shaq: shaq

In RBigData/kazaam: Tools for Tall Distributed Matrices

Description

Constructor for shaq objects.

Usage

shaq(Data, nrows, ncols, checks = TRUE)

tshaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'matrix'
shaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'float32'
shaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'vector'
shaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'integer'
shaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'double'
shaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class ''NULL''
shaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'matrix'
tshaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'float32'
tshaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class 'vector'
tshaq(Data, nrows, ncols, checks = TRUE)

## S3 method for class ''NULL''
tshaq(Data, nrows, ncols, checks = TRUE)

Arguments

Data	The local submatrix.
nrows, ncols	The GLOBAL number of rows and columns.
checks	Logical. Should some basic dimension checks be performed? Note that these require communication, and with many MPI ranks, could be expensive.

Details

If nrows and/or ncols is missing, then it will be imputed. The internal logic tries closely to mimic R's, but it may do unexpected things that you do not want. You are encouraged to specify the distributed dimension (i.e., nrows for a shaq, ncols for tshaq).

One can pass NULL for the Data argument to specify that that MPI rank owns no data. In this case, you must manually provide the non-distributed dimension (i.e., ncols for shaq). This use case is typical when reading from a subset of processors and then broadcasting out to the remainder.

Communication

If checks=TRUE, a check on the global number of rows is performed. This amounts to an allgather operation on a logical value (the local dimension check).

See Also

shaq-class and load_balance

RBigData/kazaam documentation built on Nov. 9, 2021, 9:09 a.m.