Scribd
Upload
3 views

Introduction

The Matrix package in R provides S4 classes for efficient manipulation of dense and sparse matrices, extending the basic matrix data type. It utilizes advanced linear algebra routines from libraries like BLAS and Lapack for dense matrices, and TAUCS and UMFPACK for sparse matrices, allowing for optimized computations. Notably, the package preserves matrix factorizations for reuse in subsequent operations, enhancing computational efficiency.

Uploaded by

sahril safikri
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
3 views

Introduction

The Matrix package in R provides S4 classes for efficient manipulation of dense and sparse matrices, extending the basic matrix data type. It utilizes advanced linear algebra routines from libraries like BLAS and Lapack for dense matrices, and TAUCS and UMFPACK for sparse matrices, allowing for optimized computations. Notably, the package preserves matrix factorizations for reuse in subsequent operations, enhancing computational efficiency.

Uploaded by

sahril safikri
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Introduction to the Matrix package — as of

Feb. 2005∗
Douglas Bates
R Core Development Group
[email protected]
May 17, 2023

Abstract
Linear algebra is at the core of many areas of statistical computing and
from its inception the S language has supported numerical linear algebra
via a matrix data type and several functions and operators, such as %*%,
qr, chol, and solve. However, these data types and functions do not
provide direct access to all of the facilities for efficient manipulation of
dense matrices, as provided by the Lapack subroutines, and they do not
provide for manipulation of sparse matrices.
The Matrix package provides a set of S4 classes for dense and sparse
matrices that extend the basic matrix data type. Methods for a wide va-
riety of functions and operators applied to objects from these classes pro-
vide efficient access to BLAS (Basic Linear Algebra Subroutines), Lapack
(dense matrix), TAUCS (sparse matrix) and UMFPACK (sparse matrix)
routines. One notable characteristic of the package is that whenever a ma-
trix is factored, the factorization is stored as part of the original matrix
so that further operations on the matrix can reuse this factorization.

1 Introduction
Linear algebra is at the core of many statistical computing techniques and, from
its inception, the S language has supported numerical linear algebra via a matrix
data type and several functions and operators, such as %*%, qr, chol, and solve.
Initially the numerical linear algebra functions in R called underlying Fortran
routines from the Linpack (Dongarra et al., 1979) and Eispack Smith et al.
(1976) libraries but over the years most of these functions have been switched
to use routines from the Lapack Anderson et al. (1999) library. Furthermore, R
can be configured to use accelerated BLAS (Basic Linear Algebra Subroutines),
∗ There’s an unfinished “2nd Introduction to the Matrix package” which contains partly

newer information, but is not at all self-contained. Eventually that will replace this one.

1
such as those from the Atlas Whaley et al. (2001) project or Goto’s BLAS Goto
and van de Geijn (2002).
Lapack provides routines for operating on several special forms of matrices,
such as triangular matrices and symmetric matrices. Furthermore,matrix de-
compositions like the QR decompositions produce multiple output components
that should be regarded as parts of a single object. There is some support in R
for operations on special forms of matrices (e.g. the backsolve, forwardsolve
and chol2inv functions) and for special structures (e.g. a QR structure is im-
plicitly defined as a list by the qr, qr.qy, qr.qty, and related functions) but it
is not as fully developed as it could be.
Also there is no direct support for sparse matrices in R although Koenker
and Ng (2003) have developed a contributed package for sparse matrices based
on SparseKit.
The Matrix package provides S4 classes and methods for dense and sparse
matrices. The methods for dense matrices use Lapack and BLAS. The sparse
matrix methods use TAUCS (Toledo, 2003), UMFPACK (Davis, 2003), and
Metis (Karapis, 2003).

2 Classes for dense matrices


The Matrix package will provide classes for real (stored as double precision)
and complex (stored as double precision complex) dense matrices. At present
only the real classes have been implemented. These classes are
dgeMatrix Real matrices in general storage mode
dsyMatrix Symmetric real matrices in non-packed storage
dspMatrix Symmetric real matrices in packed storage (one triangle only)

dtrMatrix Triangular real matrices in non-packed storage


dtpMatrix Triangular real matrices in packed storage (triangle only)
dpoMatrix Positive semi-definite symmetric real matrices in non-packed stor-
age

dppMatrix ditto in packed storage


Methods for these classes include coercion between these classes, when appro-
priate, and coercion to the matrix class; methods for matrix multiplication
(%*%); cross products (crossprod), matrix norm (norm); reciprocal condition
number (rcond); LU factorization (lu) or, for the poMatrix class, the Cholesky
decomposition (chol); and solutions of linear systems of equations (solve).
Further, group methods have been defined for the Arith (basic arithmetic,
including with scalar numbers) and the Math (basic mathematical functions)
group..

2
Whenever a factorization or a decomposition is calculated it is preserved as a
(list) element in the factors slot of the original object. In this way a sequence of
operations, such as determining the condition number of a matrix then solving
a linear system based on the matrix, do not require multiple factorizations of
the same matrix nor do they require the user to store the intermediate results.

3 Classes for sparse matrices


3.1 Representations of sparse matrices
Conceptually, the simplest representation of a sparse matrix is as a triplet of an
integer vector i giving the row numbers, an integer vector j giving the column
numbers, and a numeric vector x giving the non-zero values in the matrix. An
S4 class definition might be
setClass("dgTMatrix",
representation(i = "integer", j = "integer", x = "numeric",
Dim = "integer"))
The triplet representation is row-oriented if elements in the same row were
adjacent and column-oriented if elements in the same column were adjacent. The
compressed sparse row (csr) (or compressed sparse column - csc) representation
is similar to row-oriented triplet (column-oriented triplet) except that i (j) just
stores the index of the first element in the row (column). (There are a couple
of other details but that is the gist of it.) These compressed representations
remove the redundant row (column) indices and provide faster access to a given
location in the matrix because you only need to check one row (column).
The preferred representation of sparse matrices in the SparseM package
is csr. Matlab uses csc. We hope that Octave will also use this representa-
tion. There are certain advantages to csc in systems like R and Matlab where
dense matrices are stored in column-major order. For example, Sivan Toledo’s
TAUCS Toledo (2003) library and Tim Davis’s UMFPACK Davis (2003) library
are both based on csc and can both use level-3 BLAS in certain sparse matrix
computations.
The Matrix package provides the following classes for sparse matrices
dgTMatrix general, numeric, sparse matrices in (a possibly redundant) triplet
form. This can be a convenient form in which to construct sparse matrices.
dgCMatrix general, numeric, sparse matrices in the (sorted) compressed sparse
column format.
dsCMatrix symmetric, real, sparse matrices in the (sorted) compressed sparse
column format. Only the upper or the lower triangle is stored. Although
there is provision for both forms, the lower triangle form works best with
TAUCS.
dtCMatrix triangular, real, sparse matrices in the (sorted) compressed sparse
column format.

3
References
E. Anderson, Z. Bai, C. Bischof, S. Blackford, J. Demmel, J. Dongarra, J. Du
Croz, A. Greenbaum, S. Hammarling, A. McKenney, and D. Sorensen. LA-
PACK Users’ Guide. SIAM, Philadelphia, PA, 3rd edition, 1999.
Tim Davis. Umfpack: Unified multifrontal package.
http://www.cise.ufl.edu/research/sparse/umfpack, 2003.
Jack Dongarra, Cleve Moler, Bunch, and G.W. Stewart. Linpack Users’ Guide.
SIAM, 1979.

Kazushige Goto and Robert van de Geijn. On reducing tlb misses in matrix
multiplication. Technical Report TR02-55, Department of Computer Sciences,
U. of Texas at Austin, 2002.
George Karapis. Metis: Family of multilevel partioning algorithms.
http://www-users.cs.umn.edu/ karypis/metis/, 2003.
Roger Koenker and Pin Ng. SparseM: A sparse matrix package for R. J. of
Statistical Software, 8(6), 2003.
B. T. Smith, J. M. Boyle, J. J. Dongarra, B. S. Garbow, Y. Ikebe, V. C. Klema,
and C. B. Moler. Matrix Eigensystem Routines. EISPACK Guide, volume 6
of Lecture Notes in Computer Science. Springer-Verlag, New York, 1976.
Sivan Toledo. Taucs: A library of sparse linear solvers.
http://www.tau.ac.il/ stoledo/taucs/, 2003.
R. Clint Whaley, Antoine Petitet, and Jack J. Dongarra. Auto-
mated empirical optimization of software and the ATLAS project.
Parallel Computing, 27(1–2):3–35, 2001. Also available as Univer-
sity of Tennessee LAPACK Working Note #147, UT-CS-00-448, 2000
(www.netlib.org/lapack/lawns/lawn147.ps).

You might also like