-
Notifications
You must be signed in to change notification settings - Fork 10
/
RcppGSL-intro.Rnw
932 lines (767 loc) · 37.9 KB
/
RcppGSL-intro.Rnw
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
\documentclass[11pt]{article}
%\VignetteIndexEntry{RcppGSL}
%\VignetteKeywords{R,GSL,Rcpp,data transfer}
%\VignetteDepends{RcppGSL}
%\VignetteEngine{highlight::highlight}
\usepackage{url,color}
\usepackage[authoryear,round,longnamesfirst]{natbib}
\usepackage[colorlinks]{hyperref}
\definecolor{link}{rgb}{0,0,0.3} %% next few lines courtesy of RJournal.sty
\hypersetup{
colorlinks,%
citecolor=link,%
filecolor=link,%
linkcolor=link,%
urlcolor=link
}
\usepackage{vmargin}
\setmargrb{0.75in}{0.75in}{0.75in}{0.75in}
\usepackage{booktabs} % fancier \hrule
\usepackage{microtype} %% cf http://www.khirevich.com/latex/microtype/
\usepackage[T1]{fontenc} %% cf http://www.khirevich.com/latex/font/
\usepackage[bitstream-charter]{mathdesign} %% cf http://www.khirevich.com/latex/font/
\newcommand{\proglang}[1]{\textsf{#1}}
\newcommand{\pkg}[1]{{\fontseries{b}\selectfont #1}}
\newsavebox{\hlbox}
\definecolor{hlBg}{rgb}{0.950,0.950,0.950} % background
\definecolor{hlBd}{rgb}{0.666,0.666,0.666} % border
\renewenvironment{Hchunk}{\vspace{0.125em}\noindent\begin{lrbox}{\hlbox}\begin{minipage}[b]{.98\linewidth}}%
{\end{minipage}\end{lrbox}\fcolorbox{hlBd}{hlBg}{\usebox{\hlbox}}\vspace{0.125em}}
<<setup,echo=FALSE,print=FALSE>>=
require(inline)
library(RcppGSL)
options("width"=65)
rcppgsl.version <- packageDescription( "RcppGSL" )$Version
prettyDate <- format(Sys.Date(), "%B %e, %Y")
@
% closing $ needed here
\author{Dirk Eddelbuettel \and Romain Fran\c{c}ois}
\title{\pkg{RcppGSL}: Easier \pkg{GSL} use from \proglang{R} via \pkg{Rcpp}}
\date{Version \Sexpr{rcppgsl.version} as of \Sexpr{prettyDate}}
\begin{document}
\maketitle
\abstract{
\shortcites{GSL} %% so that we get Galassi et al instead of all names
\noindent
The GNU Scientific Library, or \pkg{GSL}, is a collection of numerical
routines for scientific computing \citep{GSL}. It is particularly useful for
\proglang{C} and \proglang{C++} programs as it provides a standard
\proglang{C} interface to a wide range of mathematical routines such as
special functions, permutations, combinations, fast fourier transforms,
eigensystems, random numbers, quadrature, random distributions,
quasi-random sequences, Monte Carlo integration, N-tuples, differential
equations, simulated annealing, numerical differentiation, interpolation,
series acceleration, Chebyshev approximations, root-finding, discrete
Hankel transforms physical constants, basis splines and wavelets. There
are over 1000 functions in total with an extensive test suite.
The \pkg{RcppGSL} package provides an easy-to-use interface between
\pkg{GSL} and \proglang{R}, with a particular focus on matrix and vector
data structures. \pkg{RcppGSL} relies on \pkg{Rcpp}
\citep{JSS:Rcpp,Eddelbuettel:2013:Rcpp,CRAN:Rcpp} which is
itself a package that eases the interfaces between \proglang{R} and C++.}
\section{Introduction}
The GNU Scientific Library, or \pkg{GSL}, is a collection of numerical
routines for scientific computing \citep{GSL}. It is a rigourously developed
and tested library providing support for a wide range of scientific or
numerical tasks. Among the topics covered in the \pkg{GSL} are
%% from the GSL manual
complex numbers, roots of polynomials,
special functions, vector and matrix data structures,
permutations, combinations, sorting, BLAS support,
linear algebra, fast fourier transforms, eigensystems,
random numbers, quadrature, random distributions, quasi-random sequences,
Monte Carlo integration, N-tuples,
differential equations, simulated annealing,
numerical differentiation, interpolation,
series acceleration, Chebyshev approximations,
root-finding, discrete Hankel transforms
least-squares fitting, minimization,
physical constants, basis splines and wavelets.
Support for \proglang{C} programming with the \pkg{GSL} is available as the
\pkg{GSL} itself is written in \proglang{C}, and provides a
\proglang{C}-language Application Programming Interface (API).
Access from \proglang{C++} is possible, albeit not at an
abstraction level that could be offered by dedicated \proglang{C++}
implementations. Several \proglang{C++} wrappers for the \pkg{GSL} have
been written over the years; none reached a state of completion
comparable to the \pkg{GSL} itself. %Three such wrapping library are
%\url{http://cholm.home.cern.ch/cholm/misc/gslmm/},
%\url{http://gslwrap.sourceforge.net/} and
%\url{http://code.google.com/p/gslcpp/}.
The \pkg{GSL} combines broad coverage of scientific topics, serious
implementation effort, and the use of the well-known GNU General Public
License (GPL). This has lead to fairly wide usage of the library. As a concrete
example, we can consider the Comprehensive R Archive Network (CRAN)
repository network for the \proglang{R} language and environment
\citep{R:Main}. CRAN contains over three dozen packages interfacing the
\pkg{GSL}. Of these more than half interface the vector or matrix classes as
shown in Table~\ref{tab:useOfGSLatCRAN}. This provides empirical evidence indicating
that the \pkg{GSL} is popular among programmers using either the \proglang{C}
or \proglang{C++} language for solving problems applied science.
\begin{table}[p]
\centering
\begin{small}
\begin{tabular}{lccc}
\toprule
Package & Any \texttt{gsl} header & \texttt{gsl\_vector.h} & \texttt{gsl\_matrix.h} \\
\midrule
abn & $\star$ & $\star$ & $\star$ \\
BayesLogit & $\star$ & & \\
BayesSAE & $\star$ & $\star$ & $\star$ \\
BayesVarSel & $\star$ & $\star$ & $\star$ \\
BH & $\star$ & $\star$ & \\
bnpmr & $\star$ & & \\
BNSP & $\star$ & $\star$ & $\star$ \\
cghseg & $\star$ & $\star$ & $\star$ \\
cit & $\star$ & & \\
diversitree & $\star$ & & $\star$ \\
eco & $\star$ & & \\
geoCount & $\star$ & & \\
graphscan & $\star$ & & \\
gsl & $\star$ & $\star$ & \\
gstat & $\star$ & & \\
hgm & $\star$ & & \\
HiCseg & $\star$ & & $\star$ \\
igraph & $\star$ & & \\
KFKSDS & $\star$ & $\star$ & $\star$ \\
libamtrack & $\star$ & & \\
mixcat & $\star$ & $\star$ & $\star$ \\
mvabund & $\star$ & $\star$ & $\star$ \\
outbreaker & $\star$ & $\star$ & $\star$ \\
R2GUESS & $\star$ & $\star$ & $\star$ \\
RCA & $\star$ & & \\
RcppGSL & $\star$ & $\star$ & $\star$ \\
RcppSMC & $\star$ & & \\
RcppZiggurat & $\star$ & & \\
RDieHarder & $\star$ & $\star$ & $\star$ \\
ridge & $\star$ & $\star$ & $\star$ \\
Rlibeemd & $\star$ & $\star$ & \\
Runuran & $\star$ & & \\
SemiCompRisks & $\star$ & & $\star$ \\
simplexreg & $\star$ & $\star$ & $\star$ \\
stsm & $\star$ & $\star$ & $\star$ \\
survSNP & $\star$ & & \\
TKF & $\star$ & $\star$ & $\star$ \\
topicmodels & $\star$ & $\star$ & $\star$ \\
VBLPCM & $\star$ & $\star$ & \\
VBmix & $\star$ & $\star$ & $\star$ \\
\bottomrule
\end{tabular}
\end{small}
\caption{CRAN Package Usage of \pkg{GSL} outright, for vectors and for matrices.}
\label{tab:useOfGSLatCRAN}
\begin{flushleft}
\footnotesize \textsl{Note:} Data gathered in late July 2015 by use of
\texttt{grep} searching (recursively) for inclusion of any GSL header, or
the vector and matrix headers specifically, within the \texttt{src/} or
\texttt{inst/include/} directories of expanded source code archives of
the CRAN network. Convenient (temporary) shell access to such an
expanded code archive via WU Vienna is gratefully acknowledged.
\end{flushleft}
\end{table}
At the same time, the \pkg{Rcpp} package
\citep{JSS:Rcpp,Eddelbuettel:2013:Rcpp,CRAN:Rcpp} offers a
higher-level interface between \proglang{R} and \proglang{C++}.
\pkg{Rcpp} permits \proglang{R} objects like
vectors, matrices, lists, functions, environments, $\ldots$, to be
manipulated directly at the \proglang{C++} level, and alleviates the needs for
complicated and error-prone parameter passing and memory allocation. It also
allows compact vectorised expressions similar to what can be written in
\proglang{R} directly at the \proglang{C++} level.
The \pkg{RcppGSL} package discussed here aims to close the gap. It
offers access to \pkg{GSL} functions, in particular via the vector and
matrix data structures used throughout the \pkg{GSL}, while staying closer to the
`whole object model' familar to the \proglang{R} programmer.
The rest of paper is organised as follows. The next section shows a
motivating example of a fast linear model fit routine using \pkg{GSL} functions.
The following section discusses support for \pkg{GSL} vector types, which is
followed by a section on matrices. The following two section discusses error
handling, and then use of \pkg{RcppGSL} in your own package. This is followed
by short discussions of how to use \pkg{RcppGSL} with \pkg{inline} and
\textsl{Rcpp Attributes}, respectively, before a short concluding summary.
\section{Motivation: fastLm}
Fitting linear models is a key building block of analysing and
modeling data. \proglang{R} has a very complete and feature-rich function in
\texttt{lm()} which provides a model fit as well as a number of diagnostic
measure, either directly or via the \texttt{summary()} method for linear
model fits. The \texttt{lm.fit()} function provides a faster alternative
(which is however recommend only for for advanced users) which provides
estimates only and fewer statistics for inference. This may lead to user
requests for a routine which is both fast and featureful enough. The
\texttt{fastLm} routine shown here provides such an implementation as part of
the \pkg{RcppGSL} package. It uses
the \pkg{GSL} for the least-squares fitting functions and provides a nice
example for \pkg{GSL} integration with \proglang{R}.
<<fastLm,lang=cpp,size=small>>=
#include <RcppGSL.h>
#include <gsl/gsl_multifit.h>
#include <cmath>
// declare a dependency on the RcppGSL package; also activates plugin
// (but not needed when 'LinkingTo: RcppGSL' is used with a package)
//
// [[Rcpp::depends(RcppGSL)]]
// tell Rcpp to turn this into a callable function called 'fastLm'
//
// [[Rcpp::export]]
Rcpp::List fastLm(const RcppGSL::Matrix & X, const RcppGSL::Vector & y) {
int n = X.nrow(), k = X.ncol(); // row and column dimension
double chisq; // assigned but not returned
RcppGSL::Vector coef(k); // to hold the coefficient vector
RcppGSL::Matrix cov(k,k); // and the covariance matrix
// the actual fit requires working memory which we allocate and then free
gsl_multifit_linear_workspace *work = gsl_multifit_linear_alloc (n, k);
gsl_multifit_linear (X, y, coef, cov, &chisq, work);
gsl_multifit_linear_free (work);
// assign diagonal to a vector, then take square roots to get std.error
Rcpp::NumericVector std_err;
std_err = gsl_matrix_diagonal(cov); // need two step decl. and assignment
std_err = Rcpp::sqrt(std_err); // sqrt() is an Rcpp sugar function
return Rcpp::List::create(Rcpp::Named("coefficients") = coef,
Rcpp::Named("stderr") = std_err,
Rcpp::Named("df.residual") = n - k);
}
@
The function interface defines two \pkg{RcppGSL} variables: a matrix and a
vector. Both use the standard numeric type \texttt{double} as discussed
below. The \pkg{GSL} supports other types ranging from lower precision
floating point to signed and unsigned integers as well as complex numbers.
The vector and matrix classes are templated for use with all these
\proglang{C} / \proglang{C++} types---though \proglang{R} uses only
\texttt{double} and \texttt{int}. For these latter two, we offer a shorthand
definition via a \texttt{typedef} which allows a shorter non-template use.
Having extracted the row and column dimentions, we then reserve another
vector and matrix to hold the resulting coefficient estimates as well as the
estimate of the covariance matrix. Next, we allocate workspace using a
\pkg{GSL} routine, fit the linear model and free the just-allocated
workspace. The next step involves extracting the diagonal element from the
covariance matrix, and taking the square root (using a vectorised function
from \pkg{Rcpp}). Finally we create a named list with the return values.
In earlier version of the \pkg{RcppGSL} package, we also explicitly called
\texttt{free()} to return temporary memory allocation to the operating
system. This step had to be done as the underlying objects are managed as
\proglang{C} objects. They conform to the \pkg{GSL} interface, and work
without any automatic memory management. But as we provide a \proglang{C++}
data structure for matrix and vector objects, we can manage them using
\proglang{C++} facilities. In particular, the destructor can free the memory
when the object goes out of scope. Explicit \texttt{free()} calls are still
permitted as we keep track the object status so that memory cannot
accidentally be released more than once. Another more recent addition permits
use of \texttt{const \&} in the interface. This instructs the compiler that
values of the corresponding variable will not be altered, and are passed into
the function by reference rather than by value.
We note that \pkg{RcppArmadillo}
\citep{CRAN:RcppArmadillo,Eddelbuettel+Sanderson:2013:RcppArmadillo}
implements a matching \texttt{fastLm} function using the Armadillo library by
\cite{Sanderson:2010:Armadillo}, and can do so with even more compact code due to
\proglang{C++} features. Moreover, \pkg{RcppEigen}
\citep{CRAN:RcppEigen,JSS:RcppEigen} provides a \texttt{fastLm} implementation with a
comprehensive comparison of matrix decomposition methods.
\section{Vectors}
This section details the different vector represenations, starting with their
definition inside the \pkg{GSL}. We then discuss our layering before showing
how the two types map. A discussion of read-only `vector view' classes
concludes the section.
\subsection{\pkg{GSL} Vectors}
\pkg{GSL} defines various vector types to manipulate one-dimensionnal
data, similar to \proglang{R} arrays. For example the \verb|gsl_vector| and
\verb|gsl_vector_int| structs are defined as:
<<vector_int,lang=cpp,size=small>>=
typedef struct{
size_t size;
size_t stride;
double * data;
gsl_block * block;
int owner;
} gsl_vector;
typedef struct {
size_t size;
size_t stride;
int * data;
gsl_block_int * block;
int owner;
} gsl_vector_int;
@
A typical use of the \verb|gsl_vector| struct is given below:
<<vector_free,lang=cpp,size=small>>=
int i;
gsl_vector *v = gsl_vector_alloc(3); // allocate a gsl_vector of size 3
for (i = 0; i < 3; i++) { // fill the vector
gsl_vector_set(v, i, 1.23 + i);
}
double sum = 0.0; // access elements
for (i = 0; i < 3; i++) {
sum += gsl_vector_set(v, i);
}
gsl_vector_free(v); // free the memory
@
Note that we have to explicitly free the allocated memory at the end. With
\proglang{C}-style programming, this step is always the responsibility of the
programmer.
\subsection{RcppGSL::vector}
\pkg{RcppGSL} defines the template \texttt{RcppGSL::vector<T>} to manipulate
\verb|gsl_vector| pointers taking advantage of C++ templates. Using this
template type, the previous example now becomes:
<<vector_sum,lang=cpp,size=small>>=
int i;
RcppGSL::vector<double> v(3); // allocate a gsl_vector of size 3
for (i = 0; i < 3; i++) { // fill the vector
v[i] = 1.23 + i;
}
double sum = 0.0; // access elements
for (i = 0; i < 3; i++) {
sum += v[i];
}
v.free(); // (optionally) free the memory
// also automatic when out of scope
@
The class \texttt{RcppGSL::vector<double>} is a smart pointer which can be deployed
anywhere where a raw pointer \verb|gsl_vector| can be used, such as the
\verb|gsl_vector_set| and \verb|gsl_vector_get| functions above.
Beyond the convenience of a nicer syntax for allocation (and of course the
managed release of memory either via \texttt{free()} or when going out of
scope), the \texttt{RcppGSL::vector} template faciliates interchange of
\pkg{GSL} vectors with \pkg{Rcpp} objects, and hence \pkg{R} objects. The
following example defines a \texttt{.Call} compatible function called
\verb|sum_gsl_vector_int| that operates on a \verb|gsl_vector_int| through
the \texttt{RcppGSL::vector<int>} template specialization:
<<macro,lang=cpp,size=small>>=
// [[Rcpp::export]]
int sum_gsl_vector_int(const RcppGSL::vector<int> & vec){
int res = std::accumulate(vec.begin(), vec.end(), 0);
return res;
}
@
Here we no longer need to call \texttt{free()} explicitly as the \texttt{vec}
allocation is returned automatically at the end of the function body when the
variable goes out of scope.
Once the function has created via \texttt{sourceCpp()} or
\texttt{cppFunction()} from \textsl{Rcpp Attributes} (see
section~\ref{sec:attributes} for more on this), it can then be called from
\proglang{R} :
<<inlineex1,results=hide,echo=FALSE>>=
fx <- Rcpp::cppFunction("int sum_gsl_vector_int(RcppGSL::vector<int> vec) {
int res = std::accumulate( vec.begin(), vec.end(), 0);
return res;
}", depends="RcppGSL")
<<callinlineex1,size=small>>=
sum_gsl_vector_int(1:10)
@
A second example shows a simple function that grabs elements of an
R list as \verb|gsl_vector| objects using implicit conversion mechanisms
of \pkg{Rcpp}
<< example2,lang=cpp,size=small>>=
// [[Rcpp::export]]
double gsl_vector_sum_2(const Rcpp::List & data) {
// grab "x" as a gsl_vector through the RcppGSL::vector<double> class
const RcppGSL::vector<double> x = data["x"];
// grab "y" as a gsl_vector through the RcppGSL::vector<int> class
const RcppGSL::vector<int> y = data["y"];
double res = 0.0;
for (size_t i=0; i< x->size; i++) {
res += x[i] * y[i];
}
return res; // return the result, memory freed automatically
}
@
called from \proglang{R}:
<<inlinexex2,results=hide, echo=FALSE>>=
Rcpp::cppFunction("double gsl_vector_sum_2(Rcpp::List data) {
RcppGSL::vector<double> x = data[\"x\"];
RcppGSL::vector<int> y = data[\"y\"];
double res = 0.0;
for (size_t i=0; i< x->size; i++) {
res += x[i] * y[i];
}
return res;
}", depends= "RcppGSL")
@
<<callinlineex2,size=small>>=
data <- list( x = seq(0,1,length=10), y = 1:10 )
gsl_vector_sum_2(data)
@
\subsection{Mapping}
Table~\ref{tab:mappingVectors} shows the mapping between types defined by the
\pkg{GSL} and their corresponding types in the \pkg{RcppGSL} package.
\begin{table}[htb]
\centering
\begin{small}
\begin{tabular}{ll}
\toprule
GSL vector & RcppGSL \\
\midrule
\texttt{gsl\_vector} & \texttt{RcppGSL::vector<double>} as well as \texttt{RcppGSL::Vector} \\
\texttt{gsl\_vector\_int} & \texttt{RcppGSL::vector<int>} as well as \texttt{RcppGSL::IntVector} \\
\texttt{gsl\_vector\_float} & \texttt{RcppGSL::vector<float>} \\
\texttt{gsl\_vector\_long} & \texttt{RcppGSL::vector<long>} \\
\texttt{gsl\_vector\_char} & \texttt{RcppGSL::vector<char>} \\
\texttt{gsl\_vector\_complex} & \texttt{RcppGSL::vector<gsl\_complex>} \\
\texttt{gsl\_vector\_complex\_float} & \texttt{RcppGSL::vector<gsl\_complex\_float>} \\
\texttt{gsl\_vector\_complex\_long\_double} & \texttt{RcppGSL::vector<gsl\_complex\_long\_double>} \\
\texttt{gsl\_vector\_long\_double} & \texttt{RcppGSL::vector<long double>} \\
\texttt{gsl\_vector\_short} & \texttt{RcppGSL::vector<short>} \\
\texttt{gsl\_vector\_uchar} & \texttt{RcppGSL::vector<unsigned char>} \\
\texttt{gsl\_vector\_uint} & \texttt{RcppGSL::vector<unsigned int>} \\
\texttt{gsl\_vector\_ushort} & \texttt{RcppGSL::vector<insigned short>} \\
\texttt{gsl\_vector\_ulong} & \texttt{RcppGSL::vector<unsigned long>} \\
\bottomrule
\end{tabular}
\end{small}
\caption{Correspondance between \pkg{GSL} vector types and templates defined in \pkg{RcppGSL}.}
\label{tab:mappingVectors}
\end{table}
As shown, we also define two convenient shortcuts for the very common case of
\texttt{double} and \texttt{int} vectors. First, \texttt{RcppGSL::Vector} is a
short-hand for the \texttt{RcppGSL::vector<double>} template
instantiation. Second, \texttt{RcppGSL::IntVector} does the same for
integer-valued vectors. Other types still require explicit templates.
\subsection{ Vector Views}
Several \pkg{GSL} algorithms return \pkg{GSL} vector views as their result
type. \pkg{RcppGSL} defines the template class \texttt{RcppGSL::vector\_view}
to handle vector views using \proglang{C++} syntax.
<<lang=cpp,size=small>>=
// [[Rcpp::export]]
Rcpp::List test_gsl_vector_view() {
int n = 10;
RcppGSL::vector<double> v(n);
for (int i=0 ; i<n; i++) {
v[i] = i;
}
const RcppGSL::vector_view<double> v_even =
gsl_vector_subvector_with_stride(v,0,2,n/2);
const RcppGSL::vector_view<double> v_odd =
gsl_vector_subvector_with_stride(v,1,2,n/2);
return Rcpp::List::create(Rcpp::Named("even") = v_even,
Rcpp::Named("odd" ) = v_odd);
}
@
As with vectors, \proglang{C++} objects of type
\texttt{RcppGSL::vector\_view} can be converted implicitly to their
associated \pkg{GSL} view type. Table~\ref{tab:mappingVectorViews} displays
the pairwise correspondance so that the \proglang{C++} objects can be passed
to compatible \pkg{GSL} algorithms.
\begin{table}[htb]
\centering
\begin{small}
\begin{tabular}{ll}
\toprule
gsl vector views & RcppGSL \\
\midrule
\texttt{gsl\_vector\_view} & \texttt{RcppGSL::vector\_view<double>}; \texttt{RcppGSL::VectorView} \\
\texttt{gsl\_vector\_view\_int} & \texttt{RcppGSL::vector\_view<int>}; \texttt{RcppGSL::IntVectorView} \\
\texttt{gsl\_vector\_view\_float} & \texttt{RcppGSL::vector\_view<float>} \\
\texttt{gsl\_vector\_view\_long} & \texttt{RcppGSL::vector\_view<long>} \\
\texttt{gsl\_vector\_view\_char} & \texttt{RcppGSL::vector\_view<char>} \\
\texttt{gsl\_vector\_view\_complex} & \texttt{RcppGSL::vector\_view<gsl\_complex>} \\
\texttt{gsl\_vector\_view\_complex\_float} & \texttt{RcppGSL::vector\_view<gsl\_complex\_float>} \\
\texttt{gsl\_vector\_view\_complex\_long\_double} & \texttt{RcppGSL::vector\_view<gsl\_complex\_long\_double>} \\
\texttt{gsl\_vector\_view\_long\_double} & \texttt{RcppGSL::vector\_view<long double>} \\
\texttt{gsl\_vector\_view\_short} & \texttt{RcppGSL::vector\_view<short>} \\
\texttt{gsl\_vector\_view\_uchar} & \texttt{RcppGSL::vector\_view<unsigned char>} \\
\texttt{gsl\_vector\_view\_uint} & \texttt{RcppGSL::vector\_view<unsigned int>} \\
\texttt{gsl\_vector\_view\_ushort} & \texttt{RcppGSL::vector\_view<insigned short>} \\
\texttt{gsl\_vector\_view\_ulong} & \texttt{RcppGSL::vector\_view<unsigned long>} \\
\bottomrule
\end{tabular}
\end{small}
\caption{Correspondance between \pkg{GSL} vector view types and templates defined
in \pkg{RcppGSL}.}
\label{tab:mappingVectorViews}
\end{table}
The vector view class also contains a conversion operator to automatically
transform the data of the view object to a \pkg{GSL} vector object. This
enables use of vector views where \pkg{GSL} would expect a vector. And as
before, \texttt{double} and \texttt{int} types can be accessed via the
\texttt{typedef} variants \texttt{RcppGSL::VectorView} and
\texttt{RcppGSL::IntVectorView}, respectively.
Lastly, in order to support \texttt{const \&} behaviour, all
\texttt{gsl\_vector\_XXX\_const\_view} variants are also supported (where
\texttt{XXX} stands for any of the atomistic \proglang{C} and \proglang{C++}
data types).
\section{Matrices}
The \pkg{GSL} also defines a set of matrix data types : \texttt{gsl\_matrix},
\texttt{gsl\_matrix\_int} etc ... for which \pkg{RcppGSL} defines a corresponding
convenience \proglang{C++} wrapper generated by the \texttt{RcppGSL::matrix}
template.
\subsection{Creating matrices}
The \texttt{RcppGSL::matrix} template exposes three constructors.
<<lang=cpp,size=small>>=
// convert an R matrix to a GSL matrix
matrix(SEXP x) throw(::Rcpp::not_compatible)
// encapsulate a GSL matrix pointer
matrix(gsl_matrix* x)
// create a new matrix with the given number of rows and columns
matrix(int nrow, int ncol)
@
\subsection{Implicit conversion}
\texttt{RcppGSL::matrix} defines an implicit conversion to a pointer to the
associated \pkg{GSL} matrix type, as well as dereferencing operators. This
makes the class \texttt{RcppGSL::matrix} look and feel like a pointer to a
\pkg{GSL} matrix type.
<<lang=cpp,size=small>>=
gsltype* data;
operator gsltype*() { return data; }
gsltype* operator->() const { return data; }
gsltype& operator*() const { return *data; }
@
\subsection{Indexing}
Indexing of \pkg{GSL} matrices is usually the task of the functions
\texttt{gsl\_matrix\_get}, \texttt{gsl\_matrix\_int\_get}, ... and
\texttt{gsl\_matrix\_set}, \texttt{gsl\_matrix\_int\_set}, ...
\pkg{RcppGSL} takes advantage of both operator overloading and templates
to make indexing a \pkg{GSL} matrix much more convenient.
<<lang=cpp,size=small>>=
RcppGSL::matrix<int> mat(10,10); // create a matrix of size 10x10
for (int i=0; i<10: i++) { // fill the diagonal
mat(i,i) = i; // no need for setter function
}
@
\subsection{Methods}
The \texttt{RcppGSL::matrix} type also defines the following member functions:
\begin{quote}
\begin{itemize}
\item[\texttt{nrow}] extracts the number of rows
\item[\texttt{ncol}] extract the number of columns
\item[\texttt{size}] extracts the number of elements
\item[\texttt{free}] releases the memory (also called via destructor)
\end{itemize}
\end{quote}
\subsection{Matrix views}
Similar to the vector views discussed above, the \pkg{RcppGSL} also provides
an implicit conversion operator which returns the underlying matrix stored in
the matrix view class.
%% next section contributed by Qiang Kou
\section{Error handler}
When input values for \pkg{GSL} functions are invalid, the default error
handler will abort the program after printing an error message. This
leads \proglang{R} to an abortion error. To avoid this behaviour, one needs to avoid it
first by using \texttt{gsl\_set\_error\_handler\_off()}, and then detect
error conditions by checking whether the result is \texttt{NAN} or not.
<<lang=cpp,size=small>>=
// close the GSL error handler
gsl_set_error_handler_off();
// call GSL function with some invalid values
double res = gsl_sf_hyperg_2F1(1, 1, 1.1467003, 1);
// detect the result is NAN or not
if (ISNAN(res)) {
Rcpp::Rcout << "Invalid input found!" << std::endl;
}
@
See \url{http://thread.gmane.org/gmane.comp.lang.r.rcpp/7905} for a longer
discussion of the related issues.
Starting with release 0.2.4, two new functions are available:
\texttt{gslSetErrorHandlerOff()} and \texttt{gslResetErrorHandler()} which
allow to turn off the error handler (as discussed above), and to reset to
the prior (default) value. In addition, the package now also calls
\texttt{gslSetErrorHandlerOff()} when being attached, ensuring that the
\pkg{GSL} error handler is turned off by default.
\section{Using \pkg{RcppGSL} in your package}
The \pkg{RcppGSL} package contains a complete example package providing a single
function \texttt{colNorm} which computes a norm for each column of a
supplied matrix. This example adapts a matrix example from the \pkg{GSL} manual that has
been chosen primarily as a means to showing how to set up a package to use
\pkg{RcppGSL}.
Needless to say, we could compute such a matrix norm easily in \proglang{R}
using existing facilities. One such possibility is a simple
\verb|apply(M, 2, function(x) sqrt(sum(x^2)))| as shown on the corresponding
help page in the example package inside \pkg{RcppGSL}. One point in favour of
using the \pkg{GSL} code is that it employs a BLAS function so on
sufficiently large matrices, and with suitable BLAS libraries installed, this
variant could be faster due to the optimised code in high-performance BLAS
libraries and/or the inherent parallelism a multi-core BLAS variant which
compute compute the vector norm in parallel. On all `reasonable' matrix
sizes, however, the performance difference should be neglible.
\subsection{The \texttt{configure} script}
\subsubsection{Using autoconf}
Using \pkg{RcppGSL} means employing both the \pkg{GSL} and \proglang{R}. We
may need to find the location of the \pkg{GSL} headers and library, and this
done easily from a \texttt{configure} source script which \texttt{autoconf}
generates from a \texttt{configure.in} source file such as the following:
<<lang=sh,size=small>>=
AC_INIT([RcppGSLExample], 0.1.0)
## Use gsl-config to find arguments for compiler and linker flags
##
## Check for non-standard programs: gsl-config(1)
AC_PATH_PROG([GSL_CONFIG], [gsl-config])
## If gsl-config was found, let's use it
if test "${GSL_CONFIG}" != ""; then
# Use gsl-config for header and linker arguments (without BLAS which we get from R)
GSL_CFLAGS=`${GSL_CONFIG} --cflags`
GSL_LIBS=`${GSL_CONFIG} --libs-without-cblas`
else
AC_MSG_ERROR([gsl-config not found, is GSL installed?])
fi
# Now substitute these variables in src/Makevars.in to create src/Makevars
AC_SUBST(GSL_CFLAGS)
AC_SUBST(GSL_LIBS)
AC_OUTPUT(src/Makevars)
@
A source file such as this \texttt{configure.in} gets converted into a script
\texttt{configure} by invoking the \texttt{autoconf} program.
We note that many other libraries use a similar (but somewhat newer and
by-now fairly standard) scripting frontend called \texttt{pkg-config} which
be deployed in a very similar by other packages. Calls such as the following
two can be used from \texttt{configure} in a very similar manner:
<<lang=sh,size=small>>=
pkg-config --cflags libpng
pkg-config --libs libpng
@
where \texttt{libpng} (for the png image format) is used just for illustration.
\subsubsection{Using functions provided by RcppGSL}
\pkg{RcppGSL} provides R functions (in the file \texttt{R/inline.R}) that
allow us to retrieve the same information. Therefore the configure script can
also be written as:
<<lang=bash,size=small>>=
#!/bin/sh
GSL_CFLAGS=`${R_HOME}/bin/Rscript -e "RcppGSL:::CFlags()"`
GSL_LIBS=`${R_HOME}/bin/Rscript -e "RcppGSL:::LdFlags()"`
sed -e "s|@GSL_LIBS@|${GSL_LIBS}|" \
-e "s|@GSL_CFLAGS@|${GSL_CFLAGS}|" \
src/Makevars.in > src/Makevars
@
Similarly, the configure.win for windows can be written as:
<<lang=sh,size=small>>=
GSL_CFLAGS=`${R_HOME}/bin${R_ARCH_BIN}/Rscript.exe -e "RcppGSL:::CFlags()"`
GSL_LIBS=`${R_HOME}/bin${R_ARCH_BIN}/Rscript.exe -e "RcppGSL:::LdFlags()"`
sed -e "s|@GSL_LIBS@|${GSL_LIBS}|" \
-e "s|@GSL_CFLAGS@|${GSL_CFLAGS}|" \
src/Makevars.in > src/Makevars.win
@
This allows for a simpler and more direct way to just set the compile and
link options, taking advantage of the installed \pkg{RcppGSL} package. See
the \pkg{RcppZiggurat} package for an example.
\subsection{The \texttt{src} directory}
The \proglang{C++} source file takes the matrix supplied from \proglang{R}
and applies the \pkg{GSL} function to each column.
<<lang=cpp,size=small>>=
#include <RcppGSL.h>
#include <gsl/gsl_matrix.h>
#include <gsl/gsl_blas.h>
// [[Rcpp::export]]
Rcpp::NumericVector colNorm(const RcppGSL::Matrix & G) {
int k = G.ncol();
Rcpp::NumericVector n(k); // to store results
for (int j = 0; j < k; j++) {
RcppGSL::vector_view<double> colview = gsl_matrix_const_column(G, j);
n[j] = gsl_blas_dnrm2(colview);
}
return n; // return vector
}
@
The \proglang{Makevars.in} file governs the compilation and uses the values
supplied by \texttt{configure} during build-time:
<<lang=sh,size=small>>=
# set by configure
GSL_CFLAGS = @GSL_CFLAGS@
GSL_LIBS = @GSL_LIBS@
# combine with standard arguments for R
PKG_CPPFLAGS = $(GSL_CFLAGS)
PKG_LIBS = $(GSL_LIBS)
@
The variables surrounded by \@ will be filled by \texttt{configure} during
package build-time. As discussed above, this can either rely on
\texttt{autoconf} or a possibly-simpler \texttt{Rscript}.
\subsection{The \texttt{R} directory}
The \proglang{R} source is very simply: it contains a single file created by
the \texttt{Rcpp::compileAttributes()} function implementing the wrapper to
the \texttt{colNorm()} function.
\section{Using \pkg{RcppGSL} with \pkg{inline}}
The \pkg{inline} package \citep{CRAN:inline} is very helpful for prototyping code in
\proglang{C}, \proglang{C++} or \proglang{Fortran} as it takes care of code
compilation, linking and dynamic loading directly from \proglang{R}. It has been
used extensively by \pkg{Rcpp}, for example in the numerous unit tests.
The example below shows how \pkg{inline} can be deployed with
\pkg{RcppGSL}. We implement the same column norm example, but this time as an
\proglang{R} script which is compiled, linked and loaded on-the-fly. Compared
to standard use of \pkg{inline}, we have to make sure to add a short section
declaring which header files from \pkg{GSL} we need to use; the \pkg{RcppGSL}
then communicates with \pkg{inline} to tell it about the location and names
of libraries used to build code against \pkg{GSL}.
<<>>=
require(inline)
inctxt='
#include <gsl/gsl_matrix.h>
#include <gsl/gsl_blas.h>
'
bodytxt='
RcppGSL::matrix<double> M = sM; // create data structures from SEXP
int k = M.ncol();
Rcpp::NumericVector n(k); // to store results
for (int j = 0; j < k; j++) {
RcppGSL::vector_view<double> colview = gsl_matrix_column (M, j);
n[j] = gsl_blas_dnrm2(colview);
}
return n; // return vector
'
foo <- cxxfunction(
signature(sM="numeric"),
body=bodytxt, inc=inctxt, plugin="RcppGSL")
## see Section 8.4.13 of the GSL manual: create M as a sum of two outer products
M <- outer(sin(0:9), rep(1,10), "*") + outer(rep(1, 10), cos(0:9), "*")
foo(M)
@
The \texttt{RcppGSL} inline plugin supports creation of a package skeleton
based on the inline function.
<<eval=FALSE>>=
package.skeleton("mypackage", foo)
@
\section{Using \pkg{RcppGSL} with Rcpp Attributes}
\label{sec:attributes}
\textsl{Rcpp Attributes} \citep{CRAN:Rcpp:Attributes} builds on the features
of the \pkg{inline} package described in previous section, and streamlines
the compilation, loading and linking process even further. It leverages the
existing plugins for \pkg{inline}. We already showed the corresponding
function in the previous section. Here, we show it again as a self-contained
example used via \texttt{sourceCpp()}. We stress that usage is
\texttt{sourceCpp()} is meant for interactive work at the R command-prompt,
but is not the recommended practice in a package.
<<colNormAttr,lang=cpp,size=small>>=
#include <gsl/gsl_matrix.h>
#include <gsl/gsl_blas.h>
#include <RcppGSL.h>
// declare a dependency on the RcppGSL package; also activates plugin
//
// [[Rcpp::depends(RcppGSL)]]
// declare the function to be 'exported' to R
//
// [[Rcpp::export]]
Rcpp::NumericVector colNorm(const RcppGSL::Matrix & M) {
int k = M.ncol();
Rcpp::NumericVector n(k); // to store results
for (int j = 0; j < k; j++) {
RcppGSL::VectorView colview = gsl_matrix_const_column (M, j);
n[j] = gsl_blas_dnrm2(colview);
}
return n; // return vector
}
/*** R
## see Section 8.4.13 of the GSL manual:
## create M as a sum of two outer products
M <- outer(sin(0:9), rep(1,10), "*") +
outer(rep(1, 10), cos(0:9), "*")
colNorm(M)
*/
@
With the code above stored in a file, say, ``gslNorm.cpp'' one can simply
call \texttt{sourceCpp()} to have the wrapper code added, and all of the
compilation, linking and loading done --- including the execution of the
short \proglang{R} segment at the end:
<<eval=FALSE>>=
sourceCpp("gslNorm.cpp")
@
The function \texttt{cppFunction()} is also available to convert a simple
character string argument containing a valid C++ function into a eponymous R
function. And like \texttt{sourceCpp()}, it can also use plugins. See the
vignette ``Rcpp-attributes'' \citep{CRAN:Rcpp:Attributes} of the \pkg{Rcpp}
package \citep{CRAN:Rcpp} for full details.
\section{Summary}
The GNU Scientific Library (GSL) by \citet{GSL} offers a very comprehensive
collection of rigorously developed and tested functions for applied
scientific computing under a widely-used and well-understood Open Source
license. This has lead to widespread deployment of \pkg{GSL} among a number
of disciplines.
Using the automatic wrapping and converters offered by the \pkg{RcppGSL}
package presented here, \proglang{R} users and programmers can now deploy
algorithmns provided by the \pkg{GSL} with greater ease.
\bibliographystyle{plainnat}
\bibliography{\Sexpr{Rcpp:::bib()}}
\end{document}