https://github.com/cran/spatstat
Revision 8f171b836eb1993abbdc454a2780c057b63d910e authored by Adrian Baddeley on 01 March 2013, 12:28:29 UTC, committed by cran-robot on 01 March 2013, 12:28:29 UTC
1 parent 4712eff
Tip revision: 8f171b836eb1993abbdc454a2780c057b63d910e authored by Adrian Baddeley on 01 March 2013, 12:28:29 UTC
version 1.31-1
version 1.31-1
Tip revision: 8f171b8
NEWS
CHANGES IN spatstat VERSION 1.31-1
OVERVIEW
o We thank Marcelino de la Cruz, Daniel Esser, Jason Goldstick,
Abdollah Jalilian, Ege Rubak and Fabrice Vinatier for contributions.
o Nonparametric estimation and tests for
point patterns in a linear network.
o More support for 'layered' objects.
o Find clumps in a point pattern.
o Connected component interaction model.
o Improvements to interactive plots.
o Visual debugger for Metropolis-Hastings algorithm.
o Bug fix in Metropolis-Hastings simulation of Geyer process.
o Faster Metropolis-Hastings simulation.
o Faster computation of 'envelope', 'fv' and 'fasp' objects.
o Improvements and bug fixes.
NEW FUNCTIONS
o connected.ppp
Find clumps in a point pattern.
o kstest.lpp, kstest.lppm
The spatial Kolmogorov-Smirnov test can now be applied
to point patterns on a linear network (class 'lpp') and
point processes on a linear network (class 'lppm').
o bermantest.lpp, bermantest.lppm
Berman's Z1 and Z2 tests can now be applied to point patterns
on a linear network (class 'lpp') and point processes on a linear
network (class 'lppm').
o rhohat.lpp, rhohat.lppm
Nonparametric estimation of the dependence of a point pattern
on a spatial covariate: 'rhohat' now applies to
objects of class 'lpp' and 'lppm'.
o intensity.lpp
Empirical intensity of a point pattern on a linear network.
o as.function.rhohat
Converts a 'rhohat' object to a function,
with extrapolation beyond the endpoints.
o [.layered
Subset operator for layered objects.
o shift, rotate, affine, rescale, reflect, flipxy, scalardilate
These geometrical transformations now work for 'layered' objects.
o iplot.layered
Interactive plotting for 'layered' objects.
o as.owin.layered
Method for as.owin for layered objects.
o [.owin
Subset operator for windows, equivalent to intersect.owin.
o rcellnumber
Generates random integers for the Baddeley-Silverman counterexample.
o is.lpp
Tests whether an object is a point pattern on a linear network.
o is.stationary.lppm, is.poisson.lppm
New methods for is.stationary and is.poisson for class 'lppm'
o sessionLibs
Print library names and version numbers (for use in Sweave scripts)
SIGNIFICANT USER-VISIBLE CHANGES
o iplot
iplot is now generic, with methods for 'ppp', 'layered' and 'default'.
iplot methods now support zoom and pan navigation.
o rmh.default
New argument 'snoop' allows the user to activate a visual debugger
for the Metropolis-Hastings algorithm.
o connected
connected() is now generic, with methods for 'im', 'owin' and 'ppp'.
o alltypes
Now works for lpp objects
o rlabel
Now works for lpp, pp3, ppx objects
o plot.kstest
Can now perform P-P and Q-Q plots as well.
o plot.fasp
New argument 'samey' controls whether all panels
have the same y limits.
o plot.fasp
Changed default value of 'samex'.
o Objects of class 'envelope', 'fv' and 'fasp'
Reduced computation time and storage required for these objects.
o pcfmodel.kppm
Improved calculation.
o plot.fv
Improved collision-avoidance algorithm
(for avoiding overlaps between curves and legend)
o ppm
Improved error handling
o envelope
All methods for 'envelope' now handle fun=NULL
o setminus.owin
Better handling of the case where both arguments are rectangles.
o rmh
Simulation has been further accelerated.
o lppm
Accelerated.
o vcov.ppm
Accelerated.
o marktable
Accelerated.
o Triplets() interaction
Accelerated.
o alltypes
Accelerated when envelope=TRUE.
BUG FIXES
o rmh
Simulation of the Geyer saturation process was incorrect.
[Bug introduced in previous version, spatstat 1.31-0.]
Fixed.
o rmh
Simulation of the Geyer saturation process was
incorrectly initialised, so that the results of a short run
(i.e. small value of 'nrep') were incorrect,
while long runs were correct.
[Bug introduced in spatstat 1.17-0, october 2009.]
Fixed.
o ppm
Objects fitted with use.gam=TRUE caused fatal errors in
various functions including print, summary, vcov and model.frame.
Spotted by Jason Goldstick.
Fixed.
o lpp, runiflpp, rpoislpp
Empty point patterns caused an error.
Fixed.
o rmh.default
Crashed for hybrid models, with message
'Attempt to apply non-function'.
Spotted by Ege Rubak.
Fixed.
o relrisk
Crashed when 'at="points"' for a multitype pattern
with more than 2 types.
Spotted by Marcelino de la Cruz.
Fixed.
o erosion.owin, dilation.psp, border
Ignored the arguments "..." in some cases
(namely when the window was polygonal and 'gpclib' was disabled).
Fixed.
o rsyst, rcell
Did not correctly handle the argument 'dx'.
Spotted by Fabrice Vinatier.
Fixed.
o correction="trans"
Various functions such as Kest
no longer recognised 'correction = "trans"'.
Fixed.
o istat
Crashed with an error message about envelopes.
Fixed.
o summary.ppm, print.ppm
p-values which were exactly equal to zero were reported as NA.
Fixed.
o [.im
Crashed if the intersection consisted of
a single row or column of pixels.
Fixed.
o plot.im
Sometimes incorrectly displayed an image consisting of
a single row or column of pixels.
Fixed.
o plot.layered
The plot region was determined by the first layer,
so that objects in subsequent layers could sometimes
fall outside the plot region.
Fixed.
o transect.im
If the arguments 'from' and 'to' were numeric vectors of length 2,
the result was garbled.
Fixed.
o Inhomogeneous K functions and pair correlation functions
[Kinhom, pcfinhom, Kcross.inhom, Kdot.inhom, pcfcross.inhom, etc.]
These functions reported an error 'lambda is not a vector'
if the intensity argument lambda was computed using density(, at="points").
Fixed.
o rlabel
Did not accept a point pattern with a hyperframe of marks.
Fixed.
o alltypes
Crashed when envelope=TRUE if the summary function 'fun'
did not have default values for the marks i and j.
Fixed.
o Kres, Gres, psst, psstA
Ignored the unit of length.
Fixed.
CHANGES IN spatstat VERSION 1.31-0
OVERVIEW
o We thank Frederic Lavancier and Ege Rubak for contributions.
o Major bug fix in simulation of area-interaction process.
o Metropolis-Hastings simulations accelerated.
o Rounding of spatial coordinates
o clmfires dataset corrected.
o Bug fixes and minor improvements.
NEW FUNCTIONS
o round.ppp
Round the spatial coordinates of a point pattern
to a specified number of decimal places.
o rounding
Determine whether a dataset has been rounded.
SIGNIFICANT USER-VISIBLE CHANGES
o rmh
Simulation of the following models has been accelerated:
areaint, dgs, diggra, fiksel, geyer, hardcore, lennard, multihard,
strauss, straush, straussm, strausshm.
o rmh
The transition history of the simulation (which is saved if 'track=TRUE')
now also contains the value of the Hastings ratio for each proposal.
o clmfires
The clmfires dataset has been modified to remove errors and inconsistencies.
o plot.linim
Appearance of the plot has been improved, when style='width'.
o summary.ppm
Now reports whether the spatial coordinates have been rounded.
o dclf.test, mad.test
The range of distance values ('rinterval') used in the test
is now printed in the test output, and is saved as an attribute.
BUG FIXES
o rmh
Simulation of the Area-Interaction model was completely incorrect.
Spotted by Frederic Lavancier.
The bug was introduced in spatstat version 1.23-6 or later.
Fixed.
o dclf.test
The test statistic was incorrectly scaled (by a few percent).
This did not affect the p-value of the test.
Fixed.
o ppx
If argument 'coord.type' was missing, various errors occurred:
a crash may have occurred, or the results may have depended
on the storage type of the data.
Spotted by Ege Rubak.
Fixed.
o plot.ppx
Crashed for 1-dimensional point patterns.
Spotted by Ege Rubak.
Fixed.
CHANGES IN spatstat VERSION 1.30-0
OVERVIEW
o We thank Jorge Mateu, Andrew Bevan, Olivier Flores, Marie-Colette van Lieshout, Nicolas Picard and Ege Rubak for contributions.
o The spatstat manual now exceeds 1000 pages.
o Hybrids of point process models.
o Five new datasets
o Second order composite likelihood method for kppm.
o Inhomogeneous F, G and J functions.
o Delaunay graph distance
o Fixed serious bug in 'lppm' for marked patterns.
o bug fix in some calculations for Geyer model
o Improvements to linear networks code
o Pixel images can now be displayed with a logarithmic colour map.
o spatstat now formally 'Depends' on the R core package 'grDevices'
o miscellaneous improvements and bug fixes
NEW DATASETS
o clmfires
Forest fires in Castilla-La Mancha
o gordon
People sitting on the grass in Gordon Square, London
o hyytiala
Mixed forest in Hyytiala, Finland (marked by species)
o paracou
Kimboto trees in Paracou, French Guiana (marked as adult/juvenile)
o waka
Trees in Waka national park (marked with diameters)
NEW FUNCTIONS
o Hybrid
The hybrid of several point process interactions
[Joint research with Jorge Mateu and Andrew Bevan]
o is.hybrid
Recognise a hybrid interaction or hybrid point process model.
o Finhom, Ginhom, Jinhom
Inhomogeneous versions of the F, G and J functions
[Thanks to Marie-Colette van Lieshout]
o delaunay.distance
Graph distance in the Delaunay triangulation.
o distcdf
Cumulative distribution function of the distance between
two independent random points in a given window.
o bw.frac
Bandwidth selection based on window geometry
o shortside.owin, sidelengths.owin
Side lengths of (enclosing rectangle of) a window
SIGNIFICANT USER-VISIBLE CHANGES
o ppm
Can now fit models with 'hybrid' interactions
[Joint research with Jorge Mateu and Andrew Bevan]
o kppm
Now has the option of fitting models using Guan's (2006)
second order composite likelihood.
o envelope.lpp
Now handles multitype point patterns.
o envelope.envelope
New argument 'transform' allows the user to apply a transformation
to previously-computed summary functions.
o runifpointOnLines, rpoisppOnLines, runiflpp, rpoislpp
Can now generate multitype point patterns.
o rmhmodel, rmh, simulate.ppm
Now handle point process models with 'hybrid' interactions.
o kppm
Accelerated, and more reliable, due to better choice of starting values
in the optimisation procedure.
o kppm
The internal format of kppm objects has changed.
o minimum contrast estimation
Error messages from the optimising function 'optim' are now trapped
and handled.
o rhohat
This command is now generic, with methods for ppp, quad, and ppm.
o raster.x, raster.y, raster.xy
These functions have a new argument 'drop'
o summary.ppm
Improved behaviour when the model covariates are a data frame.
o progressreport
Output improved.
o second order summary functions
(Kest, Lest, Kinhom, pcf.ppp, Kdot, Kcross, Ldot etc etc)
These functions now accept correction="translation" as an
alternative to correction = "translate", for consistency.
o plot.im
New argument 'log' allows colour map to be equally spaced on a log scale.
o as.owin.ppm, as.owin.kppm
New argument 'from' allows the user to extract the spatial window
of the point data (from="points") or the covariate images (from="covariates")
o dclf.test, mad.test
The rule for handling tied values of the test statistic has been changed.
The tied values are now randomly ordered to obtain a randomised integer rank.
o with.fv
New argument 'enclos' allows evaluation in other environments
BUG FIXES
o lppm
For multitype patterns, the fitted model was completely incorrect
due to an error in constructing the quadrature scheme.
Fixed.
o Geyer
For point process models with the 'Geyer' interaction,
vcov.ppm() and suffstat() sometimes gave incorrect answers.
[Spotted by Ege Rubak.]
Fixed.
o as.im.im
Did not correctly handle factor-valued images
if one of the arguments 'dimyx', 'eps', 'xy' was given.
Fixed.
o envelope.lppm
Crashed if the model was multitype.
Fixed.
o lpp
Did not handle empty patterns.
Fixed.
o density.ppp
If 'sigma' was a bandwidth selection function such as bw.scott()
which returned a numeric vector of length 2, a warning message was issued,
and the smoothing bandwidth was erroneously taken to be
the first element of the vector.
Fixed.
o Fest, Jcross, Jdot, Jmulti
If these functions were computed using correction = 'rs',
plotting them would sometimes give an error,
with the message "no finite x/y limits".
Fixed.
o pcfmodel.kppm
For models with clusters="VarGamma"
the value of the pcf at distance r=0 was given as NaN.
Fixed.
o vcov.ppm
Result was incorrect in rare cases, due to numerical rounding effects.
Fixed.
o rLGCP, simulate.kppm
For models fitted to point patterns in an irregular window,
simulation sometimes failed, with a message that the image 'mu'
did not cover the simulation window.
(Spotted by George Limitsios.)
Fixed.
o rLGCP, simulate.kppm
Crashed sometimes with an error about unequal x and y steps
(from 'GaussRF').
Fixed.
CHANGES IN spatstat VERSION 1.29-0
OVERVIEW
o We thank Colin Beale, Li Haitao, Frederic Lavancier, Erika Mudrak and Ege Rubak for contributions.
o random sequential packing
o Allard-Fraley estimator
o method for pooling several quadrat tests
o better control over dummy points in ppm
o more support for data on a linear network
o nearest neighbour map
o changes to subsetting of images
o improvements and bug fixes
NEW FUNCTIONS
o clusterset
Allard-Fraley estimator of high-density features in a point pattern
o pool.quadrattest
Pool several quadrat tests
o nnfun
Nearest-neighbour map of a point pattern or a line segment pattern
o as.ppm
Converts various kinds of objects to ppm
o crossdist.lpp
Shortest-path distances between pairs of points in a linear network
o nobs.lppm
Method for 'nobs' for lppm objects.
o as.linim
Converts various kinds of objects to 'linim'
o model.images.slrm
Method for model.images for slrm objects
o rotate.im
Rotate a pixel image
SIGNIFICANT USER-VISIBLE CHANGES
o "[.im" and "[<-.im"
New argument 'j' allows any type of matrix indexing to be used.
o "[.im"
Default behaviour changed in the case of a rectangular subset.
New argument 'rescue' can be set to TRUE to reinstate previous behaviour.
o rSSI
Performs 'Random Sequential Packing' if n=Inf.
o ppm
New argument 'eps' determines the spacing between dummy points.
(also works for related functions quadscheme, default.dummy, ...)
o fitted.ppm, predict.ppm
Argument 'new.coef' specifies a vector of parameter values
to replace the fitted coefficients of the model.
o lppm
Stepwise model selection using step() now works for lppm objects.
o vcov.slrm
Can now calculate correlation matrix or Fisher information matrix
as well as variance-covariance matrix.
o eval.fv
Improved behaviour when plotted.
o "[.fv"
Improved behaviour when plotted.
o lohboot
When the result is plotted, the confidence limits are now shaded.
o lohboot
New argument 'global' allows global (simultaneous) confidence bands
instead of pointwise confidence intervals.
o vcov.ppm
Accelerated by 30% in some cases.
o quadrat.test.splitppp
The result is now a single object of class 'quadrattest'
o progressreport
Improved output (also affects many functions which print progress reports)
o Full redwood data (redwoodfull)
Plot function redwoodfull.extra$plotit has been slightly improved.
o nncross
This function is now generic, with methods for 'ppp' and 'default'.
o distfun
The internal format of objects of class 'distfun' has been changed.
o duplicated.ppp, unique.ppp
New argument 'rule' allows behaviour to be consistent with package 'deldir'
BUG FIXES
o bdist.tiles
Values were incorrect in some cases due to numerical error.
(Spotted by Erika Mudrak.)
Fixed.
o vcov.ppm, suffstat
These functions sometimes gave incorrect values
for marked point process models.
Fixed.
o simulate.ppm, predict.ppm
Did not correctly handle the 'window' argument. (Spotted by Li Haitao).
Fixed.
o smooth.ppp, markmean
If sigma was very small, strange values were produced, due to
numerical underflow. (Spotted by Colin Beale).
Fixed.
o MultiHard, MultiStrauss, MultiStraussHard
Crashed if the data point pattern was empty.
(Spotted by Ege Rubak).
Fixed.
o vcov.ppm
Crashed sporadically, with multitype interactions.
(Spotted by Ege Rubak).
Fixed.
o rStrauss, rHardcore, rStraussHard, rDiggleGratton, rDGS
If the simulated pattern was empty, these functions would either
crash, or return a pattern containing 1 point.
(Spotted by Frederic Lavancier).
Fixed.
o model.matrix.slrm
Crashed if the model was fitted using split pixels.
Fixed.
o residuals.ppm, diagnose.ppm
Did not always correctly handle models that included offset terms.
Fixed.
o project.ppm
When a model was projected by project.ppm or by ppm(project=TRUE),
the edge corrections used the projected models were sometimes different
from the edge corrections in the original model, so that the
projected and unprojected models were not comparable.
Fixed.
o plot.listof, plot.splitppp
Crashed sometimes due to a scoping problem.
Fixed.
o dclf.test, mad.test
Crashed if any of the function values were infinite or NaN.
Fixed.
o psstA
Default plot did not show the horizontal line at y=0
corresponding to a perfect fit.
Fixed.
o vcov.ppm
names attribute was spelt incorrectly in some cases.
Fixed.
CHANGES IN spatstat VERSION 1.28-2
OVERVIEW
o We thank Thomas Bendtsen, Ya-Mei Chang, Daniel Esser,
Robert John-Chandran, Ege Rubak and Yong Song for contributions.
o New code for Partial Residual Plots and Added Variable Plots.
o maximum profile pseudolikelihood computations vastly accelerated.
o New dataset: cells in gastric mucosa
o now possible to capture every k-th state of Metropolis-Hastings algorithm.
o size of 'ppm' objects reduced.
o scope of 'intensity.ppm' extended.
o quadrat.test can now perform Monte Carlo tests and one/two-sided tests
o improvements to 'plot.fv'
o improvement to 'rescale'
o some datasets reorganised.
o numerous bug fixes
NEW DATASET
o mucosa
Cells in gastric mucosa
Kindly contributed by Dr Thomas Bendtsen
NEW FUNCTIONS
o parres
Partial residual plots for spatial point process models.
A diagnostic for the form of a covariate effect.
o addvar
Added variable plots for spatial point process models.
A diagnostic for the existence of a covariate effect.
SIGNIFICANT USER-VISIBLE CHANGES
o profilepl
Accelerated (typically by a factor of 5).
o rmh, rmhcontrol
It is now possible to save every k-th iteration of the Metropolis-Hastings
algorithm. The arguments 'nsave' and 'nburn' may be given to rmh
or to rmhcontrol. They specify that the point pattern will be saved
every 'nsave' iterations, after an initial burn-in of 'nburn' iterations.
o simulate.ppm
New argument 'singlerun' determines whether the simulated patterns
are generated using independent runs of the Metropolis-Hastings algorithm
or are obtained by performing one long run of the algorithm and saving
every k-th iteration.
o exactMPLEstrauss
New argument 'project' determines whether the parameter gamma
is constrained to lie in [0,1].
o intensity.ppm
Now works for stationary point process models with the interactions
DiggleGratton, DiggleGatesStibbard, Fiksel, PairPiece and Softcore.
o plot.fv
Improved algorithm for avoiding collisions between graphics and legend.
o plot.fv
New argument 'log' allows plotting on logarithmic axes.
o envelope
Can now calculate an estimate of the true significance level
of the "wrong" test (which declares the observed summary function
to be significant if it lies outside the pointwise critical boundary
anywhere). Controlled by new argument 'do.pwrong'.
o quadrat.test
New argument 'alternative' allows choice of alternative hypothesis
and returns one-sided or two-sided p-values as appropriate.
o quadrat.test
Can now perform Monte Carlo test as well
(for use in small samples where the chi^2 approximation is inaccurate)
o Softcore
Improved numerical stability.
New argument 'sigma0' for manual control over rescaling.
o rescale
If scale argument 's' is missing, then the data are rescaled
to native units. For example if the current unit is 0.1 metres,
coordinates will be re-expressed in metres.
o psst
Extra argument 'verbose=TRUE'
o is.subset.owin
Accelerated for polygonal windows
o rmh.default
'track' is no longer a formal argument of rmh.default; it is now a
parameter of rmhcontrol. However there is no change in usage:
the argument 'track' can still be given to rmh.default.
o clf.test
Has been renamed 'dclf.test'
to give proper attribution to Peter Diggle.
o betacells
This dataset has been restructured. The vector of cell profile areas,
formerly given by betacells.extra$area, has now been included
as a column of marks in the point pattern 'betacells'.
o ants
The function ants.extra$plot() has been renamed plotit()
for conformity with other datasets.
o redwoodfull
The function redwoodfull.extra$plot() has been renamed plotit()
for conformity with other datasets.
o nbfires
For conformity with other datasets, there is now an object nbfires.extra
BUG FIXES
o ripras
Expansion factor was incorrect in the rectangular case.
Fixed.
o Triplets
Crashed sometimes with error "dim(X) must have positive length".
Fixed.
o affine.im
Crashed in the case of a diagonal transformation matrix!
Spotted by Ege Rubak.
Fixed.
o envelope.envelope
Ignored the argument 'global'.
Fixed.
o MultiStraussHard
The printed output showed the hardcore radii as NULL.
Spotted by Ege Rubak.
Fixed.
o "[.psp"
Crashed if the data were generated by rpoisline().
Spotted by Marcelino de la Cruz.
Fixed.
o plot.linim
If style="colour", the main title was always "x".
Fixed.
o plot.ppx
Setting add=TRUE did not prevent the domain being plotted.
Fixed.
o rmh
Crashed if x.start was an empty point pattern.
Spotted by Ege Rubak.
Fixed.
o as.ppp.data.frame
Crashed if any points lay outside the window.
Spotted by Ege Rubak.
Fixed.
o Ripley isotropic edge correction
Divide-by-zero error in rare cases.
Spotted by Daniel Esser.
Fixed.
o summary functions
For many of the summary functions (e.g. Kest, pcf),
the result of saving the object to disc was an enormous file.
Spotted by Robert John-Chandran.
Fixed.
o pcf.fv
Default plot was wrongly coloured.
Fixed.
CHANGES IN spatstat VERSION 1.28-1
OVERVIEW
o We thank Ege Rubak, Gopal Nair, Jens Oehlschlaegel and Mike Zamboni
for contributions.
o New approximation to the intensity of a fitted Gibbs model.
o Minor improvements and bug fixes
o spatstat now 'Suggests' the package 'gsl'
NEW FUNCTIONS
o intensity, intensity.ppp, intensity.ppm
Calculate the intensity of a dataset or fitted model.
Includes new approximation to the intensity of a fitted Gibbs model
o LambertW
Lambert's W-function
SIGNIFICANT USER-VISIBLE CHANGES
o envelope
Improved plot labels for envelopes that were generated using
the 'transform' argument.
o plot.fv
Improved algorithm for collision detection.
o plot.im
Now returns the colour map used.
o plot.listof, plot.splitppp
Slight change to handling of plot.begin and plot.end
o square
Now accepts vectors of length 2
o plot.fii
Increased resolution of the plot obtained from plot(fitin(ppm(...)))
o image.listof
If equal.ribbon=TRUE, the colour ribbon will no longer be
displayed repeatedly for each panel,
but will now be plotted only once, at the right hand side of the plot array.
BUG FIXES
o vcov.ppm
Results were sometimes incorrect for a Gibbs model with non-trivial trend.
Spotted by Ege Rubak.
Fixed.
o nncross
In rare cases the results could be slightly incorrect.
Spotted by Jens Oehlschlaegel.
Fixed.
o plot.fv
When add=TRUE, the x limits were sometimes truncated.
Spotted by Mike Zamboni.
Fixed.
o plot.im
Labels for the tick marks on the colour ribbon
were sometimes ridiculous, e.g. "2.00000001".
Fixed.
CHANGES IN spatstat VERSION 1.28-0
OVERVIEW
o We thank Farzaneh Safavimanesh, Andrew Hardegen and Tom Lawrence
for contributions.
o Improvements to 3D summary functions.
o A multidimensional point pattern (ppx) can now have 'local' coordinates
as well as spatial and temporal coordinates and marks.
o Changed format for point patterns on a linear network (lpp).
Changes are backward compatible. Many computations run faster.
o More support for fitted cluster models (kppm).
o split method for multidimensional point patterns (ppx)
and point patterns on a linear network (lpp).
o Fixed bug causing errors in plot.im
o Miscellaneous improvements and bug fixes
NEW FUNCTIONS
o exactMPLEstrauss
Fits the stationary Strauss point process model
using an exact maximum pseudolikelihood technique.
This is mainly intended for technical investigation of algorithms.
o split.ppx
Method for 'split' for multidimensional point patterns (class 'ppx').
This also works for point patterns on a linear network (class 'lpp').
o model.images
This function is now generic, with methods for classes ppm, kppm, lppm
o model.frame, model.matrix
These generic functions now have methods for classes kppm, lppm
o as.owin.kppm, as.owin.lppm
New methods for 'as.owin' for objects of class kppm, lppm
o as.linnet.lppm
Extracts the linear network in which a point process model was fitted.
SIGNIFICANT USER-VISIBLE CHANGES
o class 'ppx'
An object of class 'ppx' may now include 'local' coordinates
as well as 'spatial' and 'temporal' coordinates, and marks.
o ppx
Arguments have changed.
o class 'lpp'
The internal format of lpp objects has been extended
(but is backward-compatible). Many computations run faster.
To convert an object to the new format: X <- lpp(as.ppp(X), as.linnet(X)).
o F3est
Calculation of theoretical Poisson curve ('theo') has changed,
and is now controlled by the argument 'sphere'.
o rmh, rmhstart
The initial state ('start') can now be missing or null.
o im, as.im
The pixel coordinates in an image object are now generated
more accurately. This avoids a numerical error in plot.im.
o eval.fv, eval.fasp
Evaluation is now applied only to columns that contain
values of the function itself (rather than values of the
derivative, hazard rate, etc). This is controlled by the
new argument 'dotonly'.
o spatstat.options
New option 'nvoxel'
o quad.ppm
Now accepts kppm objects.
o str
This generic function (for inspecting the internal structure of an object)
now produces sensible output for objects of class 'hyperframe', 'ppx', 'lpp'
o ppx, coords.ppx, coords<-.ppx
The arguments to these functions have changed.
o lgcp.estK, Kmodel
Computation can be greatly accelerated
by setting spatstat.options(fastK.lgcp=TRUE).
o G3est
Computation accelerated.
o envelope
Computation slightly accelerated.
o spatstat.options
New option 'fastK.lgcp'
BUG FIXES
o nndist.psp
Caused an error if length(k) > 1.
Fixed.
o plot.im
Sometimes reported an error
"useRaster=TRUE can only be used with a regular grid."
This was due to numerical rounding effects on the coordinates
of a pixel image.
Fixed.
o plot.fv
If a formula was used to specify the plot,
the names of variables in the formula were sometimes
incorrectly matched to *functions*.
Spotted by Farzaneh Safavimanesh.
Fixed.
o F3est
Took a very long time if the containing box was very flat,
due to the default value of 'vside'.
Fixed.
o rmh, rmhmodel
An erroneous warning about 'outdated format of rmhmodel object'
sometimes occurred.
Fixed.
o marks<-.ppx
Names of result were incorrect. Fixed.
o hyperframe class
Various minor bug fixes.
CHANGES IN spatstat VERSION 1.27-0
OVERVIEW
o Variance estimates are now available for all Gibbs point process models.
o Cressie-Loosmore-Ford test implemented
o plot.fv now avoids collisions between the legend and the graphics.
o Extension to predict.ppm
o Improvements to envelopes and multitype summary functions.
o Line transects of a pixel image.
o Changes to defaults in Metropolis-Hastings simulations.
o More geometrical operations
o Bug fixes.
o We thank Aruna Jammalamadaka for contributions.
NEW FUNCTIONS
o clf.test
Perform the Cressie (1991)/ Loosmore and Ford (2006) test of CSR
(or another model)
o mad.test
Perform the Maximum Absolute Deviation test of CSR (or another model).
o convolve.im
Compute convolution of pixel images.
o Kmulti.inhom
Counterpart of 'Kmulti' for spatially-varying intensity.
o rmhexpand
Specify a simulation window, or a rule for expanding the
simulation window, in Metropolis-Hastings simulation (rmh)
o transect.im
Extract pixel values along a line transect.
o affine.im
Apply an affine transformation to a pixel image.
o scalardilate
Perform scalar dilation of a geometrical object
relative to a specified origin.
o reflect
Reflect a geometrical object through the origin.
o "[.lpp", "[.ppx"
Subset operators for the classes "lpp" (point pattern on
linear network) and "ppx" (multidimensional space-time
point pattern).
o is.rectangle, is.polygonal, is.mask
Determine whether a window w is a rectangle, a domain with polygonal
boundaries, or a binary pixel mask.
o has.offset
Determines whether a fitted model object (of any kind)
has an offset.
SIGNIFICANT USER-VISIBLE CHANGES
o predict.ppm
This function can now calculate the conditional intensity of a model
relative to any point pattern X (not just the original data pattern).
o vcov.ppm
This function now handles all Gibbs point process models.
o plot.fv
Collisions between the legend box and the graphics are now detected
and avoided.
o rmh.ppm, rmh.default, simulate.ppm, qqplot.ppm, envelope.ppm
These functions now have slightly different default behaviour
because of changes to the handling of arguments to 'rmhcontrol'.
o rmhcontrol
The default value of the parameters 'periodic' and 'expand'
has changed.
o rmhcontrol
The parameter 'expand' can now be in any format
acceptable to rmhexpand().
o rmh.ppm, rmh.default, simulate.ppm
Any 'rmhcontrol' parameter can now be given directly
as an argument to rmh.ppm, rmh.default or simulate.ppm.
o Kmulti, Gmulti, Jmulti
The arguments I, J can now be any kind of subset index
or can be functions that yield a subset index.
o envelope.envelope
In envelope(E, fun=NULL) if E does not contain simulated
summary functions, but does contain simulated point patterns,
then 'fun' now defaults to Kest, instead of flagging an error.
o print.ppp, summary.ppp
If the point pattern x was generated by Metropolis-Hastings
simulation using 'rmh', then print(x) and summary(x) show
information about the simulation parameters.
o print.ppm
Standard errors for the parameter estimates,
and confidence intervals for the parameters,
can now be printed for all Gibbs models
(but are printed only for Poisson models by default).
o eval.im
Images with incompatible dimensions are now resampled
to make them compatible (if harmonize=TRUE).
o spatstat.options
New option 'print.ppm.SE' controls whether standard errors
and confidence intervals are printed for all Gibbs models,
for Poisson models only, or are never printed.
o inside.owin
Now accepts the form list(x,y) for the first argument.
o image.listof
New argument 'equal.ribbon' allows several images to be plotted
with the same colour map.
o is.subset.owin
Improved accuracy in marginal cases.
o expand.owin
Functionality extended to handle all types of expansion rule.
o default.rmhcontrol, default.expand
These functions now work with models of class 'rmhmodel'
as well as 'ppm'
o print.rmhcontrol
Output improved.
BUG FIXES
o linearK, linearKinhom
If any data points were located exactly at a vertex of the
linear network, the weights for Ang's correction were incorrect,
due to numerical error. This sometimes produced infinite
or NA values of the linear K function.
Fixed.
o predict.ppm
In some cases, predict.ppm(type="cif") generated a spurious warning
that "number of rows of result is not a multiple of vector length."
Fixed.
o crossing.psp
Results were sometimes incorrect due to numerical rounding error
associated with GCC bug #323.
Fixed.
o MultiHard, MultiStrauss, MultiStraussHard
If the mark values contained non-alphanumeric characters,
the names of the interaction coefficients in coef(ppm(...))
were sometimes garbled.
Fixed.
o profilepl
For edge corrections other than the border correction,
an error message about 'rbord' would sometimes occur.
Fixed.
o is.marked, is.multitype
These functions gave the wrong answer for 'lpp' objects.
Fixed.
o marks<-.lpp, marks<-.ppx
Format of result was garbled if new columns of marks were added.
Fixed.
o reach.rmhmodel
Gave the wrong answer for Geyer and BadGey models.
Fixed.
o envelope.envelope
Ignored the argument 'savefuns'.
Fixed.
o BadGey
Sometimes wrongly asserted that the parameter 'sat' was invalid.
Occurred only in ppm(project=TRUE).
Fixed.
CHANGES IN spatstat VERSION 1.26-1
OVERVIEW
o Variance-covariance matrix for Gibbs point process models.
o Bootstrap confidence bands for pair correlation function and K function.
o Bug fix in scan test.
o Area-interaction model accelerated.
o we thank Jean-Francois Coeurjolly and Ege Rubak for contributions.
NEW FUNCTIONS
o lohboot
Computes bootstrap confidence bands for pair correlation function
and K function using Loh's (2008) mark bootstrap.
SIGNIFICANT USER-VISIBLE CHANGES
o vcov.ppm
Now works for all Gibbs point process models, thanks to new code (and theory)
from Jean-Francois Coeurjolly and Ege Rubak
o AreaInter
Computations related to the area-interaction point process
(ppm, predict.ppm, residuals.ppm, diagnose.ppm, qqplot.ppm)
have been accelerated.
BUG FIXES
o scan.test
Results were sometimes incorrect due to numerical instability
(a 'Gibbs phenomenon').
Fixed.
CHANGES IN spatstat VERSION 1.26-0
OVERVIEW
o We thank Jens Oehlschlaegel for contributions.
o Further substantial acceleration of spatstat functions.
o Workaround for bug in RandomFields package.
o Numerous modifications to internal code.
NEW FUNCTIONS
o RandomFieldsSafe
There is a bug in the package 'RandomFields' (version <= 2.0.54)
which causes a crash to occur, in the development version of R
but not in R 2.15.0. To avoid crashing spatstat, we have written
the temporary, undocumented function RandomFieldsSafe()
which returns TRUE if it is safe to use the RandomFields package.
Examples in the spatstat help files for kppm, lgcp.estK, lgcp.estpcf
and rLGCP are only executed if RandomFieldsSafe() returns TRUE.
SIGNIFICANT USER-VISIBLE CHANGES
o Many functions
Many spatstat functions now run faster, and will handle larger datasets,
thanks to improvements in the internal code,
following suggestions from Jens Oehlschlaegel.
o Many functions
The response to an 'Interrupt' signal is slightly slower.
CHANGES IN spatstat VERSION 1.25-5
OVERVIEW
o We thank Ya-Mei Chang, Jens Oehlschlaegel and Yong Song for contributions.
o Extended functionality of 'rhohat' to local likelihood smoothing
and bivariate smoothing.
o Nearest neighbour distance computations accelerated.
o spatstat now 'Suggests:' the package 'locfit'
NEW FUNCTIONS
o rho2hat
Bivariate extension of 'rhohat' for estimating spatial residual risk,
or intensity as a function of two covariates.
SIGNIFICANT USER-VISIBLE CHANGES
o rhohat
Estimation can now be performed using local likelihood fitting
with the 'locfit' package, or using kernel smoothing.
o nncross
Substantially accelerated.
New arguments added to control the return value and the sorting of data.
BUG FIXES
o plot.msr
Crashed if the argument 'box' was given.
Fixed.
CHANGES IN spatstat VERSION 1.25-4
OVERVIEW
o We thank Jonathan Lee and Sergiy Protsiv for contributions.
o Improvements and bug fixes to K function for very large datasets
NEW FUNCTIONS
o rStraussHard
Perfect simulation for Strauss-hardcore process (with gamma <= 1)
SIGNIFICANT USER-VISIBLE CHANGES
o plot.im
The colour ribbon can now be placed left, right, top or bottom
using new argument 'ribside'
o profilepl
Does not generate warnings when some of the candidate models
have zero likelihood - for example when fitting model with a hard core.
o Kest
Now includes fast algorithm for 'correction="none"'
which will handle patterns containing millions of points.
BUG FIXES
o Kest, Lest
Gave incorrect values in very large datasets, due to numerical overflow.
`Very large' typically means about 1 million points in a random pattern,
or 100,000 points in a tightly clustered pattern.
[Overflow cannot occur unless there are at least 46,341 points.]
[Spotted by Sergiy Protsiv.]
Fixed.
o Kest, Lest
Ignored 'ratio=TRUE' if the argument 'domain' was given.
[Spotted by Jonathan Lee.]
Fixed.
o rjitter
Output was sometimes incorrect.
[Spotted by Sergiy Protsiv.]
Fixed.
CHANGES IN spatstat VERSION 1.25-3
OVERVIEW
o We thank Daniel Esser for contributions.
o Improved support for fitted point process models.
o Bug fixes.
NEW FUNCTIONS
o simulate.slrm
Method for 'simulate' for spatial logistic regression models.
o labels.ppm, labels.kppm, labels.slrm
Methods for 'labels' for fitted point process models.
o commonGrid
Determine a common spatial domain and pixel resolution
for several pixel images and/or binary masks
SIGNIFICANT USER-VISIBLE CHANGES
o effectfun
Now has argument 'se.fit' allowing calculation of standard errors
and confidence intervals.
o [.msr
Now handles character-valued indices.
o print.summary.ppm
Output gives a more precise description of the fitting method.
o ppm, kppm, slrm
Confidence intervals for the fitted trend parameters
can now be obtained using 'confint'
o predict.slrm
New argument 'window'
o union.owin
Now handles a single argument: union.owin(A) returns A.
BUG FIXES
o selfcrossing.psp
y coordinate values were incorrect.
[Spotted by Daniel Esser.]
Fixed.
o as.im.owin
Did not handle a binary mask with a 1 x 1 pixel array.
Fixed.
o predict.slrm
Results of predict(object, newdata) were incorrect
if the spatial domain of 'newdata' was larger than the original domain.
Fixed.
o ppm
If the model was the uniform Poisson process,
the argument 'rbord' was ignored.
Fixed.
o image subset assignment "[<-.im"
Generated an error if the indexing argument 'i'
was a point pattern containing zero points.
Fixed.
o hyperframe subset assignment "[<-.hyperframe"
Did not correctly handle the case where a single column
of the hyperframe was to be changed.
Fixed.
o help(bw.relrisk), help(rmh.ppm), help(plot.plotppm)
These help files had the side-effect of changing some options
in spatstat.options.
Fixed.
CHANGES IN spatstat VERSION 1.25-2
OVERVIEW
o We thank Abdollah Jalilian and Thierry Onkelinx for contributions.
o Very Important Bug fixes.
o Improved mechanism for handling 'invalid' point processes
NEW FUNCTIONS
o as.matrix.owin
Converts a window to a logical matrix.
SIGNIFICANT USER-VISIBLE CHANGES
o project.ppm
Improved algorithm.
Now handles terms in the trend formula as well as the interaction.
The projected point process is now obtained by re-fitting the model,
and is guaranteed to be the maximum pseudolikelihood fit.
o plot.im
Now handles many arguments recognised by plot.default
such as 'cex.main'. Also handles argument 'box'.
New argument 'ribargs' contains parameters controlling the
ribbon plot only.
o spatstat.options
New option 'project.fast' allows a faster shortcut for project.ppm
o spatstat.options
New options 'rmh.p', 'rmh.q', 'rmh.nrep' determine the default values
of the parameters p, q and nrep of the Metropolis-Hastings algorithm.
See rmhcontrol
o ppm
Slightly accelerated.
BUG FIXES
o nncross, distfun, AreaInter
Results of nncross were possibly incorrect
when X and Y did not have the same window.
This bug affected values of 'distfun' and may also
have affected ppm objects with interaction 'AreaInter'.
[Spotted by Thierry Onkelinx]
Bug introduced in spatstat 1.9-4 (June 2006).
Fixed.
o rCauchy
Simulations were incorrect in the sense that
the value of 'omega' was inadvertently doubled
(i.e. omega was incorrectly replaced by 2 * omega).
Bug introduced in spatstat 1.25-0.
Fixed.
o plot.im
White lines were present in the image display, on some graphics devices,
due to changes in R 2.14.
Fixed.
o update.ppm
The result of 'update(object, formula)' sometimes contained errors
in the internal format.
Bug introduced in spatstat 1.25-0.
Fixed.
o example(AreaInter), example(bw.smoothppp),
example(Kest.fft), example(plot.owin), example(predict.ppm),
example(simulate.ppm)
Executing these examples had the side-effect of
changing some of the parameters in spatstat.options.
Fixed.
CHANGES IN spatstat VERSION 1.25-1
OVERVIEW
o We thank Neba Funwi-Gabga and Jorge Mateu for contributions.
o New dataset of gorilla nest sites
o New functions for perfect simulation
o Bug fix for rare crashes in rStrauss
o Code for ensuring a fitted point process model is a valid point process
NEW DATASET
o gorillas
Gorilla nest sites in a National Park in Cameroon.
Generously contributed by Neba Funwi-Gabga
NEW FUNCTIONS
o rDiggleGratton, rDGS, rHardcore
Perfect simulation for the Diggle-Gratton process,
Diggle-Gates-Stibbard process, and Hardcore process.
o bw.scott
Scott's rule of thumb for bandwidth selection in
multidimensional smoothing
o valid.ppm
Checks whether a fitted point process model is a valid point process
o project.ppm
Forces a fitted point process model to be a valid point process
SIGNIFICANT USER-VISIBLE CHANGES
o ppm
New argument 'project' determines whether the fitted model
is forced to be a valid point process
o linnet
Substantially accelerated.
o rStrauss
Slightly accelerated.
o summary.lpp
Now prints the units of length.
BUG FIXES
o rStrauss
Crashed rarely (once every 10 000 realisations)
with a memory segmentation fault.
Fixed.
CHANGES IN spatstat VERSION 1.25-0
OVERVIEW
o Leverage and influence for point process models
o New cluster models (support for model-fitting and simulation).
o Fit irregular parameters in trend of point process model
o Third order summary statistic.
o Improvements to speed and robustness of code.
o spatstat now depends on R 2.14
o We thank Abdollah Jalilian and Rasmus Waagepetersen for contributions.
NEW FUNCTIONS
o leverage.ppm, influence.ppm, dfbetas.ppm
Leverage and influence for point process models
o ippm
Experimental extension to 'ppm' which fits irregular parameters in trend
by Fisher scoring algorithm.
o Tstat
Third order summary statistic for point patterns
based on counting triangles.
o rCauchy, rVarGamma
simulation of a Neyman-Scott process with Cauchy clusters
or Variance Gamma (Bessel) clusters.
Contributed by Abdollah Jalilian.
o rPoissonCluster
simulation of a general Poisson cluster process
o model.covariates
Identify the covariates involved in a model (lm, glm, ppm etc)
o as.im.distfun
Converts a 'distfun' to a pixel image.
o cauchy.estK, cauchy.estpcf, vargamma.estK, vargamma.estpcf
Low-level model-fitting functions for the Neyman-Scott process
with Cauchy or Variance-Gamma cluster kernel.
Contributed by Abdollah Jalilian.
SIGNIFICANT USER-VISIBLE CHANGES
o kppm
Now accepts clusters="Cauchy" or clusters="VarGamma"
for the Neyman-Scott process with Cauchy or Variance-Gamma cluster kernel.
Code contributed by Abdollah Jalilian.
o rNeymanScott
Argument 'rcluster' may now take a different format.
o psst
Argument 'funcorrection' changed to 'funargs' allowing greater flexibility.
o plot.fv, plot.envelope
New argument 'limitsonly' allows calculation of
a common x,y scale for several plots.
o overall speed
spatstat is now byte-compiled and runs slightly faster.
o user interrupt
Long calculations in spatstat now respond to the Interrupt/Stop signal.
o update.ppm
Now runs faster and uses much less memory,
when the update only affects the model formula (trend formula).
o rNeymanScott, rThomas, rMatClust
Accelerated thanks to Rasmus Waagepetersen.
o multitype data and models
Second order multitype statistics (such as Kcross, pcfcross)
and multitype interaction models (such as MultiStrauss)
now run faster, by a further 5%.
BUG FIXES
o distfun
Some manipulations involving 'distfun' objects failed if the
original data X in distfun(X) did not have a rectangular window.
Fixed.
CHANGES IN spatstat VERSION 1.24-2
OVERVIEW
o Geyer's triplet interaction
o more functionality for replicated point patterns
o changed default for simulation window in point process simulation
o changed default for edge correction in Kcom, Gcom
o data in spatstat is now lazy-loaded
o bug fixes
NEW FUNCTIONS
o Triplets
Geyer's triplet interaction, for point process models
o coef.summary.ppm
New method coef.summary.ppm
You can now type 'coef(summary(fit))' to extract a table
of the fitted coefficients of the point process model 'fit'
SIGNIFICANT USER-VISIBLE CHANGES
o data in spatstat are now lazy-loaded
so you don't have to type data(amacrine), etc.
o rmh.default, rmh.ppm, simulate.ppm
These now handle the 'triplets' interaction
o fryplot
Now has arguments 'to' and 'from', allowing
selection of a subset of points.
o fryplot, frypoints
These functions now handle marked point patterns properly.
o Kcross, Kdot, Kmulti
New argument 'ratio' determines whether the numerator and denominator
of the estimate of the multitype K-function will be stored.
This enables analysis of replicated point patterns,
using 'pool.rat()' to pool the K function estimates.
o rmh.ppm, simulate.ppm, default.expand
For point process models which have a trend depending only on x and y,
the simulation window is now taken to be the same as the original
window containing the data (by default).
That is, `expansion' does not take place, by default.
(In previous versions of spatstat the simulation window was larger
than the original data window.)
o rmh.ppm, simulate.ppm
The argument sequence for these functions has changed.
New argument 'expand' allows more explicit control over simulation domain.
o Kcom, Gcom
New argument 'conditional' gives more explicit control
over choice of edge correction in compensator.
Simplified defaults for edge correction.
o Kinhom
Improved plot labels.
o profilepl
Printed output improved.
BUG FIXES
o Lest
The variance approximations (Lotwick-Silverman and Ripley)
obtained with var.approx=TRUE, were incorrect for Lest
(although they were correct for Kest) due to a coding error.
Fixed.
o simulate.ppm
Ignored the argument 'control' in some cases.
Fixed.
o pcf and its relatives (pcfinhom, pcfcross.inhom, pcfdot.inhom)
Sometimes gave a warning about 'extra arguments ignored'.
Fixed.
CHANGES IN spatstat VERSION 1.24-1
OVERVIEW
o Spatial Scan Test
o Functionality for replicated point patterns
o Bug fixes
NEW FUNCTIONS
o scan.test
Spatial scan test of clustering
o rat
New class of 'ratio objects'
o pool.rat
New method for 'pool'.
Combines K function estimates for replicated point patterns (etc)
by computing ratio-of-sums
o unnormdensity
Weighted kernel density with weights that do not sum to 1
and may be negative.
o compatible
New generic function with methods for 'fv', 'im', 'fasp' and 'units'
SIGNIFICANT USER-VISIBLE CHANGES
o Kest
New argument 'ratio' determines whether the numerator and denominator
of the estimate of the K-function will be stored.
This enables analysis of replicated point patterns,
using 'pool.rat()' to pool the K function estimates.
o Lest
Now handles theoretical variance estimates (using delta method)
if var.approx=TRUE
o as.mask
Argument 'eps' can now be a 2-vector, specifying x and y resolutions.
o default.expand
Behaviour changed slightly.
o plot.listof, plot.splitppp, contour.listof, image.listof
The arguments 'panel.begin' and 'panel.end' can now be objects
such as windows.
BUG FIXES
o rgbim, hsvim
Did not work on images with non-rectangular domains.
Fixed.
o scaletointerval
Did not handle NA's.
Fixed.
CHANGES IN spatstat VERSION 1.24-0
OVERVIEW
o This version was not released publicly.
CHANGES IN spatstat VERSION 1.23-6
OVERVIEW
o Spatial covariance functions of windows and pixel images.
o Area-interaction models can now be fitted in non-rectangular windows
o Bug fix for envelope of inhomogeneous Poisson process
o Bug fix for raster conversion
o New vignette on 'Getting Started with Spatstat'
o Code accelerated.
NEW FUNCTIONS
o imcov
Spatial covariance function of pixel image
or spatial cross-covariance function of two pixel images
o harmonise.im
Make several pixel images compatible
by converting them to the same pixel grid
o contour.listof, image.listof
Methods for contour() and image() for lists of objects
o dummify
Convert data to numeric values by constructing dummy variables.
SIGNIFICANT USER-VISIBLE CHANGES
o setcov
Can now compute the `cross-covariance' between two regions
o AreaInter
Point process models with the AreaInter() interaction
can now be fitted to point pattern data X in any window.
o areaGain, areaLoss
These now handle arbitrary windows W.
They are now more accurate when r is very small.
o Kcom
Computation vastly accelerated, for non-rectangular windows.
o vignettes
New vignette 'Getting Started with the Spatstat Package'
o nncorr, nnmean, nnvario
These functions now handle data frames of marks.
BUG FIXES
o envelope.ppm
If the model was an inhomogeneous Poisson process,
the resulting envelope object was incorrect
(the simulations were correct, but the envelopes were calculated
assuming the model was CSR). Bug was introduced in spatstat 1.23-5.
Fixed.
o envelope.ppm
If the model was an inhomogeneous Poisson process
with intensity a function of x and y only,
overflow errors sometimes occurred ('insufficient storage'
or 'attempting to generate a large number of random points').
Fixed.
o as.im.im
The result of as.im(X, W) was incorrect if 'W' did not cover 'X'.
Fixed.
o as.mask
The result of as.mask(w, xy) was incorrect if 'xy' did not cover 'w'.
Fixed.
o plot.fv
Legend was incorrectly labelled if 'shade' variables were not included
in the plot formula.
Fixed.
o areaGain, areaLoss
Crashed if the radius r was close to zero.
Fixed.
CHANGES IN spatstat VERSION 1.23-5
OVERVIEW
o Bug fix to bandwidth selection.
o Functions to pool data from several objects of the same class.
o Improvements and bug fixes.
o We thank Michael Sumner for contributions.
NEW FUNCTIONS
o pool
Pool data from several objects of the same class
o pool.envelope
Pool simulated data from several envelope objects and create a new envelope
o pool.fasp
Pool simulated data from several function arrays and create a new array
o envelope.envelope
Recalculate an envelope from simulated data using different parameters
SIGNIFICANT USER-VISIBLE CHANGES
o bw.diggle, bw.relrisk, bw.smoothppp, bw.optim
Plot method modified.
o model.depends
Now also recognises 'offset' terms.
BUG FIXES
o bw.diggle
Bandwidth was too large by a factor of 2.
Fixed.
o plot.psp
Crashed if any marks were NA.
Fixed.
o pointsOnLines
Crashed if any segments had zero length.
Ignored argument 'np' in some cases.
Fixed.
o stieltjes
Crashed if M had only a single column of function values.
Fixed.
CHANGES IN spatstat VERSION 1.23-4
OVERVIEW
o Bandwidth selection for density.ppp and smooth.ppp
o Layered plots.
o Model-handling facilities.
o Improvements and bug fixes.
NEW FUNCTIONS
o bw.diggle
Bandwidth selection for density.ppp by mean square error cross-validation.
o bw.smoothppp
Bandwidth selection for smooth.ppp by least-squares cross-validation.
o layered, plot.layered
A simple mechanism for controlling plots that consist of
several successive layers of data.
o model.depends
Given a fitted model (of any kind), identify which of the covariates
is involved in each term of the model.
o model.is.additive
Determine whether a fitted model (of any kind) is additive,
in the sense that each term in the model involves at most one covariate.
SIGNIFICANT USER-VISIBLE CHANGES
o smooth.ppp
Bandwidth 'sigma' is now selected by least-squares cross-validation
o bw.relrisk
Computation in large datasets accelerated.
New arguments 'hmin', 'hmax' control the range of trial values of bandwidth.
o Hest, Gfox, Jfox
Improved algebraic labels for plot
o spatstat.options
New parameter 'n.bandwidth'
o density.ppp, smooth.ppp
Slightly accelerated.
o point-in-polygon test
Accelerated.
BUG FIXES
o with.fv
Mathematical labels were incorrect in some cases.
Fixed.
o bw.relrisk
Implementation of method="weightedleastsquares" was incorrect
and was equivalent to method="leastsquares".
Fixed.
o smooth.ppp
NaN values occurred if the bandwidth was very small.
Fixed.
CHANGES IN spatstat VERSION 1.23-3
OVERVIEW
o Urgent bug fix.
BUG FIXES
o crossing.psp
Crashed occasionally with a message about NA or NaN values.
Fixed.
o affine.ppp
Crashed if the point pattern was empty.
Fixed.
CHANGES IN spatstat VERSION 1.23-2
OVERVIEW
o Bug fixes.
o Several functions have been accelerated.
o We thank Marcelino de la Cruz and Ben Madin for contributions.
NEW FUNCTIONS
o sumouter, quadform
Evaluate certain quadratic forms.
o flipxy
Exchange x and y coordinates.
SIGNIFICANT USER-VISIBLE CHANGES
o vcov.ppm
Accelerated.
o owin, as.owin
Checking the validity of polygons has been accelerated.
o crossing.psp, selfcrossing.psp
Accelerated.
BUG FIXES
o split.ppp
If drop=TRUE then some of the point patterns had the wrong windows.
Spotted by Marcelino de la Cruz.
Fixed.
o split.ppp
Crashed if the tessellation did not cover the point pattern.
Fixed.
o predict.ppm
Crashed when type="se" if NA's were present.
Spotted by Ben Madin.
Fixed.
o plot.ppp
Incorrectly handled the case where both 'col' and 'cols' were present.
Fixed.
o polygon geometry
The point-in-polygon test gave the wrong answer in some boundary cases.
Fixed.
CHANGES IN spatstat VERSION 1.23-1
OVERVIEW
o Important bug fix to 'localpcf'.
o Inverse-distance weighted smoothing.
o Inhomogeneous versions of neighbourhood density functions.
o Internal repairs and bug fixes.
o We thank Mike Kuhn and Ben Madin for contributions.
NEW FUNCTIONS
o idw
Inverse-distance weighted smoothing.
o localKinhom, localLinhom, localpcfinhom
Inhomogeneous versions of localK, localL, localpcf
BUG FIXES
o localpcf
The columns of the result were in the wrong order.
[i.e. pair correlation functions were associated with the wrong points.]
Fixed.
o delaunay
If the union of several Delaunay triangles formed a triangle,
this was erroneously included in the result of delaunay().
Fixed.
o predict.ppm, plot.ppm
Sometimes crashed with a warning about 'subscript out of bounds'.
Fixed.
o point-in-polygon test
Vertices of a polygon were sometimes incorrectly classified
as lying outside the polygon.
Fixed.
o Internal code
Numerous tweaks and repairs to satisfy the package checker
for the future R version 2.14.
CHANGES IN spatstat VERSION 1.23-0
OVERVIEW
o point patterns on a linear network:
new tools including geometrically-corrected linear K function,
pair correlation function, point process models, envelopes
o changes to renormalisation of estimates in Kinhom and pcfinhom
o new dataset: Chicago street crime
o spatstat now 'Suggests:' the package RandomFields
o spatstat now has a Namespace
o we thank Mike Kuhn, Monia Mahling, Brian Ripley for contributions.
NEW DATASET
o chicago
Street crimes in the University district of Chicago.
A point pattern on a linear network.
NEW FUNCTIONS
o envelope.lpp
Simulation envelopes for point patterns on a linear network
o lineardisc
Compute the 'disc' of radius r in a linear network
o linearpcf
Pair correlation for point pattern on a linear network
o linearKinhom, linearpcfinhom
Inhomogeneous versions of the K function and pair correlation function
for point patterns on a linear network
o lppm
Fit point process models on a linear network.
o anova.lppm
Analysis of deviance for point process models on a linear network.
o predict.lppm
Prediction for point process models on a linear network.
o envelope.lppm
Simulation envelopes for point process models on a linear network.
o linim
Pixel image on a linear network
o plot.linim
Plot a pixel image on a linear network
SIGNIFICANT USER-VISIBLE CHANGES
o linearK
New argument 'correction'.
Geometrically-corrected estimation is performed by default
(based on forthcoming paper by Ang, Baddeley and Nair)
o Kinhom
New argument 'normpower' allows different types of renormalisation.
o pcfinhom
Now performs renormalisation of estimate.
Default behaviour changed - estimates are now renormalised by default.
BUG FIXES
o density.ppp
Crashed if argument 'varcov' was given.
Fixed.
CHANGES IN spatstat VERSION 1.22-4
OVERVIEW
o new diagnostics based on score residuals
o new dataset
o improvements to plotting summary functions
o We thank Ege Rubak, Jesper Moller, George Leser, Robert Lamb
and Ulf Mehlig for contributions.
NEW FUNCTIONS
o Gcom, Gres, Kcom, Kres
New diagnostics for fitted Gibbs or Poisson point process models
based on score residuals.
Gcom is the compensator of the G function
Gres is the residual of the G function
Kcom is the compensator of the K function
Kres is the residual of the K function
o psst, psstA, psstG
New diagnostics for fitted Gibbs or Poisson point process models
based on pseudoscore residuals.
psst is the pseudoscore diagnostic for a general alternative
psstA is the pseudoscore diagnostic for an Area-interaction alternative
psstG is the pseudoscore diagnostic for a Geyer saturation alternative
o compareFit
Computes and compares several point process models
fitted to the same dataset, using a chosen diagnostic.
o as.interact
Extracts the interpoint interaction structure (without parameters)
from a fitted point process model or similar object.
NEW DATASET
o flu
Spatial point patterns giving the locations of
influenza virus proteins on cell membranes.
Kindly released by Dr George Leser and Dr Robert Lamb.
SIGNIFICANT USER-VISIBLE CHANGES
o pixel images and grids
The default size of a pixel grid, given by spatstat.options("npixel"),
has been changed from 100 to 128. A power of 2 gives faster
and more accurate results in many cases.
o residuals.ppm
New arguments 'coefs' and 'quad' for advanced use
(make it possible to compute residuals from a
modified version of the fitted model.)
o relrisk
New argument 'casecontrol' determines whether a bivariate
point pattern should be treated as case-control data.
o plot.fv
Further improvements in mathematical labels.
o plot.fv
The formula can now include the symbols .x and .y
as abbreviation for the function argument and the
recommended function value, respectively.
o plot.fv
New argument 'add'
BUG FIXES
o multitype summary functions
(Kcross, Kdot, Gcross, Gdot, .....)
Plotting these functions generated an error if the
name of one of the types of points contained spaces,
e.g. "Escherichia coli".
Fixed.
CHANGES IN spatstat VERSION 1.22-3
OVERVIEW
o Important bug fix to simulation code
o Miscellaneous improvements
o spatstat now depends on R 2.13.0 or later
o We thank Ege Rubak, Kaspar Stucki, Vadim Shcherbakov,
Jesper Moller and Ben Taylor for contributions.
NEW FUNCTIONS
o is.stationary, is.poisson
New generic functions for testing whether a point process model
is stationary and/or Poisson.
Methods for ppm, kppm, slrm etc
o raster.xy
raster coordinates of a pixel mask
o zapsmall.im
'zapsmall' for pixel images
SIGNIFICANT USER-VISIBLE CHANGES
o density.ppp
New argument 'diggle' allows choice of edge correction
o rotate.owin, affine.owin
These functions now handle binary pixel masks.
New argument 'rescue' determines whether rectangles will be preserved
BUG FIXES
o rmh, simulate.ppm
Serious bug - simulation was completely incorrect in the case of
a multitype point process with an interaction that does not depend
on the marks, such as ppm(betacells, ~marks, Strauss(60))
The calling parameters were garbled.
Fixed.
o effectfun
Crashed if the covariate was a function(x,y).
Fixed.
o lurking
Gave erroneous error messages about 'damaged' models.
Fixed.
o envelope.ppm
Did not recognise when the fitted model was equivalent to CSR.
Fixed.
o plot.ppx
Crashed in some cases. Fixed.
CHANGES IN spatstat VERSION 1.22-2
OVERVIEW
o Fitting and simulation of log-Gaussian Cox processes
with any covariance function
o More support for 'kppm' and 'rhohat' objects
o K-function for point patterns on a linear network
o Metropolis-Hastings algorithm now saves its transition history
o Easier control of dummy points in ppm
o Convert an 'fv' object to an R function
o spatstat now depends on the package 'RandomFields'
o We thank Abdollah Jalilian, Shen Guochun, Rasmus Waagepetersen,
Ege Rubak and Ang Qi Wei for contributions.
NEW FUNCTIONS
o linearK
Computes the Okabe-Yamada network K-function
for a point pattern on a linear network.
o pairdist.lpp
Shortest-path distances between each pair of points
on a linear network.
o vcov.kppm
Asymptotic variance-covariance matrix for regression parameters
in kppm object.
[Contributed by Abdollah Jalilian and Rasmus Waagepetersen]
o rLGCP
Simulation of log-Gaussian Cox processes
[Contributed by Abdollah Jalilian and Rasmus Waagepetersen]
o predict.rhohat
Method for 'predict' for objects of class 'rhohat'
Computes a pixel image of the predicted intensity.
o Kmodel, pcfmodel
Generic functions that compute the K-function or pair correlation function
of a point process *model*.
So far the only methods are for the class 'kppm'.
o as.function.fv
Converts a function value table (class 'fv') to a function in R
o coef.kppm
Method for 'coef' for objects of class 'kppm'
o unitname, unitname<-
These generic functions now have methods for fitted model objects
(classes ppm, slrm, kppm, minconfit) and quadrature schemes (quad).
o nobs.ppm
Method for 'nobs' for class 'ppm'.
Returns the number of points in the original data.
SIGNIFICANT USER-VISIBLE CHANGES
o kppm
Can now fit a log-Gaussian Cox process
o simulate.kppm
Can now simulate a fitted log-Gaussian Cox process
o lgcp.estK, lgcp.estpcf
These functions previously fitted a log-Gaussian Cox process
with exponential covariance. They can now fit a log-Gaussian Cox process
with any covariance function implemented by the RandomFields package.
o rmh
If track=TRUE, the history of transitions of the Metropolis-Hastings
algorithm is saved and returned.
o ppm
New argument 'nd' controls the number of dummy points.
o as.fv
Now handles objects of class kppm or minconfit.
o rhohat
If covariate = "x" or "y", the resulting object has the same
'unitname' as the original point pattern data.
o rhohat
Now has arguments 'eps, 'dimyx' to control pixel resolution.
o MultiStrauss, MultiHard, MultiStraussHard
Default value of 'types' has been changed to NULL.
o data(ants)
The auxiliary data 'ants.extra' now includes a function called 'side'
determining whether a given location is in the scrub or field region.
Can be used as a covariate in ppm, kppm, slrm.
o print.ppm
Now has argument 'what' to allow only selected information to be printed.
BUG FIXES
o profilepl
Crashed in some cases involving multitype interactions.
Fixed.
o plot.splitppp
Behaved incorrectly if 'main' was an expression.
Fixed.
o effectfun
Crashed in trivial cases.
Fixed.
o kppm, thomas.estpcf, matclust.estpcf, lgcp.estpcf
Gave a spurious warning message.
Fixed.
o step
When applied to ppm objects this gave a spurious warning.
Fixed.
CHANGES IN spatstat VERSION 1.22-1
OVERVIEW
o marked line segment patterns can now be plotted
o multitype point process models are now 'self-starting'
o new functions to manipulate colour images
NEW FUNCTIONS
o rgbim, hsvim
Specify three colour channels.
These functions convert three pixel images with numeric values
into a single image whose pixel values are strings representing colours.
o scaletointerval
Generic utility function to rescale data (including spatial data)
to a specified interval
SIGNIFICANT USER-VISIBLE CHANGES
o plot.im
Can now plot images whose pixel values are strings representing colours.
New argument 'valuesAreColours'
o plot.psp
Now handles marked line segment patterns
and plots the marks as colours.
o MultiHard, MultiStrauss, MultiStraussHard
The argument 'types' can now be omitted;
it will be inferred from the point pattern data.
o rhohat
Improved mathematical labels (when the result of rhohat is plotted)
o plot.fv
Minor improvements in graphics
BUG FIXES
o several minor bug fixes and improvements to satisfy R-devel
CHANGES IN spatstat VERSION 1.22-0
OVERVIEW
o support for point patterns on a linear network
o 'superimpose' is now generic
o improved mathematical labels when plotting functions
NEW CLASSES
o linnet
An object of class 'linnet' represents a linear network,
i.e. a connected network of line segments, such as a road network.
Methods for this class include plot, print, summary etc.
o lpp
An object of class 'lpp' represents a point pattern on a linear network,
such as a record of the locations of road accidents on a road network.
Methods for this class include plot, print, summary etc.
NEW FUNCTIONS
o runiflpp
Uniformly distributed random points on a linear network
o rpoislpp
Poisson point process on a linear network
o clickjoin
Interactive graphics to create a linear network
o superimpose
The function 'superimpose' is now generic, with methods
for ppp, psp and a default method.
o as.ppp.psp
New method for as.ppp extracts the endpoints and marks
from a line segment pattern
NEW DATASETS
o simplenet
Simple example of a linear network
SIGNIFICANT USER-VISIBLE CHANGES
o superimposePSP
This function is now deprecated in favour of 'superimpose'
o superimpose
Now handles data frames of marks.
o plot.fv
Argument 'legendmath' now defaults to TRUE.
New argument 'legendargs' gives more control over appearance of legend.
Increased default spacing between lines in legend.
o eval.fv, with.fv
Functions computed using eval.fv or with.fv
now have better labels when plotted.
o summary functions
(Kest, Kest.fft, Kcross, Kdot, Kmulti,
Kinhom, Kcross.inhom, Kdot.inhom, Kmulti.inhom,
Lest, Lcross, Ldot,
pcf, pcfcross, pcfdot, pcfinhom, pcfcross.inhom, pcfdot.inhom,
Fest, Gest, Gcross, Gdot, Gmulti,
Jest, Jcross, Jdot, Jmulti, Iest,
localL, localK,
markcorr, markvario, markconnect, Emark, Vmark,
allstats, alltypes)
Improved plot labels.
BUG FIXES
o superimpose
If the marks components of patterns consisted of character
vectors (rather than factors or non-factor numeric vectors)
an error was triggered.
Fixed.
o plot.fv
The y axis limits did not always cover the range of values
if the argument 'shade' was used.
Fixed.
o plot.rhohat
The y axis label was sometimes incorrect.
Fixed.
o plot.rhohat
If argument 'xlim' was used, a warning was generated from 'rug'.
Fixed.
CHANGES IN spatstat VERSION 1.21-6
OVERVIEW
o A line segment pattern can now have a data frame of marks.
o Various minor extensions and alterations in behaviour
NEW FUNCTIONS
o nsegments
Number of segments in a line segment pattern
SIGNIFICANT USER-VISIBLE CHANGES
o psp class
A line segment pattern (object of class 'psp') can now have a
data frame of marks.
o density.ppp
New argument 'adjust' makes it easy to adjust the smoothing bandwidth
o plot.envelope
If the upper envelope is NA but the lower envelope is finite,
the upper limit is now treated as +Infinity
o msr
Argument 'continuous' renamed 'density'
BUG FIXES
o [.psp
In X[W] if X is a line segment pattern and W is a polygonal window,
marks were sometimes discarded, leading to an error.
Fixed.
o [.psp
In X[W] if X is a line segment pattern and W is a rectangular window,
if the marks of X were factor values, they were converted to integers.
Fixed.
o superimposePSP
If the marks were a factor, they were mistakenly converted to integers.
Fixed.
o is.marked.ppp
Did not generate a fatal error when na.action="fatal" as described
in the help file.
Fixed.
CHANGES IN spatstat VERSION 1.21-5
OVERVIEW
o Increased numerical stability.
o New 'self-starting' feature of interpoint interactions.
SIGNIFICANT USER-VISIBLE CHANGES
o ppm
Interaction objects may now be 'self-starting'
i.e. initial parameter estimates can be computed
from the point pattern dataset.
So far, only the LennardJones() interaction
has a self-starting feature.
o LennardJones
Increased numerical stability.
New (optional) scaling argument 'sigma0'.
Interpoint distances are automatically rescaled
using 'self-starting' feature.
o vcov.ppm
New argument 'matrix.action' controls what happens when
the matrix is ill-conditioned.
Changed name of argument 'gamaction' to 'gam.action'
o rmhmodel.ppm
Default resolution of trend image has been increased.
o is.poisson.ppm
Accelerated.
o ppm, kppm, qqplot.ppm
Improved robustness to numerical error
CHANGES IN spatstat VERSION 1.21-4
OVERVIEW
o Urgent bug fix
BUG FIXES
o print.summary.ppm exited with an error message,
if the model had external covariates.
Fixed.
CHANGES IN spatstat VERSION 1.21-3
OVERVIEW
o Point process model covariates may now depend on additional parameters.
o New class of signed measures, for residual analysis.
o Miscellaneous improvements and bug fixes.
NEW FUNCTIONS
o clarkevans.test
Classical Clark-Evans test of randomness
o msr
New class 'msr' of signed measures and vector-valued measures
supporting residual analysis.
o quadrat.test.quadratcount
Method for 'quadrat.test' for objects of class 'quadratcount'
(allows a chi-squared test to be performed on quadrat counts
rather than recomputing from the original data)
o tile.areas
Computes areas of tiles in a tessellation (efficiently)
SIGNIFICANT USER-VISIBLE CHANGES
o ppm
The spatial trend can now depend on additional parameters.
This is done by allowing spatial covariate functions
to have additional parameters: function(x, y, ...)
where ... is controlled by the new argument 'covfunargs' to ppm
o profilepl
Can now maximise over trend parameters as well as interaction parameters
o residuals.ppm
The value returned by residuals.ppm is now an object
of class 'msr'. It can be plotted directly.
o eval.im
When the argument 'envir' is used, eval.im() now recognises
functions as well as variables in 'envir'
o colourmap
The argument 'col' can now be any kind of colour data
o persp.im
The 'colmap' argument can now be a 'colourmap' object
o ppm
The print and summary methods for 'ppm' objects
now show standard errors for parameter estimates
if the model is Poisson.
o quadrat.test
The print method for 'quadrattest' objects
now displays information about the quadrats
o lurking
Improved format of x axis label
o distmap.ppp
Internal code is more robust.
BUG FIXES
o im
Did not correctly handle 1 x 1 arrays.
Fixed.
o as.mask, pixellate.ppp
Weird things happened if the argument 'eps' was set
to a value greater than the size of the window.
Fixed.
CHANGES IN spatstat VERSION 1.21-2
OVERVIEW
o New multitype hardcore interaction.
o Nonparametric estimation of covariate effects on point patterns.
o Output of 'Kmeasure' has been rescaled.
o Numerous improvements and bug fixes.
NEW FUNCTIONS
o MultiHard
multitype hard core interaction for use in ppm()
o coords<-
Assign new coordinates to the points in a point pattern
o rhohat
Kernel estimate for the effect of a spatial covariate
on point process intensity
SIGNIFICANT USER-VISIBLE CHANGES
o as.ppp.matrix, as.ppp.data.frame
These methods for 'as.ppp' now accept a matrix or data frame
with any number of columns (>= 2) and interpret the additional
columns as marks.
o Kmeasure
The interpretation of the output has changed:
the pixel values are now density estimates.
o rmh.ppm, rmhmodel.ppm
These functions now accept a point process model
fitted with the 'MultiHard' interaction
o rmh.default, rmhmodel.default
These functions now accept the option: cif='multihard'
defining a multitype hard core interaction.
o markcorr
Now handles a data frame of marks
o varblock
Improved estimate in the case of the K function
o colourmap, lut
New argument 'range' makes it easier to specify a colour map
or lookup table
o [<-.hyperframe
Now handles multiple columns
o plot.fv
Improved y axis labels
o spatstat.options
New option 'par.fv' controls default parameters for line plotting
o rmhmodel
More safety checks on parameter values.
o quadratresample
New argument 'verbose'
o smooth.fv
Default value of 'which' has been changed.
BUG FIXES
o Kest
If the argument 'domain' was used, the resulting estimate
was not correctly normalised.
Fixed.
o Kest
The Lotwick-Silverman variance approximation was incorrectly
calculated. (Spotted by Ian Dryden and Igor Chernayavsky).
Fixed.
o plot.owin, plot.ppp
Display of binary masks was garbled if the window was
empty or if it was equivalent to a rectangle.
Fixed.
o plot.bermantest
One of the vertical lines for the Z1 test was in the wrong place.
Fixed.
o marks<-.ppx
Crashed in some cases.
Fixed.
o is.convex
An irrelevant warning was issued (for non-convex polygons).
Fixed.
CHANGES IN spatstat VERSION 1.21-1
OVERVIEW
o Confidence intervals for K-function and other statistics
o Bug fixes for smoothing and relative risk estimation
NEW FUNCTIONS
o varblock
Variance estimation (and confidence intervals)
for summary statistics such as Kest, using subdivision technique
o bw.stoyan
Bandwidth selection by Stoyan's rule of thumb.
o which.max.im
Applied to a list of images, this determines which image
has the largest value at each pixel.
o as.array.im
Convert image to array
SIGNIFICANT USER-VISIBLE CHANGES
o smooth.ppp, markmean, sharpen.ppp, relrisk, bw.relrisk
Further acceleration achieved.
o Kest
Argument 'correction' now explicitly overrides automatic defaults
o plot.fv
More robust handling of 'shade'
BUG FIXES
o relrisk
Format of relrisk(at="points") was incorrect.
Fixed.
o bw.relrisk
Result was incorrect in the default case method="likelihood"
because of previous bug.
Fixed.
o Jdot, Jcross, Jmulti
Return value did not include the hazard function, when correction="km"
Fixed.
o Jdot, Jcross, Jmulti
Format of output was incompatible with format of Jest.
Fixed.
CHANGES IN spatstat VERSION 1.21-0
OVERVIEW
o Implemented Spatial Logistic Regression
o Implemented nonparametric estimation of relative risk
with bandwidth selection by cross-validation.
o Smoothing functions can handle a data frame of marks.
o New options in Kinhom; default behaviour has changed.
NEW FUNCTIONS
o slrm
Fit a spatial logistic regression model
o anova.slrm, coef.slrm, fitted.slrm, logLik.slrm, plot.slrm, predict.slrm
Methods for spatial logistic regression models
o relrisk
Nonparametric estimation of relative risk
o bw.relrisk
Automatic bandwidth selection by cross-validation
o default.rmhcontrol
Sets default values of Metropolis-Hastings parameters
SIGNIFICANT USER-VISIBLE CHANGES
o smooth.ppp, markmean
These functions now accept a data frame of marks.
o Kinhom
Default behaviour has changed.
New argument 'renormalise=TRUE' determines scaling of estimator
and affects bias and variance in small samples.
o residuals.ppm
Now also computes the score residuals.
o plot.im
New argument 'ribscale'
o plot.listof, plot.splitppp
New arguments panel.begin, panel.end and panel.args
o ppp
Now checks for NA/NaN/Inf values in the coordinates
o envelope.ppm
Changed default value of 'control'
New argument 'nrep'
o qqplot.ppm
Changed default value of 'control'
BUG FIXES
o marks<-.ppp, setmarks, %mark%
A matrix of marks was accepted by ppp()
but not by these assignment functions.
Fixed.
o density.ppp, smooth.ppp, sharpen.ppp, markmean
Crashed if the bandwidth was extremely small.
Fixed.
CHANGES IN spatstat VERSION 1.20-5
OVERVIEW
o Accelerated computations of kernel smoothing.
o Implemented Choi-Hall data sharpening.
NEW FUNCTIONS
o sharpen.ppp
Performs Choi-Hall data sharpening of a point pattern
SIGNIFICANT USER-VISIBLE CHANGES
o density.ppp, smooth.ppp
Computation has been vastly accelerated
for density(X, at="points") and smooth.ppp(X, at="points")
o Kinhom
Accelerated in case where lambda=NULL
o Vignette 'shapefiles' updated
CHANGES IN spatstat VERSION 1.20-4
OVERVIEW
o New functions for inhomogeneous point patterns
and local analysis.
o Pair correlation function for 3D point patterns
o Minor improvements and bug fixes to simulation code and image functions
NEW FUNCTIONS
o pcf3est
Pair correlation function for 3D point patterns.
o Kscaled, Lscaled
Estimator of the template K function (and L-function)
for a locally-scaled point process.
o localpcf
Local version of pair correlation function
o identify.psp
Method for 'identify' for line segment patterns.
o as.im.matrix
Converts a matrix to a pixel image
SIGNIFICANT USER-VISIBLE CHANGES
o rMaternI, rMaternII
New argument 'stationary=TRUE' controls whether the
simulated process is stationary (inside the simulation window).
Default simulation behaviour has changed.
o im
New arguments 'xrange', 'yrange'
o envelope
Improvements to robustness of code.
BUG FIXES
o quadratcount
If V was a tessellation created using a factor-valued image,
quadratcount(X, tess=V) crashed with the error
"Tessellation does not contain all the points of X".
Fixed.
o [.im
If Z was a factor valued image and X was a point pattern
then Z[X] was not a factor.
Fixed.
CHANGES IN spatstat VERSION 1.20-3
OVERVIEW
o minor improvements (mostly internal).
NEW FUNCTIONS
o unmark.ppx
Method for 'unmark' for general space-time point patterns
SIGNIFICANT USER-VISIBLE CHANGES
o plot.ppx
Now handles marked patterns, in two-dimensional case
o as.psp.psp
Default value of argument 'check' set to FALSE
CHANGES IN spatstat VERSION 1.20-2
OVERVIEW
o Extensions to minimum contrast estimation.
o Bug fix in simulation of Lennard-Jones model.
o More support for distance functions.
o Changes to point process simulations.
NEW FUNCTIONS
o thomas.estpcf
Fit Thomas process model by minimum contrast using the
pair correlation function (instead of the K-function).
o matclust.estpcf
Fit Matern Cluster model by minimum contrast using the
pair correlation function (instead of the K-function).
o lgcp.estpcf
Fit log-Gaussian Cox process model by minimum contrast using the
pair correlation function (instead of the K-function).
o contour.distfun, persp.distfun
Methods for 'contour' and 'persp' for distance functions
o default.expand
Computes default window for simulation of a fitted point process model.
SIGNIFICANT USER-VISIBLE CHANGES
o kppm
Models can now be fitted using either the K-function
or the pair correlation function.
o ppm
The list of covariates can now include windows
(objects of class 'owin'). A window will be treated
as a logical covariate that equals TRUE inside the window
and FALSE outside it.
o plot.distfun
Pixel resolution can now be controlled.
o envelope.ppm, qqplot.ppm
The default value of 'control' has changed;
simulation results may be slightly different.
o rmh
Slightly accelerated.
BUG FIXES
o rmh
Simulation of the Lennard-Jones model (cif = 'lennard') was incorrect
due to an obscure bug, introduced in spatstat 1.20-1.
Fixed.
o thomas.estK, matclust.estK, lgcp.estK
The value of 'lambda' (if given)
was ignored if X was a point pattern.
Fixed.
CHANGES IN spatstat VERSION 1.20-1
OVERVIEW
o Further increases in speed and efficiency of ppm and rmh
o New pairwise interaction model
NEW FUNCTIONS
o DiggleGatesStibbard
Diggle-Gates-Stibbard pairwise interaction for use in ppm()
SIGNIFICANT USER-VISIBLE CHANGES
o ppm
has been accelerated by a factor of 10 for the BadGey interaction.
o rmh
simulation of the Lennard-Jones model (cif='lennard')
has been greatly accelerated.
o rmh, rmhmodel.ppm
Point process models fitted by ppm() using the DiggleGatesStibbard
interaction can be simulated automatically using rmh.
BUG FIXES
o fitin
The plot of a fitted Hardcore interaction was incorrect.
Fixed.
CHANGES IN spatstat VERSION 1.20-0
OVERVIEW
o spatstat now contains over 1000 functions.
o Substantial increase in speed and efficiency
of model-fitting code.
o Changes to factor-valued images.
SIGNIFICANT USER-VISIBLE CHANGES
o ppm
has been accelerated by a factor of 10,
and can handle datasets with 20,000 points,
for the following interactions:
DiggleGratton, Fiksel, Geyer, Hardcore, Strauss, StraussHard
o predict.ppm
accelerated by a factor of 3 (when type = "cif")
with vastly reduced memory requirements
for the following interactions:
DiggleGratton, Fiksel, Geyer, Hardcore, Strauss, StraussHard
o pixel images (class "im")
The internal representation of factor-valued images has changed.
Existing objects in the old format should still work.
o im
The syntax for creating a factor-valued image has changed.
Argument 'lev' has been deleted.
o ppm
Some warnings have been reworded for greater clarity.
BUG FIXES
o [.im
Mishandled some factor-valued images.
Fixed.
o hist.im
Produced slightly erroneous output for some factor-valued images.
Fixed.
o plot.owin
Filled polygons appeared to contain criss-cross lines
on some graphics drivers.
Fixed.
o deltametric
Did not handle windows with different enclosing frames
(error message: 'dA and dB are incompatible')
Fixed.
o quadratcount
Crashed if the pattern was empty and the window was a rectangle.
(Noticed by Sandro Azaele)
Fixed.
o rNeymanScott
Crashed if the parent process realisation was empty.
(Noticed by Sandro Azaele)
Fixed.
CHANGES IN spatstat VERSION 1.19-3
ACKNOWLEDGEMENTS
o We thank David Dereudre for contributions.
OVERVIEW
o Urgent bug fix to Metropolis-Hastings for Lennard-Jones model.
o Miscellaneous additions to plotting and colour management.
NEW FUNCTIONS
o col2hex, rgb2hex, paletteindex, samecolour
Functions for converting and comparing colours.
o plot.envelope
New method for plotting envelopes.
By default the area between the upper and lower envelopes
is shaded in grey.
SIGNIFICANT USER-VISIBLE CHANGES
o plot.fasp
If the entries in the array are envelopes,
they are plotted using plot.envelope
(hence the envelope region is shaded grey).
o plot.fv
Now displays mathematical notation for each curve,
if legendmath=TRUE.
o print.fv
Now prints the available range of 'r' values
as well as the recommended range of 'r' values.
BUG FIXES
o rmh
Simulation of Lennard-Jones model was incorrect;
the simulations were effectively Poisson patterns.
(Spotted by David Dereudre.)
Fixed.
o plot.fv
Did not correctly handle formulas that included I( )
Fixed.
CHANGES IN spatstat VERSION 1.19-2
ACKNOWLEDGEMENTS
o We thank Jorge Mateu, Michael Sumner and Sebastian Luque
for contributions.
OVERVIEW
o More support for fitted point process models and pixel images.
o Improved plotting of pixel images and envelopes.
o Simulation algorithm for Lennard-Jones process.
o Improvements and bug fixes to envelopes.
o Bug fixes to Metropolis-Hastings simulation.
NEW FUNCTIONS
o pairs.im
Creates a scatterplot matrix for several pixel images.
o model.frame.ppm
Method for 'model.frame' for point process models.
o sort.im
Method for 'sort' for pixel images.
SIGNIFICANT USER-VISIBLE CHANGES
o plot.fv, plot.fasp
New argument 'shade' enables confidence intervals
or significance bands to be displayed as filled grey shading.
o LennardJones
The parametrisation of this interaction function has been changed.
o rmh, rmhmodel
These functions will now simulate a point process model
that was fitted using the LennardJones() interaction.
o rmh.default, rmhmodel.default
These functions will now simulate a point process model
with the Lennard-Jones interaction (cif='lennard').
o ecdf
This function now works for pixel images.
o dim, row, col
These functions now work for pixel images.
o order
This function now works for pixel images.
o [.im and [<-.im
The subset index can now be any valid subset index for a matrix.
o density.ppp, smooth.ppp
The return value now has attributes 'sigma' and 'varcov'
reporting the smoothing bandwidth.
o plot.im
The argument 'col' can now be a 'colourmap' object.
This makes it possible to specify a fixed mapping between
numbers and colours (e.g. so that it is consistent between
plots of several different images).
o rmh, spatstat.options
spatstat.options now recognises the parameter 'expand'
which determines the default window expansion factor in rmh.
o rmh
Improved handling of ppm objects with covariates.
o kstest
The 'covariate' can now be one of the characters "x" or "y"
indicating the Cartesian coordinates.
BUG FIXES
o model.matrix.ppm
For a fitted model that used a large number of quadrature points,
model.matrix.ppm sometimes reported an internal error about mismatch
between the model matrix and the quadrature scheme.
Fixed.
o plot.ppx
Minor bugs fixed.
o rmh
In rare cases, the simulated point pattern
included multiple points at the origin (0,0).
(Bug introduced in spatstat 1.17-0.)
Fixed.
o rmh, rmhmodel.ppm
Crashed when applied to a fitted multitype point process model
if the model involved more than one covariate image.
(Spotted by Jorge Mateu)
Fixed.
o density.psp
If any segment had zero length, the result contained NaN values.
(Spotted by Michael Sumner and Sebastian Luque.)
Fixed.
o envelope
Crashed with fun=Lest or fun=Linhom if the number of points in
a simulated pattern exceeded 3000.
Fixed.
o plot.kstest
Main title was corrupted if the covariate was a function.
Fixed.
CHANGES IN spatstat VERSION 1.19-1
OVERVIEW
o New dataset: replicated 3D point patterns.
o Improvements to Metropolis-Hastings simulation code.
o More support for hyperframes.
o Bug fixes.
NEW DATASETS
o osteo:
Osteocyte Lacunae data: replicated 3D point patterns
NEW FUNCTIONS
o rbind.hyperframe:
Method for rbind for hyperframes.
o as.data.frame.hyperframe:
Converts a hyperframe to a data frame.
SIGNIFICANT USER-VISIBLE CHANGES
o Fiksel:
Fitted point process models (class ppm) with the Fiksel()
double exponential interaction can now be simulated by rmh.
o rmh.default:
Point processes with the Fiksel interaction can now be
simulated by specifying parameters in rmh.default.
o logLik.ppm:
New argument 'warn' controls warnings.
o profilepl:
No longer issues spurious warnings.
BUG FIXES
o Hardcore, rmh:
Simulation of the 'Hardcore' process was incorrect.
The hard core radius was erroneously set to zero
so that the simulated patterns were Poisson.
Fixed.
o fitin:
A plot of the pairwise interaction function of a fitted model,
generated by plot(fitin(model)) where model <- ppm(...),
was sometimes incorrect when the model included a hard core.
Fixed.
CHANGES IN spatstat VERSION 1.19-0
OVERVIEW
o Numerous bugs fixed in the implementation of
the Huang-Ogata approximate maximum likelihood method.
o New interpoint interaction model.
NEW FUNCTIONS
o Fiksel:
new interpoint interaction: Fiksel's double exponential model.
SIGNIFICANT USER-VISIBLE CHANGES
o runifpoint, rpoispp, envelope
These functions now issue a warning if the number of random points
to be generated is very large. This traps a common error in
simulation experiments.
BUG FIXES
o predict.ppm, fitted.ppm:
Predictions and fitted values were incorrect for
objects fitted using ppm(..., method="ho").
Fixed.
o logLik, AIC:
Values of logLik() and AIC() were incorrect for
objects fitted using ppm(..., method="ho").
Fixed.
o profilepl:
Results were incorrect if the argument 'method="ho"' was used.
Fixed.
o fitin
The result of fitin() was incorrect for
objects fitted using ppm(..., method="ho").
Fixed.
o rmhcontrol:
rmhcontrol(NULL) generated an error.
Fixed.
CHANGES IN spatstat VERSION 1.18-4
ACKNOWLEDGEMENTS
o We thank Michael Sumner for contributions.
BUG FIXES
o pixellate.psp: segments shorter than one pixel width
were measured incorrectly if the 'weights' argument was present.
Fixed.
NEW FUNCTIONS
o pairdist.ppx, crossdist.ppx, nndist.ppx, nnwhich.ppx:
Methods for pairdist, crossdist, nndist, nnwhich
for multidimensional point patterns (class 'ppx')
o runifpointx, rpoisppx:
Random point patterns in any number of dimensions
o boxx:
Multidimensional box in any number of dimensions
o diameter.boxx, volume.boxx, shortside.boxx, eroded.volumes.boxx:
Geometrical computations for multidimensional boxes
o sum.im, max.im, min.im:
Methods for sum(), min(), max() for pixel images.
o as.matrix.ppx:
Convert a multidimensional point pattern to a matrix
SIGNIFICANT USER-VISIBLE CHANGES
o plot.ppp: New argument 'zap'
o diameter: This function is now generic,
with methods for "owin", "box3" and "boxx"
o eroded.volumes: This function is now generic,
with methods for "box3" and "boxx"
CHANGES IN spatstat VERSION 1.18-3
ACKNOWLEDGEMENTS
o We thank Michael Sumner for contributions.
BUG FIXES
o pixellate.psp: segments shorter than one pixel width
were measured incorrectly. Fixed.
o fv: 'alim' not handled correctly. Fixed.
NEW FUNCTIONS
o smooth.fv:
Applies spline smoothing to the columns of an fv object.
CHANGES IN spatstat VERSION 1.18-2
ACKNOWLEDGEMENTS
o We thank Michael Sumner for contributions.
NEW FUNCTIONS
o Gfox, Jfox:
Foxall's G and J functions
o as.owin.distfun:
New method for as.owin extracts the domain of a distfun object.
SIGNIFICANT USER-VISIBLE CHANGES
o distfun: objects of class 'distfun', when called as functions,
will now accept either two vectors (x,y) or a point pattern x.
o Hest: this function can now compute the Hanisch estimator.
It now has arguments 'r', 'breaks' and 'correction',
like other summary functions.
o Hest: new argument 'conditional'.
BUG FIXES
o pixellate.psp: Values were sometimes incorrect due to coding error.
(Spotted by Michael Sumner)
Fixed.
o kstest: Crashed if the covariate contained NA's.
Fixed.
o kstest: Crashed if X was a multitype point pattern
in which some mark values were unrepresented.
Fixed.
o lurking: Minor bug in handling of NA values.
Fixed.
o Hest: labels of columns were incorrect.
Fixed.
CHANGES IN spatstat VERSION 1.18-1
ACKNOWLEDGEMENTS
o we thank Andrew Bevan and Ege Rubak for suggestions.
NEW FUNCTIONS
o Hardcore: Hard core interaction (for use in ppm)
o envelope.pp3: simulation envelopes for 3D point patterns
o npoints: number of points in a point pattern of any kind
SIGNIFICANT USER-VISIBLE CHANGES
o rmh.ppm, rmhmodel.ppm:
It is now possible to simulate Gibbs point process models
that are fitted to multitype point patterns using a
non-multitype interaction, e.g.
data(amacrine)
fit <- ppm(amacrine, ~marks, Strauss(0.1))
rmh(fit, ...)
o rmh.ppm, rmhmodel.ppm, rmh.default, rmhmodel.default:
Hard core models can be simulated.
o rmh.default, rmhmodel.default:
The argument 'par' is now required to be a list, in all cases
(previously it was sometimes a list and sometimes a vector).
o Fest:
Calculation has been accelerated in some cases.
o summary.pp3 now returns an object of class 'summary.pp3'
containing useful summary information. It is plotted by
'plot.summary.pp3'.
o F3est, G3est, K3est: these functions now accept
'correction="best"'
o union.owin, intersect.owin:
these functions now handle any number of windows.
o envelope.ppp, envelope.ppm, envelope.kppm:
argument lists have changed slightly
BUG FIXES
o Fest: The result of Fest(X, correction="rs") had a slightly
corrupted format, so that envelope(X, Fest, correction="rs")
in fact computed the envelopes based on the "km" correction.
(Spotted by Ege Rubak).
Fixed.
o rmh (rmh.ppm, rmhmodel.ppm):
rmh sometimes failed for non-stationary point process models,
with a message about "missing value where TRUE/FALSE needed".
(Spotted by Andrew Bevan).
Fixed.
o diagnose.ppm, lurking:
Calculations were not always correct if the model
had conditional intensity equal to zero at some locations.
Fixed.
o ppm, profilepl:
If data points are illegal under the model (i.e. if any
data points have conditional intensity equal to zero)
the log pseudolikelihood should be -Inf but was sometimes
returned as a finite value. Thus profilepl did not always
work correctly for models with a hard core.
Fixed.
o F3est, G3est:
Debug messages were printed unnecessarily.
Fixed.
CHANGES IN spatstat VERSION 1.18-0
ACKNOWLEDGEMENTS
o we thank Ege Rubak and Tyler Dean Rudolph for suggestions.
HEADLINES
o A point pattern is now allowed to have a data frame of marks
(previously the marks had to be a vector).
o Extended capabilities for 'envelope' and 'kstest'.
NEW FUNCTIONS
o pixellate.psp, as.mask.psp
Convert a line segment pattern to a pixel image or binary mask
o as.data.frame.im
Convert a pixel image to a data frame
SIGNIFICANT USER-VISIBLE CHANGES
o A point pattern is now allowed to have a data frame of marks
(previously the marks had to be a vector).
o Many functions in spatstat now handle point patterns with
a data frame of marks.
These include print.ppp, summary.ppp, plot.ppp, split.ppp.
o finpines, nbfires, shapley:
The format of these datasets has changed.
They are now point patterns with a data frame of marks.
o envelope() is now generic, with methods for "ppp", "ppm" and "kppm".
o kstest() now handles multitype point patterns and
multitype point process models.
o nnclean() now returns a point pattern with a data frame of marks.
o plot.ppp() has new argument 'which.marks' to select one column
from a data frame of marks to be plotted.
o plot.ppp() now handles marks that are POSIX times.
o complement.owin now handles any object acceptable to as.owin.
BUG FIXES
o erosion(w) and opening(w) crashed if w was not a window.
Fixed.
o diameter() and eroded.areas() refused to work if w was not a window.
Fixed.
CHANGES IN spatstat VERSION 1.17-6
ACKNOWLEDGEMENTS
o We thank Simon Byers and Adrian Raftery for generous contributions.
OVERVIEW
o Nearest neighbour clutter removal algorithm
o New documentation for the 'fv' class.
o Minor improvements and bug fixes.
NEW FUNCTIONS
o nnclean: Nearest neighbour clutter removal for recognising features
in spatial point patterns. Technique of Byers and Raftery (1998)
[From original code by Simon Byers and Adrian Raftery,
adapted for spatstat.]
o marks.ppx, marks<-.ppx: Methods for extracting and changing marks
in a multidimensional point pattern
o latest.news: print news about the current version of the package
SIGNIFICANT USER-VISIBLE CHANGES
o news: spatstat now has a NEWS file which can be printed by
typing news(package="spatstat").
o areaGain, areaLoss: New algorithms in case exact=TRUE.
Syntax slightly modified.
o with.hyperframe:
- The result now inherits 'names' from the row names
of the hyperframe.
- New argument 'enclos' controls the environment
in which the expression is evaluated.
- The algorithm is now smarter at simplifying the result
when simplify=TRUE.
o update.ppm: Tweaked to improve the ability of ppm objects
to be re-fitted in different contexts.
ADVANCED USERS ONLY
o Documentation for the class 'fv' of function value tables
- fv: Creates an object of class 'fv'
- cbind.fv, collapse.fv: Combine objects of class 'fv'
- bind.fv: Add additional columns of data to an 'fv' object
BUG FIXES
o "$<-.hyperframe" destroyed the row names of the hyperframe.
Fixed.
o model.matrix.ppm had minor inconsistencies.
Fixed.
o ppm: The fitted coefficient vector had incorrect format
in the default case of a uniform Poisson process.
Fixed.
o plot.ppx: Crashed if the argument 'main' was given.
Fixed.
o envelope.ppp: Crashed if the object returned by 'fun'
did not include a column called "theo".
Fixed.
Computing file changes ...