =
TRUE.
* allPerms: Now allows unbalanced designs for "series" or "free"
permutations within levels of strata only (i.e. allows different
number of observations per level of strata). Unbalanced "grid"
designs are not supported (and are unlikely to be so in the near
future). permuted.index2 has allowed these cases from the
beginning.
* allPerms, numPerms: Corrected for situations where type = "series",
mirror = TRUE with only 2 observations. In this case, mirroring
does not double the number of permutations.
Version 1.12-5 (closed Mar 24, 2008)
* plot.procrustes: could fail if compared configurations were
numerically identical.
* allPerms: New function to enumerate all possible permutations
for a given permutation scheme and number of samples. Has 'print'
and 'summary' methods.
* numPerms: Now accepts a numeric or integer vector of length 1
as the first argument 'object'. If supplied it is expanded to
seq(from = 1, to = object). This change allows you to state the
number of observations rather than pass an object from which the
number of observations is deduced. Updated and clarified
documentation in this regard.
Version 1.12-4 (closed Mar 20, 2008)
* permDisper: Added pairwise comparisons of group dispersions
via a classical t test and via permutation test, accessed via
new argument 'pairwise = TRUE'. 'permDisper' is now documented
seperately from 'betadisper'.
* TukeyHSD.betadisper: A method for 'TukeyHSD' to calculate
Tukey's Honest Significant Differences for the grouping factor
in 'betadisper'.
* numPerms: Was incorrectly calculating the number of permutations
when permuting levels of strata was performed. numPerms was also
incorrectly reporting too many permutations in the particular
case of type="grid" and mirror=TRUE with ncol=2.
* permuted.index2: When permuting levels of 'strata'
(type = "strata"), permuted.index2 was shuffling both the levels
of strata *and* shuffling the samples within levels. The latter
was incorrect and is now fixed.
Version 1.12-3 (closed Mar 9, 2008)
* betadiver: a new function that implements all indices of beta
diversity reviewed by Koleff et al. (J. Anim. Ecol., 72, 367-382;
2003), with a plot function to produce triangular plots.
* isomap: added dynamic, spinnable 3D graphics using rgl
(rgl.isomap).
Version 1.12-2 (closed Mar 7, 2008)
* nestedtemp: Rodriguez-Girones & Santamaria (J. Biogeogr. 33,
924-935; 2006) have had a look inside the original temperature
calculator code and give a more explicit description of the
method. Following their paper, I implemented an iterative
procedure to alternately pack columns and rows for ones and
zeros. The fill line is still a parabola, but this seems to be
very similar to the R-G & S fill line with fills 0.2...0.8 and I
don't hurry with this. There still are differences in many aspects
of implementation. The iterative procedure gives slightly
different results between runs, but here tests: Rocky mountain
mammals 3.87 (A&P 3.81, was 3.59 in 1.12-1), Baja
birds (Bajabs.txt) 10.17 (A&P 7.25, was 7.75), Queen Charlotte
birds 14.78 (A&P 15.87, was 14.51). Some results are indeed more
different than they used to be...
* nestedtemp (2): After previous entry implemented the fill line
of Rodriguez-Girones & Santamaria. This means almost complete
rewrite of the function from 1.12-1. New test results: mammals
4.48, Baja birds 10.16, Queen Charlotte birds 15.85. Baja is the
most different. Atmar & Patterson have five Baja bird data sets
with these test results: Bajaball 6.33, Bajabl 14.43, Bajabn
5.98, Bajabo 11.18, Bajabs 10.13. Smaller change is that packing
is based on indices s and t directly instead of their ranks.
* vignettes: added discussion on nestedtemp and backtracking in
"Design decisions and implementation".
Version 1.12-1 (closed Mar 2, 2008)
* biplot.rda: gained argument col to select colours for sites and
species.
* bstick.princomp: used a wrong number of sticks (PCs). Reported
by Guy Yollin.
* nestedtemp: Returns row and column coordinates,
optionally labels plot.
* merged last doc fixes from branches/1.11-0/ after release.
Version 1.12-0 (Feb 15, 2008)
* Made a release branch (1.11-0) based on the the rev. 204, and
relabelled this devel branch to 1.12-0.
Version 1.10-13 (closed Feb 15, 2008)
* preparing for a release and trying to close open threads.
* nestedtemp: implemented smooth fill line. The kind of the fill
line was not specified in the original paper, but I used parabola
which is "as smooth as possible" in the sense that its second
derivative is constant. Needs checking, though.
Version 1.10-12 (closed Feb 12, 2008)
* nestedtemp: new try with smooth fill line.
* anova.cca: by = "margin" stops with informative error message if
some terms are completely aliased (used to stop with an
uninformative error message). Type III effects really do not make
sense with aliased variables.
* calibrate.cca: does correct pivoting of aliased models.
* orditkplot: Set up all basic features, and added points, text,
and scores methods. Features that should be added include zooming
to a rectangle drawn with a mouse, and adding env variables. The
latter would be trivial in a simple way, but the user should be
able to move the bunch of arrows to another place, and scale the
length of arrows with a mouse.
Version 1.10-11 (closed Feb 3, 2008)
* bgdispersal: New, more powerful test statistic (in terms of Type
1 error rate) for the McNemar test (Pierre Legendre).
* CCorA: internal functions embedded in CCorA().
* orditkplot: More resilient against user errors. New output file
types (bmp, xfig), jpeg uses 100% quality. more aware of operating
system and capabilities. Accepts graphical par in the function
call. Gains arguments xlim, ylim. Arguments col, bg, cex renamed
to pcol, pb, pcex so that the do not shadow par().
* permuted.index2: fixed a bug in permutations with strata.
* summary.prc: honours now argument 'axis'. Reported by Francois
Gillet, Lausanne.
* Documentation: added documentation for permuted.index2,
permControl, permCheck and helper functions. FAQ and intro know
now about orditkplot, anova.cca(..., by="margin") and
permuted.index2. FAQ gained new sections on using different
plotting symbols and handling congested plots in ordination
graphics.
Version 1.10-10 (closed Jan 27, 2008)
* DESCRIPTION: suggests tcltk.
* orditkplot: new function for ordination plots based on Tcl/Tk:
you can edit graph by moving labels using a mouse and save the
edited graph as an eps file, or export via R plot function to
eps, pdf, png or jpeg, or to dump to R session for further
processing.
* CCorA: removed partial analysis (which needs more work and may
reappear later).
Version 1.10-9 (closed Jan 16, 2008)
* anova.cca: new option by = "margin" to perform separate tests
for each marginal effect in a model with all other terms. This is
similar to Type III tests, but will only work with marginal terms,
and will exclude main effects when corresponding interactions are
in the model. P-value is now assessed in the true blue way as
(hits+1)/(permutations+1) in all anova.cca methods. Permutations
should be able to handle now rank deficit cases with pivoting in
the QR decomposition (permutest.cca).
* permuted.index2: New version of permuted.index() that now allows
restricted permutations. Can produce permutations for time-series
or line transects and for spatial grids. These can also be nested
within 'strata'. permuted.series() and permuted.grid() are the
relevant workhorse functions. Permutation options are set by new
function permControl(). Plan to migrate vegan functions to
permuted.index2() in the devel version, and will eventually replace
the current permuted.index().
* permCheck: New function for checking permutation schemes under
permuted.index2. Uses new function numPerms() to generate the
number of possible permutations. permCheck() has print, summary
and print.summary methods.
Version 1.10-8 (closed Dec 14 2007, at Rev. 133)
* metaMDS: gains argument wascores (defaults TRUE) to suppress
calculation of species scores.
* wascores: return now NA for missing (all zero) species instead
of failing.
* ordiplot: ignore missing scores in plotting and printing.
* betadisper: new functions for Marti Anderson's analysis of
homogeneity of multivariate dispersions.
* nestedtemp: handles now tied frequencies using ties.method =
"average" in rank(). Function is still preliminary, and may be
dropped from the release.
* taxa2dist: issues now a warning if called with 'check = FALSE'
and there are zero distances, typically meaning that basal taxa
(species) were not coded. Also explained in the help.
* varpart: failed if there were unused levels in factors (reported
by Clément Tisseuil, Toulouse).
* adonis: loses argument 'distfun'. Instead accepts now any 'dist'
object as input instead of a data frame or a matrix.
Version 1.10-7 (closed Nov 18, 2007 at Rev. 119)
* adonis: gains argument 'distfun' (default 'vegdist') so that
other standard conforming dissimilarity functions than vegdist can
be used (a user request).
* decorana: does not crash R when called with NULL row data such
as decorana(dune[FALSE,]) (reported by Richard Tellford, Bergen,
Norway), and predict(..., type="sites") works correctly with
downweighted analysis (reported by Colin Beale, Aberdeen,
Scotland).
* isomap: checks now that input data are dissimilarities or can be
changed into dissimilarities without warnings.
* ordixyplot: can now display biplot arrows and factor centroids
in constrained ordination or from envfit (ordisplom and ordicloud
cannot display these).
* ordixyplot, ordicloud and ordisplom: no longer have explicit scaling
argument, but can now pass arguments to scores() as well as to the
plotting functions.
* oecosimu: evaluates now two-sided P-value of the statistic from
simulations. Doc updates.
* vignettes: use all similar style (amsart, a4paper, sidecap,
date style).
* FAQ: tell about docs added to R-Forge versions of vegan. Tell
about predict() for having "passive" points.
Version 1.10-6 (closed Oct 31, 2007 at Rev. 100)
* Gavin Simpson joined the vegan team.
* oecosimu: new functions for indices of nestedness of patches or
islands, and simulation of null communities.
* scores: cca method does now accept display = c("species", "sites").
Previous fix (1.10-5) was for rda method only. Documentation updated
to reflect changes.
* biplot.rda: now allows a different type to be specified for each of
the species and site scores. Function now documented more completely.
* ordixyplot, ordicloud and ordisplom: now have argument 'scaling'.
* orditorp: now handles additional arguments passed as '...' more
cleanly. Arguments can be passed to scores() or points/text. This
should avoid the warning messages of previous versions.
* sipoo: a new data set on the bird communities in the Sipoo
(Sibbo) archipelago to demonstrate the oecosimu functions.
Version 1.10-5 (closed Oct 18, 2007 at Rev. 86)
* biplot.rda: new biplot function for PCA in rda() (Gavin
Simpson).
* scores: cca method accepts now display = c("species", "sites"),
when it only used to accept c("sp","wa"). More consistent
arguments in the default method.
* CCorA: Pierre Legendre submitted code for permutation test of
Pillai's trace.
* capscale: negative scaling works now in plot() and scores():
estimates species sd() from the 'comm' if 'comm' is known.
Version 1.10-4 (closed Oct 6, 2007 at Rev. 75)
* anova.cca: failed in anova(object, by = "axis") when the fitted
model had terms like poly(x, 2) or log(x).
* ordiresids: new function for plot.lm like diagnostic plots for
constrained ordination: Residuals ~ Fitted, sqrt(abs(Residuals)) ~
Fitted, and qqmath(~ Residuals) using Lattice graphics.
* fitted.cca, fitted.rda: gained agument type = "working" to get
the fitted values and residuals used internally in calculation (to
be used in plot.lm diagnostic plots residuals ~ fitted).
Version 1.10-3 (closed Oct 2, 2007 at Rev. 62)
* doc: added vignettes intro-vegan on ordination methods and
diversity-vegan.
* Trellis graphics for ordination objects with functions
ordixyplot, ordicloud and ordisplom. These functions are still
unsatisfactory. In particular, I haven't found a clean way of
having biplot arrows in the graphs (how do you know the
coordinates and the axes used?). Contributions are welcome.
* ordispantree made dysfunctional (deprecated in 1.8-1).
Version 1.10-2 (closed Sep 14, 2007 at Rev. 36)
* vegandocs: added FAQ-vegan, and renamed the old vegan-FAQ to
decision-vegan. Only primary sources (texi, Rnw) in the source
repository: derived files (.tex, .pdf) should be made when package
is built.
* CCorA: Pierre Legendre's functions for Canonical Correlation
Analysis. These provide a more robust alternative to the standard
cancor() in base R. (under work)
* scores.default: handles numeric data frames.
Version 1.10-1 (closed Sep 7, 2007 at Rev. 21)
* taxondive: Fixed dimension checking, matching names and
non-unique taxon names.
Version 1.10-0 (Sep 6, 2007)
* Moved vegan_1.8-7 (CRAN release) into R-Forge.R-Project.org and
Subversion SCM. This ChangeLog will only contain most important
news (and should be renamed to NEWS), and the real ChangeLogs are
created by the SCM. The revisions will be numbered as 1.10-1 to
1.10-x, and the next release will be 1.11-0.
VEGAN RELEASE VERSIONS at http://cran.r-project.org/
Version 1.8-8 (Oct 2, 2007)
* Minor bugfix release for upcoming R-2.6.0. Based on the
http://r-forge.r-project.org/projects/vegan/ revision 17 (= 1.8-7)
with ported bug fix revisions from 1.10-3 (see below for revision
numbers).
* anova.cca: by = "term" failed in partial model. This was broken
in 1.8-6 by introducing a test against deficit rank models (r47).
* cascadeKM: Calinski index works now when the input data is a
data.frame (r57, 58).
* flush.console: metaMDSiter and bioenv use now flush.console()
so that Windows people also see the trace (r56).
* ordispantree: made defunct, was deprecated in 1.8-1 (r38).
* scores: handles now numeric data frames (r25).
* summary.cca: failed if only one type of scores was requested
(r50, 52).
* taxondive: Fixed dim checking and matching species names in
community data and taxonomic distance data (r21).
* tweaks to pass --pedantic R CMD check, mainly in formatting
source files, unused variables in source files and superfluous
braces in help files (r39, r46, r62).
* Updated FAQ-vegan.pdf to the current version at R-Forge.
Version 1.8-7 (August 24, 2007)
* Based on devel version 1.9-34.
* DESCRIPTION: M. Henry H. Stevens (Miami University, Oxford,
Ohio) joined the vegan team.
* adonis: new function for nonparametric MANOVA that is
appropriate for even extremely wide matrices sometimes associated
with gene data and with diverse ecological communities. Author
Hank Stevens.
* taxondive: a new function for indices of taxonomic diversity and
distinctness after Clarke & Warwick (Mar Ecol Prog Ser 216,
265--278, 2001 and other papers). With a helper function
'taxa2dist' to turn taxonomies into distances with an option for
variable step length (Clarke & Warwick, Mar Ecol Prog Ser 184,
21--29, 1999), and a toy data set on the taxonomy of dune meadow
species ('dune.taxon'). With a help, testing and pressure from
Mike Cappo, James Cook Uni, Qld.
* bgdispersal: previous version was partly garbled (by me), and
P. Legendre provided a corrected one.
* designdist: keeps 'dist' attributes even when the 'method'
function drops them. Swapped the order to (terms, methods) in the
default name.
* metaMDS: issues a warning if data are disconnected. Passes extra
arguments to other 'distfun' than 'vegdist' so that you can set
'terms' in 'designdist' etc (metaMDSdist). Can now do trymax=0 or
skip random starts and give you enhanced 'isoMDS' result
(metaMDSiter).
* ordiplot: failed if number of species was equal to number of
sites (and so did plot.metaMDS and plot.isomap using this).
* plot.profile.fisherfit: corrected a harmless error detected by
checkUsagePackage().
* predict.rda: removed some dead (but heavy) code from type =
"response".
Version 1.8-6 (May 9, 2007)
* Based on devel version 1.9-23.
* as.mlm.cca, as.mlm.rda: new functions to refit constrained
ordination result (cca, rda, capscale) as a multiple response
linear model. You can find influence statistics (Cook's distance,
hat values) from the refitted model. You also can find t-values
etc., but these have the same bias as in other software and should
not be used.
* bgdispersal: a new function for dispersal direction in
biogeography (Legendre & Legendre 1998, section 13.3.4). Author
Pierre Legendre.
* designdist: a new function for defining your own dissimilarity
index or for estimating beta diversity (Koleff et al.,
J. Ecol. 72, 367-382; 2003).
* isomap: a new function for isometric feature mapping of
Tenenbaum et al. (Science 290, 2319-2323; 2000).
* screeplot, bstick: new functions to draw screeplots of vegan
ordination results with brokenstick lines, and alternative
screeplot functions for prcomp and princomp with brokenstick.
Author Gavin L. Simpson.
* swan: a new function for the degree of absence (Swan 1970,
Ecology 51, 89-102).
* anova.cca: now refuses to do 'by = "terms"' if the rank of
constraints is higher than the rank of the community matrix.
* bioenv: gains argument 'partial' to perform partial bioenv.
* cca, rda, capscale: can now handle longer expression within
'Condition()' (ordiParseFormula). Used to drop observations with
missing values in unused variables (ordiGetData).
* goodness.cca, goodness.rda. 'statistic = "distance"' was wrongly
implemented. Now refuses to find "distance" in constrained
analysis: distances of constrained and unconstrained components do
not add up to to distances in unconstrained ordination.
* metaMDS (metaMDSdist): gains argument 'distfun' to use other
dissimilarity functions than vegdist.
* renyiaccum: used a variable that was not defined as an argument.
Added support functions persp.renyiaccum and rgl.renyiaccum
(Roeland Kindt).
* stressplot: R2's renamed to 'non-metric fit' and 'linear fit'.
* Doc: Corrected reference to Hurlbert in diversity (thanks to
Ralph Grundel). Updated references (varpart, renyiaccum). Removed
discussion on t-values in cca from vignettes, because as.mlm.cca
now implements those. General cleanup and better utf-8 encoding.
Version 1.8-5 (January 11, 2007)
* Based on devel version 1.9-12.
* no.shared (manifest in metaMDS): prints thousands of lines of
debugging info that I forgot to deactive in release. Not fatal,
but extremely annoying.
* capscale: inertia name as "unknown" if the dissimilarity object
does not have a "method" name. Suggested by Roeland Kindt.
* DESCRIPTION: license is now explicitly GPL v2 (but not later).
Version 1.8-4 (January 8, 2007)
* Based on devel version 1.9-10.
* cascadeKM: a new function to wrap kmeans and optimality criteria
for classification (Sebastien Durand, Pierre Legendre & Marie
Helene Ouellette).
* renyiaccum: a new function for Renyi (and Hill) accumulation
curves (Roeland Kindt).
* bioenv: bioenv.formula uses now "na.action = NULL" in
'model.frame', and bioenv.default passes arguments to 'cor' which
means that you can set NA treatment in 'cor' using argument "use".
* cca, rda: added "..." to formula versions to satisfy tests in
R-DEVEL.
* cca, rda, capscale: used to fail if called within other
functions. Now data always evaluated in the environment of
formula using new internal function ordiGetData.
* anova.cca: checks that the model has both residual and
constrained components or stops with understandable error message
(used to stop with incomprehensible error message).
* print.summary.cca, print.summary.decorana: have now arguments
'head' and 'tail' to print only a part of species and site
scores. Suggested by Gavin Simpson.
* metaMDS: checks now that the input data ('comm') is not a 'dist'
object (like many users have had).
* ordisurf: Does not depend on package 'akima' any longer, but
directly finds fitted values in a regular grid using
'predict.gam'. Added pnpoly.c to find which of these values are
within the convex hull defined by data. Results also look neater
with sparse data now. Added argument 'labcex' passed to 'contour'
for changing size of contour labels. Setting 'labcex = 0' will
suppress drawing labels (by setting drawlabels = FALSE in
'contour()').
* orditorp: handles now vector arguments of 'col', 'pcol', 'cex',
and 'pch'.
* rad.zipfbrot: less likely to overflow to NA coefficients during
iteration.
* renyi: added a plot function, and documented together with
renyiaccum() instead of diversity().
* scores.default: Knows now about ade4 objects. Primarily looks
for scores scaled by eigenvalues both for sites and species.
* specaccum: Added new conditioned method of Colwell et
al. with estimated sd based on extrapolated richness (Roeland
Kindt).
* vegdist.c: More informative warnings with 'method' name (useful
with rankindex).
* DESCRIPTION: listed 'require()d' packages in the "Suggests:"
field to satisfy more anal tests in R-DEVEL. Mention diversity
analysis in the "Description:".
Version 1.8-3 (Sept 29, 2006)
* Based on devel version 1.9-2.
* varespec.rda, varechem.rda: saved in binary form, because old
ascii form gave warning in R-2.4.0-rc.
* vegdist: added Chao index (of Jaccard type) that should take
into account missing pairs of species. Checks that Binomial index
is non-negative. Identical sites could have dissimilarity of
magnitude 1e-17 after some standardizations, but now values <1e-15
are zapped to zero.
* estimateR: uses now standard unbiased formulation of Chao.
* renyi: should work now (really!).
* metaMDS: with zero = "add", zeros now replaced with
min(dis[dis>0])/2 (used to be 1E-4) (metaMDSdist). Sets number of
tries also when this was not set previously (metaMDSiter)
Version 1.8-2 (June 13, 2006)
* version 1.8-1 failed test in Windows because of a wrong encoding
name. Explicit \enc added for non-ascii words. Kurt Hornik and
Uwe Ligges diagnosed this and led me to see the light.
* similar to devel version 1.7-97.
Version 1.8-1 (June 12, 2006)
* Based on devel version 1.7-96.
* Pierre Legendre joined the vegan team.
* beals: a new function for Beals smoothing.
* bioenv: added 'trace' argument.
* cca/rda/capscale: accept several 'Condition' elements in the
formula.
* capscale: capscale(y ~ ., data=...) or expansion of "." on the
rhs works now. Documentation recognizes now db-RDA as the real
mother method.
* scores.cca, summary.cca etc: rewrite so that is cleaner and
easier to maintain. User visible changes are scaling by species
standard deviation (negative scaling) for 'rda', scaling=0
(no scaling) for all methods and slightly changed output and
improved user control in summary. These scalings actually were
documented in 1.6-10, although I dropped them just before the
release. predict.cca, predict.rda: work now with 'newdata' even
when not called with formula.
* anova.cca: new argument 'by' for tests of single terms or axes:
with by = "terms" performs individual test for constraints, and
with by = "axis" a separate sequential test for each axis. New
argument 'first' to analyse only the first axis instead of all
constrained variation.
* intersetcor: new function for the interset correlation or the
(weighted) correlation between individual constraints (contrasts)
and invidual axes in cca/rda/capscale. (Not recommended.)
* decostand: does not automatically convert matrix to a
data.frame. NA handling more consistent now (thanks to Tyler Smith
for diagnosis). Adds attribute 'decostand' giving the "method".
* linestack: now really uses median as the midpoint with the odd
number of cases, and does not give superfluous warnings with three
or less items. New argument 'labels' to replace the default text
in plot. The old argument 'label' renamed to 'side'. However, the
function still works with the old syntax, but gives a warning if
the old argument 'label' is used for 'side'. Returns invisibly the
shifted positions of labels.
* metaMDS: 'postMDS' sets now attributes similarly when called
independently or within 'metaMDS'. 'metaMDS' forwards arguments
to 'postMDS' except 'halfchange'. Change of phrasing in
'print'. Added handling of zero dissimilarities into
'metaMDSdist': either "fail" or "add" 1E-4 into zeros.
* mrpp: new function for the multiresponse permutation procedure
(MRPP). Code by Henry Stevens (Miami Univ, Oxford, Ohio).
* ordiarrows: draws arrowhead only in the last segment. New
argument 'startmark' for marking the starting point of the arrow.
* ordisurf: new arguments 'main' for the title and 'nlevels' and
'levels' for the number of contour or their values.
* orditorp: arguments for text colour and text character expansion
changed to standard 'col' and 'cex' from previous 'tcol' and
'tcex'.
* procrustes: 'summary' prints now rotation matrix, translation
and scale, and honours 'digits'.
* prc: new functions for Principal Response Curves (PRC) of van
den Brink and ter Braak (Envir. Toxicol. Chem, 18, 138-148; 1999).
This is a special rda() model with dedicated summary and plot
functions. New data set 'pyrifos' to demonstrate 'prc'.
* radfit: added brokenstick as a null model (rad.null), removed
rad.veil (as it was a bad idea originally), corrected minor bugs
in rad.preempt (which did not fail gracefully). Line colours
start from the point colour in plot.radfit.frame. 'print' uses
"g" format for coefficients and adds 'digits' argument. Added
'summary.radfit.frame' that simply prints each model.
* rankindex: uses now cluster:::daisy when 'grad' includes
factors.
* spantree: now a method function with 'plot', 'lines' and
'cophenetic' methods. 'lines' replaces 'ordispantree'. The plot
has a weird, new way of finding configuration for a spanning tree
from cophenetic distances (unpublished). Documented separately.
* specaccum: 'plot' honours now 'ylim'.
* specpool: Chao richness was wrongly defined, but now uses the
biased formula (error introduced in 1.6-5, correct earlier).
Failed with zero species or if there were no species that occurred
exactly one in the 'pool' (thanks to Emmanuel Castella, Geneve CH,
for the bug report).
* varpart: new functions for unbiased partitioning of variation by
two to four explanatory tables in RDA or linear regression. The
author of these functions is Pierre Legendre & co (Univ
Montreal). New data sets 'mite', 'mite.env' and 'mite.pcnm' to
demonstrate the functions.
* vegandocs: new function to display *all* pdf documentation and
ChangeLog. This really should be in base R -- this is a kluge to
fill the hole.
* vegemite: added argument to 'select' a subset of sites. Drops
missing species from the table. Prints number of species and
number of sites and the used cover scale at the end of the table.
Passes arguments (i.e., 'maxabund') to coverscale().
* coverscale: added argument 'maxabund' to 'scale = "log"'.
Returns the name of the cover scale as an attribute for vegemite()
to print.
* vegdist: now first checks input and then transforms (if
needed). Thanks to Tyler Smith,
* Internal changes: permuted.index acceptes NULL strata as an
alternative to missing strata. ordispantreee deprecated.
spider.cca removed.
* Documentation: general cleanup in help files. New chapter on
t-values in cca/rda/capscale in vegan-FAQ. New pdf document on
partioning with varpart by Pierre Legendre & co. Non-latin
characters now use UTF-8 in documentation. R manual says that you
should not use non-latin characters in help files, but that was
written by Englishmen. However, this seems to cause distress to
some users of a US West Coast OS (Windows), but OK with mainstream
OS's (Linux, MacOS). .
Version 1.6-10 (September 26, 2005)
* Based on devel version 1.7-77. Checked with R 2.1.1 (stable)
and R 2.2.0 (alpha).
* rda: negative scalings explicitly ignored and treated as
corresponding positive values. Function summary.rda used to fail
with NA centroids.
* permutest.cca & anova.cca: permutation of 'cca' result now
re-weights environmental data properly with permuted community
weights (this breaks compliance with popular proprietary
software). New default method 'direct' that always permutes the
data instead of residuals. Now clearly faster basic routines, but
re-weighting in 'cca' is costly, and permutations may even be
slower than earlier in small data sets. The permutest.cca returns
more data: constrained and residual total inertia, degrees of
freedom in the input model, and .Random.seed used in iterations.
Thanks to Pierre Legendre for pushing me to make this faster.
* ordination plot functions: obey now xlim and ylim.
* ordination text and points functions: a new argument 'select'
that can be used to select a subset of items, and a new argument
'labels' for text used instead of the default row names.
* points.cca, text.cca: biplot arrows will be scaled automatically
to fit the current graph if 'arrow.mul' is not given. The new
behaviour is similar to the default in plot.cca and in
plot.envfit.
* orditorp: a new "ordination text or points" function to add text
or points to an existing plot: adds text if this can be done
without overwriting other text labels, and points otherwise.
* linestack: a new function to draw labelled one-dimensional
diagrams without overwriting the labels (a primitive one, and
could easily be improved: submissions are welcome).
* ordirgl, orgltext: adapted to changes in rgl package version
0.65. Workaround for older rgl packages, too.
* decostand: Added Hellinger transformation. Empty columns and
rows become now 0 instead of NaN in most methods, except when the
input data contains negative values. Warns on input with negative
entries or on output containing NaN. Has now argument
'range.global' for method 'range' based on the code supplied by
Tyler Smith.
* plot.radfit: puts now legend "topright" in R >= 2.1.0 (which has
this keyword).
* read.cep: issues a warning if vegan was compiled with gfortran,
which has a bug that may corrupt the result. The bug concerns
multiline input with T format modifier, and it was corrected in
http://gcc.gnu.org/ml/gcc-patches/2005-09/msg00126.html, but still
bugs most released versions of gcc.
* vegdist: new indices 'raup' and 'binomial'. Method 'raup'
implements probabilistic Raup-Crick index and is based on the code
submitted by Michael Bedward. Method 'binomial' implements
Millar's index, officially published as "Binomial deviance as a
dissimilarity measure" (the C code has been in vegan for about two
years, but now I tell about it and add the public interface).
Method 'mountford' will give NA for any comparison involving an
empty site. Warns about empty sites or negative entries with all
methods except 'euclidean' and 'manhattan'.
* Documentation: added documentation of the cca/rda/capscale
result object. Dontruns used more neatly.
Version 1.6-9 (April 22, 2005)
* Maintenance release: 1.6-8 failed in R 2.1.0 patched and R 2.2.0
devel (works in R 2.1.0 release) due to problems in
as.preston. Based on devel version 1.7-62.
* as.fisher, as.preston: used table() in a way that failed in R
2.1.0 patched
* calibrate.cca: new function to predict or calibrate or
bioindicate the values of environmental constraints from
community composition (ordination).
* decostand: new argument na.rm (defaults FALSE) for ignoring
missing values in row, column or matrix standardizations.
* vegdist: new argument na.rm (defaults FALSE) for pairwise
deletion of missing vaues in dissimilarity calculation.
Version 1.6-8 (April 18, 2005)
* Based on devel version 1.7-59. Adapted to R 2.1.0 beta.
* DESCRIPTION: gives due credit to Roeland Kindt and Bob O'Hara as
co-authors.
* documentation: updates in capscale, vegdist. vegan-FAQ adapted
to changes in Sweave in R 2.1.0.
* several methods assumed that input is count data, but silently
accepted floating point numbers and gave wrong results. Now they
stop with error with non-integer input: fisherfit, prestonfit,
prestondistr, rarefy, fisher.alpha, estimateR.
* bioenv: uses now 'cor' instead of 'cor.test', and does not give
so many superfluous warnings and is marginally faster. Changed
printed output, so that gives 'call' instead of names of community
and environmental data, since bioenv.formula could not handle
these cleanly.
* capscale: has now argument 'add' to use an additive constant to
all dissimilarities so that all computed eigenvalues are
non-negative. This is an argument of underlying 'cmdscale'
function, which implements the "Correction method 2" of Legendre &
Legendre (1998), p. 434.
* decorana: checks now that there are no negative data entries.
* dune: cleaner site names.
* envfit: 'plot' now automatically scales arrows similarly as
'plot.cca' if 'arrow.mul' is not specified and arrows are added to
an old plot. Has now a 'scores' function.
* goodness.cca, predict.cca: documented separately.
* goodness.metaMDS: new function to assess pointwise goodness of
fit in metaMDS or isoMDS.
* humpfit: user can now give starting values of parameters.
'summary.humpfit' returns 'cov.unscaled' so that users can apply
'ellipse' function of 'ellipse' package to display approximate
confidence ellipses for pairs of parameters (with Normal
assumptions).
* make.cepnames: No longer duplicates one component names, but
'abbreviate's them to eight characters.
* metaMDS: Split to independent metaunits 'metaMDSdist' and
'metaMDSiter'. New function 'metaMDSredist' tries to reconstruct
dissimilarity matrices and transformations. 'postMDS' skips
halfchange scaling with too few points (with a warning) instead of
stopping with an error. Prints now info about 'postMDS'
operations. New 'scores' function. Improved 'plot' function with
'display' argument and labelling of axes. Argument 'shrink' to
undo expansion of species scores in plot or scores functions.
Workaround for a future bug in 'isoMDS' which drops names of
points in R 2.1.0. Updates number of 'tries' with
'previous.best'.
* ordiplot: has now 'display' argument, so that only species or
sites can be plotted.
* ordiplot3d: a new function for 3D static ordination diagrams.
* ordirgl: new function for dynamic 3D graphics of ordination
results. With support functions orglpoints, orgltext,
orglsegments and orglspider to add graphical items to dynamic
plots. Needs 'rgl' package.
* predict.cca, predict.rda: type = "wa" with 'newdata' works now
even when some species were removed from the original fit because
they were all-zero (missing). Default now to 'model = "CA"' in
unconstrained analysis without "CCA" component.
* predict.decorana: a new function for 'decorana' results. Similar
to 'predict.cca' (but refuses to give impossible results).
* procrustes: obeys now 'choices' even when input is a matrix.
Plots now projections of rotated axes. New argument 'scores' so
that can handle other than "site" scores.
* protest: passes now arguments to 'scores' so that now can handle
other than "site" scores on more than two dimensions.
* rda: handles now species with constant values (typically
missing or all zeros).
* spenvcor: new function to find the "species -- environment
correlation" in contrained ordination (cca, rda, capscale).
* stressplot: a new function to plot Shepard diagram for 'metaMDS'
or 'isoMDS'.
* summary.cca/summary.rda: print now call.
* vegemite: can now 'use' 'dendrogram' objects.
Version 1.6-7 (Jan 24, 2005)
* Based on devel version 1.7-42.
* plot.envfit did not plot vectors. Thanks to Roeland Kindt and
Ron E. VanNimwegen for bug reports and fixes.
* ordisurf obeys now keyword 'display', and '...' will transfer
arguments to 'scores'.
* ordisegments: obeys now keyword 'display'.
* ordigrid: works now.
* ordihull, ordiarrows, ordisegments, ordispider, ordiellipse:
have a new keyword 'show.groups' to show only specified groups.
With the help of this argument it is possible to use different
colours and linetypes for each group. Further, it makes possible
to plot results of logical comparisons (such as A1 > 4) by setting
show.groups = T. Two user requests.
* new diagnostic and helper functions for 'cca', 'rda' and
'capscale': 'goodness' to estimate the proportion of inertia
accounted for or residuals for sites or species; 'inertcomp' to
decompose species and site inertia for conditioned, contrained and
residual componets; 'vif.cca' to estimate the variance inflation
factors for constraints and conditions; 'fitted' and 'residuals'
to approximate data by ordination scores; 'predict' to approximate
data or find site or species scores, possibly with 'newdata';
'coef' to find the regression coefficients. 'alias.cca' is now
similar to 'alias.lm' (simplified version of the latter).
Version 1.6-6 (Jan 7, 2005)
* Based on version 1.7-34, but without still experimental
functions predict, fitted, vif & goodness for cca and rda
objects.
* mantel: implemented partial mantel test (function
'mantel.partial'). Both 'mantel' and 'mantel.partial' should be
marginally faster and not give so many superfluous warnings.
* plot.envfit: should be more reliable now. A new keyword 'at' to
position the bunch of arrows in other places than the origin.
Thanks to Roeland Kindt for fixes.
* rankindex: Default correlation now "spearman" since the older
default ("kendall") was far too slow in large data sets. Should
not give so many superfluous warnings.
* summary.rda: Site scores were wrong with 'scaling = 3'.
Influences 'plot' and 'scores' commands for 'rda' and 'capscale'
results with 'scaling = 3'.
* vegdist: Issues a warning if 'method = "morisita"' is used with
non-integer data.
Version 1.6-5 (Oct 12, 2004)
* Based on version 1.7-27. Checked with R-1.9.1 (Linux, MacOS X)
and R-2.0.0 (Linux, patched version in Windows XP). Passed other
tests, but some examples in 'humpfit' failed in Windows XP, and
are not run on that platfrom. General cleanup in documentation.
Does not 'require(mva)' any longer.
* anosim: corrected the equation in docs (function was
correct). Thanks to Yong Cao for notifying this.
* bioenv.formula: finds now variables from attach'ed data.frames
and 'data' need not be specified.
* capscale & deviance.rda: capscale was modified so that its
result could be handled with step (added a terms component).
Function deviance.capscale was deleted so that deviance.rda will
be used. A literature reference to AIC in CCA/RDA was added to the
documentation.
* cca/rda/capscale: Small eigenvalues are made to zero and rank
reduced accordingly. Most often influences 'capscale'.
* cca/rda/capscale: Now alias constraints that are collinear with
other constraints or conditions in partial analysis. New function
'alias.cca' to print aliased contrasts. NB this does not influence
the proper ordination results, but only to the selection of biplot
scores and centroids displayed. 'summary' no longer gives a
redundant warning.
* decorana: removes empty species with a warning instead of
stopping with error. Still error with empty sites. The behaviour
is similar to 'cca'.
* envfit (vectorfit, factorfit): Has now a formula interface.
Bottlenecks in permutation changed now into C functions. Much
faster, in particular 'factorfit' which was 25 times faster in my
tests. plot.envfit could drop names (dimensions). Now honours
'scaling' argument in cca/rda/capscale (used to 'scaling = 2'
always).
* envfit, ordisurf: have now weights parameter 'w'. Weights are
used for the results of 'cca' and 'decorana' as default, or they
can be supplied by the user. In this way, envfit gives equal
results to biplot scores and centroids in 'cca', and ordisurf is
consistent with envfit. Both can now access 'lc' scores of 'cca'
etc.
* estimateR: Abundance-based estimators of extrapolated richness,
written by Bob O'Hara .
* humpfit: Asesses now the dispersion parameter in non-Poisson
models (summary.humpfit). Added a profile method that inherits
from profile.glm, so that you can use plot.profile.glm,
pairs.profile.glm and confint.profile.glm of MASS for displaying
profile and finding the confidence intervals from the profile
likelihood.
* fisherfit: prints now standard error of the estimate. Added
'profile' and 'confint' methods.
* metaMDS: A new function to collect all Peter Minchin's
recommendations to run nonmetric multidimensional scaling: (1)
adequate data transformation with Wisconsin double and sqrt if
needed, (2) use of adequate dissimilarity measure, (3) possible
stepacross if needed (this Minchin didn't have), (4) run NMDS with
several random starts and stop after finding two similar
solutions, (5) scale results with postMDS, and (6) add species
scores with wascores. Function has print and plot methods.
Function scores.default changed to understand metaMDS.
* specpool: Chao is now based on unbiased equation. Standard
errors now estimated for Chao, 1st-order jackknife and bootstrap
richness, but SE of 2nd-order jackknife still missing.
* vegdist: an option for binary indices, since some users believed
these are not in vegan, although you can get them with
'decostand'.
Version 1.6-4 (June 10, 2004)
* Based on 1.7-12.
* anova.cca: Changed defaults to beta=0.01 and perm.max=10000.
* bioenv: A new function implementing the BIO-ENV procedure of
Clarke & Ainsworth (Mar. Ecol. Prog. Ser. 92, 205-219; 1993). The
function finds the subset of scaled environmental variables with
maximum (rank) correlation with community data.
* cca: added ter Braak scaling, a.k.a. Hill scaling, with
negative values of `scaling'. This scaling should approximate
Hill's non-linear rescaling of `decorana', but it the scaling is
linear and does not rescale. The scaling is very simple: the
results from corresponding positive scaling are multipiled with
sqrt(1/(1-lambda)).
* plot.cca: Better heuristics used for the length of biplot
arrows, and they are now longer in general. Axes are drawn for
biplot arrows also with `text.cca' and `points.cca' functions.
* deviance.cca and extractAIC.cca: auxilliary functions which
allow automatic model building with `step' or `stepAIC' (MASS)
functions in cca and rda. Unfortunately, cca and rda do not have
deviance or AIC, and these functions are certainly wrong and
dangerous. However, with continuous candidate variables they
select the variables in the same order as Canoco.
* humpfit: A new family of functions to fit the no-interaction
model (Oksanen, J., J. Ecol. 84, 293-295; 1996) to species
richness -- biomass data.
* vegdist: corrected a bug in Gower index -- range standardization
was made to sites, but it should be done to species. Thanks to
Brett Melbourne for the bug report.
Version 1.6-3 (Mar 22, 2004)
* Still based on 1.7-3, but ported some changes from version
1.7-7 (envfit, procrustes). This version was produced to prepare
for incompatible changes in coming R-1.9.0.
* vegan-FAQ: Does not show the vegan and R versions the
documentations were built, because of stupid incompatible change
in R-1.9.0 of the future. package.description() function was
unnecessarily replaced with packageDescription, and to accomodate
recent, present and future versions of R seemed to be too much
hassle.
* plot.envfit: Added option 'p.max' to display only environmental
variables assessed to have P values less or equal to the given
limit.
* plot.procrustes: added option kind = 0 to draw only axes and
functions points.procrustes and lines.procrustes to add points and
line segments or arrows to the plot.
Version 1.6-2: Based on devel version 1.7-3.
* renyi: should work now (again?).
* prestondistr: The truncation level was badly chosen. Now default
level is set to -1, or log2(-1) = 0.5 which might be the left
limit of the first octave. This is now a parameter, leaving the
choice as the responsibility of the user.
* cca.default: Removes missing species and issues a warning
instead of stopping with error.
Version 1.6-1: Based on devel version 1.7-1.
* Minor maintanenance release to satisfy tests in R-devel. There
was a buglet in ordisegments that caused a warning. I had
introduced a trick to handle printCoefmat before it was invented
to replace print.coefmat, but this failed in R-devel.
* Vignettes included in this release. These were only in devel
versions of my web pages previously.
Version 1.6-0 (Fisher)
* Based on devel version 1.5-58. Passes checks with R-1.8.0 and
R-1.7.0 and compiles into a Windows binary. Many functions were
contributed by Roeland Kindt (Nairobi, Kenya).
* BCI: A new data set on tree counts on 1-ha plots in the Barro
Colorado Island. Thanks to Roeland Kindt and Richard S. Condit.
* capscale: A new function for [partial] constrained analysis of
principal coordinates (CAP).
* cca, rda: Can now handle cases where the number of constraints
is higher than the number of species.
* cca, decorana: Find now if data have empty rows or columns.
* decostand: Added Chi-square transformation.
* distconnected: A new function to find if vegetation data are
disconnected below given threshold dissimilarity level. Utility
function no.shared gives a logical dissimilarity matrix with
values TRUE for cases with no shared species.
* fisher.alpha: Should work with large sample sizes now (bug
report thanks to Ariel Bergamini).
* fisherfit, prestonfit, prestondistr: New functions to fit
Fisher's logarithmic series and Preston's log-normal model.
* make.cepnames: New utility function to change Latin names into
4+4 character CEP names.
* ordicluster, ordispantree: New functions in the ordihull family
to overlay a cluster dendrogram or a spanning tree onto an
ordination. Weights are now really used with cca and decorana
plots.
* radfit: New function to fit Ranked Abundance Dominance or
Dominance/ Diversity models with maximum likelihood: Pre-emption,
log-normal, veiled log-normal, Zipf and Zipf--Mandelbrot models.
* rankindex: Can now use stepacross and new dissimilarity indices
of vegdist.
* rarefy: Can now optionally find SE of rarefied richness.
* renyi: A new function to find Rényi diversities or Hill numbers
with any scale (thanks to Roeland Kindt).
* scores.ordiplot: should be more robust now.
* spantree: A new function to find a minimum spanning tree using
only dissimilarities below a threshold or disregarding NA
dissimilarities.
* specaccum: A new function for species accumulation
curves. Thanks to Roeland Kindt.
* specnumber: A very simple utility function to find species
richness.
* specpool: New function for extrapolated species richness in a
species pool, or for estimating the number of unseen species.
* stepacross: New function for estimation of dissimilarities
between sites that have no shared species. Implements both
flexible shortest paths and their approximation known as extended
dissimilarities.
* vegdist: New dissimilarity indices Jaccard (finally), Morisita,
Horn--Morisita and Mountford.
Version 1.4-4
* Based on devel version 1.5-35.
* decorana: Finds now eigenvalues from the final solution as ratio
of biased weighted variances of site and species scores. The
values returned by the legacy Fortran function are called
``decorana values''.
* downweight: passes the downweighting fraction as an attribute,
and decorana catches and prints the fraction.
* wascores: Uses now biased variances for expading WAs and returns
the shrinkage factors as an attribute "shrinkage". Shrinkage
factors are equal to eigenvalues in CCA when only this one
variable is used as constraint. Function `eigengrad' returns only
these eigenvalues.
* rda summary/plot bugfix: Failed with non-default scaling (1 or
3) when there were no factor constraints. Corrected a statement
giving a harmless warning in plot.cca.
* ordiplot knows now option type = "text". New simpleminded
functions points.ordiplot and text.ordiplot for adding items in an
ordiplot graph.
* ordiarrows, ordisegments could drop dimensions (when will I
learn!) and fail if only one arrow or one segment should be
drawn. One argument was missing in a `gl' command in `ordigrid'.
Version 1.4-3 (Luova)
* Based on devel version 1.5-30
* `cca' and `rda' use canonical expansion of the formula and
return the `terms' component. Functions like `terms', `formula'
and `update' magically started to work with cca and rda (and eight
lines of code would allow `step' and `stepAIC' to work magically
in model building... but it's so much magic that I don't trust
it). Some of the allowed things are now `mod <- cca(dune ~ . -
Use, dune.env)'
(all variables but `Use' in dune.env), and `update(mod, . ~ . -
Manure)' (remove `Manure' from the previous).
* `cca' and `rda' find centroids of factor levels with factor
constraints (only with formula interface). These can be accessed
with `display="cn"' in `scores', `summary' and `plot'. If
`display="cn"' is requested in `plot', biplot arrows are still
drawn unless there is a class centroid with the same name:
continuous variables are still shown as arrows and ordered factors
both as arrows as centroids.
* Enhanced user control of many ordination plots. Some accept more
graphical parameters (`ordiplot', `plot.procrustes'). New
functions `points.cca' and `text.cca' to add these items into
CCA/RDA plots (documented separately).
* `identify.ordiplot' knows more alternatives, among them
`plot.procrustes'. This was helped with new, very generic
`scores.ordiplot'.
* New functions to add graphical items in ordination diagrams:
`ordihull' draws convex hulls for groups, `ordiarrows' draws
arrows, `ordisegments' segments, `ordigrid' combines points in to
a grid, `ordispider' combines points to their (weighted) centroids
or WA scores to the corresponding LC score in cca/rda,
`ordiellipse' draws dispersion or confidence ellipses for
points. All these take a groups argument for selecting subsets of
points. `ordispider' obsoletes `spider.cca' (introduced in the
previous release).
* Implemented simple ``deshrinking'' of weighted averages in
`wascores'.
* A new diversity function `fisher.alpha' that estimates the
diversity as the alpha parameter of Fisher's log-series. Thanks to
Bob O'Hara for the main code.
* `summary.decorana' now in canonical form so that printing is
done by `print.summary.decorana' and user can intercept and catch
the result. Several users had been surprised of the earlier
non-canonical behaviour. `summary' knows again the prior weights
used in downweighting of rare species (these evaporated in 1.4-0).
* rda summary/scores/plot bugfix: Scaling of site scores was wrong
with scaling alternatives 1 and 3.
* envfit (vectorfit, factorfit) bug fixes: Can now handle single
vectors (labelling, permutation). Recognize now `choices' so that
the user can select ordination axes.
* Checked with released R-1.7.0.
Version 1.4-2
Based on devel version 1.5-18
* New functions and data sets:
* dune: A new data set -- the classic Dune Meadow data with the
environmental data.
* ordiplot, identify.ordiplot: New functions to plot "any"
ordination result and identify plotted points. Intended to provide
similar functionality as `plot', `plotid' and `specid' functions
in `labdsv' without name clashes. Functions `plot.cca' and
`plot.decorana' return (invisibly) `ordiplot' objects as well, and
these can be used by `identify' to label species and sites.
* rda: Redundancy Analysis, or optionally, yet another PCA. This
is a spin-off from cca(), and it is documented together with
cca. The only new functions are rda, rda.default, rda.formula and
summary.rda. Otherwise rda uses cca methods which were changed to
be aware of rda.
* read.cep: A function to read Cornell Ecology Package or CEP
formatted files into R. This has been in devel versions since
1.1-1, but it never made to the releases, since this worked only
in Linux, but crashed R in Windows. It seemed to work in R-1.6.1
with gcc 3.2
(MinGW) in Windows, so it is included. Treat with caution.
* spider.cca: A tiny function that joins each WA score to the
corresponding LC score in cca/rda plots.
Minor changes:
* Checked with R-1.7.0 (unstable devel version, snapshot
2003-02-05) and corrected as needed.
* Added symmetric scaling (=3) to alternitves in cca and rda.
* Minor improvements and bug fixes in vegemite.
Version 1.4-1 (Logan)
Based on devel version 1.5-10.
New features:
* Permutation tests added to envfit (vectorfit, factorfit) and
procrustes (called protest).
* All permutation tests take now argument strata: If given,
permutations made only within strats. Concerns anosim, anova.cca
(permutest.cca), mantel, envfit (factorfit, envfit) and protest
(in procrustes).
* fitted.procrustes(): A new function.
* rarefy(): A new function for rarefaction species richness.
Other changes:
* cca.default(): Handles now NULL matrices X and Z: skips them.
* cca.formula(): Knows now cca(X ~ 1) and permforms unconstrained
CA, and cca(X ~ ., data) and perfors CCA using all variables in
`data' as constraints. Has now na.action=na.fail so that cca
crashes more informatively (used to crash mysteriously). A more
graceful na.action may come. Assignment "=" corrected to the true
blue "<-" (used to crash in old R).
* envfit (factorfit, vectorfit): Use `scores' now. Various
bugfixes and should work now with single factors or vectors.
* plot.envfit(): Options `choices' works now like documented.
* ordisurf(): A new name to surf() to avoid name clash with
labdsv.
* procrustes(): added option `symmetric' for symmetric rotation
and goodness of fit statistic.
* plot.procrustes(): Keyword `axes' changed to `choices' to be
consistent with other functions.
* vegemite(): A new name to vegetab. The name was chosen because
the output is so compact (and to avoid confusion with function
vegtab in labdsv). Function checks now that the cover codes fit in
one column and splits the output if it does not fit into used page
width.
Version 1.4-0
Based on devel version 1.3-19.
New functions:
* `anosim': Analysis of Similarities.
* `cca': [Partial] [Constrained] Correspondence Analysis. This is
an alternative to Dray's `CAIV' (package `CoCoAn'). Differences
include formula interface, WA scores in addition to LC scores,
partial CCA, residual CA after constraints, algorithm based on svd
(and much faster), more standard support functions (`plot',
`summary', `print').
* `anova.cca' and `permutest.cca': Permutation tests for `cca'.
* `envfit': A new wrapper function which calls either `vectorfit'
or new `factorfit' function, and `plot.envfit' for easier display
of results in graphs.
* `mantel': Mantel test for two
dissimilarity matrices.
* `scores': A generic function to extract scores from `cca',
`decorana' or even from some common ordination methods in R.
* `surf': Surface fitting for ordination, intended as an
alternative to `vectorfit'.
* `vegetab': Prints a compact, ordered vegetation table, together
with `coverscale' to transform data into compact format.
Other changes:
* `downweight': directly callable instead of being embedded in
`decorana'. Now it can be used with other downweighting thresholds
or with other CA's than `decorana', e.g., `cca', `ca', `CAIV'.
* `plot.decorana': new keyword `type', removed keyword `tol'.
* `vectorfit': centres now ordination before fitting, so it will
work with other ordinations than MDS.
* Mainly user invisible changes in `plot.decorana',
`plot.procrustes', `postMDS', `print.summary.procrustes',
`vectorfit', `procrustes'. One of the main changes was that
`eqscplot' of `MASS' library was replaced with `plot(..., asp=1)'
of standard R (but this is not S-plus compliant).
* Edited help files. More examples run from help files.
* Tried to improve LabDSV compliance, see
http://labdsv.nr.usu.edu/.
* Checked with R-1.5.0. However, `log' in diversity not changed
into `logb' (yet), although base is specified, because `logb'
won't work in R pre-1.5.0.
Version 1.2-1
* Checked with R-1.4-0 (frozen snapshot) tools: Documentation and
method consistency corrected (summary.decorana.R, decorana.Rd),
and file permissions changed.
* decorana.Rd: removed dependence on package `multiv' in the
example, since it is not a `recommended package' -- replaced with
`decorana(..., ira=1)'.
Version 1.2-0
Based on experimental version 1.1-6. Doesn't include read.cep() of the
experimental version, because the function is certainly unportable to
all platforms (works on Red Hat Linux 7.1, but perhaps nowhere else).
Major new feature:
* decorana(): R port of Mark Hill's DECORANA for Detrended
correspondence analysis, with methods print(), summary() and
plot().
Minor changes:
* postMDS(): finds isoMDS scores and doesn't destroy list
structure.
* vectorfit: finds isoMDS scores and decorana row scores (which
are uncentred, though).
* Proof reading in documentation.
Version 1.0-3
*`plot.procrustes': selection of axes plotted, improved scaling in
graphics, control of axis labels.
* `postMDS': added plot to demonstrate half-change scaling.
* `procrustes': target can now have lower dimensionality than
rotated matrix.
Version 1.0-2
* Fixed bugs in documentation.
* `plot.procrustes': Fixed passing `...' to graphics functions.
Version 1.0-1
* The first public version.