ddmatrix-constructors: Distributed Matrix Creation

Description

Methods for simple construction of distributed matrices.

Usage

ddmatrix(data, ...)
# S4 method for ddmatrix
ddmatrix(data, nrow = 1, ncol = 1,
  byrow = FALSE, ..., bldim = .pbd_env$BLDIM, ICTXT = .pbd_env$ICTXT)
# S4 method for missing
ddmatrix(data, nrow = 1, ncol = 1, byrow = FALSE,
  ..., bldim = .pbd_env$BLDIM, ICTXT = .pbd_env$ICTXT)
# S4 method for vector
ddmatrix(data, nrow = 1, ncol = 1, byrow = FALSE,
  ..., bldim = .pbd_env$BLDIM, ICTXT = .pbd_env$ICTXT)
# S4 method for matrix
ddmatrix(data, nrow = 1, ncol = 1, byrow = FALSE,
  ..., bldim = .pbd_env$BLDIM, ICTXT = .pbd_env$ICTXT)
# S4 method for character
ddmatrix(data, nrow = 1, ncol = 1,
  byrow = FALSE, ..., min = 0, max = 1, mean = 0, sd = 1,
  rate = 1, shape, scale = 1, bldim = .pbd_env$BLDIM,
  ICTXT = .pbd_env$ICTXT)
ddmatrix.local(data, ...)
# S4 method for missing
ddmatrix.local(data, nrow = 1, ncol = 1,
  byrow = FALSE, ..., bldim = .pbd_env$BLDIM, ICTXT = .pbd_env$BLDIM)
# S4 method for vector
ddmatrix.local(data, nrow = 1, ncol = 1,
  byrow = FALSE, ..., bldim = .pbd_env$BLDIM, ICTXT = .pbd_env$ICTXT)
# S4 method for matrix
ddmatrix.local(data, nrow = 1, ncol = 1,
  byrow = FALSE, ..., bldim = .pbd_env$BLDIM, ICTXT = .pbd_env$ICTXT)
# S4 method for character
ddmatrix.local(data, nrow = 1, ncol = 1,
  byrow = FALSE, ..., min = 0, max = 1, mean = 0, sd = 1,
  rate = 1, shape, scale = 1, bldim = .pbd_env$BLDIM,
  ICTXT = .pbd_env$ICTXT)

Arguments

data

A global value: a string (for random generation) or an optional data vector. In the data vector case, the data should be the same across all processes.

...

Extra arguments

nrow

number of rows. Global rows for ddmatrix(). Local rows for ddmatrix.local(). See details below.

ncol

number of columns. Global columns for ddmatrix(). Local columns for ddmatrix.local(). See details below.

byrow

logical. If FALSE then the distributed matrix will be filled by column major storage, otherwise row-major.

bldim

blocking dimension.

ICTXT

BLACS context number.

min, max

Min and max values for random uniform generation.

mean, sd

Mean and standard deviation for random normal generation.

rate

Rate for random exponential generation.

shape, scale

Shape and scale parameters for random weibull generation.

Value

Returns a distributed matrix.

Details

These methods are simplified methods of creating distributed matrices, including random ones. These methods involve only local computations, i.e., no communication is performed in the construction of a ddmatrix using these methods (in contrast to using as.ddmatrix() et al).

For non-character inputs, the methods attempt to mimic R as closely as possible. So ddmatrix(1:3, 5, 7) produces the distributed analogue of matrix(1:3, 5, 7).

For character inputs, you may also specify additional parametric family information.

The functions predicated with .local generate data with a fixed local dimension, i.e., each processor gets an identical amount of data. Likewise, the remaining functions generate a fixed global amount of data, and each processor may or may not have an identical amount of local data.

To ensure good random number generation, you should only consider using the character methods with the comm.set.seed() function from pbdMPI which uses the method of L'Ecuyer via the rlecuyer package.