Arithmetic reductions for distributed matrices.
rowMin(x, ...)rowMax(x, ...)
colMin(x, ...)
colMax(x, ...)
# S4 method for ddmatrix
rowSums(x, na.rm = FALSE)
# S4 method for ddmatrix
colSums(x, na.rm = FALSE)
# S4 method for ddmatrix
rowMeans(x, na.rm = FALSE)
# S4 method for ddmatrix
colMeans(x, na.rm = FALSE)
# S4 method for ddmatrix
rowMin(x, na.rm = FALSE)
# S4 method for matrix
rowMin(x, na.rm = FALSE)
# S4 method for ddmatrix
colMin(x, na.rm = FALSE)
# S4 method for matrix
colMin(x, na.rm = FALSE)
# S4 method for ddmatrix
rowMax(x, na.rm = FALSE)
# S4 method for matrix
rowMax(x, na.rm = FALSE)
# S4 method for ddmatrix
colMax(x, na.rm = FALSE)
# S4 method for matrix
colMin(x, na.rm = FALSE)
numeric distributed matrix
additional arguments
logical. Should missing (including NaN
) be removed?
Returns a global numeric vector.
Performs the reduction operation on a distributed matrix.
There are several legitimately new operations, including rowMin()
,
rowMax()
, colMin()
, and colMax()
. These
implementations are not really necessary in R because one can easily (and
reasonably efficiently) do something like
apply(X=x, MARGIN=1L, FUN=min, na.rm=TRUE)
But apply()
on a ddmatrix
is very costly, and should be
used sparingly.
sd()
will compute the standard deviations of the columns, equivalent
to calling apply(x, MARGIN=2, FUN=sd)
(which will work for
distributed matrices, by the way). However, this should be much faster and
use less memory than apply()
. If reduce=FALSE
then the return
is a distributed matrix consisting of one (global) row; otherwise, an
R
vector is returned, with ownership of this vector determined by
proc.dest
.