| Title: | Sparse Matrix Format with Data on Disk |
| Version: | 0.7.3 |
| Description: | Provide a sparse matrix format with data stored on disk, to be used in both R and C++. This is intended for more efficient use of sparse data in C++ and also when parallelizing, since data on disk does not need copying. Only a limited number of features will be implemented. For now, conversion can be performed from a 'dgCMatrix' or a 'dsCMatrix' from R package 'Matrix'. A new compact format is also now available. |
| License: | GPL-3 |
| Encoding: | UTF-8 |
| RoxygenNote: | 7.2.3 |
| URL: | https://github.com/privefl/bigsparser |
| BugReports: | https://github.com/privefl/bigsparser/issues |
| Depends: | R (≥ 3.1) |
| LinkingTo: | Rcpp, RcppEigen, rmio |
| Imports: | Rcpp, bigassertr, methods, Matrix, rmio (≥ 0.4) |
| Suggests: | testthat (≥ 2.1.0) |
| NeedsCompilation: | yes |
| Packaged: | 2024-09-06 08:24:21 UTC; au639593 |
| Author: | Florian Privé [aut, cre] |
| Maintainer: | Florian Privé <florian.prive.21@gmail.com> |
| Repository: | CRAN |
| Date/Publication: | 2024-09-06 15:40:06 UTC |
bigsparser: Sparse Matrix Format with Data on Disk
Description
Provide a sparse matrix format with data stored on disk, to be used in both R and C++. This is intended for more efficient use of sparse data in C++ and also when parallelizing, since data on disk does not need copying. Only a limited number of features will be implemented. For now, conversion can be performed from a 'dgCMatrix' or a 'dsCMatrix' from R package 'Matrix'. A new compact format is also now available.
Author(s)
Maintainer: Florian Privé florian.prive.21@gmail.com
See Also
Useful links:
Class SFBM
Description
A reference class for storing and accessing sparse matrix-like data stored in files on disk.
Convert a 'dgCMatrix' or 'dsCMatrix' to an SFBM.
Usage
as_SFBM(spmat, backingfile = tempfile(), compact = FALSE)
Arguments
spmat |
A 'dgCMatrix' (non-symmetric sparse matrix of type 'double') or 'dsCMatrix' (symmetric sparse matrix of type 'double'). |
backingfile |
Path to file where to store data. Extension |
compact |
Whether to use a compact format? Default is |
Details
An object of class SFBM has many fields:
-
$address: address of the external pointer containing the underlying C++ object to be used as aXPtr<SFBM>in C++ code -
$extptr: (internal) use$addressinstead -
$nrow: number of rows -
$ncol: number of columns -
$nval: number of non-zero values -
$p: vector of column positions -
$backingfileor$sbk: File with extension 'sbk' that stores the data of the SFBM -
$rds: 'rds' file (that may not exist) corresponding to the 'sbk' file -
$is_saved: whether this object is stored in$rds?
And some methods:
-
$save(): Save the SFBM object in$rds. Returns the SFBM. -
$add_columns(): Add new columns from a 'dgCMatrix' or a 'dsCMatrix'. -
$dense_acc(): Equivalent toas.matrix(.[ind_row, ind_col]). Use with caution;ind_rowandind_colmust be positive indices within range.
Value
The new SFBM.
Examples
spmat2 <- Matrix::Diagonal(4, 0:3)
spmat2[4, 2] <- 5
spmat2[1, 4] <- 6
spmat2[3, 4] <- 7
spmat2
# Stores all (i, x) for x != 0
(X2 <- as_SFBM(spmat2))
matrix(readBin(X2$sbk, what = double(), n = 100), 2)
# Stores only x, but all (even the zero ones) from first to last being not 0
(X3 <- as_SFBM(spmat2, compact = TRUE))
X3$first_i
readBin(X3$sbk, what = double(), n = 100)
Class SFBM_compact
Description
A reference class for storing and accessing sparse matrix-like data stored in files on disk, in a compact format (when non-zero values in columns are contiguous).
Details
It inherits the fields and methods from class SFBM.
Class SFBM_corr_compact
Description
A reference class for storing and accessing from disk a sparse correlation matrix where non-zero values in columns are mostly contiguous. It rounds correlation values with precision 1/32767 to store them using 2 bytes only. This class has been specifically designed for package 'bigsnpr'.
Convert a 'dgCMatrix' or 'dsCMatrix' to an SFBM_corr_compact.
Usage
as_SFBM_corr_compact(spmat, backingfile = tempfile())
Arguments
spmat |
A 'dgCMatrix' (non-symmetric sparse matrix of type 'double') or 'dsCMatrix' (symmetric sparse matrix of type 'double'). |
backingfile |
Path to file where to store data. Extension |
Details
It inherits the fields and methods from class SFBM_compact.
Value
The new SFBM_corr_compact.
Examples
spmat2 <- as(cor(iris[1:4]), "dsCMatrix")
(X2 <- as_SFBM_corr_compact(spmat2))
(bin <- readBin(X2$sbk, what = integer(), size = 2, n = 100))
matrix(bin / 32767, 4)
spmat2
Accessor methods for class SFBM.
Description
Accessor methods for class SFBM.
Usage
## S4 method for signature 'SFBM,ANY,ANY,ANY'
x[i, j, ..., drop = FALSE]
## S4 method for signature 'SFBM_compact,ANY,ANY,ANY'
x[i, j, ..., drop = FALSE]
## S4 method for signature 'SFBM_corr_compact,ANY,ANY,ANY'
x[i, j, ..., drop = FALSE]
Arguments
x |
A SFBM object. |
i |
A vector of indices (or nothing). You can use positive and negative indices, and also logical indices (that are recycled). |
j |
A vector of indices (or nothing). You can use positive and negative indices, and also logical indices (that are recycled). |
... |
Not used. Just to make nargs work. |
drop |
Not implemented; always return a sparse matrix ( |
Examples
spmat <- Matrix::Diagonal(4, 0:3)
spmat[4, 2] <- 5
spmat[1, 4] <- 6
spmat[3, 4] <- 7
spmat
X <- as_SFBM(spmat)
X[1:3, 2:3]
X[, 4] # parameter drop is not implemented
X[-1, 3:4]
X$dense_acc(2:4, 3:4)
X2 <- as_SFBM(spmat, compact = TRUE)
X2[1:3, 2:3]
X2$dense_acc(1:3, 2:3)
Dimension and type methods for class SFBM.
Description
Dimension and type methods for class SFBM.
Usage
## S4 method for signature 'SFBM'
dim(x)
## S4 method for signature 'SFBM'
length(x)
## S4 method for signature 'SFBM'
diag(x)
## S4 method for signature 'SFBM_compact'
diag(x)
## S4 method for signature 'SFBM_corr_compact'
diag(x)
Arguments
x |
An object of class SFBM. |
Products with a vector
Description
Products between an SFBM and a vector.
Usage
sp_prodVec(X, y)
sp_cprodVec(X, y)
Arguments
X |
An SFBM. |
y |
A vector of same size of the number of columns of |
Value
-
sp_prodVec(): the vector which is equivalent toX %*% yifXwas a dgCMatrix. -
sp_cprodVec(): the vector which is equivalent toMatrix::crossprod(X, y)ifXwas a dgCMatrix.
Examples
spmat <- Matrix::rsparsematrix(1000, 1000, 0.01)
X <- as_SFBM(spmat)
sp_prodVec(X, rep(1, 1000))
sp_cprodVec(X, rep(1, 1000))
Solver for symmetric SFBM
Description
Solve Ax=b where A is a symmetric SFBM, and b is a vector.
Usage
sp_solve_sym(
A,
b,
add_to_diag = rep(0, ncol(A)),
tol = 1e-10,
maxiter = 10 * ncol(A)
)
Arguments
A |
A symmetric SFBM. |
b |
A vector. |
add_to_diag |
Vector (or single value) to virtually add to
the diagonal of |
tol |
Tolerance for convergence. Default is |
maxiter |
Maximum number of iterations for convergence. |
Value
The vector x, solution of Ax=b.
Examples
N <- 100
spmat <- Matrix::rsparsematrix(N, N, 0.01, symmetric = TRUE)
X <- bigsparser::as_SFBM(as(spmat, "dgCMatrix"))
b <- runif(N)
test <- tryCatch(as.vector(Matrix::solve(spmat, b)), error = function(e) print(e))
test2 <- tryCatch(sp_solve_sym(X, b), error = function(e) print(e))
test3 <- as.vector(Matrix::solve(spmat + Matrix::Diagonal(N, 1:N), b))
test4 <- sp_solve_sym(X, b, add_to_diag = 1:N)
all.equal(test3, test4)