1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
|
%% $Id: csdacm.Rnw,v 1.44 2008-04-06 20:47:04 roger Exp $
% -*- mode: noweb; noweb-default-code-mode: R-mode; -*-
% merge all depth coefficients!!
% Customising spatial data analysis classes and methods
%\SweaveOpts{echo=TRUE}
%*Intro and plan*
%motivate spatial data handling
\documentclass{article}
% \VignetteIndexEntry{Customising spatial data classes and methods}
\usepackage{graphicx}
\usepackage[colorlinks=true,urlcolor=blue]{hyperref}
\usepackage{color}
\usepackage{Sweave}
\SweaveOpts{keep.source=FALSE}
\usepackage{natbib}
\newcommand{\strong}[1]{{\normalfont\fontseries{b}\selectfont #1}}
\let\pkg=\strong
\newcommand\code{\bgroup\@codex}
\def\@codex#1{\small {\normalfont\ttfamily\hyphenchar\font=45 #1}\egroup}
\usepackage{xspace}
\def\RR{\textsf{R}\xspace}
\def\SP{\texttt{S-PLUS}\texttrademark\xspace}
\def\SS{\texttt{S}\xspace}
\def\SIII{\texttt{S3}\xspace}
\def\SIV{\texttt{S4}\xspace}
\title{{\bf Customising spatial data classes and methods}\footnote{This vignette formed pp. 127--148 of the first edition of Bivand, R. S.,
Pebesma, E. and G\'{o}mez-Rubio V. (2008) Applied Spatial Data Analysis with R,
Springer-Verlag, New York. It was retired from the second edition (2013) to
accommodate material on other topics, and is made available in this form
with the understanding of the publishers. It has been updated to the 2013
state of the software, e.g. using \code{over}.} }
\author{Edzer Pebesma\footnote{Institute for Geoinformatics,
University of Muenster, Weseler Strasse 253, 48151 M\"{u}nster, Germany.
{\tt edzer.pebesma@uni-muenster.de}}}
\date{Feb 2008}
\begin{document}
\maketitle
\tableofcontents
<<echo=FALSE>>=
owidth <- getOption("width")
options("width"=90)
.PngNo <- 0
@
<<label=figreset,echo=FALSE,eval=FALSE>>=
.iwidth <- 5
.iheight <- 6
.ipointsize <- 12
@
<<echo=FALSE,results=hide>>=
<<figreset>>
@
<<label=afig,echo=FALSE,eval=FALSE>>=
.PngNo <- .PngNo + 1; file <- paste("Fig-bitmap-", .PngNo, ".pdf", sep="")
pdf(file=file, width = .iwidth, height = .iheight, pointsize = .ipointsize)
opar <- par(mar=c(3,3,1,1)+0.1)
@
<<label=zfig,echo=FALSE,eval=FALSE>>=
dev.null <- dev.off()
cat("\\includegraphics[width=0.95\\textwidth]{", file, "}\n\n", sep="")
@
Although the classes defined in the \pkg{sp} package cover many needs,
they do not go far beyond the most typical GIS data models. In applied
research, it often happens that customised classes would suit the
actual data coming from the instruments better. Since \SIV classes have
mechanisms for inheritance, it may be attractive to build on the \pkg{sp}
classes, so as to utilise their methods where appropriate. Here, we will
demonstrate a range of different settings in which \pkg{sp} classes can
be extended. Naturally, this is only useful for researchers with specific
and clear needs, so our goal is to show how (relatively) easy it may be
to prototype classes extending \pkg{sp} classes for specific purposes.
\section{Programming with classes and methods}
\label{sect:cls}
This section will explain the elementary basics of programming with
classes and methods in \RR. The \SS language (implemented in \RR and
\SP) contains two mechanisms for creating classes and methods: the
traditional \SIII system and the more recent \SIV system (see Section 2.2 in
\cite{bivand}, in which classes were described for the use{\RR} ---
here they are described for the develope{\RR}). This chapter is not a
full introduction to \RR programming \citep[see][for more details]{braun+murdoch:07}, but it will try to give some
feel of how the \code{Spatial} classes in package \pkg{sp} can be
extended to be used for wider classes of problems. For full details,
the interested reader is referred to e.g.\ \cite{R:Venables+Ripley:2000}
and \cite{R:Chambers:1998}, the latter being a reference for new-style
\SIV classes and methods. Example code is for example found in the source
code for package \pkg{sp}, available from CRAN.
Suppose we define myfun as
\begin{footnotesize}
<< >>=
myfun <- function(x) {
x + 2
}
@
\end{footnotesize}
then, calling it with the numbers 1, 2 and 3 results in
\begin{footnotesize}
<< >>=
myfun(1:3)
@
\end{footnotesize}
or alternatively using a named argument:
\begin{footnotesize}
<< >>=
myfun(x=1:3)
@
\end{footnotesize}
The return value of the function is the last expression evaluated.
Often, we want to wrap existing functions, such as a plot function:
\begin{footnotesize}
<< >>=
plotXplus2Yminus3 <- function(x, y, ...) {
plot(x = x + 2, y = y - 3, ...)
}
@
\end{footnotesize}
In this case, the \code{...} is used to pass information to the
\code{plot} function without explicitly anticipating what it will be:
named arguments \code{x} and \code{y}, or the first two arguments if they
are unnamed are processed, remaining arguments are passed on.
The plot function is a generic method, with
an instance that depends on the class of its first (\SIII) or first $n$
arguments (\SIV). The available instances of \code{plot} are shown for
\SIII-type methods by
\begin{footnotesize}
<< >>=
methods("plot")
@
\end{footnotesize}
and for \SIV-type methods by
\begin{footnotesize}
<< >>=
library(sp)
showMethods("plot")
@
\end{footnotesize}
where we first loaded \pkg{sp} to make sure there are some \SIV plot
methods to show.
\subsection{\SIII-style classes and methods}
Building \SIII-style classes is simple. Suppose we want to
build an object of class \code{foo}:
\begin{footnotesize}
<< >>=
x <- rnorm(10)
class(x) <- "foo"
x
@
\end{footnotesize}
If we plot this object, e.g., by \code{plot(x)} we get the same plot as
when we would not have set the class to \code{foo}. If we know, however, that
objects of class \code{foo} need to be plotted without symbols but with
connected lines, we can write a plot method for this class:
\begin{footnotesize}
<< >>=
plot.foo <- function(x, y, ...) {
plot.default(x, type = 'l', ...)
}
@
\end{footnotesize}
after which \code{plot(x)} will call this particular method, rather than
a default plot method.
Class inheritance is obtained in \SIII when an object is given multiple
classes, as in
\begin{footnotesize}
<< >>=
class(x) <- c("foo", "bar")
@
<<eval=FALSE>>=
plot(x)
@
\end{footnotesize}
For this plot, first function \code{plot.foo} will be looked for, and
if not found the second option \code{plot.bar} will be looked for. If
none of them is found, the default \code{plot.default} will be used.
The \SIII class mechanism is simple and powerful. Much of \RR works with
it, including key functions such as \code{lm}.
\begin{footnotesize}
<< >>=
data(meuse)
class(meuse)
class(lm(log(zinc)~sqrt(dist), meuse))
@
\end{footnotesize}
There is, however, no checking that a class with a particular name does
indeed contain the elements that a certain method for it expects.
It also has design flaws, as method specification by dot separation
is ambiguous in case of names such as \code{as.data.frame}, where one
cannot tell whether it means that the method \code{as.data} acts on
objects of class \code{frame}, or the method \code{as} acts on objects
of class \code{data.frame}, or none of them (the answer is: none). For
such reasons, \SIV-style classes and methods were designed.
\subsection{\SIV-style classes and methods}
\SIV-style classes are formally defined, using \code{setClass}. As an example,
somewhat simplified versions of classes \code{CRS} and \code{Spatial}
in \pkg{sp} are
\begin{footnotesize}
<<echo=FALSE>>=
options("width"=60)
@
<<eval=FALSE>>=
setClass("CRS", representation(projargs = "character"))
setClass("Spatial",
representation(bbox = "matrix", proj4string = "CRS"),
# NOT TOO WIDE
validity <- function(object) {
bb <- bbox(object)
if (!is.matrix(bb))
return("bbox should be a matrix")
n <- dimensions(object)
if (n < 2)
return("spatial.dimension should be 2 or more")
if (any(is.na(bb)))
return("bbox should never contain NA values")
if (any(!is.finite(bb)))
return("bbox should never contain infinite values")
if (any(bb[,"max"] < bb[,"min"]))
return("invalid bbox: max < min")
TRUE
}
)
@
<<echo=FALSE>>=
options("width"=70)
@
\end{footnotesize}
The command \code{setClass} defines a class name as a formal class, gives
the names of the class elements (called slots), and their type---type
checking will happen upon construction of an instance of the class. Further
checking, e.g., on valid dimensions and data ranges can be done in the
\code{validity} function. Here, the validity function retrieves the
bounding box using the generic \code{bbox} method.
Generics, if not defined in the base R system, e.g.,
\begin{footnotesize}
<< >>=
isGeneric("show")
@
\end{footnotesize}
can be defined with \code{setGeneric}. Defining a specific instance of a
generic is done by \code{setMethod}:
\begin{footnotesize}
<<eval=FALSE>>=
setGeneric("bbox", function(obj) standardGeneric("bbox"))
setMethod("bbox", signature = "Spatial", function(obj) obj@bbox)
@
\end{footnotesize}
where the signature tells the class of the first (or first $n$) arguments.
Here, the \code{@} operator is used to access the \code{bbox} slot in
an \SIV object, not to be confused with the \code{\$} operator to access
list elements.
We will now illustrate this mechanism by providing a few examples of
classes, building on those available in package \pkg{sp}.
\section{Animal track data in package \pkg{trip}}
CRAN Package \pkg{trip}, written by Michael Sumner
\citep{kirkwood06,page06}, provides a class for animal tracking
data. Animal tracking data consist of sets of ($x,y,t$) stamps,
grouped by an identifier pointing to an individual animal, sensor or
perhaps isolated period of monitoring. A strategy for this
(slightly simplified from that of \pkg{trip}) is to extend the {\tt
SpatialPointsDataFrame} class by a length 2 character vector carrying
the names of the time column and the trip identifier column in the {\tt
SpatialPointsDataFrame} attribute table.
Package \pkg{trip} does a lot of work to read and analyse tracking
data from data formats typical for tracking data (Argos DAT), removing
duplicate observations and validating the objects, e.g., checking that
time stamps increase and movement speeds are realistic. We ignore this
and stick to the bare bones.
We now define a class called \code{trip} that extends \code{SpatialPointsDataFrame}:
\begin{footnotesize}
<< >>=
library(sp)
setClass("trip", representation("SpatialPointsDataFrame", TOR.columns = "character"),
validity <- function(object) {
if (length(object@TOR.columns) != 2)
stop("Time/id column names must be of length 2")
if (!all(object@TOR.columns %in% names(object@data)))
stop("Time/id columns must be present in attribute table")
TRUE
}
)
showClass("trip")
@
\end{footnotesize}
that checks, upon creation of objects, that indeed two variable names
are passed and that these names refer to variables present in the
attribute table.
\subsection{Generic and constructor functions}
It would be nice to have a constructor function, just like \code{data.frame} or
\code{SpatialPoints}, so we now create it and set it as the generic function to
be called in case the first argument is of class \code{SpatialPointsDataFrame}.
\begin{footnotesize}
<< >>=
trip.default <- function(obj, TORnames) {
if (!is(obj, "SpatialPointsDataFrame"))
stop("trip only supports SpatialPointsDataFrame")
if (is.numeric(TORnames))
TORnames <- names(obj)[TORnames]
new("trip", obj, TOR.columns = TORnames)
}
if (!isGeneric("trip"))
setGeneric("trip", function(obj, TORnames)
standardGeneric("trip"))
setMethod("trip", signature(obj = "SpatialPointsDataFrame", TORnames = "ANY"), trip.default)
@
\end{footnotesize}
We can now try it out, with turtle data:
\begin{footnotesize}
<< >>=
turtle <- read.csv(system.file("external/seamap105_mod.csv", package="sp"))
@
<< >>=
timestamp <- as.POSIXlt(strptime(as.character(turtle$obs_date), "%m/%d/%Y %H:%M:%S"), "GMT")
turtle <- data.frame(turtle, timestamp = timestamp)
turtle$lon <- ifelse(turtle$lon < 0, turtle$lon+360, turtle$lon)
turtle <- turtle[order(turtle$timestamp),]
coordinates(turtle) <- c("lon", "lat")
proj4string(turtle) <- CRS("+proj=longlat +ellps=WGS84")
turtle$id <- c(rep(1, 200), rep(2, nrow(coordinates(turtle)) - 200))
turtle_trip <- trip(turtle, c("timestamp", "id"))
summary(turtle_trip)
@
\end{footnotesize}
\subsection{Methods for trip objects}
The summary method here is not defined for \code{trip}, but is the
default summary inherited from class \code{Spatial}. As can be seen,
nothing special about the trip features is mentioned, such as what the
time points are and what the identifiers. We could alter this by writing a
class-specific summary method
\begin{footnotesize}
<< >>=
summary.trip <- function(object, ...) {
cat("Object of class \"trip\"\nTime column: ")
print(object@TOR.columns[1])
cat("Identifier column: ")
print(object@TOR.columns[2])
print(summary(as(object, "Spatial")))
print(summary(object@data))
}
setMethod("summary", "trip", summary.trip)
summary(turtle_trip)
@
\end{footnotesize}
As \code{trip} extends \code{SpatialPointsDataFrame},
subsetting using {\small \verb+"["+}
and column selection or replacement using {\small \verb+"[["+}
or {\small \verb+"$"+} all work,
as these are inherited. Creating invalid trip objects can be prohibited by
adding checks to the validity function in the class definition, e.g.,
%<<eval=FALSE,echo=TRUE>>=
%x <- turtle_trip[1]
%@
will not work because the time and/or id column are not present any more.
A custom plot method for trip could be written, for example
using colour to denote a change in identifier:
\begin{footnotesize}
<< >>=
setGeneric("lines", function(x, ...) standardGeneric("lines"))
setMethod("lines", signature(x = "trip"),
function(x, ..., col = NULL) {
# NOT TOO WIDE
tor <- x@TOR.columns
if (is.null(col)) {
l <- length(unique(x[[tor[2]]]))
col <- hsv(seq(0, 0.5, length = l))
}
coords <- coordinates(x)
lx <- split(1:nrow(coords), x[[tor[2]]])
for (i in 1:length(lx))
lines(coords[lx[[i]], ], col = col[i], ...)
}
)
@
\end{footnotesize}
Here, the \code{col} argument is added to the function header so that a
reasonable default can be overridden, e.g., for black/white plotting.
\section{Multi-point data: \texttt{SpatialMultiPoints}}
One of the feature types of the OpenGeospatial Consortium (OGC) simple feature
specification that has
not been implemented in \pkg{sp} is the \code{MultiPoint} object. In a
\code{MultiPoint} object, each feature refers to a {\em set of} points. The
\pkg{sp} classes \code{SpatialPointsDataFrame} only provide reference to a
single point. Instead of building a new class up from scratch, we'll try
to re-use code and build a class \code{SpatialMultiPoint} from the {\tt
SpatialLines} class. After all, lines are just sets of ordered points.
In fact, the \code{SpatialLines} class implements the \code{MultiLineString}
simple feature, where each feature can refer to multiple lines. A special
case is formed if each feature only has a single line:
\begin{footnotesize}
<<echo=FALSE>>=
options("width"=50)
@
<< >>=
setClass("SpatialMultiPoints", representation("SpatialLines"),
validity <- function(object) {
if (any(unlist(lapply(object@lines, function(x) length(x@Lines))) != 1))
# NOT TOO WIDE
stop("Only Lines objects with one Line element")
TRUE
}
)
SpatialMultiPoints <- function(object) new("SpatialMultiPoints", object)
@
<<echo=FALSE>>=
options("width"=70)
@
\end{footnotesize}
As an example, we can create an instance of this class for two MultiPoint
features each having three locations:
\begin{footnotesize}
<< >>=
n <- 5
set.seed(1)
x1 <- cbind(rnorm(n),rnorm(n, 0, 0.25))
x2 <- cbind(rnorm(n),rnorm(n, 0, 0.25))
x3 <- cbind(rnorm(n),rnorm(n, 0, 0.25))
L1 <- Lines(list(Line(x1)), ID="mp1")
L2 <- Lines(list(Line(x2)), ID="mp2")
L3 <- Lines(list(Line(x3)), ID="mp3")
s <- SpatialLines(list(L1,L2,L3))
smp <- SpatialMultiPoints(s)
@
\end{footnotesize}
If we now plot object \code{smp}, we get the same plot as when we plot
\code{s}, showing the two lines. The \code{plot} method for a
\code{SpatialLines} object is not suitable, so we write a new one:
\begin{footnotesize}
<< >>=
plot.SpatialMultiPoints <- function(x, ..., pch = 1:length(x@lines), col = 1, cex = 1) {
n <- length(x@lines)
if (length(pch) < n)
pch <- rep(pch, length.out = n)
if (length(col) < n)
col <- rep(col, length.out = n)
if (length(cex) < n)
cex <- rep(cex, length.out = n)
plot(as(x, "Spatial"), ...)
for (i in 1:n)
points(x@lines[[i]]@Lines[[1]]@coords, pch = pch[i], col = col[i], cex = cex[i])
}
setMethod("plot", signature(x = "SpatialMultiPoints", y = "missing"),
function(x, y, ...) plot.SpatialMultiPoints(x, ...))
@
\end{footnotesize}
Here we chose to pass any named \code{...} arguments to the plot method
for a \code{Spatial} object. This function sets up the axes and controls the
margins, aspect ratio, etc. All arguments that need to be passed
to \code{points} (\code{pch} for symbol type, \code{cex} for symbol size and
\code{col} for symbol colour) need explicit naming and sensible defaults,
as they are passed explicitly to the consecutive calls to \code{points}.
According to the documentation of \code{points}, in addition to \code{pch},
\code{cex} and \code{col}, the arguments \code{bg} and \code{lwd} (symbol
fill colour and symbol line width) would need a similar treatment to
make this plot method completely transparent with the base \code{plot}
method---something an end user would hope for.
Having \code{pch}, \code{cex} and \code{col} arrays the length of the number
of \code{MultiPoints} {\em sets} rather than the number of points to be
plotted is useful for two reasons. First, the whole point of {\tt
MultiPoints} object is to distinguish {\em sets} of points. Second,
when we extend this class to \code{SpatialMultiPointsDataFrame}, e.g., by
\begin{footnotesize}
<< >>=
cName <- "SpatialMultiPointsDataFrame"
setClass(cName, representation("SpatialLinesDataFrame"),
validity <- function(object) {
lst <- lapply(object@lines, function(x) length(x@Lines))
if (any(unlist(lst) != 1))
stop("Only Lines objects with single Line")
TRUE
}
)
SpatialMultiPointsDataFrame <- function(object) {
new("SpatialMultiPointsDataFrame", object)
}
@
\end{footnotesize}
then we can pass symbol characteristics by (functions of) columns in the attribute table:
\begin{footnotesize}
<<fig=FALSE>>=
df <- data.frame(x1 = 1:3, x2 = c(1,4,2), row.names = c("mp1", "mp2", "mp3"))
smp_df <- SpatialMultiPointsDataFrame(SpatialLinesDataFrame(smp, df))
setMethod("plot", signature(x = "SpatialMultiPointsDataFrame", y = "missing"),
function(x, y, ...) plot.SpatialMultiPoints(x, ...))
grys <- c("grey10", "grey40", "grey80")
@
<<eval=FALSE>>=
plot(smp_df, col = grys[smp_df[["x1"]]], pch = smp_df[["x2"]], cex = 2, axes = TRUE)
@
\end{footnotesize}
for which the plot is shown in Figure \ref{fig:smpdf}.
\begin{figure}
%<<fig=TRUE,echo=FALSE>>=
<<echo=FALSE,results=tex>>=
.iwidth <- 6
.iheight <- 2.5
.ipointsize <- 10
<<afig>>
plot(smp_df, col = grys[smp_df[["x1"]]], pch = smp_df[["x2"]], cex = 2, axes = TRUE)
<<zfig>>
<<figreset>>
@
\caption{Plot of the \code{SpatialMultiPointsDataFrame} object.}
\label{fig:smpdf}
\end{figure}
Hexagonal grids are like square grids, where grid points are centres
of matching hexagons, rather than squares. Package \pkg{sp} has no
classes for hexagonal grids, but does have some useful functions for
generating and plotting them. This could be used to build a class. Much
of this code in \pkg{sp} is based on postings to the R-sig-geo mailing
list by Tim Keitt, used with permission.
The spatial sampling method \code{spsample} has a method for sampling points
on a hexagonal grid:
\begin{footnotesize}
<<>>=
data(meuse.grid)
gridded(meuse.grid)=~x+y
xx <- spsample(meuse.grid, type="hexagonal", cellsize=200)
class(xx)
@
\end{footnotesize}
gives the points shown in the left side of Figure \ref{fig:hex}. Note
that an alternative hexagonal representation is obtained by rotating
this grid 90 degrees; we will not further consider that here.
\begin{footnotesize}
<<fig=FALSE>>=
HexPts <- spsample(meuse.grid, type="hexagonal", cellsize=200)
@
<<eval=FALSE>>=
spplot(meuse.grid["dist"], sp.layout = list("sp.points", HexPts, col = 1))
@
<<fig=FALSE>>=
HexPols <- HexPoints2SpatialPolygons(HexPts)
df <- over(HexPols, meuse.grid)
HexPolsDf <- SpatialPolygonsDataFrame(HexPols, df, match.ID = FALSE)
@
<<eval=FALSE>>=
spplot(HexPolsDf["dist"])
@
\end{footnotesize}
for which the plots are shown in Figure \ref{fig:hex}.
\begin{figure}
<<echo=FALSE,results=tex>>=
.iwidth <- 6
.iheight <- 4
<<afig>>
library(lattice)
# RSB quietening greys
grys <- grey.colors(11, 0.95, 0.55, 2.2)
print(spplot(meuse.grid["dist"], cuts=10, col.regions=grys, sp.layout = list("sp.points", HexPts, col = 1)),
split = c(1, 1, 2, 1), more = TRUE)
print(spplot(HexPolsDf["dist"], cuts=10, col.regions=grys),
split = c(2, 1, 2, 1), more = FALSE)
<<zfig>>
<<figreset>>
@
\caption{Hexagonal points (left) and polygons (right).}
\label{fig:hex}
\end{figure}
We can now generate and plot hexagonal grids, but need to deal with two
representations: as points and as polygons, and both representations do
not tell by themselves that they represent a hexagonal grid.
For designing a hexagonal grid class we will extend \code{SpatialPoints},
assuming that computation of the polygons can be done when needed without
a prohibitive overhead.
\begin{footnotesize}
<< >>=
setClass("SpatialHexGrid", representation("SpatialPoints", dx = "numeric"),
validity <- function(object) {
if (object@dx <= 0)
stop("dx should be positive")
TRUE
}
)
@
<<echo=FALSE>>=
options("width"=40)
@
<< >>=
setClass("SpatialHexGridDataFrame", representation("SpatialPointsDataFrame", dx = "numeric"),
# NOT TOO WIDE
validity <- function(object) {
if (object@dx <= 0)
stop("dx should be positive")
TRUE
}
)
@
<<echo=FALSE>>=
options("width"=70)
@
\end{footnotesize}
Note that these class definitions do not check that instances actually do
form valid hexagonal grids; a more robust implementation could provide
a test that distances between points with equal $y$ coordinate are
separated by a multiple of \code{dx}, that the $y$-separations are
correct and so on.
It might make sense to adapt the generic \code{spsample} method in
package \pkg{sp} to return \code{SpatialHexGrid} objects; we can also
add \code{plot} and \code{spsample} methods for them. Method \code{over}
should work with a \code{SpatialHexGrid} as its first argument,
by inheriting from \code{SpatialPoints}. Let us first see
how to create the new classes. Without a constructor function we can use
\begin{footnotesize}
<< >>=
HexPts <- spsample(meuse.grid, type="hexagonal", cellsize=200)
Hex <- new("SpatialHexGrid", HexPts, dx = 200)
df <- over(Hex, meuse.grid)
spdf <- SpatialPointsDataFrame(HexPts, df)
HexDf <- new("SpatialHexGridDataFrame", spdf, dx = 200)
@
\end{footnotesize}
Because of the route taken to define both HexGrid classes, it is not
obvious that the second extends the first. We can tell the \SIV system
this by \code{setIs}:
\begin{footnotesize}
<< >>=
is(HexDf, "SpatialHexGrid")
setIs("SpatialHexGridDataFrame", "SpatialHexGrid")
is(HexDf, "SpatialHexGrid")
@
\end{footnotesize}
to make sure that methods for \code{SpatialHexGrid} objects work as well
for objects of class \code{SpatialHexGridDataFrame}.
When adding methods, several of them will need conversion to the
polygon representation, so it makes sense to add the conversion
function such that e.g. \code{as(x, "SpatialPolygons")} will work:
\begin{footnotesize}
<<echo=FALSE>>=
options("width"=50)
@
<< >>=
# NOT TOO WIDE
setAs("SpatialHexGrid", "SpatialPolygons",
function(from)
HexPoints2SpatialPolygons(from, from@dx)
)
setAs("SpatialHexGridDataFrame", "SpatialPolygonsDataFrame",
function(from)
SpatialPolygonsDataFrame(as(obj, "SpatialPolygons"), obj@data, match.ID = FALSE)
)
@
<<echo=FALSE>>=
options("width"=70)
@
\end{footnotesize}
We can now add \code{plot}, \code{spplot}, \code{spsample} and
\code{over} methods for these classes:
\begin{footnotesize}
<< >>=
setMethod("plot", signature(x = "SpatialHexGrid", y = "missing"),
function(x, y, ...) plot(as(x, "SpatialPolygons"), ...)
)
setMethod("spplot", signature(obj = "SpatialHexGridDataFrame"),
function(obj, ...)
spplot(SpatialPolygonsDataFrame( as(obj, "SpatialPolygons"), obj@data, match.ID = FALSE), ...)
)
setMethod("spsample", "SpatialHexGrid", function(x, n, type, ...)
spsample(as(x, "SpatialPolygons"), n = n, type = type, ...)
)
setMethod("over", c("SpatialHexGrid", "SpatialPoints"), function(x, y, ...)
over(as(x, "SpatialPolygons"), y)
)
@
\end{footnotesize}
After this, the following will work:
\begin{footnotesize}
<<eval=FALSE>>=
spplot(meuse.grid["dist"], sp.layout = list("sp.points", Hex, col = 1))
spplot(HexDf["dist"])
@
\end{footnotesize}
Coercion to a data frame is done by
\begin{footnotesize}
<<eval=FALSE>>=
as(HexDf, "data.frame")
@
\end{footnotesize}
Another detail not mentioned is that the bounding box of the hexgrid
objects only match the grid centre points, not the hexgrid cells:
\begin{footnotesize}
<< >>=
bbox(Hex)
bbox(as(Hex, "SpatialPolygons"))
@
\end{footnotesize}
One solution for this is to correct for this in a constructor function,
and check for it in the validity test. Explicit coercion functions
to the points representation would have to set the bounding box back
to the points ranges. Another solution is to write a bbox method for
the hexgrid classes, taking the risk that someone still looks at the
incorrect bbox slot.
\section{Spatio-temporal grids}
Spatio-temporal data can be represented in different ways. One simple
option is when observations (or model-results, or predictions) are given
on a regular space-time grid.
Objects of class or extending \code{SpatialPoints}, \code{SpatialPixels}
and \code{SpatialGrid} do not have the constraint that they represent a
two-dimensional space; they may have arbitrary dimension; an example for
a three-dimensional grid is
\begin{footnotesize}
<< >>=
n <- 10
x <- data.frame(expand.grid(x1 = 1:n, x2 = 1:n, x3 = 1:n), z = rnorm(n^3))
coordinates(x) <- ~x1+x2+x3
gridded(x) <- TRUE
fullgrid(x) <- TRUE
summary(x)
@
\end{footnotesize}
We might assume here that the third dimension, \code{x3}, represents
time. If we are happy with time somehow represented by a real number
(in double precision), then we are done. A simple representation is that of
decimal year, with e.g. 1980.5 meaning the 183rd day of 1980, or
e.g. relative time in seconds after the start of some event.
When we want to use the \code{POSIXct} or \code{POSIXlt} representations,
we need to do some more work to see the readable version. We will now
devise a simple three-dimensional space-time grid with the \code{POSIXct}
representation.
\begin{footnotesize}
<<echo=FALSE>>=
options("width"=50)
@
<< >>=
# NOT TOO WIDE
setClass("SpatialTimeGrid", "SpatialGrid",
validity <- function(object) {
stopifnot(dimensions(object) == 3)
TRUE
}
)
@
<<echo=FALSE>>=
options("width"=70)
@
\end{footnotesize}
Along the same line, we can extend the \code{SpatialGridDataFrame}
for space-time:
\begin{footnotesize}
<< >>=
setClass("SpatialTimeGridDataFrame", "SpatialGridDataFrame",
validity <- function(object) {
stopifnot(dimensions(object) == 3)
TRUE
}
)
setIs("SpatialTimeGridDataFrame", "SpatialTimeGrid")
x <- new("SpatialTimeGridDataFrame", x)
@
\end{footnotesize}
A crude summary for this class could be written along these lines:
\begin{footnotesize}
<< >>=
summary.SpatialTimeGridDataFrame <- function(object, ...) {
cat("Object of class SpatialTimeGridDataFrame\n")
x <- gridparameters(object)
t0 <- ISOdate(1970,1,1,0,0,0)
t1 <- t0 + x[3,1]
cat(paste("first time step:", t1, "\n"))
t2 <- t0 + x[3,1] + (x[3,3] - 1) * x[3,2]
cat(paste("last time step: ", t2, "\n"))
cat(paste("time step: ", x[3,2], "\n"))
summary(as(object, "SpatialGridDataFrame"))
}
@
<<echo=FALSE>>=
options("width"=50)
@
<< >>=
# NOT TOO WIDE
setMethod("summary", "SpatialTimeGridDataFrame", summary.SpatialTimeGridDataFrame)
summary(x)
@
<<echo=FALSE>>=
options("width"=70)
@
\end{footnotesize}
Next, suppose we need a subsetting method that selects on the time.
When the first subset argument is allowed to be a time range, this is done by
\begin{footnotesize}
<< >>=
subs.SpatialTimeGridDataFrame <- function(x, i, j, ..., drop=FALSE) {
t <- coordinates(x)[,3] + ISOdate(1970,1,1,0,0,0)
if (missing(j))
j <- TRUE
sel <- t %in% i
if (! any(sel))
stop("selection results in empty set")
fullgrid(x) <- FALSE
if (length(i) > 1) {
x <- x[i = sel, j = j,...]
fullgrid(x) <- TRUE
as(x, "SpatialTimeGridDataFrame")
} else {
gridded(x) <- FALSE
x <- x[i = sel, j = j,...]
cc <- coordinates(x)[,1:2]
p4s <- CRS(proj4string(x))
# NOT TOO WIDE
SpatialPixelsDataFrame(cc, x@data, proj4string = p4s)
}
}
setMethod("[", c("SpatialTimeGridDataFrame", "POSIXct", "ANY"),
subs.SpatialTimeGridDataFrame)
t1 <- as.POSIXct("1970-01-01 0:00:03", tz = "GMT")
t2 <- as.POSIXct("1970-01-01 0:00:05", tz = "GMT")
summary(x[c(t1,t2)])
summary(x[t1])
@
\end{footnotesize}
The reason to only convert back to \code{SpatialTimeGridDataFrame} when
multiple time steps are present is that the time step (``cell size''
in time direction) cannot be found when there is only a single step. In
that case, the current selection method returns an object of class
\code{SpatialPixelsDataFrame} for that time slice.
Plotting a set of slices could be done using levelplot, or writing
another \code{spplot} method:
\begin{footnotesize}
<< >>=
spplot.stgdf <- function(obj, zcol = 1, ..., format = NULL) {
# NOT TOO WIDE
if (length(zcol) != 1)
stop("can only plot a single attribute")
if (is.null(format)) format <- "%Y-%m-%d %H:%M:%S"
cc <- coordinates(obj)
df <- unstack(data.frame(obj[[zcol]], cc[,3]))
ns <- as.character(coordinatevalues(getGridTopology(obj))[[3]] + ISOdate(1970,1,1,0,0,0), format = format)
cc2d <- cc[cc[,3] == min(cc[,3]), 1:2]
obj <- SpatialPixelsDataFrame(cc2d, df)
spplot(obj, names.attr = ns,...)
}
setMethod("spplot", "SpatialTimeGridDataFrame", spplot.stgdf)
@
\end{footnotesize}
\begin{figure}
\begin{center}
<<echo=FALSE,results=tex>>=
.iwidth <- 6
.iheight <- 4
<<afig>>
print(spplot(x, format = "%H:%M:%S", as.table=TRUE))
<<zfig>>
<<figreset>>
@
\end{center}
\caption{ \code{spplot} for an object of class \code{SpatialTimeGridDataFrame},
filled with random numbers.}
\label{fig:stgdf}
\end{figure}
Now, the result of
\begin{footnotesize}
<<eval=FALSE>>=
library(lattice)
trellis.par.set(canonical.theme(color = FALSE))
spplot(x, format = "%H:%M:%S", as.table=TRUE, cuts=6,
col.regions=grey.colors(7, 0.55, 0.95, 2.2))
# RSB quietening greys
@
\end{footnotesize}
is shown in Figure \ref{fig:stgdf}. The format argument passed controls the
way time is printed; one can refer to the help of
\begin{footnotesize}
<<eval=FALSE>>=
?as.character.POSIXt
@
\end{footnotesize}
for more details about the \code{format} argument.
\section{Analysing spatial Monte Carlo simulations}
\label{sec:simquant}
Quite often, spatial statistical analysis results in a large number
of spatial realisations or a random field, using some Monte Carlo
simulation approach. Regardless whether individual values refer to
points, lines, polygons or grid cells, we would like to write some
methods or functions that aggregate over these simulations, to get
summary statistics such as the mean value, quantiles, or cumulative
distributions values. Such aggregation can take place in two ways.
Either we aggregate over the probability space, and compute summary
statistics for each geographical feature over the set of realisations
(i.e., the rows of the attribute table), or for each realisation we
aggregate over the complete geographical layer or a subset of it (i.e.,
aggregate over the columns of the attribute table).
Let us first generate, as an example, a set of 100 conditional Gaussian
simulations for the zinc variable in the meuse data set:
\begin{footnotesize}
<<eval=TRUE,echo=TRUE>>=
library(gstat)
data(meuse)
coordinates(meuse) <- ~x+y
v <- vgm(.5, "Sph", 800, .05)
sim <- krige(log(zinc)~1, meuse, meuse.grid, v, nsim=100, nmax=30)
sim@data <- exp(sim@data)
@
\end{footnotesize}
where the last statement back-transforms the simulations from the log scale
to the observation scale. A quantile method for Spatial object attributes
can be written as
\begin{footnotesize}
<< >>=
quantile.Spatial <- function(x, ..., byLayer = FALSE) {
stopifnot("data" %in% slotNames(x))
apply(x@data, ifelse(byLayer, 2, 1), quantile, ...)
}
@
\end{footnotesize}
after which we can find the sample lower and upper 95\%
confidence limits by
\begin{footnotesize}
<< >>=
sim$lower <- quantile.Spatial(sim[1:100], probs = 0.025)
sim$upper <- quantile.Spatial(sim[1:100], probs = 0.975)
@
\end{footnotesize}
To get the sample distribution of the areal median, we can aggregate over
layers:
\begin{footnotesize}
<< >>=
medians <- quantile.Spatial(sim[1:100], probs = 0.5, byLayer = TRUE)
@
<<eval=FALSE>>=
hist(medians)
@
\end{footnotesize}
It should be noted that in these particular cases, the quantities computed
by simulations could have been obtained faster and exact by working
analytically with ordinary (block) kriging and the normal distribution
(Section 8.7.2 in \cite{bivand}).
A statistic that cannot be obtained analytically is the sample distribution
of the area fraction that exceeds a threshold. Suppose that 500 is a
crucial threshold, and we want to summarise the sampling distribution of
the area fraction where 500 is exceeded:
\begin{footnotesize}
<<echo=FALSE>>=
options("width"=50)
@
<< >>=
fractionBelow <- function(x, q, byLayer = FALSE) {
stopifnot(is(x, "Spatial") || !("data" %in% slotNames(x)))
apply(x@data < q, ifelse(byLayer, 2, 1), function(r) sum(r)/length(r))
# NOT TOO WIDE
}
@
<<echo=FALSE>>=
options("width"=70)
@
<< >>=
over500 <- 1 - fractionBelow(sim[1:100], 200, byLayer = TRUE)
summary(over500)
quantile(over500, c(0.025, 0.975))
@
\end{footnotesize}
For space-time data, we could write methods that aggregate over space,
over time or over space and time.
\section{Processing massive grids}
Up to now we have made the assumption that gridded data can be completely read,
and are kept by \RR in memory. In some cases, however, we need to process grids
that exceed the memory capacity of the computers available. A method for analysing
grids without fully loading them into memory then seems useful. Note that
package \pkg{rgdal} allows for partial reading of grids, e.g.,
<<echo=FALSE,results=hide>>=
run <- require(rgdal, quietly=TRUE)
@
\begin{footnotesize}
<<>>=
if (run) {
fn <- system.file("pictures/erdas_spnad83.tif", package = "rgdal")[1]
}
@
<<eval=FALSE>>=
x <- readGDAL(fn, output.dim = c(120, 132))
x$band1[x$band1 <= 0] <- NA
spplot(x, col.regions=bpy.colors())
@
\end{footnotesize}
reads a downsized grid, where 1\% of the grid cells remained. Another
option is reading certain rectangular sections of a grid, starting at
some offset.
Yet another approach is to use the low-level opening routines and then
subset:
\begin{footnotesize}
<<>>=
if (run) {
library(rgdal)
x <- GDAL.open(fn)
class(x)
}
if (run) {
x.subs <- x[1:100, 1:100, 1]
class(x.subs)
}
if (run) {
gridparameters(x.subs)
}
@
\end{footnotesize}
An object of class \code{GDALReadOnlyDataset} only contains a file handle.
The subset method {\small \verb+"["+}
for it does not, as it quite often does,
return an object of the same class but actually reads the data requested,
with arguments interpreted as {\em rows}, {\em columns} and raster {\em
bands}, and returns a \code{SpatialGridDataFrame}. We will now extend this
approach to allow partial writing through {\small \verb+"["+}
as well. As the
actual code is rather lengthy and involves a lot of administration,
it will not all be shown and details can be found in the \pkg{rgdal}
source code.
We will define two classes,
\begin{footnotesize}
<<echo=FALSE>>=
options("width"=50)
@
<<>>=
if (run) {
setClass("SpatialGDAL",
representation("Spatial", grid = "GridTopology", grod = "GDALReadOnlyDataset",
# NOT TOO WIDE
name = "character"))
setClass("SpatialGDALWrite", "SpatialGDAL")
}
@
<<echo=FALSE>>=
options("width"=70)
@
\end{footnotesize}
that derive from \code{Spatial}, contain a \code{GridTopology}
and a file handle in the \code{grod} slot. Next, we can define a
function \code{open.SpatialGDAL} to open a raster file, returning a
\code{SpatialGDAL} object and a function \code{copy.SpatialGDAL}
that returns a writable copy of the opened raster. Note that some GDAL
drivers only allow copying, some only writing and some both.
\begin{footnotesize}
<<eval=FALSE>>=
x <- open.SpatialGDAL(fn)
nrows <- GDALinfo(fn)["rows"]
ncols <- GDALinfo(fn)["columns"]
xout <- copy.SpatialGDAL(x, "erdas_spnad83_out.tif")
bls <- 20
for (i in 1:(nrows/bls - 1)) {
r <- 1+(i-1)*bls
for (j in 1:(ncols/bls - 1)) {
c <- 1+(j-1)*bls
x.in <- x[r:(r+bls),c:(c+bls)]
xout[r:(r+bls),c:(c+bls)] <- x.in$band1 + 10 #$
}
cat(paste("row-block", i, "\n"))
}
close(x)
close(xout)
@
\end{footnotesize}
This requires the functions {\small \verb+"["+}
and {\small \verb+"[<-"+}
to be present. They are set by
\begin{footnotesize}
<<eval=FALSE>>=
setMethod("[", "SpatialGDAL",
function(x, i, j, ... , drop = FALSE)
x@grod[i = i, j = j, ...]
)
setReplaceMethod("[", "SpatialGDALWrite", function(x, i, j, ..., value) {
...
})
@
\end{footnotesize}
where, for the latter, the implementation details are here omitted. It
should be noted that single rows or columns cannot be read this way,
as they cannot be converted sensibly to a grid.
It should be noted that flat binary representations such as the
Arc/Info Binary Grid allow much faster random access than ASCII
representations or compressed formats such as jpeg varieties. Also,
certain drivers in the GDAL library suggest an optimal
block size for partial access (e.g., typically a single row), which is
not used here\footnote{An attempt to use this block size is, at time of
writing, found in the \code{blockApply} code, found in the THK
branch of the \pkg{rgdal} project on R-forge.}.
This chapter has sketched developments beyond the base \pkg{sp} classes
and methods used otherwise in this book. Although we think that the base
classes cater for many standard kinds of spatial data analysis, it is
clear that specific research problems will call for specific solutions,
and that the \RR environment provides the high-level abstractions needed
to help busy researchers get their work done.
<<echo=FALSE>>=
options("width"=owidth)
@
\begin{thebibliography}{}
\bibitem[Bivand et al., 2008]{bivand}
Roger S. Bivand, Edzer J. Pebesma and Virgilio Gomez-Rubio (2008).
\newblock {\em Applied spatial data analysis with {\RR}}
\newblock Springer, NY
\bibitem[Braun and Murdoch, 2007]{braun+murdoch:07}
Braun, W.~J. and Murdoch, D.~J. (2007).
\newblock {\em A first course in statistical programming with {\RR}}.
\newblock Cambridge University Press, Cambridge.
\bibitem[Chambers, 1998]{R:Chambers:1998}
Chambers, J.M. (1998).
\newblock {\em Programming with Data}.
\newblock Springer, New York.
\bibitem[Kirkwood et al., 2006]{kirkwood06}
Kirkwood, R., Lynch, M., Gales, N., Dann, P., and Sumner, M. (2006).
\newblock At-sea movements and habitat use of adult male {A}ustralian fur seals
({A}rctocephalus pusillus doriferus).
\newblock {\em Canadian Journal of Zoology}, 84:1781--1788.
\bibitem[Page et al., 2006]{page06}
Page, B., McKenzie, J., Sumner, M., Coyne, M., and Goldsworthy, S. (2006).
\newblock Spatial separation of foraging habitats among {N}ew {Z}ealand fur
seals.
\newblock {\em Marine Ecology Progress Series}, 323:263--279.
\bibitem[Venables and Ripley, 2000]{R:Venables+Ripley:2000}
Venables, W. N. and Ripley, B. D. (2000).
\newblock {\em {\SS} Programming}.
\newblock Springer, New York.
\end{thebibliography}
\end{document}
|